Radiologists

29-1224.00
Median wage $420,860/yr26,770 employed (US)Rank #577 of 923 scored · top 63% by substitution

Diagnose and treat diseases and injuries using medical imaging techniques, such as x rays, magnetic resonance imaging (MRI), nuclear medicine, and ultrasounds. May perform minimally invasive medical procedures and tests.

Sub-scores

0–100 · band = confidence interval from rater disagreement

Substitution24
Exposure25
Augmentation63

Substitution — the headline: capability discounted by cost, barriers and adoption.

Exposure — technical capability alone, regardless of whether anyone deploys it.

Augmentation — how much AI assists without replacing. High here + moderate substitution = a changing job, not a disappearing one.

Tasks on the substitution scale

30 rated tasks, binned by substitution score.

Position among all scored occupations

Distribution of 923 occupation scores; the marker is this occupation.

Tasks with substitution ≥ 70

0%

Run 1.0.0-draft.1 · computed 2026-08-05 · rater panel: claude-sonnet-5, claude-haiku-4-5-20251001 · intervals span rater disagreement.

Why this score

The five weighted dimensions of the composite, averaged across this occupation's tasks (importance-weighted, panel mean). Exact weights and formulas: /api/v1/methodology.

Task automatabilityw 35%26

panel mean rating 2.0/5 → substitution pressure 26/100

Technical feasibility todayw 20%28

panel mean rating 2.1/5 → substitution pressure 28/100

Cost vs. human wagew 15%27

panel mean rating 2.1/5 → substitution pressure 27/100

Adoption barriersw 20%inverted — strong barriers lower the score16

panel mean rating 4.4/5 (barrier strength) → substitution pressure 16/100

Sector adoption velocityw 10%33

panel mean rating 2.3/5 → substitution pressure 33/100

Task breakdown (30 tasks)

Substitution pressure per task, weighted by O*NET importance in the composite. Expand a task for the full rater audit trail — every rating, every model, every rationale.

Review or transmit images and information using picture archiving or communications systems.

61

CI 47–75 · exposure 67 · augmentation 100 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510014/5Large radiology departments and hospital systems have rapidly adopted PACS automation and AI-assisted image prioritization over the past 5–10 years; smaller practices lag, but the trend is fast in mainstream healthcare IT.
Sector adoption velocityclaude-sonnet-54/5Radiology is among the fastest-adopting medical specialties for AI, with numerous FDA-cleared tools integrated into PACS workflows at many hospitals and health systems.
Augmentation potentialclaude-haiku-4-5-202510015/5AI significantly augments radiologist productivity by automatically organizing, retrieving, and prioritizing images, allowing radiologists to focus on diagnostic interpretation rather than manual archival and transmission tasks, which directly accelerates workflow.
Augmentation potentialclaude-sonnet-55/5AI-driven worklist prioritization, anomaly flagging, and image enhancement integrated into PACS meaningfully speed up and improve radiologists' review process while they remain the decision-maker.
Task automatabilityclaude-haiku-4-5-202510014/5Current AI systems can automatically transmit, route, and archive medical images with high reliability, and PACS-integrated AI can flag and prioritize images for review, saving substantial time. However, the core diagnostic interpretation still requires radiologist oversight, preventing a full 5 rating, though the transmission and organizational workflow aspects are nearly fully automatable.
Task automatabilityclaude-sonnet-53/5Transmitting and organizing images via PACS is largely a routine IT/workflow function that software already automates, but the 'review' component involves clinical interpretation that current AI cannot fully replace end-to-end at equal quality across the range of cases.
Adoption barriersclaude-haiku-4-5-202510013/5While radiology data is protected by HIPAA and GDPR, the transmission and archival functions themselves are not legally restricted—radiologists do not need to manually certify image filing. However, integration friction and institutional preference for human verification of image routing create moderate adoption friction.
Adoption barriersclaude-sonnet-54/5Final interpretation and sign-off legally require a licensed radiologist, and liability for missed diagnoses creates a strong barrier to full automation of the review portion of this task.
Cost vs. human wageclaude-haiku-4-5-202510014/5Cloud-based PACS and AI-assisted image routing cost a fraction of a radiologist's time per image cycle; the all-in AI cost (inference, integration, storage, oversight) is substantially lower than the loaded wage for manual image filing and transmission tasks.
Cost vs. human wageclaude-sonnet-53/5AI-assisted triage and image routing tools can be cheap per use, but integration, validation, and required radiologist oversight for final review keep total costs closer to parity with human-driven workflows.
Technical feasibility todayclaude-haiku-4-5-202510015/5Picture archiving and communication systems (PACS) with integrated AI routing, tagging, and transmission capabilities are mature, deployed products in hospital systems worldwide. Automated image transmission, storage retrieval, and basic sorting are production-standard in radiology departments.
Technical feasibility todayclaude-sonnet-53/5PACS systems themselves are mature and universally deployed, and AI triage/worklist prioritization tools are in production at some sites, but fully autonomous review across all modalities and pathologies is not yet reliable in deployed products.

Document the performance, interpretation, or outcomes of all procedures performed.

53

CI 41–65 · exposure 62 · augmentation 88 · importance 4.9/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510014/5Large hospital systems and radiology groups have begun deploying AI-assisted documentation in production (especially for structured reports on CT/X-ray), though uptake remains concentrated in well-resourced academic and corporate environments rather than small practices.
Sector adoption velocityclaude-sonnet-53/5Radiology is a relatively fast-adopting medical specialty with AI-assisted reporting tools in growing use, but full deployment across health systems remains uneven and pilot-heavy in many institutions.
Augmentation potentialclaude-haiku-4-5-202510015/5AI excels at drafting templates, extracting measurements, and flagging key findings, substantially accelerating radiologist throughput and reducing dictation fatigue while the human physician reviews, refines, and certifies the final report.
Augmentation potentialclaude-sonnet-54/5AI-powered dictation, structured reporting templates, and draft-generation tools meaningfully speed up documentation while the radiologist remains responsible for final interpretation and sign-off.
Task automatabilityclaude-haiku-4-5-202510014/5Current AI can generate structured reports of imaging findings with high accuracy, extracting relevant measurements and pathology descriptions from images and prior notes. However, some nuanced clinical judgment about significance and follow-up recommendations may still require human review, preventing a perfect 5.
Task automatabilityclaude-sonnet-53/5AI can draft structured reports and boilerplate documentation from imaging findings, but final documentation still requires radiologist verification and clinical judgment, so only partial time savings are achievable end-to-end today.
Adoption barriersclaude-haiku-4-5-202510014/5Regulatory and liability barriers are significant: radiologists must legally sign and take responsibility for reports, and malpractice risk deters full automation without human attestation. Institutional credentialing and quality-control mandates also slow substitution.
Adoption barriersclaude-sonnet-54/5Medical documentation of diagnostic interpretations legally requires a licensed radiologist's attestation and carries significant liability exposure, creating strong regulatory and professional barriers to full automation.
Cost vs. human wageclaude-haiku-4-5-202510014/5AI documentation tools cost substantially less per report than radiologist time spent on dictation and transcription, typically $0.50–$3 per report versus $30–$80 in radiologist labor, yielding a favorable cost ratio even accounting for oversight.
Cost vs. human wageclaude-sonnet-52/5AI documentation tools reduce transcription time but still require licensed radiologist review and sign-off, so overall cost savings are moderate rather than order-of-magnitude given required oversight infrastructure and liability review.
Technical feasibility todayclaude-haiku-4-5-202510014/5Commercial radiology AI platforms (e.g., Aidoc, Zebra Medical, clinical LLMs) routinely perform automated report generation in production hospital systems, though most workflows retain mandatory radiologist review and sign-off rather than fully autonomous documentation.
Technical feasibility todayclaude-sonnet-53/5Speech recognition and AI-assisted reporting templates (e.g., Nuance DAX, PowerScribe with AI suggestions) are deployed in many radiology departments, but they assist rather than autonomously document outcomes reliably without physician review.

Participate in quality improvement activities including discussions of areas where risk of error is high.

51

CI 20–82 · exposure 53 · augmentation 88 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510014/5Radiology is a high-digitization, information-intensive specialty with strong incentives for quality improvement and error reduction; many academic and large hospital systems have deployed AI-assisted quality monitoring. Adoption is accelerating in professional healthcare networks, though smaller practices lag.
Sector adoption velocityclaude-sonnet-52/5Healthcare, especially clinical quality governance, adopts AI slowly due to regulatory, liability, and cultural factors, with pilots for error-pattern analytics but rare deep integration into QI committee processes.
Augmentation potentialclaude-haiku-4-5-202510015/5AI dramatically augments radiologist capability in quality improvement by automatically surfacing high-risk cases, quantifying error patterns, and highlighting systematic gaps that humans might miss in manual review. This transforms the productivity and comprehensiveness of quality discussions while keeping radiologists firmly in the oversight and decision-making role.
Augmentation potentialclaude-sonnet-54/5AI can meaningfully augment this task by analyzing error logs, flagging high-risk patterns, and summarizing case data to inform and enrich the human-led quality improvement discussions.
Task automatabilityclaude-haiku-4-5-202510015/5AI systems can now identify high-risk error areas through pattern analysis of imaging datasets, flagging cases with diagnostic uncertainty, anatomical variants, and known pitfalls. This analytical component can be fully automated, achieving >50% time savings while maintaining or improving quality through systematic detection of problem areas that might otherwise require human review.
Task automatabilityclaude-sonnet-52/5This involves collaborative discussion, professional judgment, and organizational context-setting that current AI cannot conduct end-to-end; AI can inform but not replace the participatory process.'
Adoption barriersclaude-haiku-4-5-202510013/5While quality improvement itself is not strictly licensed, clinical governance standards, peer-review expectations, and institutional credentialing requirements typically mandate that radiologists lead or oversee quality discussions. Liability and accreditation concerns create moderate friction against full automation, though AI-assisted preparation is widely adopted.
Adoption barriersclaude-sonnet-54/5Quality improvement in medicine is often mandated by accreditation bodies (e.g., Joint Commission) and requires licensed physician participation and accountability, creating strong institutional and regulatory barriers to full automation.
Cost vs. human wageclaude-haiku-4-5-202510015/5AI inference costs for analyzing imaging datasets and generating risk summaries are orders of magnitude cheaper than radiologist time spent in quality improvement committee discussions and case review, particularly when amortized across large institutions.
Cost vs. human wageclaude-sonnet-52/5Since AI cannot perform the core participatory/discussion task, cost comparison mostly applies to supporting analytics, which are cheap but do not substitute for the full task, keeping overall ratio unfavorable to full automation.
Technical feasibility todayclaude-haiku-4-5-202510014/5Deployed AI systems (e.g., CAD software, AI-assisted diagnostic platforms) successfully identify high-risk patterns and anomalies in radiology workflows at scale in production environments. However, the facilitation and discussion aspects of quality improvement require some human synthesis, so full end-to-end automation is not yet seamless in all organizational contexts.
Technical feasibility todayclaude-sonnet-51/5No deployed product runs quality-improvement meetings or error-risk discussions autonomously for radiology departments; at best AI provides analytics inputs to human-led discussions.

