ABSTRACT
Lung ultrasound with an artificial intelligence (AI) application provides a low-cost, non-invasive diagnostic that can play a supporting role in diagnosing COVID-19, especially in areas without PCR/CT access. [1][2] Especially throughout the COVID-19 pandemic fast, safe and highly sensitive diagnostic tools are crucial. [3] The goal of this work was twofold: 1. create a publicly available dataset of lung ultrasound images/videos and 2. train an AI algorithm to detect and classify COVID-19 on lung ultrasound images and videos. The largest publicly available COVID-19 lung ultrasound dataset was created from a variety of sources, with > 200 videos and > 50 images. The dataset is heterogeneous, mostly acquired with a convex transducer and according to BLUE protocol. Using available additional patient information, lung ultrasound images in the dataset were categorized as COVID-19, bacterial pneumonia, other viral pneumonia, and healthy. In addition, two independent reviewers evaluated the visible pathologies in the lung ultrasound images. On the dataset, an in-depth study of deep learning methods for differential diagnosis of lung pathologies was performed. In the COVID-19 ultrasound images and videos lung ultrasound signs of a nonspecific pneuomia (fragmented pleural lines, B-lines, (subpleural) consolidations, aero bronchograms and pleural effusions) were visible.The frame-based model correctly distinguished COVID-19 lung ultrasound images from healthy and bacterial pneumonia with a sensitivity of 0.90 ± 0.08 and a specificity of 0.96 ± 0.04. Our work shows promising results of AI application in the field of lung sonography using COVID-19 as an example. Currently, the AI model is in the clinical trial phase. The data set as well as the code for the CNN are publicly available: https://github.com/BorgwardtLab/covid19%5fultrasound. The provided dataset facilitates the validation of lung ultrasound based neural networks to develop fast, accessible screening methods for pulmonary diseases. [ FROM AUTHOR] Copyright of Ultrasound in Medicine & Biology is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full . (Copyright applies to all s.)