Deep learning for ultrasound data-rate reduction

Abstract

US devices generate a set of signals that are carried from a transducer probe to a computer for further processing in order to obtain images. Those signals are transmitted between both ends through a set of cables, making up a high capacity data transmission channel. In order to achieve a portable US device, it will be required to transfer the data through a much lower capacity channel. To reduce the data - rate, deep/convolutional neural networks are used for this purpose in this master thesis, showing that it is possible to reduce remarkably the data rates generated by those devices while keeping a high quality in the final reconstructed ultrasound images.