Related Experiment Video
Updated: Jun 3, 2025
09:57
Four-Dimensional Computed Tomography-Guided Valve Sizing for Transcatheter Pulmonary Valve Replacement
Published on: January 20, 2022
2.5K
Automatic 4D mitral valve segmentation from transesophageal echocardiography: a semi-supervised learning approach
Riccardo Munafò1, Simone Saitta2,3,4, Davide Tondi2
1Department of Electronics, Information and Bioengineering, Politecnico di Milano, Milan, Italy. riccardo.munafo@polimi.it.
Medical & Biological Engineering & Computing
|January 11, 2025
Summary
This study introduces a semi-supervised method for segmenting mitral valves (MV) in 4D transesophageal echocardiography (TEE) using a Teacher-Student framework. The approach enhances accuracy and efficiency in MV analysis, reducing manual annotation needs.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Imaging
Background:
- Automatic 4D transesophageal echocardiography (TEE) segmentation and mitral valve (MV) analysis face challenges due to echocardiography limitations and limited annotated data.
- Accurate MV segmentation is crucial for diagnosing and planning treatments for mitral valve disease.
Purpose of the Study:
- To develop a semi-supervised strategy for robust MV segmentation in 4D TEE using pseudo-labeling.
- To improve the accuracy, temporal consistency, and efficiency of MV analysis in 4D TEE.
Main Methods:
- A Teacher-Student framework was employed for semi-supervised mitral valve segmentation in 4D TEE.
- A Teacher model (ensemble of CNNs) generated pseudo-labels on intermediate cardiac frames.
- Pseudo-annotated data augmented the Student model's training set, enhancing segmentation performance.
Main Results:
- The Student model achieved a Dice score of 0.82, average surface distance of 0.37 mm, and 95% Hausdorff distance of 1.72 mm for MV leaflets.
- The method demonstrated reliable frame-by-frame MV segmentation, capturing leaflet morphology and dynamics.
- Significant reduction in inference time compared to the ensemble Teacher model was observed.
Conclusions:
- The proposed semi-supervised approach significantly reduces manual annotation workload for 4D TEE MV analysis.
- This method ensures reliable, repeatable, and time-efficient MV segmentation, with potential for clinical application in diagnostics and treatment planning.