Advise other physicians of the clinical indications, limitations, assessments, or risks of diagnostic and therapeutic applications of radioactive materials.

50

CI 25–75 · exposure 58 · augmentation 88 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Healthcare and radiology departments are adopting AI-assisted decision support and documentation tools at moderate pace, with many pilots underway but fewer mature production deployments at scale. Regulatory scrutiny, legacy EHR integration challenges, and physician skepticism have slowed broader substitution compared to technology sectors.
Sector adoption velocityclaude-sonnet-52/5Healthcare, especially radiology consultative practice, adopts AI tools cautiously due to liability, regulatory scrutiny, and the interpersonal nature of physician consultations.
Augmentation potentialclaude-haiku-4-5-202510015/5AI systems can dramatically augment radiologists' productivity by auto-generating evidence-based advisories, retrieving relevant guidelines, and flagging key risks or contraindications, allowing radiologists to review, customize, and approve communications far faster than manual composition. The human radiologist remains responsible and in control while AI handles the knowledge synthesis.
Augmentation potentialclaude-sonnet-54/5AI can rapidly surface relevant literature, dosing guidelines, contraindications, and similar case data to help radiologists formulate more thorough and faster advice to colleagues.
Task automatabilityclaude-haiku-4-5-202510015/5Modern AI-powered clinical decision support systems, image analysis tools, and large language models can now draft comprehensive clinical advisories on radioactive materials' indications, limitations, and risks, incorporating current guidelines and literature. With appropriate integration into clinical workflows, these systems can generate evidence-based communications that meet or exceed the time-efficiency threshold, especially for routine or well-established advisory scenarios.
Task automatabilityclaude-sonnet-52/5This requires synthesizing patient-specific clinical context, current guidelines, and radiation safety judgment in a consultative dialogue, which AI can support but not fully replace end-to-end today.
Adoption barriersclaude-haiku-4-5-202510014/5Radiologists are licensed physicians required by law to provide certain clinical guidance; liability for incorrect or incomplete advice remains with the physician, not the AI vendor. Organizational risk-aversion and the expectation that a named radiologist sign advisory communications create meaningful friction against full automation.
Adoption barriersclaude-sonnet-54/5This is a licensed physician-to-physician clinical consultation involving radiation safety and liability, requiring a credentialed radiologist to advise and be accountable for the guidance given.
Cost vs. human wageclaude-haiku-4-5-202510015/5The cost of AI inference, document generation, and minimal oversight is orders of magnitude cheaper than the loaded salary of a radiologist ($200k+/year) performing this advisory task. A single AI system can advise thousands of clinicians per year at negligible marginal cost.
Cost vs. human wageclaude-sonnet-52/5Human radiologist consultation remains the standard; AI tools require significant oversight and integration, so cost savings are limited despite some efficiency gains in information retrieval.
Technical feasibility todayclaude-haiku-4-5-202510014/5Deployed AI products (e.g., clinical NLP systems, guideline-based decision support, and generative models fine-tuned on medical literature) reliably generate clinical guidance on radiation risks and therapeutic applications in production settings. However, current systems still require radiologist review and refinement for complex or novel cases, preventing a full 5-rating.
Technical feasibility todayclaude-sonnet-52/5AI decision-support tools exist for imaging appropriateness and radiopharmaceutical guidance, but no deployed product independently advises physicians on these clinical judgments in production at scale.

Prepare comprehensive interpretive reports of findings.

47

CI 31–62 · exposure 58 · augmentation 100 · importance 5.0/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Radiology is a digitized, information-intensive field that has begun AI adoption in pilots and some production settings, but broad displacement remains limited by regulatory requirements and physician gatekeeping; adoption is faster than manufacturing but slower than software sectors.
Sector adoption velocityclaude-sonnet-53/5Radiology is among the more digitized medical specialties with notable AI pilot and production tool adoption (e.g., decision support, draft reporting), but full-scale deployment for report generation remains limited and cautious due to liability concerns.
Augmentation potentialclaude-haiku-4-5-202510015/5AI substantially augments radiologists by auto-generating structured initial drafts, flagging suspicious regions, and enabling faster review cycles, allowing humans to focus on complex cases and clinical correlation while remaining fully in the loop.
Augmentation potentialclaude-sonnet-55/5AI substantially augments radiologists by pre-populating structured findings, flagging abnormalities, and speeding report drafting, while the radiologist remains essential for verification and sign-off.
Task automatabilityclaude-haiku-4-5-202510014/5Current AI can generate structured interpretive reports from imaging data with high accuracy on common pathologies, achieving ≥50% time savings on routine cases. However, complex cases with rare findings or nuanced clinical context may still require significant human review and revision, preventing a full 5 rating.
Task automatabilityclaude-sonnet-53/5AI can draft structured findings and preliminary impressions from imaging, saving significant drafting time, but comprehensive interpretive reports require clinical correlation, judgment on ambiguous findings, and final accountability that current systems cannot fully replicate end-to-end.4
Adoption barriersclaude-haiku-4-5-202510014/5Legal and regulatory barriers are significant: radiologists must typically sign off on AI-generated reports in most jurisdictions, and liability concerns mean AI output cannot fully replace physician judgment. Liability asymmetry and regulatory requirements (FDA oversight of AI devices) create material friction.
Adoption barriersclaude-sonnet-55/5Only a licensed physician can legally sign off on a diagnostic radiology report, creating a hard regulatory and liability barrier that prevents full substitution regardless of AI capability.
Cost vs. human wageclaude-haiku-4-5-202510014/5AI inference costs for report generation are a fraction of a radiologist's hourly wage; integration and oversight add cost but the all-in ratio still favors AI by several-fold on high-volume routine reads.
Cost vs. human wageclaude-sonnet-52/5AI drafting tools reduce time somewhat but licensing, integration, and mandatory physician oversight keep effective cost per finalized report still dominated by radiologist time, so savings are moderate rather than order-of-magnitude.
Technical feasibility todayclaude-haiku-4-5-202510014/5Deployed AI-powered radiology reporting systems (e.g., from Zebra Medical Vision, IBM Watson for Oncology, and others) demonstrably generate interpretive reports in production clinical settings. Performance is reliable for common conditions but still requires radiologist review, so it does not yet reach full autonomous deployment at scale.
Technical feasibility todayclaude-sonnet-52/5FDA-cleared AI tools exist for detection/triage and some report-drafting assistance, but no deployed product autonomously produces final comprehensive interpretive reports without radiologist review and correction at scale.

Obtain patients' histories from electronic records, patient interviews, dictated reports, or by communicating with referring clinicians.

43

CI 36–50 · exposure 38 · augmentation 75 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Radiology practices have digitized EHRs and are experimenting with AI-assisted documentation, but adoption remains pilot-heavy. Deep, production-wide integration of history-gathering automation is slower than in pure IT/finance sectors due to clinical risk aversion and workflow inertia.
Sector adoption velocityclaude-sonnet-53/5Healthcare and radiology are adopting AI at a moderate pace, with EHR summarization and clinical documentation tools in pilot or partial deployment but not yet universal in production workflows.
Augmentation potentialclaude-haiku-4-5-202510014/5AI excels at surfacing relevant prior records, summarizing dictations, and flagging missing data before the radiologist interviews the patient or clinician, meaningfully reducing search time and improving completeness. This assistive role is already demonstrated in clinical NLP products.
Augmentation potentialclaude-sonnet-54/5AI tools that summarize EHR records, prior reports, and dictated notes significantly speed up history-gathering for radiologists, even though human interviews and clinician communication remain manual.
Task automatabilityclaude-haiku-4-5-202510013/5AI can extract and summarize structured data from electronic records and dictated reports with reasonable accuracy, but patient interviews and nuanced clinician communication require contextual judgment and clarification that AI struggles with. Partial automation (60-70% of history gathering from records) is feasible, but end-to-end replacement falls short of the 50% time-saving threshold.
Task automatabilityclaude-sonnet-52/5AI can help aggregate and summarize patient history from EHR data, but conducting patient interviews and nuanced communication with referring clinicians still requires human judgment and interaction, limiting full automation.
Adoption barriersclaude-haiku-4-5-202510013/5HIPAA compliance and data governance policies add overhead to deployment; clinicians often prefer direct communication with radiologists for nuanced history. No strict legal requirement for a licensed human to gather histories, but organizational and patient-contact norms create moderate friction to full automation.
Adoption barriersclaude-sonnet-53/5No strict licensing barrier prevents AI from assisting with history gathering, but clinical liability and the need for accurate synthesis before diagnosis create moderate oversight requirements.
Cost vs. human wageclaude-haiku-4-5-202510013/5AI-powered extraction from EHRs and transcribed reports costs far less per instance than radiologist time, but integration with existing clinical workflows and the need for human verification of extracted histories reduces the cost advantage to rough parity overall.
Cost vs. human wageclaude-sonnet-53/5EHR summarization tools are cheap to run, but integration, verification, and the interview/communication components still require radiologist time, keeping overall cost comparable rather than dramatically lower.
Technical feasibility todayclaude-haiku-4-5-202510013/5EHR data extraction and report parsing tools exist in production, but clinician communication typically requires human judgment. Current systems handle structured record review reliably but struggle with unstructured dictation and real-time clarification of ambiguous clinical details.
Technical feasibility todayclaude-sonnet-52/5Some clinical NLP tools can extract and summarize EHR data reliably, but no deployed product handles the full multi-source task (interviews, dictation, clinician communication) end-to-end in production.

Perform or interpret the outcomes of diagnostic imaging procedures including magnetic resonance imaging (MRI), computer tomography (CT), positron emission tomography (PET), nuclear cardiology treadmill studies, mammography, or ultrasound.

