Basic Science and Instrumentation Scientific Session 2
A Deep Learning Approach to Pulse-echo Sound Speed Imaging: Model Training and Generalizability
Sunday, April 7, 2024
2:28pm – 2:40pm
Location: 410
Authors: Haotian Chen, University of Illinois Urbana-Champaign Aiguo Han, Virginia Tech
Accurately imaging the spatial distribution of longitudinal sound speed may have a profound impact on the image quality and diagnostic value of medical ultrasound. Knowledge of sound speed distribution allows effective aberration correction to significantly improve ultrasound image quality. Sound speed imaging also provides a new mechanism of image contrast that will facilitate the differential diagnosis of various diseases, such as liver steatosis. However, sound speed imaging is challenging in the pulse-echo mode. Deep learning is a promising approach, but often requires a large amount of training data and has limitations in model generalizability. To address these issues, we developed a deep learning model to estimate the spatial distribution of sound speed using pulse-echo data acquired from multiple angles. Instead of using the raw echo signals as the input to the deep learning model, we transformed the echo signals into a system-independent form based on correlations between echoes acquired from various angles. Using this approach, our model was shown to be generalizable when trained and tested using different scan settings in simulation studies. Furthermore, the simulation-trained model successfully reconstructed the sound speed maps of phantoms using experimental data. This study provides a potentially practical approach to pulse-echo sound speed imaging.