Anatomical Motion Estimation Using Boundary-Aware Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep learning-based motion estimation techniques for anatomical structures often violate anatomical integrity due to reliance on image content, failing to distinguish structures with similar appearances, such as the ventricle and atrium or myocardium and papillary muscle.
Innovation Solution
A machine-learned model is trained using medical images and segmentation masks to predict motion while maintaining correspondence between the target anatomical structure and adjacent structures, emphasizing boundary regions and using non-binary heat maps to adjust parameters for anatomical integrity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional deep learning-based motion estimation techniques are used that rely on image content, then the motion estimation process is simple and fast, but the anatomical integrity of structures is violated when structures have similar appearances
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple structures (e.g., ventricle, atrium, myocardium, papillary muscle) and generating separate segmentation masks for each. The motion estimation model is trained to maintain correspondence between these segmented regions, ensuring that structures with similar appearances are correctly distinguished and their anatomical integrity is preserved during motion tracking.
2Productivity
If the ML model focuses on image content matching, then the transformation is computationally efficient, but structures with similar appearance (e.g., ventricle and atrium) are confused
Solution Approach 1:
The patent applies local quality by assigning different weights to different regions during training. Specifically, boundary regions between adjacent structures are given higher weights in the loss function, forcing the model to pay special attention to these critical areas. This enables the model to maintain computational efficiency while accurately discriminating between structures with similar appearances by focusing on locally distinctive features.
3Reliability
If the ML model uses heavy weighting of boundary regions, then the anatomical correspondence is improved, but the training complexity and computation time increase
Solution Approach 1:
The patent applies preliminary action by pre-computing boundary region masks before training the motion estimation model. These boundary masks are generated based on anatomical knowledge or preliminary segmentation, and then used to create weighted loss functions during training. This preparation step simplifies the training process by providing ready-to-use weight maps, reducing the complexity of dynamically computing weights during each training iteration while still achieving improved anatomical correspondence.
Data Source
AI summary
The motion estimation of an anatomical structure may be performed using a machine-learned (ML) model trained based on medical training images of the anatomical structure and corresponding segmentation masks for the anatomical structure. During the training of the ML model, the model may be used to predict a motion field that may indicate a change between a first training image and a second training image, and to transform the first training image and a corresponding first segmentation mask based on the motion field. The parameters of the ML model may then be adjusted to maintain a correspondence between the transformed first training image and the second training image and between the transformed first segmentation mask or a second segmentation mask associated with the second training image. The correspondence may be assessed based on at least a boundary region shared by the anatomical structure and one or more other anatomical structures.