43

CI 23–62 · exposure 50 · augmentation 88 · importance 4.9/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Adoption is active in large health systems and specialized imaging centers with pilots and some production deployments, but penetration remains limited outside major academic and hospital networks; many smaller facilities lag adoption.
Sector adoption velocityclaude-sonnet-53/5Radiology is among the most digitized and AI-adopting medical specialties, with many pilots and some FDA-cleared tools in routine use, though full autonomous deployment lags behind pilot enthusiasm.
Augmentation potentialclaude-haiku-4-5-202510015/5AI substantially augments radiologist productivity by flagging abnormalities, prioritizing worklists, and reducing time on routine analysis, allowing radiologists to focus on complex cases and clinical correlation while remaining in full control of interpretation and reporting.
Augmentation potentialclaude-sonnet-54/5AI substantially assists radiologists via computer-aided detection, triage prioritization, and quantitative measurements, meaningfully improving throughput and catch rates while the radiologist remains the final interpreter.
Task automatabilityclaude-haiku-4-5-202510014/5AI systems can detect and classify many radiological findings (tumors, fractures, pneumonia) with performance matching or exceeding human radiologists on specific imaging modalities. However, end-to-end automation still requires radiologist oversight for rare findings, clinical integration, and liability—falling short of full replacement while exceeding 50% time savings on routine studies.
Task automatabilityclaude-sonnet-52/5AI image analysis tools can flag abnormalities and assist triage in specific modalities like mammography, but comprehensive interpretation across MRI, CT, PET, nuclear cardiology and ultrasound with clinical correlation, differential diagnosis, and reporting remains far from full automation.'
Adoption barriersclaude-haiku-4-5-202510014/5Strong regulatory and liability barriers exist: FDA oversight of diagnostic imaging AI, malpractice risk, clinician certification requirements, and organizational reliance on radiologist sign-off and clinical judgment for patient safety limit full substitution.
Adoption barriersclaude-sonnet-55/5Diagnostic imaging interpretation legally requires a licensed physician's sign-off, with high liability exposure for misdiagnosis, making this one of the most protected clinical tasks.
Cost vs. human wageclaude-haiku-4-5-202510014/5AI inference and integration costs for diagnostic imaging are a small fraction of radiologist salaries ($300k+), with software costs typically $100–500k per site amortized across high volume; the cost advantage is substantial.
Cost vs. human wageclaude-sonnet-52/5Narrow AI screening tools are cheap per scan, but liability, integration, and required radiologist oversight for final interpretation keep effective all-in cost comparable to or only modestly below human cost for the full task.
Technical feasibility todayclaude-haiku-4-5-202510014/5Deployed AI products (FDA-cleared and CE-marked systems) perform detection and analysis on CT, mammography, and other modalities in real clinical workflows, though primarily as flagging and worklist-prioritization tools rather than autonomous diagnostic decision-making.
Technical feasibility todayclaude-sonnet-52/5FDA-cleared AI products exist for narrow tasks (e.g., detecting specific findings in mammograms or stroke CTs) but no deployed system reliably performs comprehensive multi-modality diagnostic interpretation independently in production.

Calculate, measure, or prepare radioisotope dosages.

41

CI 20–61 · exposure 50 · augmentation 88 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Nuclear medicine departments have incorporated algorithmic dosage guidance and dose optimization tools, but adoption remains cautious and variable across institutions. Integration is typical in larger academic medical centers but inconsistent in smaller facilities; autonomous or minimally supervised automation is not yet standard practice.
Sector adoption velocityclaude-sonnet-52/5Medical imaging/nuclear medicine is a highly regulated, specialized field where AI adoption for dosing decisions remains slow and mostly limited to decision-support tools rather than autonomous calculation.
Augmentation potentialclaude-haiku-4-5-202510015/5AI dosage calculation tools significantly augment radiologist productivity by automating tedious calculations, accounting for patient-specific factors, and suggesting optimized protocols, allowing radiologists to focus on clinical judgment and safety verification rather than arithmetic.
Augmentation potentialclaude-sonnet-54/5AI-based calculators and dosimetry software meaningfully speed up and reduce errors in dose calculations, letting radiologists verify rather than compute manually.
Task automatabilityclaude-haiku-4-5-202510014/5AI can reliably perform radioisotope dosage calculations and measurements based on standardized protocols, imaging, and patient parameters with minimal human intervention. The task is largely computational and rule-based, achievable with current systems and integration into hospital information systems, though final verification by a licensed professional remains standard practice.
Task automatabilityclaude-sonnet-52/5Dosage calculation involves patient-specific variables (weight, renal function, isotope decay, clinical context) requiring clinical judgment and accountability that current AI cannot fully replicate end-to-end without physician oversight.'
Adoption barriersclaude-haiku-4-5-202510015/5Regulatory barriers are substantial: radioisotope dosage must be prescribed and verified by a licensed physician or nuclear medicine specialist, and liability for incorrect dosing rests with the responsible clinician. Regulatory bodies (FDA, NRC) control radiopharmaceutical use, and automated systems cannot independently issue dosages without human authorization.
Adoption barriersclaude-sonnet-55/5Radioisotope dosing is tightly regulated and typically requires a licensed physician or authorized user to calculate and approve dosages, with significant patient-safety and legal liability implications.
Cost vs. human wageclaude-haiku-4-5-202510015/5AI inference for dosage calculation is negligible in cost—fractions of a cent per calculation. Compared to the loaded hourly rate of a radiologist (often $150–300/hr), even accounting for integration and oversight, the cost ratio heavily favors automation.
Cost vs. human wageclaude-sonnet-52/5Software-assisted calculation is cheap to run, but the need for physician verification, liability, and safety checks keeps overall cost closer to human-comparable rather than order-of-magnitude cheaper.
Technical feasibility todayclaude-haiku-4-5-202510014/5Deployed products and modules within clinical decision support systems already perform dosage calculations in nuclear medicine workflows; examples include dose optimization software used in major hospital networks. Performance is mature and reliable, though most implementations retain a verification step by a radiologist.
Technical feasibility todayclaude-sonnet-52/5Some dose-calculation software and decision-support tools exist in nuclear medicine, but they are calculators/aids rather than autonomous systems performing the full task reliably in production.

Check and approve the quality of diagnostic images before patients are discharged.

29

CI 20–39 · exposure 38 · augmentation 75 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare is a laggard sector for AI automation due to regulatory caution, liability concerns, and physician resistance to full delegation. While some vendors market AI quality-check tools, adoption remains limited to pilot programs and quality-flagging assistance roles rather than autonomous discharge approval in most institutions.
Sector adoption velocityclaude-sonnet-52/5Radiology is an early-adopting specialty for AI in image analysis, but formal quality-approval-for-discharge workflows remain largely manual and tied to regulatory/liability structures, slowing deployment of this specific sub-task.
Augmentation potentialclaude-haiku-4-5-202510014/5AI-powered image quality assessment tools significantly augment radiologists by automatically flagging artifacts, exposure problems, and positioning errors, reducing manual review burden and enabling faster, more consistent quality gates. Radiologists remain in the loop for final judgment, but AI transforms the speed and consistency of the preliminary quality check.
Augmentation potentialclaude-sonnet-54/5AI tools increasingly pre-screen images for technical adequacy and flag potential issues, helping radiologists work faster while they retain final approval authority.
Task automatabilityclaude-haiku-4-5-202510013/5AI systems can detect certain image quality defects (motion artifacts, exposure errors, positioning issues) automatically, but final approval for patient discharge requires clinical judgment about diagnostic adequacy in context. Current systems can automate roughly half the workflow (flagging obvious quality problems), but human radiologists must still review and approve discharge, limiting end-to-end automation.
Task automatabilityclaude-sonnet-52/5AI can flag technical image quality issues (positioning, artifacts, exposure) but final approval requires clinical judgment about diagnostic adequacy tied to the specific patient presentation, which current systems cannot fully replace end-to-end.
Adoption barriersclaude-haiku-4-5-202510014/5Radiologists are licensed professionals; regulatory bodies (FDA, ACR) and liability frameworks require qualified human approval of diagnostic image quality before patient discharge. Hospitals face malpractice and accreditation risk if they delegate final approval to unvalidated AI, creating strong legal and organizational barriers to full automation.
Adoption barriersclaude-sonnet-55/5Radiologist approval of diagnostic images is a licensed medical function with direct patient safety and legal liability implications, requiring a credentialed physician's sign-off.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI inference on image quality is cheap, but integration into hospital workflows, regulatory compliance, and mandatory radiologist oversight add significant costs. The net cost advantage is modest because radiologist time for final approval remains non-negotiable, keeping total cost closer to human baseline than to order-of-magnitude savings.
Cost vs. human wageclaude-sonnet-52/5AI-assisted quality checks are cheap to run, but the requirement for radiologist sign-off means the human cost is largely retained, limiting overall savings.
Technical feasibility todayclaude-haiku-4-5-202510013/5Commercial AI products exist for image quality assessment and defect detection (e.g., from GE, Siemens, Philips vendors), and some are deployed in clinical workflows. However, they operate with material false-positive and false-negative rates, and most require radiologist oversight rather than fully autonomous approval, limiting production-scale reliability.
Technical feasibility todayclaude-sonnet-52/5Some QA software exists to detect motion artifacts or exposure problems in radiographs, but no deployed product autonomously approves image quality for discharge decisions at scale in production workflows.

Develop or monitor procedures to ensure adequate quality control of images.

29

CI 25–32 · exposure 30 · augmentation 75 · importance 4.6/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While some large health systems are piloting AI quality monitoring tools, widespread production adoption remains limited; most radiology practices still rely on manual QA protocols and human-led procedure development, reflecting slow sector-wide adoption.
Sector adoption velocityclaude-sonnet-53/5Healthcare imaging is adopting AI tools steadily for image analysis, but QC procedure development and governance functions lag behind diagnostic AI adoption.
Augmentation potentialclaude-haiku-4-5-202510014/5AI systems that automatically flag technical defects, artifact types, and protocol deviations meaningfully augment radiologists' ability to monitor and refine QA procedures, enabling faster identification of systematic issues without removing human oversight.
Augmentation potentialclaude-sonnet-54/5AI-based image quality metrics and automated flagging of technical errors can meaningfully assist radiologists and technologists in monitoring quality control processes.
Task automatabilityclaude-haiku-4-5-202510012/5Current AI can assist with detecting image artifacts and flagging technical defects, but developing comprehensive quality control procedures and making contextual judgments about adequacy requires human expertise and domain knowledge that AI cannot fully replicate end-to-end at the 50% time-saving threshold.
Task automatabilityclaude-sonnet-52/5Developing and monitoring QA procedures involves ongoing judgment, protocol design, and institutional oversight that AI can support but not fully execute end-to-end today.dev
Adoption barriersclaude-haiku-4-5-202510014/5Regulatory requirements (FDA oversight of diagnostic imaging QA, accreditation standards like ACR), liability concerns around missed quality issues, and the need for radiologist sign-off on quality procedures create significant legal and organizational barriers to full automation.
Adoption barriersclaude-sonnet-54/5Radiology QC programs are subject to accreditation (ACR) and regulatory requirements often requiring physician/physicist sign-off, creating substantial compliance barriers to full automation.
Cost vs. human wageclaude-haiku-4-5-202510012/5The combination of algorithm licensing, integration with PACS systems, radiologist oversight time, and validation infrastructure means the all-in cost is comparable to or potentially exceeds the cost of dedicated QA personnel, especially given ongoing human review requirements.
Cost vs. human wageclaude-sonnet-52/5AI tools for image QC add cost on top of existing PACS/QA infrastructure and still require radiologist/physicist oversight, so savings versus current staff time are modest.
Technical feasibility todayclaude-haiku-4-5-202510013/5AI-based quality assurance tools exist and are deployed in some radiology departments to flag artifacts and technical failures, but they operate within narrow scopes and often require human review, falling short of fully reliable autonomous procedure development and monitoring at scale.
Technical feasibility todayclaude-sonnet-52/5Some AI-based image quality assessment tools exist (e.g., automated artifact detection, exposure checks) but comprehensive QC program development and monitoring is still largely manual and institution-specific.

Coordinate radiological services with other medical activities.

