CAD Training Data Compilation via Image and Report Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-aided diagnosis systems face challenges in extracting and utilizing all relevant information for training data, particularly missing elements from electronic documents, which affects the accuracy of machine training and inference in medical diagnoses.
Innovation Solution
An information processing apparatus that acquires and compiles training data by combining first image findings from interpretation reports and second image findings obtained through medical image analysis, ensuring comprehensive data for machine training in CAD systems, especially for diagnosing chest diseases from X-ray CT images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If training data is extracted only from electronic documents, then data extraction is simple, but completeness of training data is insufficient
Solution Approach 1:
The patent combines multiple data sources (electronic documents and medical image analysis results) into a unified training data set. The determination unit integrates finding elements from documents with image findings to create comprehensive training data, resolving the contradiction between data completeness and acquisition complexity.
Solution Approach 2:
The determination unit performs multiple functions: it extracts finding elements from electronic documents, analyzes medical images, and synthesizes both into unified training data. This multi-functional approach enables the system to overcome the limitations of single-source data extraction while managing complexity through integrated processing.
2Quantity of substance
If all case information is used as training data, then training data quantity increases, but data quality decreases
Solution Approach 1:
The determination unit changes the selection parameters for training data by evaluating multiple criteria including presence of finding elements, image analysis results, and diagnostic accuracy. This parameter-based filtering approach enables selective inclusion of cases, maintaining data quality while accumulating sufficient quantity for effective machine training.
3Extent of automation
If finding elements are extracted from electronic documents, then data extraction is automated, but elements not in documents cannot be captured
Solution Approach 1:
The determination unit acts as an intermediary that bridges electronic document extraction and medical image analysis. It receives finding elements from documents, supplements them with image analysis results, and produces comprehensive training data. This intermediary function captures information that would be lost in pure document extraction while maintaining automation.
Data Source
AI summary
An information processing apparatus is configured to acquire case information containing first information that is information on a feature of a case of a patient and that is obtained as a result of a diagnosis of the patient and determine, based on the first information that is the information contained in the case information, whether the case information is used as training data for machine training.