Tracking and Predicting Covid-19 Radiological Trajectory Using Deep Learning on Chest X-Rays: Initial Accuracy Testing

Abstract

Background - Decision scores and ethically mindful algorithms are being established to adjudicate mechanical ventilation in the context of potential resources shortage due to the current onslaught of COVID-19 cases. There is a need for a reproducible and objective method to provide quantitative information for those scores. Purpose - Towards this goal, we present a retrospective study testing the ability of a deep learning algorithm at extracting features from chest x-rays (CXR) to track and predict radiological evolution. Materials and Methods - We trained a repurposed deep learning algorithm on the CheXnet open dataset (224,316 chest X-ray images of 65,240 unique patients) to extract features that mapped to radiological labels. We collected CXRs of COVID-19-positive patients from two open-source datasets (last accessed on April 9, 2020)(Italian Society for Medical and Interventional Radiology and MILA). Data collected form 60 pairs of sequential CXRs from 40 COVID patients (mean age +/- standard deviation: 56 +/- 13 years; 23 men, 10 women, seven not reported) and were categorized in three categories: Worse, Stable, or Improved on the basis of radiological evolution ascertained from images and reports. Receiver operating characteristic analyses, Mann-Whitney tests were performed. Results - On patients from the CheXnet dataset, the area under ROC curves ranged from 0.71 to 0.93 for seven imaging features and one diagnosis. Deep learning features between Worse and Improved outcome categories were significantly different for three radiological signs and one diagnostic (Consolidation, Lung Lesion, Pleural Effusion and Pneumonia; all P < 0.05). Features from the first CXR of each pair could correctly predict the outcome category between Worse and Improved cases with 82.7% accuracy. Conclusion - CXR deep learning features show promise for classifying the disease trajectory. Once validated in studies incorporating clinical data and with larger sample sizes, this information may be considered to inform triage decisions.

Multi Deep Learning to Diagnose COVID-19 in Lung X-Ray Images with Majority Vote Technique

“…This is also supported by several contributions that have been given by the development of AI in the medical world [12]- [14]. One of them is the problem that we are facing now, namely digital image processing for classifying X-Ray images in the lungs [15][16][17][18].…”

Applications of artificial intelligence in battling against covid-19: A literature review

“…Other examples of deep neural networks in processing CT scan images of chest include [200] , [201] , [202] , [203] , [204] , [205] , [206] , [207] , [208] , [209] , [210] , [211] , [212] , [213] , [214] , [215] , [216] , [217] , [218] , [219] , [220] , [221] , [222] , [223] , [224] , [225] , [226] , [227] , [228] , [229] , [230] , [231] , [232] , [233] , [234] , [235] .…”

Deep Learning Approaches for Detecting COVID-19 From Chest X-Ray Images: A Survey

“…In the iplementation of transfer learning, a new model requires a pre-trained network chosen from among the widely adopted networks that are trained on the ImageNet dataset as a starting point. Although most studies exploited architectures trained on ImageNet, Duchesne et al [58] and Bassi and Attux [59] applied transfer learning using ChexNet [60]. ChexNet is a 121-layer dense CNN model (DenseNet) trained on the ChestX-ray14 dataset [22], which contains 112,120 frontalview CXR images labeled with 14 different thoracic diseases, including pneumonia.…”

“…Razzak et al [34], Castiglioni et al [77], El-Din Hemdan et al [46], Hall et al [44], and De Moura et al [76] also considered using a fixed number of samples for each class. Khobahi et al [53] and Duchesne et al [58] used a classweighted entropy loss function. Kumar et al [57] used the SMOTE technique; Medhi and Hussain [82] and Han et al [84] employed cost-sensitive learning.…”

“…DL studies that aim to predict, track, and assess the progress and severity of COVID-19 patients help in efficiently triaging patients. Duchesne et al [58] and Islam and Fleischer [80] considered tracking the progress of COVID-19 patients in their studies. Duchesne et al [58] monitored and predicted patient progress using the extracted DL features, which can predict whether a patient's case would "worsen" or "improve" with an accuracy of 82.7%.…”