AI-assisted CT imaging analysis for COVID-19 screening: Building and deploying a medical AI system in four weeks

Abstract

These authors contributed equally to this work. ⇤ Corresponding authors.The sudden outbreak of novel coronavirus 2019 (COVID-19) increased the diagnostic burden of radiologists. In the time of an epidemic crisis, we hoped artificial intelligence (AI) to help reduce physician workload in regions with the outbreak, and improve the diagnosis accuracy for physicians before they could acquire enough experience with the new disease. Here, we present our experience in building and deploying an AI system that automatically analyzes CT images to detect COVID-19 pneumonia features. Different from conventional medical AI, we were dealing with an epidemic crisis. Working in an interdisciplinary team of over 30 people with medical and / or AI background, geographically distributed in Beijing and Wuhan, we were able to overcome a series of challenges in this particular situation and deploy the system in four weeks. Using 1,136 training cases (723 positives for COVID-19) from five hospitals, we were able to achieve a sensitivity of 0.974 and specificity of 0.922 on the test dataset, which included a variety of pulmonary diseases. Besides, the system automatically highlighted all lesion regions for faster examination. As of today, we have deployed the system 2 All rights reserved. No reuse allowed without permission.

Review of Artificial Intelligence Techniques in Imaging Data Acquisition, Segmentation, and Diagnosis for COVID-19

“…Qi et al [53] delineate the lesions in the lung using U-Net with the initial seeds given by a radiologist. Several other works used diagnostic knowledge and identified the infection regions by the attention mechanism [57]. Weakly-supervised machine learning methods are also used when the training data are insufficient for segmentation.…”

“…To segment ROIs in CT, deep learning methods are widely used. The popular segmentation networks for COVID-19 include classic U-Net [50][51][52][53][54][55], UNet++ [56,57], VB-Net [58]. Compared with CT, X-ray is more easily accessible around the world.…”

“…The lung-region-oriented methods aim to separate lung regions, i.e., whole lung and lung lobes, from other (background) regions in CT or X-ray, which is considered as a pre-requisite step in COVID-19 applications [50-54, 57, 58, 60]. For example, Jin et al [57] propose a two-stage pipeline for screening COVID-19 in CT images, in which the whole lung region is first detected by an efficient segmentation network based on UNet++. The lung-lesion-oriented methods aim to separate lesions (or metal and motion artifacts) in the lung from lung regions [51-58, 60, 61].…”

Systematic review and critical appraisal of prediction models for diagnosis and prognosis of COVID-19 infection

“…23 26 29 69-71 One model is deployed in 16 hospitals, but the authors do not provide any usable tools in their study. 33 Prognostic models. To assist in the prognosis of mortality, a nomogram (a graphic aid to calculate mortality risk), 7 a decision tree, 21 and a CT-based scoring rule are available in the articles.…”

“…Barstugan, Ozkaya, et al 31 unclear unclear unclear high Chen, Wu, et al 26 high unclear low high *1 Gozes, Frid-Adar, et al 25 unclear unclear high high Jin, Chen, et al 11 high unclear unclear high *2 Jin, Wang, et al 33 high unclear high high *1 Li, Qin, et al 34 low unclear low high Shan, Gao, et al 28 unclear unclear high high *2 Shi, Xia, et al 36 high unclear low high Wang, Kang, et al 29 high unclear low high Xu, Jiang, et al 27 high unclear high high Ying, Zheng, et al 23 unclear unclear low high Zheng, Deng, et al 38 unclear unclear high high…”