Graph Neural Network for Anatomical Tree Disease Quantification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning-based methods for disease quantification of anatomical tree structures face challenges due to the scarcity of invasive ground truth measurements, relying on inaccurate simulated values for training, which limits their performance.
Innovation Solution
The use of graph neural networks that directly train with measured invasive FFR values, propagating information from labeled to unlabeled data points, and incorporating both implicit representations and explicit relationships to learn disease models of the whole anatomical tree structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If simulated FFR values are used as ground truth for training, then training data quantity is increased, but measurement precision deteriorates
Solution Approach 1:
The patent introduces a graph neural network as an intermediary that propagates information from limited measured FFR values to unlabeled nodes. The GNN acts as a mediator that transfers accurate measurement information through the graph structure, enabling the model to learn from scarce ground truth data without relying on inaccurate simulated values.
Solution Approach 2:
The patent creates a graph-based representation copy of the anatomical tree structure where nodes represent centerline points and edges represent spatial relationships. This graph copy allows information to be propagated throughout the entire structure from limited measured points, effectively replicating the influence of accurate measurements across all unlabeled nodes.
2Quantity of substance
If invasive FFR measurements are performed at multiple locations, then ground truth data quantity is increased, but ease of operation deteriorates
Solution Approach 1:
The patent applies partial action by performing invasive measurements at only one or a few strategic locations rather than throughout the entire tree. The graph neural network compensates for this partial measurement by propagating information from the limited measured nodes to all other nodes, achieving complete tree coverage without proportional invasive intervention.
Solution Approach 2:
The system enables self-service by allowing the graph neural network to automatically propagate measurement information throughout the entire anatomical tree structure without requiring additional manual measurements. The model uses the graph topology and learned relationships to fill in unlabeled nodes autonomously.
3Productivity
If conventional machine learning methods are used with simulated data, then productivity is improved, but reliability deteriorates
Solution Approach 1:
The patent changes the fundamental parameter of training data quality from simulated to measured values. By using measured FFR values as ground truth and employing graph neural network information propagation, the system maintains training efficiency while dramatically improving prediction reliability through the use of accurate clinical measurements.
Data Source
AI summary
A method and system can be used for disease quantification modeling of an anatomical tree structure. The method may include obtaining a centerline of an anatomical tree structure and generating a graph neural network including a plurality of nodes based on a graph. Each node corresponds to a centerline point and edges are defined by the centerline, with an input of each node being a disease related feature or an image patch for the corresponding centerline point and an output of each node being a disease quantification parameter. The method also includes obtaining labeled data of one or more nodes, the number of which is less than a total number of the nodes in the graph neural network. Further, the method includes training the graph neural network by transferring information between the one or more nodes and other nodes based on the labeled data of the one or more nodes.