DSWE-Net: A deep learning approach for shear wave elastography and lesion segmentation using single push acoustic...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Ultrasound-based elasticity imaging, like Shear Wave Elastography (SWE), is crucial for tissue characterization.
• Conventional SWE methods struggle with image quality in deeper tissues due to limitations of single focused ultrasound push beams.
• Improving non-invasive elasticity imaging is vital for accurate diagnosis and treatment planning.

Purpose of the Study:

• To develop a novel deep learning approach, DSWE-Net, for enhanced Young's modulus mapping using ultrasound velocity data.
• To improve image quality and enable simultaneous inclusion segmentation in elasticity imaging.
• To overcome limitations of conventional SWE, particularly in regions distant from the acoustic radiation force (ARF) push location.

Main Methods:

• Proposed DSWE-Net utilizes a 3D convolutional encoder and recurrent ConvLSTM layers to process spatio-temporal velocity data.
• A multi-task learning loss function was developed for end-to-end network training using simulated phantom data.
• Network performance was evaluated against the Local Phase Velocity Based Imaging (LPVI) method using synthetic and phantom datasets.

Main Results:

• DSWE-Net achieved superior imaging performance compared to LPVI, with an average SSIM of 0.90, RMSE of 0.10, and PSNR of 20.69 dB.
• The method demonstrated strong inclusion segmentation capabilities, achieving an average IoU score of 0.81.
• DSWE-Net showed significant improvements over LPVI on unseen phantom data, including a 0.09 increase in SSIM and a 0.09 decrease in RMSE.

Conclusions:

• DSWE-Net offers a robust deep learning solution for high-quality ultrasound elastography and inclusion segmentation.
• The proposed method demonstrates excellent generalization capabilities, suggesting potential for real-world clinical applications.
• DSWE-Net advances non-invasive tissue characterization and localization of abnormalities through improved elasticity imaging.