Deep and Portable Ultrasound Imaging

In this talk we present a method to obtain high quality ultrasound images using a compact portable device combined with deep learning. We begin by showing that an ultrasound image can be formed while using two orders of magnitude less data than what is used today. This paves the way to designing a compact portable probe which enables access directly to the raw channel data.

Due to the low rate, the channel data may be transmitted to the cloud, or any other computing device, for further processing. This enables deep learning methods operating directly on channel data, giving rise to improved imaging and diagnostics while using far less resources.

By applying deep adaptive beamforming directly on the channel data we show improved imaging quality with respect to standard methods.

The talk is based on joint work with our lab at the Weizmann Institute and with Eindhoven University.