Medical Image Segmentation with Shape Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current depth learning segmentation algorithms in medical imaging fail to accurately predict organ shapes due to pixel-level segmentation without considering shape constraints, leading to inconsistencies with prior medical knowledge, which lowers the accuracy of image segmentation, especially in complex anatomical structures like the head and neck.
Innovation Solution
An image processing method and device that employs a primary image segmentation model for feature extraction and a target image segmentation model obtained through adversarial training with an image encoding model to place shape constraints on the segmentation results, improving the accuracy of image segmentation by aligning with prior medical knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If pixel-level segmentation based on image features is performed without considering shapes, then the segmentation process is simple and fast, but the segmentation accuracy deteriorates due to outliers and inconsistencies with prior medical knowledge
Solution Approach 1:
An image encoding model is introduced as an intermediary to encode shape information from training images and provide shape constraints to the segmentation model. This mediator enables the segmentation model to incorporate prior medical knowledge about organ shapes without directly modifying the segmentation architecture, thereby improving segmentation accuracy while maintaining relative model simplicity.
Solution Approach 2:
The solution segments the overall segmentation task into two independent components: a primary segmentation model that performs pixel-level segmentation based on image features, and a separate image encoding model that handles shape constraint encoding. This segmentation allows each component to specialize in one function, improving overall accuracy without excessively increasing complexity.
2Measurement precision
If shape constraints from prior medical knowledge are incorporated into segmentation, then segmentation accuracy improves, but the complexity of the segmentation system increases
Solution Approach 1:
Shape information is extracted and encoded from training images during the training phase before actual segmentation is performed. The image encoding model learns shape constraints in advance from labeled training data, so that during inference, the segmentation model can directly apply these pre-encoded shape constraints without real-time computation, improving shape prediction accuracy while minimizing additional computational complexity.
Solution Approach 2:
The image encoding model serves as an intermediary that transforms complex shape constraints from prior medical knowledge into encoded representations that the segmentation model can efficiently utilize. This intermediary layer simplifies the integration of shape constraints by handling the complexity of shape representation separately from the segmentation process.
3Stability of the object's composition
If conventional loss functions (cross-entropy or Dice loss) are used without shape constraints, then the training process is simple, but the segmentation results produce inconsistent shapes with prior medical knowledge
Solution Approach 1:
The solution maintains continuous and effective training by integrating shape constraint losses with conventional loss functions in a unified training framework. The shape constraint loss continuously guides the segmentation model to produce anatomically plausible shapes throughout training, ensuring shape consistency with prior medical knowledge without requiring separate training phases or interrupting the conventional training process.
Data Source
AI summary
The disclosure provides an image processing method and device, a computer-readable storage medium. The method includes: performing feature extraction on an image to be processed through a primary image segmentation model to obtain a feature map of the image to be processed; and performing image segmentation on a segmentation target according to the feature map through a target image segmentation model to obtain a target segmentation result, herein the target image segmentation model is obtained from an adversarial training with an image encoding model, and the image encoding model is used for placing a shape constraint on the target segmentation result. The disclosure may improve an accuracy of the image segmentation.