Medical Image Learning With Attention Maps and Organ Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for disease detection in medical images using machine learning are either costly due to the need for annotating disease positions or lack accuracy when using attention mechanisms without position labels.
Innovation Solution
A learning apparatus and method that utilizes an attention mechanism to generate an attention map for regions of interest in medical images, specifies these regions using organ labeling information, and adjusts learning through error calculation or constraint imposition to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If semi-supervised learning method is used to detect disease from medical image using class label and position label, then detection accuracy is improved, but annotation cost (time and labor) increases
Solution Approach 1:
The system uses the attention mechanism to automatically generate attention maps that highlight regions of interest, which are then processed by the specification unit to automatically determine lesion positions without requiring manual annotation. This self-service approach enables the system to perform semi-supervised learning using only class labels while automatically deriving position information from the attention maps, thereby eliminating the need for time-consuming manual annotation while maintaining detection accuracy
Solution Approach 2:
The patent replaces the mechanical process of manual annotation with an automated computational process. The attention mechanism generates attention maps through neural network processing, and the specification unit automatically interprets these maps to determine lesion positions. This substitution of manual mechanical annotation with automated computational processing achieves both high detection accuracy and low annotation cost
2Loss of time
If learning method using only class label is used, then annotation cost is reduced, but detection accuracy decreases
Solution Approach 1:
The system incorporates feedback from the attention mechanism to improve detection accuracy. The attention maps generated during the learning process provide feedback information about the regions of interest, which is then used by the specification unit to accurately determine lesion positions. This feedback loop enables the system to maintain high detection accuracy while using only class labels for training, avoiding the need for costly manual annotation
Solution Approach 2:
The attention map serves as an intermediary between the class label input and the lesion position output. Instead of directly mapping class labels to positions (which would require manual annotation), the attention mechanism generates intermediate attention maps that capture spatial information, and the specification unit uses these intermediaries to accurately determine lesion positions. This intermediary approach enables accurate detection without requiring manual position annotation
Data Source
AI summary
Provided are a learning apparatus, a learning method, a trained model, and a program capable of efficiently performing learning for disease detection with high accuracy while suppressing a cost. A learning apparatus (100) includes a processor (129), a memory (114) that stores a data set of a medical image and lesion information included in the medical image, and a learning model (126) with an attention mechanism (128) that estimates a disease from an input medical image. The processor performs processing of specifying a position of a region of interest indicated by an attention map (208) in organ labeling information (206), and outputting a specification result (210), processing of calculating an error by comparing an estimation result (212) with lesion information (204), processing of setting the error on the basis of the specification result (210), and processing of causing the learning model (126) to perform learning by using the set error.