Multi-task Learning for Coronary Artery Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning approaches for coronary artery assessment in angiography images suffer from inconsistency and error propagation across tasks, lacking interpretability and explainability due to separate model training for each task.
Innovation Solution
A multi-task learning framework using a single end-to-end machine learning network that extracts shared features from temporal sequences of medical images, performing multiple vessel assessment tasks such as stenosis localization, grading, and segment labeling, with confidence measures determined using Gaussian processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate machine learning models are trained individually for each task, then each task can be performed independently, but inconsistency between results of different tasks and error propagation occur
Solution Approach 1:
The patent combines multiple separate machine learning models into a single unified model that performs multiple vessel assessment tasks simultaneously. This unified model processes angiography images through shared feature extraction networks and task-specific networks, ensuring consistent results across stenosis detection, grading, and segment classification while eliminating error propagation between independent models.
Solution Approach 2:
The unified machine learning model is designed to perform multiple functions: it detects stenosis, grades stenosis severity, and classifies vessel segments all within a single model architecture. This multi-functional approach ensures that all tasks benefit from consistent feature representations and shared learned patterns, resolving the inconsistency issue while maintaining comprehensive assessment capabilities.
2Measurement precision
If individual machine learning models are used for each task, then task-specific performance can be optimized, but interpretability and explainability across models are lost
Solution Approach 1:
The unified model is segmented into distinct functional components: shared feature extraction networks that process input images, task-specific networks for each assessment task, and a loss function module. This segmentation allows each component to be understood and interpreted individually while maintaining their coordinated function within the unified framework, preserving interpretability across tasks.
Solution Approach 2:
The model employs a comprehensive loss function that provides feedback signals to guide training across all tasks simultaneously. This feedback mechanism ensures that task-specific accuracy is optimized while maintaining consistency with other tasks, and the structured loss components enable interpretation of how each task contributes to overall model performance.
3Adaptability or versatility
If multiple separate models are deployed, then comprehensive vessel assessment can be achieved, but system complexity and computational overhead increase
Solution Approach 1:
The patent merges multiple assessment capabilities into a single unified model architecture that processes angiography images through shared feature extraction networks and task-specific networks simultaneously. This consolidation achieves comprehensive vessel assessment including stenosis detection, grading, and segment classification while reducing system complexity compared to deploying multiple separate models.
4Manufacturing precision
If separate models are trained independently, then training data can be specialized for each task, but error propagation occurs across tasks
Solution Approach 1:
The unified model combines task-specific training optimization with error prevention by processing all tasks through shared feature representations. The model can be trained with task-specific loss functions while ensuring that errors in one task do not propagate to others, as all tasks benefit from consistent and correlated feature extractions from the shared networks.
Data Source
AI summary
Systems and methods for automatic assessment of a vessel are provided. A temporal sequence of medical images of a vessel of a patient is received. A plurality of sets of output embeddings is generated using a machine learning based model trained using multi-task learning. The plurality of sets of output embeddings is generated based on shared features extracted from the temporal sequence of medical images. A plurality of vessel assessment tasks is performed by modelling each of the plurality of sets of output embeddings in a respective probabilistic distribution. Results of the plurality of vessel assessment tasks are output.