Tissue Image Analysis Using Betti Numbers for Explainable Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image analysis techniques using artificial intelligence struggle to provide understandable inference results for medical images, as the grounds for outputting these results are often incomprehensible to medical professionals, making it difficult to utilize them in current medical treatment.
Innovation Solution
An image analyzing method and device that extracts a target region from a tissue image, generates multiple binarized images with different binarization reference values, calculates Betti numbers, and determines changes in tissues based on maximized Betti numbers and reference values, utilizing the concept of homology to enhance accuracy and understanding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If artificial intelligence autonomously outputs inference results by learning enormous amounts of image data, then determination accuracy is improved, but comprehensibility of the inference grounds deteriorates
Solution Approach 1:
The patent segments the image analysis process into multiple binarized images with different reference values, calculating Betti numbers for each segment. This allows the system to maintain high determination accuracy through comprehensive analysis while providing comprehensible inference grounds by showing which specific structural features (holes, connected components) at which binarization levels contributed to the final determination.
Solution Approach 2:
The patent changes the binarization reference value parameter across multiple images to extract different structural characteristics. By analyzing how Betti numbers vary with different reference values, the system achieves accurate determination of tissue changes while making the inference process comprehensible through the visible relationship between parameter changes and structural feature extraction.
2Measurement precision
If multiple binarized images with different reference values are generated and analyzed, then determination accuracy is improved, but device complexity increases
Solution Approach 1:
The patent extracts only the essential topological features (Betti numbers representing holes and connected components) from each binarized image, rather than processing all image data. This extraction approach maintains high determination accuracy by focusing on structurally significant features while reducing processing complexity by eliminating redundant information.
Solution Approach 2:
The patent replaces complex mechanical image analysis with topological data analysis using Betti numbers. This substitution simplifies the processing system by using mathematical topology invariants that are computationally efficient to calculate and compare, thereby maintaining accuracy while reducing device complexity.
Data Source
AI summary
The disclosed feature makes it possible to accurately determine a change that has occurred in a tissue. The feature includes: a binarizing section (41) that generates, from an image to be analyzed, a plurality of binarized images having respective binarization reference values different from each other; a Betti number calculating section (42) that calculates, for each of the plurality of binarized images, a one-dimensional Betti number indicating the number of hole-shaped regions each of which is surrounded by pixels each having a first pixel value obtained by binarization and is constituted by pixels each having a second pixel value obtained by binarization; and a determining section (44) that determines a change that has occurred in the tissue, based on a binarization reference value and a one-dimensional Betti number in a binarized image in which the one-dimensional Betti number is maximized.