Joint Learning Model for Coronary Artery Disease Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for diagnosing coronary artery disease (CAD) rely on separate models for anatomical abnormality evaluation and physiological parameter estimation, leading to inconsistent predictions and suboptimal use of annotations, as they do not account for the correlation between these two tasks.
Innovation Solution
A joint learning model is applied to medical images to simultaneously detect anatomical abnormalities and estimate functional physiological parameters, using a multi-task learning framework that incorporates a predetermined constraint relationship between the two tasks, thereby improving prediction consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate models are used for anatomical abnormality evaluation and physiological parameter estimation, then each model can be trained independently on dedicated annotations, but the predictions from the two models may be inconsistent and the training procedure does not make maximal use of annotations
Solution Approach 1:
The patent merges two separate models (anatomical abnormality evaluation model and physiological parameter estimation model) into a single joint learning model. This unified model simultaneously performs both tasks, ensuring prediction consistency by eliminating the independence assumption between tasks. The joint model processes both anatomical and physiological annotation types through a shared architecture, resolving the contradiction between prediction consistency and model complexity.
Solution Approach 2:
The joint learning model is designed as a multi-functional system that can perform both anatomical abnormality evaluation and physiological parameter estimation. By making the model universal, it can handle multiple task types (classification for abnormalities, regression for physiological parameters) within a single framework, thereby improving prediction consistency while avoiding the need for separate specialized models.
2Productivity
If separate models are trained on separate annotation sets, then each model can be optimized for its specific task, but the training procedure does not make maximal use of the annotations available
Solution Approach 1:
The patent combines multiple annotation types (anatomical abnormality annotations and physiological parameter annotations) into a single unified training process. The joint learning model processes both annotation sets simultaneously, making maximal use of all available annotations to improve both anatomical evaluation and physiological parameter estimation. This approach increases annotation utilization efficiency while maintaining or improving measurement precision through the synergistic effect of multi-task learning.
3Ease of manufacture
If independent models are used for anatomical and physiological tasks, then model training can be simplified and parallelized, but the correlation between the two tasks is not captured
Solution Approach 1:
The patent integrates anatomical and physiological task training into a single unified model, capturing the correlations between tasks that independent models cannot detect. The joint learning model learns shared representations and relationships between anatomical structures and physiological parameters, improving prediction consistency while maintaining training feasibility through modern deep learning frameworks that support multi-task optimization.
Data Source
AI summary
Embodiments of the disclosure provide methods and systems for joint abnormality detection and physiological condition estimation from a medical image. The exemplary method may include receiving, by at least one processor, the medical image acquired by an image acquisition device. The medical image includes an anatomical structure. The method may further include applying, by the at least one processor, a joint learning model to determine an abnormality condition and a physiological parameter of the anatomical structure jointly based on the medical image. The joint learning model satisfies a predetermined constraint relationship between the abnormality condition and the physiological parameter.