Deep COVID DeteCT: an international experience on COVID-19 lung detection and prognosis using chest CT

Abstract

The Coronavirus disease 2019 (COVID-19) presents open questions in how we clinically diagnose and assess disease course. Recently, chest computed tomography (CT) has shown utility for COVID-19 diagnosis. In this study, we developed Deep COVID DeteCT (DCD), a deep learning convolutional neural network (CNN) that uses the entire chest CT volume to automatically predict COVID-19 (COVID+) from non-COVID-19 (COVID−) pneumonia and normal controls. We discuss training strategies and differences in performance across 13 international institutions and 8 countries. The inclusion of non-China sites in training significantly improved classification performance with area under the curve (AUCs) and accuracies above 0.8 on most test sites. Furthermore, using available follow-up scans, we investigate methods to track patient disease course and predict prognosis.

COVID-19 Diagnosis from chest CT Scans: A Weakly Supervised CNN-LSTM Approach

“…Most of the related studies use pre-trained classification models [9,10,14,24]; however, these architectures are not particularly designed for classification of CT Volumes. The architectures such as ResNet [25] and Inception [26], are designed for ImageNet dataset [27] where the images contain a wide variety of objects.…”

“…The healthcare providers are facing intense workload due to the pandemic [8]. To relieve the overwhelming workload, AI systems are being used to detect and identify COVID-19 using medical imaging technologies [9][10][11][12][13][14][15][16][17]. Recent studies on radiology demonstrate promising results on COVID-19 pneumonia classification using chest CTs with the help of deep learning methodologies.…”

“…Although there are works proposing deep Convolutional Neural Networks (CNNs) without performing segmentation on CT slices [9,13], performing segmentation of CT slices is usually preferred as a preprocessing step before classification networks. The reason why the segmentation before classification is commonly preferred is that lung regions in CT images can easily be extracted with segmentation networks which are efficacious in removing redundant information and eliminating noise.…”

“…Binary classification is usually applied as Covid pneumonia and non-Covid pneumonia [12,17] or Covid positive and Covid negative [14,16]. Multiclass classification separates Covid negative further into Covid pneumonia, non-Covid pneumonia and no pneumonia [9][10][11]13,15]. For radiologists diagnosing hundreds of patients with COVID pneumonia and distinguishing Covid pneumonia and non-Covid pneumonia in a short time is challenging.…”

“…While most of the COVID-19 classification studies exploit slice-based CT image labels [10,12,13,[15][16][17] because of its convenience, it has a big labour cost requiring manual labelling of slices by radiologists. Therefore, patient-level labelling has also been used [9,11,14] instead of slice-based labelling where slice-based features can be obtained with the help of deep learning methods.…”