Diagnosis Support Apparatus With Rejection-Based Reason Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Doctors face uncertainty when relying on medical AI diagnosis results due to a lack of transparency in the reasoning behind the outputs, leading to difficulties in adopting or rejecting these results without clear conviction.
Innovation Solution
A diagnosis support apparatus that uses processing circuitry to output identification results and reasons, allowing for modification of the identification processing based on operator feedback, enabling the exclusion of rejected reasons and re-execution of diagnosis with adjusted algorithms or additional information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If medical AI program outputs identification result with identification reason, then doctor's understanding of the diagnosis is improved, but the complexity of the system increases
Solution Approach 1:
The identification reason is segmented into multiple individual pieces of information, each corresponding to a specific feature or finding from the examination data. This allows doctors to review and reject specific reasons individually rather than treating the reasoning as a monolithic block, thereby improving understanding without proportionally increasing overall system complexity.
Solution Approach 2:
The system dynamically adjusts the identification reasons based on the specific examination data and the AI model's analysis. The reasons are generated adaptively rather than being fixed or predetermined, allowing the system to provide tailored explanations for each diagnosis case, improving reliability while maintaining manageable complexity through on-demand generation.
2Reliability
If doctor rejects identification reason, then diagnostic accuracy is improved by incorporating human judgment, but additional time is required for review and re-execution
Solution Approach 1:
The system implements a feedback mechanism where doctors can review and reject identification reasons, and the AI program re-executes the analysis incorporating this feedback. This iterative process allows continuous improvement of diagnostic accuracy by combining AI capabilities with human judgment, while the structured feedback loop ensures that time is invested only in cases where human review adds value.
Solution Approach 2:
The AI program performs preliminary analysis and generates identification reasons before doctor review, filtering out obvious cases that don't require extensive human intervention. This preliminary action reduces the overall time loss by pre-processing cases and only requiring full doctor engagement when necessary, thereby balancing accuracy improvement with time efficiency.
3Measurement precision
If AI program re-executes identification processing after rejection, then more accurate identification result is obtained, but computational resources are consumed
Solution Approach 1:
When re-executing identification processing after rejection, the AI program performs partial re-analysis focusing only on the specific rejected reasons rather than completely re-processing all examination data. This selective re-execution maintains improved identification accuracy by re-evaluating problematic areas while consuming fewer computational resources than a full re-analysis would require.
Data Source
AI summary
According to one embodiment, a diagnosis support apparatus includes processing circuitry. The processing circuitry executes identification processing by using medical information as an input to output a first identification result and a first identification reason for providing the first identification result. The processing circuitry modifies the identification processing to refrain from outputting the first identification reason in response to an instruction to reject the first identification reason. The processing circuitry executes the modified identification processing by using the medical information as an input to output a second identification result and a second identification reason for providing the second identification result.