Medical Answer Reliability Scoring Using Entity and Similarity Checks

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Generative models in the medical field often produce unreliable or incoherent answers, leading to potential harm to users due to hallucinations and lack of context awareness.

Innovation Solution

A medical information processing apparatus that generates prompts for generative models, extracts named entities from prompts and answers, searches medical databases for similarities, and determines reliability based on vector similarity and hit counts to ensure coherent and accurate responses.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a generative model is used to automatically generate medical answers, then productivity is improved, but reliability deteriorates due to hallucinations and incoherence

Engineering Contradiction:

Improveautomated answer generationVSAvoidaccuracy of medical information

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system incorporates a feedback mechanism where the reliability determination unit continuously evaluates the generative model's output against the prompt and extracts named entities to verify accuracy. This feedback loop identifies hallucinations and incoherent responses, allowing the system to flag unreliable information while maintaining automated generation capabilities.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

An intermediary reliability determination unit is introduced between the prompt generation and answer output stages. This intermediary component extracts named entities from both the prompt and answer, searches medical databases for verification, and calculates similarity scores to mediate the trustworthiness of the generated medical information.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If the reliability determination process includes database searching and vector similarity calculation, then reliability is improved, but device complexity increases

Engineering Contradiction:

Improveverification of medical informationVSAvoidnumber of processing units

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The output unit performs multiple functions: extracting named entities from text, searching medical databases, calculating vector similarities, and determining reliability. By consolidating these functions into a single multi-functional unit, the system reduces overall device complexity while maintaining comprehensive reliability verification capabilities.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system merges the reliability determination functions with the existing prompt and answer processing workflow. The output unit combines named entity extraction, database searching, and similarity calculation into an integrated process that runs concurrently with the generative model operations, avoiding the need for separate complex processing pipelines.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentEP4657279A1Medical information processing apparatus, medical information processing method, and medical information processing program

Publication Date: 2025.12.03 CANON MEDICAL SYST CORP

Figure 1
Figure 2
Figure 3

AI summary

A medical information processing apparatus (30) according to one embodiment includes a generator unit (34a), an obtaining unit (34b), and an output unit (34c). The generator unit (34a) generates a prompt to be input to a first generative model (11), in accordance with a user's input. The obtaining unit (34b) obtains an answer including medical information from the first generative model (11), in response to an input of the prompt to the first generative model (11). The output unit (34c) outputs reliability of the answer based on the prompt and the answer.