Single-channel, ultraportable, real-time imaging system based on deep learning: a proof-of-concept

There is a growing demand for user-friendly technologies that may empower individuals to independently monitor their physiological parameters. Ultrasound systems are highly promising for healthcare self-monitoring. To address these challenges, in this work, we present the proof of concept (PoC) of a real-time ultraportable system with a single-channel transmission and reception strategy and a deep learning-based image reconstruction method. The PoC uses the ULA-OP 256 scanner, employing a monostatic synthetic aperture focusing technique, along with a convolutional neural network (CNN) trained to generate B-mode images. Testing was carried out using a CNN running on a mid-range graphics processing unit (GPU), both on phantoms and in vivo scenarios, comparing the image quality achieved by the CNN with that of a delay-and-sum (DAS) beamformer. The results highlight that the CNN outperformed the DAS, showing a better image quality. Furthermore, the CNN achieved a real-time frame rate of 37.0 frames per second, proving that ultraportable ultrasound systems are highly promising for the future of self-monitoring instruments.