26

CI 25–28 · exposure 25 · augmentation 63 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Adoption of AI for service coordination remains limited; most healthcare organizations still rely on manual scheduling, email, and institutional protocols. While some large health systems pilot workflow automation, production deployment at scale is rare and slow due to regulatory caution and integration complexity.
Sector adoption velocityclaude-sonnet-53/5Healthcare is adopting AI for image analysis and workflow support at moderate pace, but adoption in cross-departmental coordination tasks specifically remains largely pilot-stage.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can provide meaningful assistance through automated scheduling suggestions, alerts for bottlenecks, and predictive workload analysis, helping radiologists and coordinators optimize workflows and reduce administrative burden. However, the human must retain primary responsibility for final coordination decisions.
Augmentation potentialclaude-sonnet-54/5AI-driven worklist prioritization, natural language processing of reports, and scheduling optimization tools can meaningfully assist radiologists in coordinating care with other medical activities, improving efficiency while humans retain final coordination responsibility.
Task automatabilityclaude-haiku-4-5-202510012/5AI cannot currently handle the full scope of coordinating radiological services with other medical activities, which involves dynamic scheduling, provider communication, care prioritization, and integration with downstream clinical workflows. While AI might assist with some scheduling components, the task requires judgment and real-time negotiation across departments.
Task automatabilityclaude-sonnet-52/5Coordinating radiological services requires interfacing with clinicians, scheduling, prioritizing cases, and integrating clinical context, which involves judgment and interpersonal/organizational coordination that current AI cannot fully replicate end-to-end.rica
Adoption barriersclaude-haiku-4-5-202510014/5Significant barriers exist: radiologists and care coordinators are licensed professionals whose judgment on prioritization and clinical sequencing carries liability weight, healthcare systems operate under regulatory requirements (HIPAA, state licensing), and coordination decisions affect patient safety outcomes. Hospitals require human sign-off on care-coordination decisions.
Adoption barriersclaude-sonnet-54/5Care coordination decisions carry high liability, require clinical judgment, and are generally expected to be performed or approved by licensed physicians in a healthcare organizational structure, creating strong barriers to full automation.
Cost vs. human wageclaude-haiku-4-5-202510012/5Implementing AI-driven coordination systems requires substantial infrastructure, integration, and ongoing human oversight costs. For the complex, context-dependent nature of medical service coordination, the total cost of an AI solution currently exceeds the cost of having a radiologist or administrative coordinator perform the task.
Cost vs. human wageclaude-sonnet-52/5AI tools can reduce some administrative overhead, but human radiologists still must oversee coordination decisions, so the all-in cost savings versus a human coordinator are modest rather than transformative.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed product reliably performs end-to-end coordination of radiological services across an entire medical system. Some healthcare systems use workflow management and scheduling tools with AI components, but these operate within narrow domains and require significant human oversight and manual intervention.
Technical feasibility todayclaude-sonnet-52/5Some scheduling and workflow-optimization tools exist (e.g., worklist prioritization software), but no deployed product autonomously coordinates radiology services with broader clinical care pathways reliably.

Review procedure requests and patients' medical histories to determine applicability of procedures and radioisotopes to be used.

26

CI 20–32 · exposure 30 · augmentation 75 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While radiology has adopted AI for image interpretation, decision-support tools for procedure selection have seen slower, more cautious adoption due to liability concerns and the entrenched role of radiologist judgment in pre-procedure assessment.
Sector adoption velocityclaude-sonnet-53/5Healthcare/radiology is adopting AI decision support and triage tools at a moderate pace, with growing pilots but cautious, regulated rollout for diagnostic decision-making.
Augmentation potentialclaude-haiku-4-5-202510014/5AI tools that summarize patient history, flag contraindications, and suggest relevant guidelines substantially assist radiologists in this task, reducing review time and improving consistency while the radiologist retains decision authority.
Augmentation potentialclaude-sonnet-54/5AI can efficiently summarize patient history, flag prior imaging/allergies, and check protocol appropriateness against guidelines, meaningfully speeding up the radiologist's review while they retain final authority.
Task automatabilityclaude-haiku-4-5-202510012/5AI can extract and summarize medical history and flag obvious contraindications, but determining procedure applicability requires nuanced clinical judgment integrating patient context, comorbidities, and risk-benefit analysis that current systems perform inconsistently without radiologist review.
Task automatabilityclaude-sonnet-52/5AI can surface relevant history and flag contraindications, but final determination of procedure/radioisotope appropriateness requires clinical judgment integrating nuanced patient context, so it doesn't yet meet the 50% end-to-end automation bar.
Adoption barriersclaude-haiku-4-5-202510014/5Radiologists are licensed professionals whose medical judgment on procedure applicability carries legal responsibility and liability; regulatory frameworks (FDA, ACR) and institutional protocols require physician decision-making authority on these determinations.
Adoption barriersclaude-sonnet-55/5This is a physician-level clinical determination tied to licensure, liability for radiation/contrast risks, and legal requirement that a radiologist authorize the study and isotope choice.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI-assisted review tools have meaningful development and oversight costs, and radiologist time for validation remains substantial; the cost advantage over radiologist review alone is modest and partially offset by required human sign-off.
Cost vs. human wageclaude-sonnet-52/5AI decision-support software has licensing and integration costs plus mandatory physician oversight, so total cost is not dramatically cheaper than physician review time for this specific judgment task.
Technical feasibility todayclaude-haiku-4-5-202510013/5Clinical decision support systems exist that flag contraindications and suggest alternatives, but they operate with material error rates and require radiologist validation; no deployed product independently makes these determinations at the reliability expected for patient safety.
Technical feasibility todayclaude-sonnet-52/5Clinical decision support tools exist that suggest appropriate imaging per guidelines (e.g., ACR Appropriateness Criteria integrations), but they are advisory, narrow in scope, and not autonomous determiners of procedure/radioisotope selection in production.

Confer with medical professionals regarding image-based diagnoses.

25

CI 23–28 · exposure 25 · augmentation 75 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Healthcare is adopting AI-assisted image interpretation at a moderate pace, with pilots and early production deployments, but radiologist-to-radiologist consultation remains human-centered. Adoption is faster in large academic centers and slower in smaller practices.
Sector adoption velocityclaude-sonnet-53/5Radiology is among the most AI-active clinical specialties with many FDA-cleared tools, but adoption is concentrated in image analysis, not in automating physician-to-physician consultation.
Augmentation potentialclaude-haiku-4-5-202510014/5AI excels at augmenting radiologists during conferencing by rapidly retrieving prior comparable cases, suggesting differential diagnoses, and highlighting ambiguous regions—substantially increasing the speed and depth of expert consultation without removing the radiologist from the decision loop.
Augmentation potentialclaude-sonnet-54/5AI-generated preliminary findings, highlighted regions of interest, and structured reports can meaningfully speed up and inform the radiologist's conversation with other clinicians.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can generate diagnostic suggestions from images, the task requires conferring with other medical professionals—a collaborative communication and reasoning process that demands human judgment, context-sharing, and accountability. Current systems cannot reliably replace this interpersonal consultation loop.
Task automatabilityclaude-sonnet-52/5This is an interactive, judgment-heavy consultative task requiring real-time dialogue, clinical context integration, and accountability that current AI cannot conduct end-to-end.dealer AI can support prep but not replace the conference itself.'
Adoption barriersclaude-haiku-4-5-202510014/5Radiologists are licensed professionals; their clinical judgment, interpretation, and communication with peers carry legal and liability weight. Regulatory and professional standards require a licensed radiologist to own diagnoses and participate in peer consultation, creating hard barriers to full substitution.
Adoption barriersclaude-sonnet-55/5Only licensed physicians can render and discuss diagnostic interpretations with other clinicians; liability, licensure, and legal requirements for physician sign-off make this a hard barrier.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI image analysis reduces some preparation costs, but the cost of integrating AI-generated reports into a conferencing workflow, plus human oversight and liability, does not yet undercut the human radiologist's labor for this communication-heavy task.
Cost vs. human wageclaude-sonnet-52/5Because the task is not automatable end-to-end, the relevant comparison is AI-as-tool cost added to radiologist time rather than substitution, so no major cost savings materialize versus human labor.
Technical feasibility todayclaude-haiku-4-5-202510012/5AI diagnostic support tools exist in production, but 'conferring' involves bidirectional discussion, negotiation of clinical findings, and shared decision-making that current AI cannot perform autonomously. No deployed system reliably conducts multi-stakeholder clinical conferences without human radiologists leading.
Technical feasibility todayclaude-sonnet-52/5AI diagnostic aids (e.g., for fracture or nodule detection) exist in production but no deployed product conducts the actual peer consultation and discussion between physicians.

Develop treatment plans for radiology patients.

25

CI 23–28 · exposure 25 · augmentation 75 · importance 4.3/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Healthcare is digitizing rapidly and AI adoption in radiology is accelerating, but adoption remains concentrated in image analysis aids rather than autonomous planning. Most radiology departments use AI as decision support, not replacement for treatment planning responsibility.
Sector adoption velocityclaude-sonnet-53/5Healthcare/radiology is adopting AI for image analysis and triage relatively quickly, but adoption for full treatment planning specifically remains in pilot/limited-support stages.
Augmentation potentialclaude-haiku-4-5-202510014/5AI can meaningfully assist radiologists by summarizing imaging findings, suggesting differential diagnoses, and retrieving relevant clinical evidence, thereby accelerating and improving the quality of human-led treatment planning while the radiologist retains full decision authority.
Augmentation potentialclaude-sonnet-54/5AI tools assist by highlighting relevant findings, prior cases, and guideline-based options, meaningfully speeding up the radiologist's planning process while the physician retains final judgment.
Task automatabilityclaude-haiku-4-5-202510012/5AI systems can assist with image interpretation and flagging abnormalities, but developing comprehensive treatment plans requires integrating clinical context, patient history, comorbidities, and multidisciplinary input—tasks that demand human judgment and accountability. Current systems cannot autonomously perform the full end-to-end planning at equal quality.
Task automatabilityclaude-sonnet-52/5Treatment planning requires integrating patient history, imaging findings, comorbidities, and clinical judgment about risks/benefits; current AI can suggest options but cannot reliably produce full plans end-to-end at equal quality without physician oversight.
Adoption barriersclaude-haiku-4-5-202510014/5Treatment plan development carries high liability and regulatory burden; radiologists typically must personally review, approve, and sign off on plans. Medical licensing, malpractice exposure, and standard-of-care requirements create substantial legal and organizational barriers to automation.
Adoption barriersclaude-sonnet-55/5Treatment planning is a licensed medical act with direct liability implications; regulations require a physician to authorize and be accountable for treatment decisions.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI inference for image analysis is cheap, but integrating these outputs into credible treatment planning still requires expert radiologist time for review, modification, and accountability. All-in costs remain comparable to or exceed the cost of human planning.
Cost vs. human wageclaude-sonnet-52/5AI-assisted drafting tools reduce some time but still require substantial physician review and integration cost, so overall cost savings versus a radiologist's judgment-driven work are modest rather than order-of-magnitude.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI tools for image analysis are deployed, they function as screening or decision-support aids rather than autonomous treatment planners. No production system today reliably develops complete treatment plans without significant radiologist oversight and modification.
Technical feasibility todayclaude-sonnet-52/5Deployed decision-support tools exist that surface differential diagnoses or flag findings, but no product autonomously generates complete treatment plans in routine clinical production.

