Medical AI Inference Visualization for Confounding Factor Reduction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine learning models in medical imaging are prone to low accuracy due to confounding factors, leading to false correlations and reduced performance on unknown data, and current methods to address this either reduce data quantity excessively or burden doctors with manual labeling.

Innovation Solution

A medical information processing device and method that visualizes the inference basis of a machine learning model, allowing doctors to edit and update the model to align with clinical intuition, thereby improving accuracy by emphasizing clinically valid information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If images highly similar to the designated image are extracted to eliminate confounding factors, then the model becomes more robust against confounding factors, but the number of training data is excessively reduced

Engineering Contradiction:

Improverobustness against confounding factorsVSAvoidnumber of training data

Core Design Contradiction:

ReliabilityVSQuantity of substance

Solution Approach 1:

The patent replaces the manual mechanical process of doctors labeling clinical concepts with an automated computational system. The system automatically identifies and labels clinical concepts in training data using image recognition and natural language processing algorithms, eliminating the need for manual annotation while maintaining accuracy in identifying clinically valid information.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables self-service by allowing the training data to automatically highlight and label its own clinical concepts through automated analysis. The training data itself generates the annotations needed for robust training without requiring external manual intervention, thus maintaining data quantity while achieving reliability.

Inventive Principle:

Principle #25Self-service

2Reliability

If doctors manually label clinical concepts for each training data to update the model, then the model emphasizes clinically valid information, but the burden on doctors is heavy

Engineering Contradiction:

Improveemphasis on clinically valid informationVSAvoidtime burden on doctors

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces an intermediary automated system that acts as a bridge between the training data and the machine learning model. This intermediary system automatically extracts and labels clinical concepts, translating raw training data into structured, clinically-valid training examples without requiring direct doctor involvement in the labeling process.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Ease of operation

If unsupervised learning is used to identify and label clinical concepts by clustering, then manual labeling burden is reduced, but it is not easy to check whether the inference basis matches clinical concepts

Engineering Contradiction:

Improvereduction of manual labeling burdenVSAvoidverification of inference basis alignment with clinical concepts

Core Design Contradiction:

Ease of operationVSDifficulty of detecting and measuring

Solution Approach 1:

The patent implements feedback mechanisms where the system generates explanations for its inferred clinical concepts and allows doctors to verify and correct these inferences. The system provides feedback loops where doctors can review the automatically identified clinical concepts, confirm their accuracy, and provide corrections that are used to improve future automatic labeling, thus enabling easy verification while maintaining automation.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20250225439A1Medical information processing device, medical information processing method, and storage medium

Publication Date: 2025.07.10 CANON KK

AI summary

A medical information processing device of an embodiment includes processing circuitry. The processing circuitry is configured to acquire a machine learning model and training data used to train the machine learning model, determine an inference basis for each piece of the training data using the machine learning model to generate inference basis visualization results, determine a concept emphasized by the machine learning model during inference based on the inference basis visualization results, calculate a concept reflection degree of each piece of the training data related to the concept, and generate visualization information of a dependency between the concept and a feature interpretable by a user in the training data based on the concept reflection degree and the feature interpretable by the user.