Data-Driven Speed-of-Sound Reconstruction for Medical Ultrasound Using Deep Neural Networks

B-mode imaging is widely used in medical examinations. However, it's a qualitative approach that's highly dependent on the operator’s expertise. Quantitative information of the tissue can potentially improve diagnosis — for example, elasticity and speed-of-sound information for breast imaging. Fundamentally, acoustic properties of tissues, including speed-of-sound, are available in raw data. We'll present a deep learning-based approach to extract speed-of-sound information directly from ultrasound raw data. Our method benefits greatly from GPU acceleration, both in generating training data and training the deep neural network.