Single-Shot CNN-Based Ultrasound Imaging with Sparse Linear Arrays

Abstract

Sparse arrays are a topic of high interest within the ultrasound (US) imaging community, because of their promising ability to reduce costs, complexity, energy consumption, and data transfer requirements of US systems, thus addressing the main challenges of 3-D and portable 2-D systems. Undersampling a transducer array usually results in a significant increase in imaging artifacts, caused primarily by higher grating lobe (GL) levels. Thus, state-of-the-art sparse arrays design strategies focus on avoiding GLs, while compromising on the resulting image resolution and uniformity. In this work, we investigated the applicability of convolutional neural network (CNN)-based image reconstruction, having recently proven its potential in reducing GL artifacts, for reconstructing images from single unfocused acquisitions using uniformly undersampled linear array configurations on receive. The proposed reconstruction method consists of first computing a low-quality estimate from the undersampled single-shot acquisitions using a delay-and-sum (DAS) algorithm, followed by applying a real-time-capable CNN, trained specifically to reduce diffraction artifacts. Experiments were conducted within a simulation environment, in the context of plane wave imaging on a numerical test phantom dedicated to US image quality assessment. The proposed approach achieved an image comparable or better to that obtained from conventional DAS beamforming using the full array with uniformly undersampled arrays up to a factor of three, demonstrating a promising potential for sparse array imaging in general.

Branched Convolutional Neural Networks for Receiver Channel Recovery in High-Frame-Rate Sparse-Array Ultrasound Imaging

“…Alternatively, channel multiplexing [8,9] can be used to receive a full set of RF channel data over multiple transmissions. As well, sparse arrays with optimized layouts [10], specialized image formation algorithms [11][12][13][14], and postbeamforming artifact reduction methods [15] can be used to form high-frame-rate ultrasound images with fewer receiving channels than a fully populated array. Nevertheless, these methods are inherently incompatible with imaging algorithms that operate on prebeamforming RF channel data.…”

Deep learning assisted sparse array ultrasound imaging

“…The resolution and SNR of sparse array images via deep learning were comparable to traditional dense array imaging [ 20 ]. However, these existing works are only demonstrated on simulations or imaging phantoms with fixed structures rather than real tissue [ 21 , 22 ]. Other works showed in vivo/ex vivo experiments were performed but with a pitch reduction of only two-fold ( S1 Table ) [ 4 , 19 , 23 ].…”

Learning velocity model for complex media with deep convolutional neural networks

“…Several cases demonstrate the convolutional network's feasibility for different biomedical imaging problems. The paper [3] studied an application of deep learning to artifacts correction on single-shot ultrasound images obtained with sparse linear arrays. The initial results demonstrated an image quality comparable or better to that obtained from conventional beamforming.…”