E-Scopics Ultrasound SaaS: App-Based Computerized Ultrasonography for Ultraportable Probes
DescriptionAt E-Scopics, we aim at democratizing ultrasound imaging with a simple concept: providing a minimal ultrasound probe and a host on which nearly all processings are performed. The challenge of hyper-portability, the availability of massive amounts of raw data and the unlimited computing power enabled by our software-based architecture has led us to deeply investigate inverse problems. In this presentation, I will discuss several applications ranging from AI-based B-Mode imaging to quantitative imaging and describe the path from research ideas to a certified product. AI-based B-Mode imaging is performed in collaboration with LTS5 at EPFL where convolutional neural networks (CNN) are used to denoise the back projection solution obtained with few insonifications. By clever choices of network architectures and rigorous training routines, we have managed to shrink the heaviness of the CNNs in order to make them operate in real time. Inverse problems in quantitative imaging are used to infer quantities related to tissue and tissue-wave interactions from ultrasound raw-data. As an example, speed of sound estimation exploits remarkable properties of ultrasound speckle when analyzed using a fully angular point of view. These properties allow for a direct and real-time estimation of the local speed of sound with few insonifications.
TimeWednesday, June 2812:00 - 12:30 CEST
Computer Science, Machine Learning, and Applied Mathematics