Medical Image Segmentation via Multi-Branch Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical image segmentation technologies focus on strong expressive features, neglecting weak expressive features, resulting in incomplete information and poor segmentation results.
Innovation Solution
An image processing method using a combination of a coding network, decoding network, and generative network to code, decode, and generate predictive images based on sample medical images, training the model on differences between predictive and tag images to enhance feature extraction and segmentation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network model focuses on strong expressive features of medical images, then the model can effectively capture prominent image characteristics, but weak expressive features are neglected resulting in incomplete information and poor segmentation results
Solution Approach 1:
The patent segments the feature extraction process into multiple parallel branches: a strong feature extraction branch and a weak feature extraction branch. Each branch independently processes the input medical image to capture different types of features, preventing the dominance of strong features from overshadowing weak features during training and inference.
Solution Approach 2:
The patent merges the feature maps from the strong feature extraction branch and the weak feature extraction branch through concatenation or element-wise addition. This combination integrates both prominent and subtle image characteristics, ensuring that weak expressive features are preserved and contribute to the final segmentation result.
2Device complexity
If only a single medical image mode is used for segmentation, then the processing complexity is reduced, but the comprehensiveness of segmentation information is insufficient
Solution Approach 1:
The patent designs a unified image processing model that can process multiple medical image modes (such as T1-weighted, T2-weighted, FLAIR images) through a common architecture. The model accepts multi-mode images as input and produces comprehensive segmentation results, making the system versatile for different imaging modalities without requiring separate specialized models for each mode.
Data Source
AI summary
A method for processing a medical image performed by a computer device. The method includes: calling a first coding network in an image processing model to code a first sample image of a first mode of a target medical object, to obtain a first feature map of the first sample image; calling a decoding network to obtain, based on the first feature map, a predictive segmentation image used for indicating at least one predicted specified type region within the first sample image; calling a generative network to generate a predictive generation image of a second mode based on the first feature map; and training the image processing model based on a difference between the predictive segmentation image and a tag image of the target medical object and a difference between the predictive generation image and a second sample image of a second mode of the target medical object.