AI-Driven Radiation Dosimeter Overdose

Hazard Definition

AI-driven radiation dosimeter overdose refers to incidents where artificial intelligence systems involved in radiation therapy planning, dose calculation, or delivery control produce incorrect dosing that results in patient overexposure. These overdoses can cause acute radiation syndrome, severe tissue damage, organ injury, and in extreme cases, patient death.

Mechanism of Harm

Radiation therapy requires precise dose delivery within narrow therapeutic windows. AI involvement at multiple workflow stages creates distinct overdose pathways.

Treatment planning errors: AI systems used to optimize radiation treatment plans may generate dose distributions that appear clinically acceptable but contain localized hot spots or systematic calculation errors. Complex tumor geometries and proximity to critical structures can expose limitations in AI planning algorithms.

Auto-segmentation failures: AI tools that automatically identify tumor boundaries and organs at risk from imaging data may incorrectly delineate target volumes. Undersegmentation of healthy tissue or oversegmentation of tumor margins can result in inappropriate dose delivery.

Adaptive therapy miscalculation: AI-driven adaptive radiation therapy systems that adjust treatment based on daily imaging may compound errors across fractions. If the AI misinterprets anatomical changes, cumulative dose errors can reach harmful levels before detection.

Documented Incident Patterns

Radiation therapy has a documented history of catastrophic overdose incidents predating AI involvement, providing context for emerging AI-specific concerns.

Historical overdose precedent: High-profile radiation overdose incidents have resulted from software errors, calibration failures, and workflow breakdowns. These incidents demonstrate the severe consequences of radiation dosing errors and inform risk assessment for AI-enabled systems.

Auto-contouring deviations: Published analyses have identified clinically significant differences between AI-generated contours and expert physician delineations for certain anatomical structures and tumor types.

Quality assurance findings: Radiation oncology quality assurance programs have identified AI treatment plan recommendations that would have delivered inappropriate doses if implemented without physics review.

Regulatory Status

AI software used in radiation therapy planning is regulated by the FDA as a medical device. Treatment planning systems and auto-contouring tools have received 510(k) clearance.

Mandatory incident reporting exists for radiation therapy misadministrations exceeding defined thresholds, though attribution to AI-specific failures may not be captured in reports.

Known Data Gaps

• Number of radiation overdose incidents attributable to AI planning or segmentation errors
• Comparative error rates between AI-assisted and traditional radiation therapy workflows
• Detection rate of AI planning errors by quality assurance processes before patient treatment
• Long-term patient outcomes following AI-planned radiation therapy across institutions