Compare nuclear medicine procedures with other types of procedures, such as computed tomography, ultrasonography, nuclear magnetic resonance imaging, and angiography.

24

CI 20–28 · exposure 25 · augmentation 63 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While radiology is digitally sophisticated, adoption of AI for high-stakes comparative modality selection remains limited to early adopters and pilots; most institutions still rely on radiologist judgment and departmental protocols rather than AI-driven recommendations.
Sector adoption velocityclaude-sonnet-53/5Radiology is a leading sector for AI adoption with many FDA-cleared tools in clinical pilots and some production use, but full comparative multi-modal reasoning tools are not yet widely deployed.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can usefully support radiologists by surfacing comparative data, highlighting relevant protocols, or flagging contraindications, improving speed and consistency of deliberation without removing the radiologist from the decision loop.
Augmentation potentialclaude-sonnet-54/5AI substantially augments radiologists by pre-analyzing individual modality images, flagging findings, and quantifying features, speeding up the comparative synthesis process while the physician retains final judgment.
Task automatabilityclaude-haiku-4-5-202510012/5Comparing imaging modalities requires understanding procedural protocols, indications, and trade-offs between modalities. While AI can retrieve and display procedural information, synthesizing comparative clinical reasoning—weighing radiation dose, soft-tissue contrast, speed, cost, and clinical context—remains substantially dependent on human judgment and domain expertise.
Task automatabilityclaude-sonnet-52/5Comparing across modalities requires integrating clinical context, prior history, and nuanced judgment about diagnostic tradeoffs that current AI cannot fully replicate end-to-end, though AI can assist with image analysis components.
Adoption barriersclaude-haiku-4-5-202510014/5Clinical decision-making about imaging modality selection carries liability risk and typically requires a licensed radiologist or physician to justify and sign off on the choice, creating legal and regulatory barriers to unsupervised AI automation.
Adoption barriersclaude-sonnet-55/5Radiology diagnosis and reporting require a licensed radiologist's sign-off due to liability, malpractice exposure, and regulatory requirements (e.g., FDA-cleared AI tools are adjunctive, not autonomous).
Cost vs. human wageclaude-haiku-4-5-202510012/5Comparative analysis and decision support tools exist but require expert validation, oversight, and integration into clinical workflows, making the all-in cost of AI assistance comparable to or exceeding a radiologist's time investment for this cognitive task.
Cost vs. human wageclaude-sonnet-52/5AI tools for image analysis are cheap per-scan, but the comparative diagnostic task still requires substantial radiologist oversight and integration costs, keeping all-in cost roughly comparable to human-only workflows.
Technical feasibility todayclaude-haiku-4-5-202510012/5Some AI tools can extract features from imaging reports or suggest modality choices from structured data, but no mature product reliably performs end-to-end comparative analysis of procedures across different modalities in production. Most offerings remain at pilot or research stage.
Technical feasibility todayclaude-sonnet-52/5Deployed AI products exist for single-modality image analysis (e.g., detecting nodules on CT) but no production system reliably performs cross-modality comparative diagnostic reasoning at physician-level scope.

Communicate examination results or diagnostic information to referring physicians, patients, or families.

24

CI 20–28 · exposure 25 · augmentation 63 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Adoption of AI for autonomous communication of diagnostic results to patients and families remains minimal. Most healthcare organizations maintain radiologist involvement in patient communication; adoption is limited to narrow back-office report drafting in some centers, not replacement of the communication task itself.
Sector adoption velocityclaude-sonnet-53/5Healthcare is adopting AI for report drafting and image analysis at a moderate pace, but direct patient communication automation remains rare and cautious due to regulatory and trust concerns.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can usefully assist radiologists by generating report drafts, organizing findings, and suggesting clear language structures that the radiologist then refines and delivers. This augmentation improves efficiency and clarity but does not transform the communication task since the radiologist retains central responsibility for the message.
Augmentation potentialclaude-sonnet-54/5AI can meaningfully speed up drafting of reports, summarizing findings, and preparing communication materials, letting radiologists focus on the interpersonal delivery.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can draft reports from imaging, communicating results to physicians and especially to patients/families requires nuanced judgment, empathy, and responsiveness to individual circumstances that current systems cannot reliably replicate end-to-end. AI might assist with report generation but cannot fully replace the physician's role in explaining findings, addressing concerns, and making clinical recommendations.
Task automatabilityclaude-sonnet-52/5AI can draft summaries of findings, but conveying results to patients/families and referring physicians requires contextual judgment, empathic communication, and handling questions that current systems cannot reliably manage end-to-end.'
Adoption barriersclaude-haiku-4-5-202510014/5Significant regulatory and liability barriers exist: radiologists have professional and legal responsibility for communicating findings accurately and appropriately; malpractice and standard-of-care expectations require a licensed physician sign-off; patients often expect direct physician communication, especially for serious diagnoses. Regulatory bodies and institutional risk policies reinforce human oversight.
Adoption barriersclaude-sonnet-55/5Only licensed physicians can legally communicate diagnostic results and bear liability for accuracy, especially for significant or life-altering findings.
Cost vs. human wageclaude-haiku-4-5-202510012/5Even if AI could generate draft communications, the radiologist must still review, personalize, and deliver the message, so labor cost savings are marginal. Integration and oversight overhead further reduce the cost advantage, keeping the all-in cost comparable to or higher than direct radiologist communication.
Cost vs. human wageclaude-sonnet-52/5While drafting tools are cheap, the actual communication task still requires physician time and liability oversight, keeping all-in costs comparable to human-only workflows.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed product reliably performs the full task of communicating diagnostic information to patients and families at the standard expected of radiologists. AI tools can generate text summaries of findings, but production systems do not independently handle the interpersonal, contextual, and clinical judgment aspects required.
Technical feasibility todayclaude-sonnet-52/5Some products generate structured report language or patient-friendly summaries, but no deployed system autonomously communicates results directly to patients or physicians in clinical practice.

Provide counseling to radiologic patients to explain the processes, risks, benefits, or alternative treatments.

20

CI 11–29 · exposure 17 · augmentation 63 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Despite digitization in radiology, patient counseling remains a direct human interaction with significant regulatory and liability consequences; adoption of AI-led counseling is minimal because radiologists must retain personal accountability and cannot legally delegate informed-consent responsibility.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall lags in AI adoption for patient-facing clinical communication due to regulatory, liability, and trust concerns, despite faster uptake in administrative and imaging-analysis tasks.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can assist by drafting personalized patient education summaries, highlighting key risk-benefit points, and generating multiple communication formats; however, the radiologist must still deliver and adapt the message, so augmentation is meaningful but not transformative.
Augmentation potentialclaude-sonnet-54/5AI can generate patient-friendly explanations, translate materials, and draft counseling scripts that radiologists can use to prepare and streamline in-person consultations, meaningfully boosting efficiency.
Task automatabilityclaude-haiku-4-5-202510011/5Counseling patients about medical processes, risks, and alternatives requires personalized communication, empathy, and the ability to adapt explanations to individual patient concerns and medical literacy—capabilities that current AI systems cannot reliably perform end-to-end in a clinical setting.
Task automatabilityclaude-sonnet-52/5Patient counseling requires empathetic, personalized dialogue addressing individual fears, medical history, and nuanced risk-benefit tradeoffs that current AI cannot fully replicate end-to-end with equal quality.
Adoption barriersclaude-haiku-4-5-202510015/5Informed consent and patient counseling are core legal and ethical obligations in medical practice; radiologists must personally document and take responsibility for risk communication, and liability for inadequate counseling falls squarely on the clinician, creating a hard barrier to full automation.
Adoption barriersclaude-sonnet-54/5Informed consent and risk disclosure for medical procedures typically require a licensed physician's direct involvement, creating strong legal and liability barriers to full automation.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI-assisted patient education tools exist but still require significant oversight and human review by radiologists; the combination of AI inference, integration, and mandatory human review makes the all-in cost comparable to or higher than direct radiologist counseling.
Cost vs. human wageclaude-sonnet-53/5AI-generated educational materials or chatbot pre-counseling are cheap to produce, but since a licensed physician must still deliver final counseling, overall cost savings are moderate rather than dramatic.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can generate patient education materials and draft explanations, no deployed product reliably performs the full counseling task with patient-appropriate personalization, risk clarification, and documented informed consent at production scale in radiology practices.
Technical feasibility todayclaude-sonnet-52/5Chatbots and AI patient-education tools exist to explain procedures generically, but no deployed product reliably conducts full clinical counseling conversations replacing physician judgment in production.

Participate in continuing education activities to maintain and develop expertise.

19

CI 11–28 · exposure 17 · augmentation 63 · importance 4.3/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5Medical education is moderately digitizing (online CME platforms, AI-assisted content discovery), but participation itself remains a human responsibility. Adoption is gradual across health systems, with pilots common but full automation not expected.
Sector adoption velocityclaude-sonnet-52/5Medical education platforms are slowly incorporating AI-driven personalization, but adoption in formal CME/CE compliance systems remains nascent.
Augmentation potentialclaude-haiku-4-5-202510014/5AI can significantly assist by recommending targeted continuing education content, summarizing recent literature, identifying knowledge gaps through case analysis, and personalizing learning paths—all while the radiologist remains the active learner and decision-maker.
Augmentation potentialclaude-sonnet-53/5AI tools can recommend relevant case studies, summarize literature, and create practice quizzes, meaningfully aiding a radiologist's self-directed learning.
Task automatabilityclaude-haiku-4-5-202510011/5Continuing education requires human judgment about knowledge gaps, learning style preferences, and career development trajectory. AI cannot autonomously decide what expertise a radiologist needs to maintain or develop.
Task automatabilityclaude-sonnet-52/5AI can help curate and summarize CME content but cannot itself 'participate' in required credentialed education activities on behalf of a licensed radiologist.','rationale2':''},
Adoption barriersclaude-haiku-4-5-202510014/5Medical licensing boards and professional standards (e.g., ACR, ARRS) require radiologists to personally participate in and attest to continuing education. Legal and regulatory frameworks mandate human engagement, not just content consumption.
Adoption barriersclaude-sonnet-55/5Continuing education requirements are tied to individual licensure and board certification, so a human must personally complete and attest to these activities.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI can assist with content curation and recommendation at low cost, but the core task—a radiologist's time spent learning—remains dominated by human labor cost. AI reduces ancillary costs only marginally.
Cost vs. human wageclaude-sonnet-52/5AI can cheaply supply study materials or quizzes, but the human must still spend time engaging in accredited CE, so overall cost savings are limited.
Technical feasibility todayclaude-haiku-4-5-202510012/5AI can recommend educational content and curate learning materials with some reliability, but cannot replace the human act of participating in, absorbing, and integrating education into practice. No deployed product performs the full participation task.
Technical feasibility todayclaude-sonnet-52/5Some AI-driven learning platforms and case-review tools exist for radiology education, but no product performs the accreditation-required CE participation itself.

