Motion Estimation by Deep Learning in 2D Echocardiography: Synthetic Dataset and Validation.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate motion estimation in echocardiography is crucial for assessing cardiac function and calculating myocardial deformation.
• Current clinical methods face limitations in measurement accuracy and robustness, hindering precise cardiac diagnostics.

Purpose of the Study:

• To develop and validate a novel deep learning solution for robust and accurate motion estimation in echocardiography.
• To address the limitations of existing techniques in clinical practice for cardiac function analysis.

Main Methods:

• A modified PWC-Net deep learning architecture was employed for high-performance motion estimation on ultrasound sequences.
• A novel simulation pipeline generated realistic B-mode ultrasound sequences for training and validation.
• The deep learning model was trained and optimized using synthetic data and advanced training/inference strategies.

Main Results:

• The deep learning method achieved an average endpoint error of 0.07 ± 0.06 mm/frame on simulated data.
• On a clinical dataset, the method yielded a mean absolute error of 2.5 ± 2.1% for global longitudinal strain (GLS) with a correlation of 0.77 against manual segmentation.
• Performance remained robust across different echocardiographic systems and patient cohorts, demonstrating generalizability.

Conclusions:

• The proposed deep learning solution offers a significant advancement in echocardiographic motion estimation.
• The method demonstrates superior accuracy and robustness compared to state-of-the-art techniques like FFT-Xcorr.
• This approach holds promise for improving the clinical assessment of cardiac function and myocardial deformation.