Myocardial strain analysis of echocardiography based on deep learning

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Myocardial strain analysis offers detailed spatiotemporal insights into heart contraction, crucial for early cardiac insufficiency detection.
• Deep learning (DL) shows promise for automated myocardial strain measurement from echocardiograms, but segmentation and motion estimation remain challenging.
• This work addresses these challenges by developing a novel DL framework for myocardial segmentation and motion estimation.

Purpose of the Study:

• To develop and validate a novel deep learning (DL) framework for automated myocardial segmentation and motion estimation from echocardiogram videos.
• To generate accurate myocardial strain measures using the developed DL framework.
• To compare the performance of the DL method against traditional speckle tracking algorithms.

Main Methods:

• A 3D Convolutional Neural Network (CNN) was employed for myocardial segmentation to define the region of interest (ROI).
• An optical flow network was utilized for motion estimation within the ROI, with an architecture search optimizing performance.
• The DL framework was validated against a traditional speckle tracking echocardiography (STE) method using independent clinical data.

Main Results:

• The DL method achieved successful automatic segmentation, motion estimation, and global longitudinal strain (GLS) measurements.
• The 3D segmentation demonstrated superior spatio-temporal smoothness (Dice correlation of 0.82) compared to 2D networks.
• The DL approach for GLS measurement showed no significant difference from the traditional method (Spearman correlation 0.90, mean bias -1.2 ± 1.5%).

Conclusions:

• The developed DL framework offers enhanced segmentation and motion estimation performance for myocardial strain analysis.
• The study confirms the feasibility of DL for automated strain analysis, reducing time and human effort.
• This approach holds significant potential for translational research and precision medicine in cardiology.