Deep Learning for Ultrasound Image Formation: CUBDL Evaluation Framework and Open Datasets

Abstract

Deep learning for ultrasound image formation is rapidly garnering research support and attention, quickly rising as the latest frontier in ultrasound image formation, with much promise to balance both image quality and display speed. Despite this promise, one challenge with identifying optimal solutions is the absence of unified evaluation methods and datasets that are not specific to a single research group. This paper introduces the largest known international database of ultrasound channel data and describes associated evaluation methods that were initially developed for the Challenge on Ultrasound Beamforming with Deep Learning (CUBDL), which was offered as a component of the 2020 IEEE International Ultrasonics Symposium. We summarize the challenge results and present qualitative and quantitative assessments using both the initially closed CUBDL evaluation test dataset (which was crowd-sourced from multiple groups around the world) and additional in vivo breast ultrasound data contributed after the challenge was completed. As an example quantitative assessment, single plane wave images from the CUBDL Task 1 dataset produced a mean generalized contrast-to-noise ratio (gCNR) of 0.67 and a mean lateral resolution of 0.42 mm when formed with delay-andsum beamforming, compared to a mean gCNR as high as 0.81 and a mean lateral resolution as low as 0.32 mm *CUBDL Organizers **Post-CUBDL Evaluator and Post-CUBDL Data Curator M.A.L.B. and A.W. acknowledge the National Institutes of Health (NIH) Trailblazer Award R21 EB025621 for partial support of their time on this project. H.R. and S.G. acknowledge the support of the Natural Sciences and Engineering Research Council of Canada (NSERC) RGPIN-2020-04612.

Lung Sonography in Critical Care Medicine

“…Deep learning is rapidly progressing in medical imaging. It is used for technical optimization of ultrasound image formation [ 145 ] and for automatic image analysis in machine and deep learning methods. In case of the COVID-19 infection, machine learning has been successfully used to assist clinicians in detecting COVID-19-associated imaging patterns on point-of-care lung sonography, with the possibility of simultaneous disease severity score prediction [ 146 ].…”

Plane-Wave Fourier-Domain Beamforming with CNN-Assisted Resolution Enhancement

“…Deep learning in recent years has shown a lot of promise in ultrasound imaging research [8,25,60,62,[64][65][66]. As a possible extension of this work, future efforts can be focused on fine-tuning ESPCN architecture [37] and weights by experimenting with different layers (depth separable convolutions [61], for example), network size, loss functions, and other hyperparameters to further improve its performance in the ultrasound imaging domain.…”

“…Plane-wave ultrasound imaging enables data acquisition at very high frame rates (due to unfocused beam transmissions), but it may negatively affect the resulting image quality [14]. The latter can be improved by using image-enhancing coherent compounding of multiple anglespecific beamformed datasets [13,17,18,19] or advanced beamforming algorithms to a singleangle raw dataset to achieve noticeable improvements in image resolution and contrast (e.g., see [20]- [25]). Another obvious alternative is to apply an image-enhancing convolutional neural network (CNN) to a single-angle beamformed dataset, which is explored here.…”

Optimally-weighted non-linear beamformer for conventional focused beam ultrasound imaging systems

“…In-vivo data. The developed methods were tested on an example in vivo data from the "Challenge on Ultrasound Beamforming with Deep Learning (CUBDL)" data set available online [32][33][34] . The data was acquired by taking informed consent form from the volunteers after the approval of the Johns Hopkins Medicine Institutional Review Board (IRB00127110).…”