Multi-task learning framework for coronary artery disease assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for automatically assessing coronary artery disease from computed tomography angiography images lack consistency across different medical imaging analysis tasks and are prone to error propagation, leading to inconsistent results.
Innovation Solution
A multi-task learning framework is employed to train a single machine learning-based model that performs multiple vessel assessment tasks using shared features extracted from input medical images, ensuring consistency and reliability across tasks such as stenosis grading, lumen segmentation, and hemodynamic index determination, without relying on segmentation results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If separate machine learning models are trained for each medical imaging analysis task, then each task can be optimized independently, but consistency between tasks and overall results deteriorates
Solution Approach 1:
The patent merges multiple separate machine learning models into a single multi-task learning model that simultaneously performs lesion detection, classification, artifact detection, stenosis grading, and lumen segmentation. This unified model architecture ensures consistency across all tasks by sharing common feature extraction layers while maintaining task-specific processing capabilities, directly resolving the contradiction between independent task optimization and inter-task consistency.
Solution Approach 2:
The machine learning model is designed with universal functionality to handle multiple vessel assessment tasks through a single integrated framework. The model extracts shared features from coronary CTA images and applies them across different assessment tasks, enabling one model to perform multiple functions consistently without requiring separate specialized models for each task.
2Adaptability or versatility
If separate machine learning models are trained for each medical imaging analysis task, then each task can be optimized independently, but error propagation to downstream tasks increases
Solution Approach 1:
By combining multiple task models into a single unified model with shared feature extraction layers, the patent prevents error propagation that occurs in sequential pipelines where errors in one task propagate to downstream tasks. The simultaneous execution of all tasks within one model ensures that errors are contained within individual task outputs rather than propagating through the entire assessment workflow.
3Reliability
If a single multi-task learning model is used for all vessel assessment tasks, then consistency and reliability improve, but model complexity increases
Solution Approach 1:
The multi-task learning model is segmented into distinct functional components: shared feature extraction layers, task-specific processing branches, and separate output modules for each assessment task. This segmentation allows the model to maintain consistency through shared architecture while managing complexity by organizing different tasks in modular, independent branches that can be trained and optimized separately.
Solution Approach 2:
Different parts of the model architecture have different qualities tailored to specific tasks. The shared feature extraction layers provide general-purpose representations, while task-specific branches provide specialized processing for each assessment type. This local quality differentiation allows the model to handle complexity efficiently by applying appropriate processing depth and complexity only where needed for each specific task.
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 are performed by the machine learning based model based on shared features extracted from the one or more input medical images. Results of the plurality of vessel assessment tasks or a combination of the results of the plurality of vessel assessment tasks are output.