Coronary Artery Segmentation Using Multi-Task Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for assessing coronary artery disease (CAD) through coronary CTA imaging struggle with accurately determining stenoses, especially when plaque buildup is diffuse or occurs at bifurcations, and AI-based systems face limitations due to reliance on diameter or area profiles of the lumen.
Innovation Solution
A machine learning-based model trained using multi-task learning performs segmentation of both the reference wall and lumen of the vessel, incorporating regularization techniques for anatomical tapering and consistency with ground truth markers, enabling accurate stenosis grading and uncertainty estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional methods interpolate diameter or area of lumen between healthy ends or fit a line to diameter/area profile, then the assessment is simple to perform, but accuracy deteriorates when plaque buildup is diffuse and long or occurs at bifurcations
Solution Approach 1:
The method segments the coronary artery into multiple sections along its centerline, identifying healthy proximal and distal segments separately from the lesion-containing segment. This segmentation allows the reference diameter to be derived from healthy segments rather than attempting to measure through diffuse plaque or bifurcations, thereby maintaining both simplicity and accuracy.
Solution Approach 2:
The method creates a reference model of the healthy coronary artery by copying and extrapolating the diameter or area from healthy proximal and distal segments to the lesion-containing segment. This reference model serves as a baseline for comparing actual lumen measurements, enabling accurate stenosis assessment even when direct measurement is unreliable.
2Use of energy by moving object
If AI-based systems are trained on synthetically generated vessel trees using only diameter or area profile, then training is computationally efficient, but accuracy deteriorates due to limited feature representation
Solution Approach 1:
The method transitions from using only scalar diameter or area profiles to incorporating spatial coordinates along the vessel centerline as an additional dimension. By representing the coronary artery as a series of points with (x, y, z) coordinates and corresponding diameter/area values, the system captures the three-dimensional geometry and tortuosity of the vessel, providing richer features for AI training while maintaining computational efficiency.
3Measurement precision
If a single machine learning model performs multiple vessel assessment tasks, then the system becomes more complex to train, but overall assessment accuracy improves through consistency across tasks
Solution Approach 1:
The patent implements a universal machine learning model that simultaneously performs multiple vessel assessment tasks including lumen segmentation, reference wall segmentation, and stenosis grading. This multi-functional model shares common feature extraction and processing across all tasks, reducing overall system complexity while improving consistency and accuracy through joint training and shared representations.
Data Source
AI summary
Systems and methods for automated assessment of a vessel are provided. One or more input medical images of a vessel of a patient are received. A plurality of vessel assessment tasks for assessing the vessel is performed using a machine learning based model trained using multi-task learning. The plurality of vessel assessment tasks comprises segmentation of reference walls of the vessel from the one or more input medical images and segmentation of lumen of the vessel from the one or more input medical images. Results of the plurality of vessel assessment tasks are output.