Automated Medical Image Data Generation for Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the medical field, generating high-quality learning data for deep learning applications is challenging due to the need for manual discrimination of correct answer data from large amounts of medical image data, which is inefficient and limits the enhancement of recognition accuracy for disease diagnosis.
Innovation Solution
A learning data generation support apparatus and method that analyzes character strings in interpretation reports to retrieve relevant data, performs image processing on medical images to extract anatomic regions differing from standard sizes or shapes, and registers this information as correct answer data, facilitating automated data acquisition for deep learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual discrimination of correct answer data is performed from large amounts of medical image data, then data quality can be ensured, but processing time and labor cost increase significantly
Solution Approach 1:
The system enables automated self-service by having the computer automatically retrieve interpretation reports, extract anatomic region information, compare with standard data, and identify correct answer data without human intervention. The retrieval unit, extraction unit, comparison unit, and identification unit work together to create a self-operating pipeline that eliminates manual discrimination while maintaining data quality.
Solution Approach 2:
The patent replaces the mechanical manual discrimination process with an automated information processing system. The computer executes programmed operations including report retrieval, image processing, region extraction, comparison, and identification, substituting human cognitive and manual labor with algorithmic automation that processes data much faster while consistent quality standards.
2Measurement precision
If manual discrimination of correct answer data is performed, then accurate learning data can be obtained, but productivity decreases due to labor-intensive process
Solution Approach 1:
The system performs automated self-service through a sequence of computer-executed operations: retrieving interpretation reports containing anatomic region information, processing medical images to extract region data, comparing extracted data with standard anatomic region data, and automatically identifying correct answer data. This eliminates manual labor while maintaining accuracy through systematic automated comparison and selection processes.
Solution Approach 2:
The system extracts only the necessary anatomic region information from interpretation reports and medical images, separating this critical data from the large volume of unrelated medical data. By focusing extraction on specific relevant features and comparing only these extracted elements against standards, the system achieves high productivity without sacrificing data accuracy.
3Stability of the object's composition
If uniform correct answer data is used for deep learning, then training consistency is maintained, but recognition accuracy enhancement is limited
Solution Approach 1:
The system incorporates feedback mechanisms by comparing extracted anatomic region data against standard anatomic region data, using this comparison to identify cases that deviate from norms. This feedback loop enables the systematic collection of both uniform and variant correct answer data, allowing the learning system to train on diverse real-world variations while maintaining structured consistency through the comparison framework.
Solution Approach 2:
The system changes the parameter diversity of training data by deliberately identifying and selecting correct answer data that exhibits variations in anatomic region characteristics. By adjusting the selection criteria to include data with different sizes, shapes, or features compared to standards, the system enriches the training dataset with parameter variations that enhance recognition accuracy while maintaining overall data quality.
Data Source
AI summary
Retrieval means analyzes character strings of a plurality of interpretation reports to retrieve an interpretation report in which a retrieval keyword is included. Registration means performs image processing with respect to a medical image corresponding to the retrieved interpretation report, extracts an anatomic region related to the retrieval keyword, and registers information indicating the anatomic region and the medical image as correct answer data in a case where the size of the extracted anatomic region is different from a standard size of the anatomic region or in a case where the shape of the extracted anatomic region is different from a standard shape of the anatomic region.