Recognize or treat complications during and after procedures, including blood pressure problems, pain, oversedation, or bleeding.

14

CI 3–25 · exposure 13 · augmentation 50 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Radiology departments are cautious adopters of AI for critical decision-making in acute scenarios. While detection aids are spreading, the real-time intervention aspect has seen slow adoption; radiologists remain gatekeepers, particularly in interventional settings where complications arise.
Sector adoption velocityclaude-sonnet-52/5While healthcare broadly is adopting AI for diagnostics and monitoring alerts, autonomous treatment of complications remains essentially unadopted due to safety and regulatory constraints.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can provide useful assistance by flagging potential complications on imaging in real time, alerting radiologists to vital signs trends, or recommending protocols, thereby raising clinician awareness and response speed. However, the cognitive and manual demands of treatment remain largely human-driven.
Augmentation potentialclaude-sonnet-53/5AI-based monitoring systems can help detect early warning signs (e.g., vital sign anomalies) that assist the radiologist in identifying complications faster, though the human must still diagnose and treat.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can assist with detection of some complications (e.g., bleeding on imaging), the task requires real-time clinical decision-making, patient assessment, and intervention selection that current AI systems cannot reliably perform end-to-end. Manual intervention by a clinician remains essential for the majority of cases.
Task automatabilityclaude-sonnet-51/5This requires real-time physical patient monitoring, physical examination, and hands-on intervention (e.g., managing bleeding, administering medications) that current AI systems cannot perform end-to-end.'
Adoption barriersclaude-haiku-4-5-202510014/5Strong regulatory and liability barriers exist: a licensed physician must legally evaluate and treat acute complications; malpractice liability for missed or mistreated complications is high, creating asymmetric error costs. Patient safety requirements mean human oversight is non-negotiable.
Adoption barriersclaude-sonnet-55/5Treating medical complications requires a licensed physician's judgment and legal authority to act, with high liability exposure and mandatory human oversight.
Cost vs. human wageclaude-haiku-4-5-202510012/5Even with AI assistance for detection, radiologists must remain present and responsible for decision-making and intervention, so labor costs remain high. AI monitoring tools add infrastructure cost without eliminating the need for expensive clinical personnel.
Cost vs. human wageclaude-sonnet-51/5AI cannot substitute for the physician's hands-on emergency response, so there is no viable cost comparison for full task substitution.
Technical feasibility todayclaude-haiku-4-5-202510012/5Deployed products can flag some complications on imaging (e.g., hemorrhage detection), but end-to-end recognition and treatment of acute complications like blood pressure problems, oversedation, or pain management requires human clinical judgment. No mature system reliably handles the full scope in production.
Technical feasibility todayclaude-sonnet-51/5No deployed product autonomously recognizes and treats acute procedural complications; clinical decision support tools may flag abnormal vitals but do not act or treat.

Test dosage evaluation instruments and survey meters to ensure they are operating properly.

12

CI 5–19 · exposure 13 · augmentation 38 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Adoption of autonomous equipment testing in radiology departments remains minimal; most facilities rely on trained technicians and periodic vendor calibration services. The regulated, high-liability nature of dosimetry means adoption velocity for AI automation is slow even in digitized healthcare settings.
Sector adoption velocityclaude-sonnet-51/5Physical equipment testing in radiology/radiation safety is a low-digitization, hands-on task with minimal AI agent deployment or displacement in this specific niche.
Augmentation potentialclaude-haiku-4-5-202510013/5AI could assist by automating data logging from meter readings, flagging out-of-spec values, or scheduling maintenance alerts based on sensor drift patterns. However, augmentation is limited because the technician must still physically perform the test and make the final judgment on equipment operability.
Augmentation potentialclaude-sonnet-52/5AI could help log results, flag anomalies in historical calibration data, or schedule maintenance, but it offers limited direct assistance to the physical testing action itself.
Task automatabilityclaude-haiku-4-5-202510012/5Testing and calibrating dosage instruments requires hands-on physical manipulation, sensor reading interpretation, and judgment about operational thresholds that current AI cannot perform autonomously. While AI could assist in data analysis or flagging anomalies in meter readings, the core task of physically testing equipment and ensuring proper operation remains fundamentally manual.
Task automatabilityclaude-sonnet-51/5This is a hands-on physical equipment testing and calibration verification task requiring manual manipulation of instruments and physical inspection, which current AI cannot perform end-to-end.imediately usable AI cannot substitute for the physical checking process.
Adoption barriersclaude-haiku-4-5-202510014/5Significant regulatory barriers exist: dosage measurement equipment validation is governed by strict FDA and clinical safety regulations requiring documented human oversight and certification. Liability for equipment failure is substantial, and most healthcare facilities require qualified medical physicists or technicians to sign off on equipment testing.
Adoption barriersclaude-sonnet-54/5Radiation safety equipment testing is often governed by regulatory and institutional protocols requiring qualified personnel to verify proper function, creating strong procedural and compliance barriers to automation.
Cost vs. human wageclaude-haiku-4-5-202510011/5Current AI systems cannot perform this task autonomously, so a cost comparison favors human technicians who are purpose-trained for equipment maintenance and calibration. Any AI assistance would be supplementary, not substitutive.
Cost vs. human wageclaude-sonnet-51/5There is no AI substitute performing this physical equipment check, so AI cost is not comparable; a human technician remains necessary and cheaper than any hypothetical automated physical system today.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed AI products reliably perform end-to-end physical equipment testing and calibration. While computer vision and sensor data analysis exist as research domains, production systems for autonomous dosimetry equipment validation are not mature or widely deployed in clinical radiology settings.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product performs physical calibration checks or operates survey meters; this remains a manual technical task performed by humans with physical instruments.

Instruct radiologic staff in desired techniques, positions, or projections.

12

CI 7–16 · exposure 5 · augmentation 38 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare institutions have been slow to adopt autonomous AI for instructional tasks; most teaching remains relational and hierarchical, with radiologists directly mentoring staff rather than delegating to AI systems or even structured digital platforms.
Sector adoption velocityclaude-sonnet-52/5While healthcare imaging adopts AI diagnostic tools at moderate pace, the specific function of training and instructing staff on technique remains largely untouched by AI adoption trends.
Augmentation potentialclaude-haiku-4-5-202510013/5AI could assist radiologists by preparing instructional materials, annotating reference images, or generating scenario-based learning modules that the radiologist then delivers or reviews—moderately useful but not transformative to the core teaching interaction.
Augmentation potentialclaude-sonnet-52/5AI-based training materials, simulation software, or protocol reference tools could support instruction, but they only marginally assist the core interpersonal teaching task.
Task automatabilityclaude-haiku-4-5-202510011/5Instructing radiologic staff requires real-time demonstration, interpersonal communication, and adaptive feedback based on individual learner needs and comprehension—capabilities current AI systems cannot reliably perform end-to-end in a supervised setting.
Task automatabilityclaude-sonnet-51/5This is an in-person supervisory and teaching task requiring live demonstration, physical positioning guidance, and real-time feedback to staff, which current AI cannot perform end-to-end.:
Adoption barriersclaude-haiku-4-5-202510014/5Regulatory standards (e.g., ARRT certification bodies, medical imaging standards) typically require credentialed radiologists to direct and validate training of technical staff; liability and quality assurance also create strong institutional preferences for human expert oversight of instruction.
Adoption barriersclaude-sonnet-54/5This involves professional oversight and training responsibilities typically tied to a radiologist's credentialed role and departmental authority, creating strong organizational and professional barriers to substitution.
Cost vs. human wageclaude-haiku-4-5-202510012/5Creating AI-driven instructional systems requires significant setup and curation; the cost per instructional interaction remains comparable to or higher than a radiologist's time spent teaching, especially when factoring in oversight and refinement.
Cost vs. human wageclaude-sonnet-51/5There is no viable AI substitute for this instructional role, so cost comparison favors the human by default.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can generate instructional content or static guidance on radiologic techniques, no deployed product reliably substitutes for a radiologist actively teaching staff positioning, correcting errors in real time, or adapting instruction to individual competency—this remains largely human-dependent in clinical practice.
Technical feasibility todayclaude-sonnet-51/5No deployed product instructs radiologic technologists on positioning or technique in a hands-on clinical training capacity today.

Formulate plans and procedures for nuclear medicine departments.

10

CI 0–20 · exposure 8 · augmentation 38 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Strategic departmental planning is performed by senior radiologists and administrators, not a high-volume standardized task. Adoption of AI for this leadership function has been minimal; healthcare organizations prioritize human expertise and accountability in governance roles.
Sector adoption velocityclaude-sonnet-52/5Healthcare administrative planning functions, especially in specialized radiology subfields, show slow AI adoption due to regulatory complexity and low digitization of policy-setting workflows.
Augmentation potentialclaude-haiku-4-5-202510012/5AI could provide modest assistance by drafting template procedures or summarizing best practices, but the core work of synthesizing organizational strategy, risk assessment, and regulatory compliance requires experienced human judgment. The augmentation potential is limited to information synthesis rather than transformative productivity gains.
Augmentation potentialclaude-sonnet-53/5AI can help draft procedure documents, summarize regulations, or benchmark against best practices, meaningfully assisting radiologists who retain final judgment and accountability.
Task automatabilityclaude-haiku-4-5-202510011/5Formulating departmental plans and procedures requires strategic decision-making, organizational knowledge, and stakeholder alignment that current AI cannot perform end-to-end. While AI can draft procedure documents or analyze data to inform planning, the core task of synthesizing clinical, operational, and regulatory requirements into coherent departmental strategy remains fundamentally a human leadership function.
Task automatabilityclaude-sonnet-52/5Formulating departmental plans and procedures requires integrating regulatory knowledge, institutional context, staffing, and safety considerations that current AI cannot reliably synthesize end-to-end without heavy human oversight.atoire
Adoption barriersclaude-haiku-4-5-202510015/5Nuclear medicine departmental procedures and plans must be authored and signed by licensed radiologists or department heads due to regulatory, liability, and accreditation requirements (ACR, ASNM, CLIA standards). Legal and professional accountability for safety protocols creates hard barriers to autonomous automation.
Adoption barriersclaude-sonnet-54/5Nuclear medicine departments operate under strict radiation safety and regulatory frameworks (e.g., NRC licensing) requiring qualified physician oversight and sign-off on procedures.
Cost vs. human wageclaude-haiku-4-5-202510011/5The cost of AI systems that could assist with departmental planning, combined with the oversight and revision required from senior radiologists, likely exceeds the cost of direct human planning by experienced department leaders.
Cost vs. human wageclaude-sonnet-52/5While AI could draft template language cheaply, the human review, regulatory compliance checking, and institutional customization still dominate cost, keeping overall savings modest.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed product reliably performs strategic departmental planning and procedure formulation autonomously. Current AI tools cannot independently create governance structures, quality standards, or operational workflows that would satisfy the accountability and clinical governance requirements of a nuclear medicine department.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product autonomously drafts or manages nuclear medicine department policies and procedures; this remains a human administrative/clinical leadership function.

