Learning Data Creation Support Apparatus for Medical Image Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual process of designating regions for learning data in medical images is time-consuming and inefficient, making it difficult to generate large amounts of high-quality learning data for deep learning applications in the medical field.
Innovation Solution
A learning data creation support apparatus and method that automatically displays candidate lesion regions on a schematic diagram of the human body, allowing radiologists to easily confirm or deny these regions, thereby registering them as correct or incorrect answer data, facilitating the creation of learning data without additional effort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual designation of lesion regions is performed on medical images, then learning data can be created with high accuracy, but the process requires a lot of time and effort
Solution Approach 1:
The system performs preliminary image analysis to automatically detect and display candidate lesion regions on the schema diagram before the radiologist creates the interpretation report. This preliminary action provides pre-processed candidate positions that guide the radiologist's attention, reducing the time needed for manual region designation while maintaining accuracy.
Solution Approach 2:
The schema diagram serves as an intermediary between the complex medical image and the final learning data. Instead of directly manipulating regions on the detailed medical image, the radiologist interacts with simplified candidate position markers on the schema diagram, which then map back to the original image for accurate learning data creation.
2Manufacturing precision
If manual region designation is required for learning data creation, then data quality can be ensured, but productivity decreases
Solution Approach 1:
The image analysis process automatically performs preliminary detection of candidate lesion regions and displays them on the schema diagram before the radiologist begins work. This pre-processing step provides a structured set of candidates that the radiologist can quickly review and confirm, enabling faster generation of high-quality learning data at scale.
Solution Approach 2:
The system segments the learning data creation process into distinct stages: automatic candidate detection by image analysis, display of candidate positions on schema diagram, radiologist confirmation/denial operations, and final learning data generation. This segmentation allows parallel processing of multiple cases and improves overall productivity while maintaining quality through focused human review at the confirmation stage.
3Productivity
If automatic image analysis is used to detect lesion regions, then data creation speed increases, but the process lacks human verification
Solution Approach 1:
The system implements feedback by allowing radiologists to review candidate lesion regions automatically detected by image analysis and perform confirmation or denial operations. The radiologist's decisions serve as feedback that validates or corrects the automatic detection results, ensuring high reliability while maintaining the speed benefits of automated processing.
Solution Approach 2:
The schema diagram with candidate position markers acts as an intermediary that presents automatically detected regions in a simplified, easy-to-review format. This intermediary representation allows radiologists to efficiently verify multiple candidate regions without directly analyzing complex medical images, balancing automation speed with human verification reliability.
Data Source
AI summary
Provided is a technique that generates learning data required for learning without performing a complicated operation.Candidate positions of a plurality of lesion candidate region images obtained by performing an image analysis process for a medical image are displayed on schematic diagrams of a human body. Lesion candidate region images other than a lesion candidate region image corresponding to a denied candidate position where a denial operation has been received are registered as correct answer data or the lesion candidate region images corresponding to confirmed candidate positions where a confirmation operation has been received are registered as the correct answer data.