3D Image Segmentation With Surface Point Cloud Shape Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image segmentation methods in medical imaging, such as CT and MRI, struggle to accurately determine the portion of images corresponding to objects of interest, leading to potential misdiagnosis and sub-optimal clinical outcomes due to limited ability to encode high-level shape priors.
Innovation Solution
A dual-branch deep learning architecture that concurrently produces segmentation masks and key points on organ surfaces, utilizing a coarse-to-fine feature projection and local smooth regularization, to constrain organ shape learning and improve segmentation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image segmentation methods are used, then the segmentation process is simple, but the segmentation accuracy is insufficient leading to potential misdiagnosis
Solution Approach 1:
The patent segments the segmentation task into two independent branches: a first branch for generating segmentation masks and a second branch for generating surface point clouds. This segmentation allows each branch to specialize in its specific function, with the point cloud branch providing shape prior constraints that improve the overall segmentation accuracy without complicating the mask generation process
Solution Approach 2:
The patent introduces surface point clouds as an intermediary representation that bridges the segmentation mask generation and shape prior knowledge. The point clouds serve as a mediator that encodes high-level shape information and provides constraints to the segmentation process, enabling accurate segmentation without requiring complex direct shape modeling
2Reliability
If conventional segmentation approaches are used, then the processing is fast, but the ability to encode high-level shape priors is limited
Solution Approach 1:
By dividing the processing into separate branches, the patent allows the point cloud generation branch to efficiently encode shape priors without slowing down the overall segmentation process. The segmentation mask branch can operate independently with the benefit of shape constraints provided by the point cloud branch
Solution Approach 2:
The patent creates a representation copy of the organ surface in the form of surface point clouds, which captures high-level shape information. This point cloud copy serves as a reusable shape prior that can constrain the segmentation process without requiring re-computation of complex shape models
Data Source
AI summary
Automatic volumetric quantification can be performed for various parameters of an object by providing volumetric data, such as three-dimensional image data, to at least one neural network. A network can extract features from the data that can be used to infer a point cloud representative of the surface of the object. One or more loss functions can be used to adjust the relevant network parameters. The network can also attempt to infer a segmentation mask for the object, indicating which data values correspond to the object of interest. Since the network performs the segmentation and point cloud generation in parallel, updates to the network parameters can impact the segmentation process, effectively constraining the segmentation based on the inferred shape of the object. Ensuring that the segmentation mask corresponds closely to the surface of the object can cause the segmentation process to be more accurate than conventional segmentation processes alone.