Teach nuclear medicine, diagnostic radiology, or other specialties at graduate educational level.

8

CI 0–16 · exposure 5 · augmentation 63 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Medical education institutions have strong institutional commitment to human faculty mentorship and regulatory requirements prevent wholesale substitution; adoption of AI in graduate radiology teaching remains at the pilot/supplemental stage.
Sector adoption velocityclaude-sonnet-52/5Academic medical education is adopting AI slowly for supplementary tools (quiz generation, case simulations) but instructor-of-record teaching remains almost entirely human-led.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can assist radiologist educators by generating case libraries, creating practice exams, or providing content summaries, but the core teaching function—clinical mentorship, real-time feedback, and adaptive instruction—remains human-led.
Augmentation potentialclaude-sonnet-54/5AI can generate teaching cases, quiz questions, literature summaries, and illustrative images to support radiology educators, meaningfully enhancing lecture and curriculum preparation.
Task automatabilityclaude-haiku-4-5-202510011/5Teaching at graduate level requires live instruction, dynamic interaction with learners, assessment of understanding, and adaptive pedagogy—tasks fundamentally dependent on human judgment and relational presence that current AI systems cannot perform end-to-end.
Task automatabilityclaude-sonnet-51/5Teaching at a graduate medical education level involves mentorship, live case discussion, clinical reasoning demonstration, and personalized feedback that current AI cannot replicate end-to-end.'},
Adoption barriersclaude-haiku-4-5-202510015/5Graduate medical education is heavily regulated; educators must hold board certification and be licensed physicians, and medical accreditation bodies require human faculty with clinical credentials to direct training and sign off on competency.
Adoption barriersclaude-sonnet-54/5Graduate medical education requires accredited, licensed faculty (ACGME requirements) with credentialing tied to specific board-certified physicians, creating strong institutional and regulatory barriers to substitution.
Cost vs. human wageclaude-haiku-4-5-202510011/5Replacing a radiologist educator would require multiple AI systems (content generation, student evaluation, interaction simulation) plus significant oversight, making total cost exceed the loaded salary of an experienced medical educator.
Cost vs. human wageclaude-sonnet-52/5AI content-generation tools are cheap, but since AI cannot perform the actual teaching role, the relevant cost comparison for genuine substitution is unfavorable to AI given need for continued expert oversight.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can generate educational content and simulate some teaching scenarios, no deployed product reliably replaces graduate-level medical instruction or provides the mentorship, clinical reasoning modeling, and real-time feedback that defines this role.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product substitutes for a radiology attending teaching residents/fellows in clinical and didactic settings; AI is at most a supplementary reference tool.

Prescribe radionuclides and dosages to be administered to individual patients.

7

CI 3–11 · exposure 5 · augmentation 50 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Medical imaging and nuclear medicine departments use dosing software and decision-support tools, but autonomous or semi-autonomous AI prescription is not in production use. Adoption remains limited to advisory functions, not autonomous ordering.
Sector adoption velocityclaude-sonnet-52/5While radiology broadly is a fast AI-adopting field for image analysis, the specific act of prescribing radionuclide dosing remains largely untouched by AI deployment due to regulatory constraints.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can meaningfully assist by retrieving dosing guidelines, calculating body-weight-adjusted doses, flagging contraindications, and surfacing relevant prior imaging. These augment a radiologist's workflow, but the final prescribing decision remains human-centered.
Augmentation potentialclaude-sonnet-53/5AI/dosimetry software can assist by calculating suggested dosages or flagging interactions based on patient data, improving efficiency, but the radiologist retains full decision authority and oversight.
Task automatabilityclaude-haiku-4-5-202510011/5This task requires authoritative clinical judgment about individual patient risk-benefit profiles, contraindications, and dose adjustments based on medical history. Current AI cannot independently prescribe therapeutically—it can suggest or draft, but cannot meet the ≥50% time-saving bar for end-to-end autonomous prescribing without clinician review and override authority.
Task automatabilityclaude-sonnet-51/5This is a licensed medical decision requiring clinical judgment about patient-specific factors (renal function, allergies, indication, weight-based dosing); no current AI system independently prescribes radiopharmaceuticals end-to-end.4Off-the-shelf tools cannot legally or reliably replace this decision-making step.
Adoption barriersclaude-haiku-4-5-202510015/5Prescription of radionuclides is explicitly a regulated, licensed professional act; only physicians (or authorized specialists) can legally issue prescriptions. Regulatory oversight, liability asymmetry, and legal mandate create hard barriers to substitution.
Adoption barriersclaude-sonnet-55/5Prescribing radioactive materials is tightly regulated (e.g., NRC/FDA authorized user requirements) and requires a licensed physician's legal authorization, creating a hard regulatory barrier to automation.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI systems can provide dosing reference data and calculation assistance at low cost, but the task itself—responsible prescription with liability—is a physician decision for which the human cost remains the dominant factor. AI augmentation is cost-effective, but cannot replace the human in the decision.
Cost vs. human wageclaude-sonnet-51/5Since AI cannot legally or practically perform this prescribing function independently, there is no viable cost comparison—a licensed radiologist must still make and sign off on the decision.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed AI system independently prescribes radionuclides and dosages in clinical practice; such prescription remains a licensed physician function. Support tools exist (dosing calculators, databases), but these are decision aids, not autonomous prescribers.
Technical feasibility todayclaude-sonnet-51/5No deployed product autonomously prescribes radionuclide type and dosage in clinical practice; decision-support tools exist but do not replace physician prescribing authority.

Direct nuclear medicine technologists or technicians regarding desired dosages, techniques, positions, and projections.

6

CI 0–11 · exposure 8 · augmentation 38 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Healthcare adoption of autonomous clinical decision-making remains minimal, especially in licensed, patient-safety-critical domains. Radiologists directing technicians is deeply embedded in clinical workflows and regulatory structure, with no observable shift toward AI delegation.
Sector adoption velocityclaude-sonnet-52/5Radiology imaging interpretation has seen AI tool adoption, but the specific supervisory/directive task with technologists remains largely untouched by AI deployment in clinical practice.
Augmentation potentialclaude-haiku-4-5-202510012/5AI might assist with archival image retrieval or protocol suggestions, but the core task—issuing real-time clinical directives—does not lend itself to meaningful AI assistance without introducing unacceptable diagnostic risk. The human radiologist remains the sole decision-maker.
Augmentation potentialclaude-sonnet-53/5AI decision-support tools can suggest dosage protocols or imaging parameters based on patient data, assisting radiologists' directives, though the human retains full decision authority.
Task automatabilityclaude-haiku-4-5-202510011/5This task fundamentally requires real-time clinical judgment, interpersonal communication, and real-time adjustments based on patient anatomy and imaging needs. Current AI cannot reliably substitute for a radiologist's diagnostic expertise and decision-making in directing technical staff during live procedures.
Task automatabilityclaude-sonnet-52/5This involves interpersonal clinical direction, judgment calls based on patient-specific factors, and real-time supervision that current AI cannot autonomously perform end-to-end.'
Adoption barriersclaude-haiku-4-5-202510015/5Regulatory and legal requirements mandate that a licensed radiologist direct nuclear medicine procedures and sign off on techniques and dosages. This is a hard licensing barrier—only a qualified radiologist can legally make these clinical decisions.
Adoption barriersclaude-sonnet-55/5Radiologists are legally required to supervise and direct dosage and imaging protocols under radiation safety regulations and medical licensing, making this a hard legal barrier to automation.
Cost vs. human wageclaude-haiku-4-5-202510011/5The cost of AI inference plus safety oversight would likely exceed the time savings, since a radiologist must remain in the loop for patient safety and liability reasons. Removing the radiologist from directing technical decisions would violate standard practice and create unacceptable clinical risk.
Cost vs. human wageclaude-sonnet-51/5There is no AI system performing this direction task, so no cost comparison favors AI; the physician's licensed judgment is currently irreplaceable at any AI price point.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed AI product reliably performs the clinical decision-making and real-time directive function required to guide technologists in nuclear medicine imaging. This requires licensed radiologist expertise and cannot be delegated to autonomous AI systems in production settings.
Technical feasibility todayclaude-sonnet-51/5No deployed product directs nuclear medicine technologists in clinical workflow today; this remains a physician-led supervisory task with no commercial substitute in production.

Monitor handling of radioactive materials to ensure that established procedures are followed.

3

CI 0–5 · exposure 5 · augmentation 25 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Radiological safety is a heavily regulated domain with safety-critical requirements; adoption of AI for independent material handling monitoring is minimal given mandatory human licensing and oversight requirements.
Sector adoption velocityclaude-sonnet-51/5Radiation safety compliance monitoring in healthcare is a highly regulated, low-digitization physical safety function with minimal AI adoption reported.
Augmentation potentialclaude-haiku-4-5-202510012/5AI could assist by flagging anomalies in recorded video or alerting to deviations from protocol, but the core task of live monitoring and compliance verification remains a human responsibility with limited scope for meaningful augmentation.
Augmentation potentialclaude-sonnet-52/5AI can assist with logging, dosimetry data analysis, or flagging anomalies in records, but it offers limited direct assistance to the core in-person compliance monitoring task.
Task automatabilityclaude-haiku-4-5-202510011/5Monitoring radioactive material handling requires real-time physical presence, visual inspection of procedures, and immediate intervention—tasks that demand embodied awareness and decision-making in regulated environments that current AI cannot perform end-to-end.
Task automatabilityclaude-sonnet-51/5This is a physical safety-monitoring and compliance oversight task requiring on-site presence, judgment, and accountability that current AI cannot perform end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Nuclear Regulatory Commission and state regulations mandate that licensed professionals directly oversee radioactive materials; legal and liability requirements prevent substitution of AI monitoring for credentialed human oversight.
Adoption barriersclaude-sonnet-55/5Radiation safety monitoring is governed by strict regulatory and licensing requirements (e.g., NRC/state agreement state rules) requiring a qualified, licensed individual to be responsible, creating a hard legal barrier.
Cost vs. human wageclaude-haiku-4-5-202510011/5The task requires a licensed radiologist or radiation safety officer on-site for compliance; AI oversight would be supplementary, not replacement, so the human wage remains the binding cost.
Cost vs. human wageclaude-sonnet-51/5There is no viable AI-only substitute for this role, so cost comparison favors the human since AI cannot deliver the required physical presence and regulatory accountability.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can analyze static images or video logs of handling procedures post-hoc, no deployed system reliably monitors live radioactive material handling in real-time with the safety-critical judgment required in production nuclear or medical settings.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product independently monitors radioactive material handling compliance in clinical settings; sensor-based tracking exists but not as an autonomous monitor replacing human oversight.

Perform interventional procedures such as image-guided biopsy, percutaneous transluminal angioplasty, transhepatic biliary drainage, or nephrostomy catheter placement.

