Unsupervised deep learning-based displacement estimation for vascular elasticity imaging applications.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Cardiovascular Research

Background:

• Arterial wall stiffness is a key indicator of cardiovascular health.
• Ultrasound elasticity imaging offers a non-invasive method for assessing arterial stiffness.
• Accurate arterial wall displacement estimation is crucial for these imaging techniques.

Purpose of the Study:

• To develop an unsupervised deep learning approach for improved arterial wall displacement estimation.
• To enhance the quality of arterial stiffness and pulse wave velocity (PWV) mapping.
• To validate the method using phantom experiments and in vivo human carotid arteries.

Main Methods:

• An unsupervised deep learning model, adapted from image registration techniques, was employed.
• Models were trained using ultrasound RF signals and B-mode images with different loss functions, including mean square error (MSE).
• Performance was evaluated based on signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and pulse wave velocity (PWV) accuracy.

Main Results:

• Training on B-mode images with MSE yielded the highest SNR (30.36 ± 1.14 dB).
• Training on RF signals with MSE resulted in the highest CNR (32.84 ± 1.89 dB).
• The RF-trained model accurately mapped PWV with a mean relative error (MRE_PWV) of 3.32 ± 1.80% in phantoms and 3.86 ± 2.69% in vivo, with R² values of 0.97 and 0.95, respectively.

Conclusions:

• The proposed deep learning approach significantly improves arterial wall displacement estimation.
• The method demonstrates high accuracy in both phantom studies and in vivo human common carotid arteries.
• This technique holds promise for advancing vascular elasticity imaging and cardiovascular assessment.