Fast and Automated Segmentation for the Three-Directional Multi-Slice Cine Myocardial Velocity Mapping.

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Deep Learning in Medicine

Background:

• Accurate myocardial delineation is essential for analyzing cardiac motion using three-directional cine multi-slice left ventricular myocardial velocity mapping (3Dir MVM).
• Standard U-Net methods may not fully leverage multi-channel CMR data (magnitude and phase velocity mapping).

Purpose of the Study:

• To develop a novel, fast, and automated deep learning framework for improved myocardial delineation in 3Dir MVM CMR data.
• To enhance the accuracy of epicardial and endocardial contour detection using multi-channel CMR data.

Main Methods:

• A novel deep learning framework incorporating cross-channel fusion with an attention module was developed.
• The framework utilizes multi-channel CMR data (magnitude and phase velocity mapping) and shape information-based post-processing.
• Standard U-Net-based methods were used as a baseline for comparison.

Main Results:

• The proposed multi-channel network demonstrated superior performance compared to single-channel U-Net networks.
• Evaluation using Dice Scores and peak myocardial velocities indicated improved accuracy in myocardial delineation.
• The framework achieved accurate delineation of both epicardial and endocardial contours.

Conclusions:

• The novel deep learning framework effectively utilizes multi-channel 3Dir MVM CMR data for accurate myocardial delineation.
• This automated approach shows significant promise for improving cardiac motion analysis and clinical applications.
• The findings support the design of advanced multi-channel image analysis techniques for CMR data.