3

CI 0–5 · exposure 5 · augmentation 50 · importance 4.7/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Interventional radiology is a highly specialized, regulated clinical domain with slow organizational adoption of autonomous automation. Pilot AI tools focus on planning and guidance rather than autonomous execution; no significant displacement of procedures has occurred, and adoption remains concentrated in academic medical centers and pilot programs.
Sector adoption velocityclaude-sonnet-51/5Interventional radiology is a highly physical, procedural medical specialty with minimal AI-driven displacement; adoption is limited to decision-support and imaging analysis tools, not the procedural task itself.
Augmentation potentialclaude-haiku-4-5-202510013/5AI-assisted image segmentation, vessel tracking, optimal needle trajectory suggestions, and real-time anatomical overlays demonstrably aid radiologists during procedure planning and execution, improving speed and reducing complications. However, the radiologist remains the primary decision-maker and performer; augmentation is meaningful but not transformative of the core task.
Augmentation potentialclaude-sonnet-53/5AI aids pre-procedure planning, image guidance, lesion detection, and navigation overlays, improving precision and efficiency, though the physician remains fully in control of the physical procedure.
Task automatabilityclaude-haiku-4-5-202510011/5Interventional procedures require real-time physical manipulation, haptic feedback, and dynamic decision-making in response to patient anatomy and procedure complications. Current AI cannot perform hands-on catheter placement, needle guidance, or device deployment—these remain firmly in the domain of trained interventional radiologists.
Task automatabilityclaude-sonnet-51/5This is a hands-on physical procedure requiring catheter/needle manipulation, real-time tactile feedback, and management of complications; no current AI system can perform the physical intervention itself.
Adoption barriersclaude-haiku-4-5-202510015/5Interventional procedures are inherently license-gated: only credentialed radiologists can perform them under radiologist supervision; malpractice liability and adverse-event costs are severe; and regulatory bodies (FDA, state medical boards) require human physician oversight and sign-off. Patient safety and legal responsibility form hard adoption barriers.
Adoption barriersclaude-sonnet-55/5Invasive procedures require licensed physician performance, informed consent, sterile technique, and legal/malpractice accountability, making this among the most heavily regulated and human-contact-dependent tasks.
Cost vs. human wageclaude-haiku-4-5-202510011/5Interventional procedures command high reimbursement and require expensive expertise, equipment, and anesthesia support. AI image analysis or planning assistance remains a small fraction of total procedure cost, and the human radiologist's loaded wage (including training, liability, and OR time) far exceeds the cost of AI assistance.
Cost vs. human wageclaude-sonnet-51/5There is no AI substitute performing the procedure, so cost comparison is moot—human interventionalist cost is the only viable option; any AI-assisted robotics add cost rather than reduce it.
Technical feasibility todayclaude-haiku-4-5-202510012/5AI systems exist for procedure planning and image analysis (segmentation, trajectory suggestion), but no deployed product autonomously executes the full procedure. Research prototypes show image-guided needle placement in controlled settings, but production systems only assist human operators; they do not substitute.
Technical feasibility todayclaude-sonnet-51/5No deployed product performs interventional radiology procedures autonomously; AI is confined to pre-procedural planning, image analysis, or robotic assistance research, not full task execution.

Establish and enforce radiation protection standards for patients and staff.

1

CI 0–3 · exposure 0 · augmentation 38 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5This is a governance and compliance task that remains tightly bound to human professional judgment and legal responsibility; adoption of AI for autonomous policy-making in radiation safety is essentially non-existent because regulators require human accountability.
Sector adoption velocityclaude-sonnet-52/5Healthcare imaging departments are adopting AI for image analysis and dose optimization, but administrative/regulatory enforcement functions like this remain largely untouched by AI adoption trends.
Augmentation potentialclaude-haiku-4-5-202510012/5AI could assist by summarizing regulatory updates or drafting documentation templates, but the core task—establishing institutional standards and enforcing compliance—remains overwhelmingly a human responsibility requiring expert judgment and legal authority.
Augmentation potentialclaude-sonnet-53/5AI-based dose monitoring and analytics tools can help track exposure data and flag anomalies, assisting radiologists in maintaining standards, though the actual establishment and enforcement remains human-driven.
Task automatabilityclaude-haiku-4-5-202510011/5Establishing and enforcing radiation protection standards requires authoritative judgment about institutional policy, regulatory compliance, and organizational implementation—tasks fundamentally dependent on human decision-making and accountability. AI cannot autonomously create or enforce organizational policy.
Task automatabilityclaude-sonnet-51/5This is a governance and enforcement task requiring judgment, authority, and accountability for physical safety standards; AI cannot establish policy or enforce compliance among staff.The task is inherently organizational and regulatory, not a data-processing task.
Adoption barriersclaude-haiku-4-5-202510015/5Strict regulatory frameworks (NCRP, ICRP, ACR guidelines) and legal liability requirements mandate that licensed radiologists or certified radiation safety professionals establish and sign off on protection standards; AI cannot legally replace this function.
Adoption barriersclaude-sonnet-55/5Radiation safety enforcement is tightly regulated (NRC, state agencies, Joint Commission) and requires a licensed radiologist or physicist to establish and be accountable for protocols, creating strong legal and liability barriers to automation.
Cost vs. human wageclaude-haiku-4-5-202510011/5A radiologist or radiation safety officer commanding professional salary and legal authority must perform this task; AI tools cannot substitute for the human authority and accountability required, making any AI assistance economically inferior to human execution.
Cost vs. human wageclaude-sonnet-51/5There is no AI system performing this task, so no meaningful cost comparison exists; the human cost is the only viable option currently.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed product independently establishes and enforces institutional radiation protection protocols; this is an administrative and governance function that requires qualified human personnel with legal and regulatory authority.
Technical feasibility todayclaude-sonnet-51/5No deployed product establishes or enforces radiation safety protocols; existing AI tools may monitor dose data but do not set or enforce standards.This remains a human administrative and regulatory responsibility.

Establish or enforce standards for protection of patients or personnel.

1

CI 0–3 · exposure 0 · augmentation 38 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Even in digitized healthcare sectors, the governance and enforcement of patient/personnel protection standards remains a core human management and compliance function with no active AI displacement.
Sector adoption velocityclaude-sonnet-52/5Healthcare administrative and safety-compliance functions show slow AI adoption due to regulatory oversight and liability concerns, despite healthcare's growing use of AI in diagnostics.
Augmentation potentialclaude-haiku-4-5-202510012/5AI could assist by surfacing policy gaps or compliance data (e.g., incident pattern analysis), but the task's core—establishing norms and enforcing accountability—remains predominantly a human leadership responsibility with limited augmentation potential.
Augmentation potentialclaude-sonnet-53/5AI tools can help track compliance data, flag anomalies in safety metrics, or draft policy documents, aiding but not replacing the radiologist's standard-setting role.
Task automatabilityclaude-haiku-4-5-202510011/5This task requires exercising professional judgment about organizational policy, institutional accountability, and regulatory compliance—functions fundamentally centered on human authority and legal responsibility that AI cannot perform end-to-end today.
Task automatabilityclaude-sonnet-51/5This is a governance and policy-setting task requiring judgment about institutional standards, regulatory compliance, and safety protocols; AI cannot autonomously establish or enforce organizational standards.'
Adoption barriersclaude-haiku-4-5-202510015/5This task is legally vested in licensed healthcare professionals and organizational management; regulatory bodies (CMS, state boards, HIPAA) mandate human accountability for institutional safety policies and enforcement, creating hard legal barriers to automation.
Adoption barriersclaude-sonnet-55/5Radiologists have legal and licensing responsibility for radiation safety standards and personnel protection under regulatory bodies (e.g., NRC, Joint Commission), making this a hard barrier to automation.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI has no meaningful role in replacing this task, so direct cost comparison is inapplicable; any deployment would require substantial human oversight, making total cost higher than traditional human management.
Cost vs. human wageclaude-sonnet-51/5There is no AI substitute performing this function, so cost comparison favors the human entirely; AI cannot replace the accountable decision-maker.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed product autonomously establishes or enforces institutional standards for patient/personnel protection; this requires organizational authority, policy-making, and disciplinary capacity that remains entirely human-driven in production radiography environments.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product creates or enforces radiation safety or patient protection standards; this remains a human administrative/leadership function.

Administer radioisotopes to clinical patients or research subjects.

0

CI 0–0 · exposure 0 · augmentation 25 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5This task involves direct patient contact and licensed clinical procedure delivery, which falls outside automation patterns even in highly digitized healthcare settings.
Sector adoption velocityclaude-sonnet-51/5Physical, licensed clinical procedures in nuclear medicine show minimal AI adoption for the hands-on administration step itself, reflecting the broader lag in physically-embodied healthcare tasks.
Augmentation potentialclaude-haiku-4-5-202510012/5AI could assist with dose calculations, patient screening, or documentation preparation, but the core act of administering radioisotopes remains entirely human-performed with limited augmentation opportunity.
Augmentation potentialclaude-sonnet-52/5AI can assist with dose calculation, scheduling, or documentation around the procedure, but offers negligible assistance for the physical act of administering the radioisotope itself.
Task automatabilityclaude-haiku-4-5-202510011/5Administering radioisotopes to patients or research subjects requires direct physical handling, injection, and patient contact that AI systems cannot perform. This is a hands-on clinical procedure with no meaningful automation pathway.
Task automatabilityclaude-sonnet-51/5This is a hands-on clinical procedure requiring physical administration of radioactive substances to a patient's body, which current AI systems cannot physically perform.It also requires real-time judgment about patient condition and IV access.
Adoption barriersclaude-haiku-4-5-202510015/5Radioisotope administration is heavily regulated and requires a licensed radiologist or nuclear medicine technologist to physically perform and sign off on the procedure under strict regulatory (NRC, state medical boards) and institutional oversight.
Adoption barriersclaude-sonnet-55/5Administering radioisotopes requires licensure (e.g., authorized user status under nuclear regulatory rules), strict radiation safety protocols, and legal accountability for handling radioactive materials on patients.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI cannot perform this task at any cost, so cost comparison is not applicable; human radiologists remain the only option for this procedure.
Cost vs. human wageclaude-sonnet-51/5There is no AI substitute performing this physical task, so no cost comparison is meaningful; a human must be present and paid to perform it.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed AI product can physically administer radioisotopes to patients. This task fundamentally requires embodied action by a trained human professional in a clinical setting.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product administers radioisotopes; this remains a manual clinical procedure performed by licensed personnel (physicians, nuclear medicine technologists, or nurses).

Related occupations — Healthcare Practitioners & Technical

How to read this

A high substitution score does not mean this job disappears — it means a large share of its current tasks face replacement pressure, so the mix of tasks is likely to change. High augmentation alongside substitution typically means the occupation reorganizes around the protected tasks. Wide confidence intervals mean the rater panel disagreed: treat those scores as open questions, not verdicts.

What would change this score

New model capabilities (automatability, feasibility), falling inference costs (cost ratio), regulation and licensing shifts (barriers), and measured sector adoption (velocity) all re-enter at every index release. Each release is recomputed, versioned and kept queryable — scores are claims with a date on them, not permanent labels.