Interpretation Management Apparatus for Medical AI Discrepancy Notification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In medical image diagnosis, there is a need for active notification of interpretation results that differ between AI analysis and primary interpretation doctor findings, particularly to prevent overlooking by secondary interpretation doctors in QA-type interpretation styles, where AI positivity may not align with doctor positivity, and to ensure priority confirmation of discrepancies.
Innovation Solution
An interpretation management apparatus that acquires both AI-generated and doctor-created findings, determines if a combination requires active notification, and proactively notifies relevant personnel, ensuring that discrepancies and high-priority findings are promptly addressed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If only AI-positive cases are notified to doctors (triage type), then notification efficiency is improved, but the risk of overlooking AI-negative but doctor-positive cases increases
Solution Approach 1:
The notification system segments different notification types based on the combination of AI and doctor findings: AI-positive/doctor-positive cases receive one type of notification, while AI-negative/doctor-positive cases receive another type of notification. This segmentation allows the system to maintain high notification efficiency for confirmed cases while separately addressing the reliability concern for potentially overlooked cases.
Solution Approach 2:
The system applies different notification strategies to different local situations (case types). For AI-positive/doctor-positive cases, efficient notification is applied. For AI-negative/doctor-positive cases, enhanced notification is applied to prevent overlooking. This local quality approach resolves the contradiction by tailoring notification behavior to specific case characteristics.
2Reliability
If all interpretation results are notified (comprehensive notification), then diagnosis accuracy is improved, but notification burden and time consumption increase
Solution Approach 1:
The system segments notification into priority-based categories: high-priority notifications for AI-positive/doctor-positive and AI-negative/doctor-positive cases, and standard notifications for other cases. This segmentation maintains diagnosis accuracy for critical cases while reducing overall notification time by not treating all cases uniformly.
Solution Approach 2:
The system applies partial notification action to high-priority cases that require immediate attention, while applying standard notification to other cases. This partial action approach ensures that time-sensitive diagnostic accuracy is maintained for critical cases without unnecessarily consuming time on all cases.
3Reliability
If AI and doctor findings are compared for every case (QA type), then diagnosis reliability is improved, but interpretation time increases
Solution Approach 1:
The system segments the comparison process into mandatory comparisons (AI-positive/doctor-positive and AI-negative/doctor-positive cases) and optional comparisons (other cases). This segmentation maintains diagnosis reliability for critical cases while reducing overall interpretation time by not requiring full QA comparison for all cases.
Solution Approach 2:
The system performs preliminary identification of high-priority cases using AI analysis before full doctor comparison. By pre-sorting cases based on AI findings, the system can focus mandatory QA comparison on the most critical cases, thereby maintaining reliability where it matters most while improving overall productivity.
Data Source
AI summary
An interpretation management apparatus including a hardware processor that: acquires an automatically generated finding obtained by computer processing on medical information; acquires a first interpretation finding created by a user based on the medical information; determines whether or not a result combination is a combination for which an active notification is to be performed, the result combination being a combination of a result of the automatically generated finding and a result of the first interpretation finding; and actively notifies predetermined information in response to determination that the result combination is the combination for which the active notification is to be performed.