Cascaded CNN Segmentation for Cardiac MRI Biomarker Quantification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for segmenting cardiac MRI images, particularly for hypertrophic cardiomyopathy, are time-consuming and prone to inter-observer variability, lacking robust automatic segmentation techniques that can efficiently handle the high variability in heart chamber shapes and sizes characteristic of HCM patients.
Innovation Solution
The use of cascaded convolutional neural networks (CNNs) for segmenting epicardium and endocardium layers in both left and right ventricles from cine and T1 image data sets, enabling efficient extraction of biomarker data and assessment of hypertrophic cardiomyopathy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual segmentation is performed by experienced cardiologists, then segmentation accuracy is maintained, but the process is very time-consuming and suffers from inter-observer variability
Solution Approach 1:
The patent replaces the manual mechanical segmentation process performed by cardiologists with an automated deep learning system using 3D U-Net convolutional neural networks. This substitution eliminates inter-observer variability and dramatically reduces segmentation time while maintaining consistent accuracy across all cases.
Solution Approach 2:
The system enables self-service automation where the segmentation process performs itself through trained neural networks that automatically process cardiac MRI images without requiring manual intervention. The model learns from training data and independently segments ventricles, myocardium, and other cardiac structures.
2Loss of information
If only end-systole and end-diastole phases are segmented manually, then key information is captured, but cardiac wall motion details are lost and the process remains time-consuming
Solution Approach 1:
The automated segmentation system processes all cardiac phases continuously rather than selecting only specific phases. This continuous processing captures cardiac wall motion dynamics throughout the entire cardiac cycle, providing comprehensive information while eliminating the time constraint of manual phase selection.
Solution Approach 2:
The system dynamically segments all cardiac phases to capture the temporal dynamics of cardiac wall motion. By processing the entire cardiac cycle rather than static phases, the system reveals motion patterns and functional information that are critical for understanding cardiac physiology and pathology.
3Ease of manufacture
If generic segmentation methods are used on HCM patients, then standard procedures are followed, but segmentation quality deteriorates due to high variability in heart chamber shapes and sizes
Solution Approach 1:
The system adapts to HCM variability by learning from diverse training data that includes various heart chamber shapes, sizes, and configurations. The neural network adjusts its parameters and features to accommodate the high variability characteristic of hypertrophic cardiomyopathy, maintaining segmentation quality across different patient phenotypes.
Solution Approach 2:
The 3D U-Net model is designed to handle multiple cardiac structures (ventricles, myocardium, atria) and various pathologies within a single unified framework. This universal approach allows the system to process both normal and HCM hearts effectively, adapting to different anatomical variations without requiring separate specialized methods.
Data Source
AI summary
In one aspect the disclosed technology relates to embodiments of a method which, includes acquiring magnetic resonance imaging data, for a plurality of images, of the heart of a subject. The method also includes segmenting, using cascaded convolutional neural networks (CNN), respective portions of the images corresponding to respective epicardium layers and endocardium layers for a left ventricle (LV) and a right ventricle (RV) of the heart. The segmenting is used for extracting biomarker data from segmented portions of the images and, in one embodiment, assessing hypertrophic cardiomyopathy from the biomarker data. The method further includes segmenting processes for T1 MRI data and LGE MRI data.