Deep Learning Model Training with Segmented Medical Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning models trained on color images experience performance degradation when applied to medical images, as medical images such as X-ray, CT, and MRI are grayscale, leading to suboptimal lesion diagnosis.
Innovation Solution
A method and apparatus that prepare segmented medical images and train deep learning models by inputting both medical images and segmented images into multiple channels, optimizing the model for medical image processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing deep learning models trained on color images are applied to medical grayscale images, then the model structure can be reused, but the diagnostic performance is degraded
Solution Approach 1:
The patent applies local quality by assigning different functional roles to different channels: the first channel processes the original grayscale medical image while the second channel processes the segmented image highlighting specific tissues or lesions. This allows each channel to specialize in extracting different types of features, improving overall diagnostic accuracy without changing the fundamental multi-channel model structure.
Solution Approach 2:
The patent segments the medical image into multiple components (original image and segmented image highlighting specific tissues/lesions) and feeds them into separate channels. This segmentation of image information allows the model to simultaneously analyze both overall structure and specific regions of interest, resolving the performance degradation issue.
2Ease of manufacture
If a single grayscale medical image is input to multiple RGB channels, then the model can be trained using existing architectures, but the performance is degraded due to redundant information
Solution Approach 1:
Instead of redundantly copying the same grayscale image to multiple channels, the patent segments the image into different meaningful representations (original grayscale image and segmented image with highlighted tissues/lesions). This provides diverse, non-redundant information to each channel while maintaining training feasibility with existing multi-channel architectures.
Solution Approach 2:
The patent changes the parameter representation in different channels by transforming the same medical image into different forms: the original grayscale intensity values in the first channel and segmented intensity values highlighting specific anatomical structures or pathologies in the second channel. This parameter transformation enriches the information content without requiring new model architectures.
Data Source
AI summary
Provided are a method and apparatus for improving the performance of a deep learning model. The apparatus for improving the performance of a deep learning model prepares a medical image and at least one segmented image obtained by segmenting at least one human tissue from the medical image and trains the deep learning model by respectively inputting the medical image and the at least one segmented image to a plurality of channels.