Medical Data Similarity Evaluation via Deviance Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer Aided Diagnosis (CAD) systems face challenges in evaluating medical data with features not sufficiently learned by the classifier, especially when the attributes of the input data differ from the training data, leading to difficulties in accurately classifying medical images and determining the reliability of classification results.
Innovation Solution
An information processing apparatus that includes a likelihood acquisition unit to determine class likelihood, a first evaluation unit to assess the degree of deviance between the class likelihood and the correct answer label, and a second evaluation unit to evaluate the similarity between medical data based on a predetermined criterion, allowing for the identification of data that may not have been adequately learned by the classifier.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the classifier is trained with available teacher data, then classification can be performed, but the classifier cannot reliably evaluate medical data with features not sufficiently learned
Solution Approach 1:
The patent introduces an intermediary evaluation process between the classifier and the final classification result. The classifier first outputs class likelihood for input medical data, then this likelihood is evaluated against reference data to determine reliability. This intermediary step allows the system to identify when the classifier has not sufficiently learned certain features, even though the classifier itself is functioning normally.
Solution Approach 2:
The system implements feedback by comparing the classifier's output (class likelihood) with reference data from a database. This feedback mechanism enables the system to assess whether the classification result is reliable based on how well the input data matches previously learned patterns, thereby identifying cases where features were not sufficiently learned during training.
2Productivity
If the classifier processes all input medical data, then productivity is maintained, but accuracy decreases for data with insufficiently learned features
Solution Approach 1:
The patent segments the classification process into two distinct phases: (1) rapid classification by the trained classifier to maintain productivity, and (2) reliability evaluation by comparing class likelihood with reference data to ensure accuracy. This segmentation allows the system to quickly process all input data while separately identifying cases where accuracy may be compromised due to insufficient learning.
Solution Approach 2:
The system performs partial verification by evaluating only the reliability aspect (comparing class likelihood with reference data) rather than re-processing the entire classification pipeline. This partial action approach maintains high productivity while providing sufficient accuracy checks for identifying insufficiently learned features.
Data Source
AI summary
An information processing apparatus according to an exemplary embodiment includes a likelihood acquisition unit, a first evaluation unit, a determination unit, and a second evaluation unit. The likelihood acquisition unit is configured to acquire class likelihood for first medical data given a correct answer label, by using a first classifier for classifying medical data. The first evaluation unit is configured to evaluates a degree of deviance based on the class likelihood acquired by the likelihood acquisition unit and a class answer corresponding to the correct answer label. The determination unit is configured to determine whether the degree of deviance by the first evaluation unit satisfies a predetermined criterion. The second evaluation unit is configured to, based on second medical data satisfying the predetermined criterion, out of the first medical data, evaluate a similarity between the second medical data and third medical data to be subjected to classification.