Mitral Valve Segmentation and Tracking from Transthoracic Echocardiography Using Deep Learning.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Valvular heart diseases (VHDs) present a major health challenge.
• Accurate assessment of heart valve function is crucial for treatment planning.
• Transthoracic echocardiography (TTE) is vital for VHD evaluation but often subjective and time-consuming.

Purpose of the Study:

• To automate the extraction of mitral valve (MV) leaflets and annular hinge points from echocardiograms using deep learning.
• To improve standardization and reduce the workload in quantitative assessment of MV disease.

Main Methods:

• Annotation of MV leaflets and annulus points in 2931 images from 127 patients.
• Development of an Attention UNet model with deep supervision and attention coefficient scheduling for feature segmentation.
• Extraction of quantitative biomarkers from segmentation masks for MV leaflet scallops throughout the cardiac cycle.

Main Results:

• Achieved a Dice score of 0.63 ± 0.14, annulus error of 3.64 ± 2.53 mm, and leaflet angle error of 8.7 ± 8.3°.
• Attention UNet improved robustness of clinically relevant metrics, reducing standard deviations compared to standard UNet.
• Successfully identified MV prolapse, stenosis, and healthy cases using derived biomarkers.

Conclusions:

• Deep learning segmentation and tracking of MV morphology and motion are feasible using attention gates and deep supervision.
• This approach shows promise for enhancing VHD diagnosis and treatment monitoring.
• Automated quantitative assessment can lead to more standardized and efficient VHD evaluation.