Siamese pyramidal deep learning network for strain estimation in 3D cardiac cine-MR.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Myocardial strain quantifies regional tissue deformation and is a sensitive indicator of cardiac function.
• Estimating myocardial strain requires precise tracking of cardiac motion throughout the cardiac cycle.
• Conventional indices like left ventricle ejection fraction (LVEF) may miss subtle myocardial abnormalities.

Purpose of the Study:

• To develop and validate a novel deep learning-based pipeline for automated and accurate myocardial strain estimation.
• To precisely quantify local and global myocardial strain using 3D cine-MR images.
• To compare the pipeline's performance against commercial software regarding reliability and variability.

Main Methods:

• A deep learning pipeline utilizing a supervised Convolutional Neural Network (CNN) for cardiac muscle segmentation.
• An unsupervised CNN for robust left ventricle motion tracking to enable strain estimation.
• Validation using artificial phantoms and real cine-MR images, with comparisons to commercial software.

Main Results:

• The proposed pipeline demonstrated reliable myocardial strain quantification.
• It showed reduced divergence levels compared to two commercial software packages.
• The approach is independent of user data, eliminating potential user bias.

Conclusions:

• The novel deep learning pipeline provides accurate and reliable myocardial strain estimation from 3D cine-MR images.
• This method enhances cardiac function assessment by offering a sensitive and unbiased alternative.
• The automated approach simplifies strain analysis and improves consistency.