Congratulations to Prof. Ashik and collaborators on the acceptance of their paper, AutoSOUL: Deep Convolutional Neural Network for Autonomous Selection of Ultrasound Speckle Tracking Parameters, in IEEE Transactions on Ultrasonics (TUSON).
This work introduces AutoSOUL, a deep learning-based framework that eliminates manual parameter tuning in ultrasound strain elastography. By integrating convolutional neural networks with physics-based speckle tracking optimization, the method autonomously evaluates strain image quality and identifies optimal regularization parameters.
The approach was trained on simulated datasets and fine tuned using experimental phantom data, and was validated on simulated, phantom, and in vivo breast ultrasound measurements. Results demonstrate consistent strain image quality, strong agreement with manually optimized reconstructions, and over 80× reduction in parameter selection time, significantly improving workflow efficiency and reproducibility. This publication represents an important step toward automated, robust, and clinically scalable ultrasound elastography.