Autonomous Selection of Energy-Based Ultrasound Speckle Tracking Parameters Using Deep Learning
Affiliation Type:
Academia
Keywords:
Ultrasound elastography, Energy-based tracking, Deep learning, Automatic parameter selection, User-independence
Abstract:
SOUL is a speckle-tracking algorithm for strain imaging, but its manual parameter selection limits efficiency and usability. This study automates the process using a custom CNN trained on simulated and phantom data and tested on simulated, phantom, and in vivo breast ultrasound datasets. The model classifies strain images as acceptable or unacceptable and identifies optimal SOUL parameters by selecting the highest-quality images. Automated tuning produced strain images comparable to manual selection, demonstrating the potential to improve accessibility and consistency in ultrasound elastography.
