Anatomical Labeling via Centerline-Guided Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for automatically labeling anatomical structures in medical images face challenges due to large individual variability, as they are not end-to-end and require pre-defined features, limiting accuracy and robustness.
Innovation Solution
A deep learning network comprising a branched network, Graph Neural Network, Recurrent Neural Network, and Probability Graph Model connected in series, which predicts anatomical labels based on the centerline of anatomical structures, incorporating geometric and image features through multiple branch networks and joint embedding, enabling robust and accurate labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If learning-based methods are used to estimate labels, then the model can process anatomical structures, but the model is not end-to-end and requires pre-defined features, limiting accuracy and robustness
Solution Approach 1:
The model is divided into multiple independent branch networks (first branch network, second branch network, etc.), each responsible for extracting specific features from different parts of the anatomical structure. This segmentation allows each branch to specialize in particular anatomical regions or feature types, improving overall reliability while maintaining manageable complexity through modular design
Solution Approach 2:
Multiple branch networks are merged into a unified deep learning model that processes anatomical structures end-to-end. The outputs from all branch networks are combined and fed into the Graph Neural Network, creating an integrated system that leverages the strengths of each branch while achieving comprehensive feature extraction without requiring pre-defined features
2Adaptability or versatility
If previous methods are used, then labeling can be performed, but the model does not model for the correlation of each label, limiting performance in dealing with large individual variability
Solution Approach 1:
The model transitions from processing anatomical structures in isolation to processing them as nodes in a graph structure. The Graph Neural Network operates on the graph representation of the anatomical tree, adding a relational dimension that captures correlations between different anatomical labels and branches, enabling the model to handle individual variability while maintaining labeling accuracy
Solution Approach 2:
The Probability Graph Model incorporates feedback mechanisms that allow the model to iteratively refine label predictions based on correlations between different anatomical labels. This feedback loop enables the model to adjust predictions in light of relationships between labels, improving accuracy in cases of large individual variability
Data Source
AI summary
The present disclosure relates to a method and a system for generating anatomical labels of an anatomical structure. The method includes receiving an anatomical structure with an extracted centerline, or a medical image containing the anatomical structure with the extracted centerline; and predicting the anatomical labels of the anatomical structure based on the centerline of the anatomical structure, by utilizing a trained deep learning network. The deep learning network includes a branched network, a Graph Neural Network, a Recurrent Neural Network and a Probability Graph Model, which are connected sequentially in series. The branched network includes at least two branch networks in parallel. The method in the disclosure can automatically generate the anatomical labels of the whole anatomical structure in medical image end to end and provide high prediction accuracy and reliability.