Related Experiment Video
Updated: Sep 29, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
An Improved 3D Deep Learning-Based Segmentation of Left Ventricular Myocardial Diseases from Delayed-Enhancement MRI
Khawla Brahim1,2,3, Tewodros Weldebirhan Arega1, Arnaud Boucher1
1ImViA EA 7535 Laboratory, University of Burgundy, 21078 Dijon, France.
Sensors (Basel, Switzerland)
|March 26, 2022
Summary
Accurate segmentation of myocardial scar tissue using the novel ICPIU-Net improves predictions for ventricular arrhythmias in cardiovascular disease patients. This deep learning approach enhances segmentation of left ventricle tissues from LGE-MR images.
Area of Science:
- Cardiovascular Imaging
- Medical Image Analysis
- Artificial Intelligence in Medicine
Background:
- Accurate segmentation of myocardial scar is crucial for predicting and managing ventricular arrhythmias in cardiovascular disease.
- Late gadolinium enhancement magnetic resonance (LGE-MR) imaging is a key modality for visualizing myocardial scar.
- Current segmentation methods may struggle with complex pathological tissue delineation.
Purpose of the Study:
- To propose and evaluate the Inclusion and Classification of Prior Information U-Net (ICPIU-Net) for segmenting left ventricle (LV) myocardium, myocardial infarction (MI), and microvascular obstruction (MVO) tissues.
- To improve the accuracy and topological consistency of scar segmentation in LGE-MR images.
- To compare the performance of ICPIU-Net against other deep learning methods in a challenge setting.
Main Methods:
- Development of a cascaded two-subnet architecture for initial LV cavity and myocardium segmentation.
- Integration of inclusion and classification constraint networks to refine segmentation of diseased regions.
- Utilizing majority voting for fusing outputs from multiple segmentation networks during testing.
- Validation against expert manual contouring on 50 LGE-MR images.
Main Results:
- The ICPIU-Net achieved accurate segmentation of LV myocardium, MI, and MVO tissues.
- The method demonstrated improved segmentation accuracy and topological consistency for pathological areas.
- ICPIU-Net showed superior agreement with expert manual contouring compared to other deep learning methods in the EMIDEC challenge.
Conclusions:
- ICPIU-Net offers an effective deep learning framework for precise myocardial scar segmentation in LGE-MR images.
- The proposed architecture successfully incorporates prior information to maintain topological constraints, enhancing segmentation reliability.
- This advancement holds significant potential for improving risk stratification and treatment strategies for patients with cardiovascular disease and arrhythmias.
