Data Augmentation and CNN Classification For Automatic COVID-19 Diagnosis From CT-Scan Images On Small Dataset (preprint)

We present an automatic COVID1-19 diagnosis framework from lung CT images. The focus is on signal processing and classification on small datasets with efforts putting into exploring data preparation and augmentation to improve the generalization capability of the 2D CNN classification models. We propose a unique and effective data augmentation method using multiple Hounsfield Unit (HU) normalization windows. In addition, the original slice image is cropped to exclude background, and a filter is applied to filter out closed-lung images. For the classification network, we choose to use 2D Densenet and Xception with the feature pyramid network (FPN). To further improve the classification accuracy, an ensemble of multiple CNN models and HU windows is used. On the training/validation dataset, we achieve a patient classification accuracy of 93.39%.

A 3D CNN Network with BERT For Automatic COVID-19 Diagnosis From CT-Scan Images (preprint)

We present an automatic COVID1-19 diagnosis framework from lung CT-scan slice images. In this framework, the slice images of a CT-scan volume are first proprocessed using segmentation techniques to filter out images of closed lung, and to remove the useless background. Then a resampling method is used to select one or multiple sets of a fixed number of slice images for training and validation. A 3D CNN network with BERT is used to classify this set of selected slice images. In this network, an embedding feature is also extracted. In cases where there are more than one set of slice images in a volume, the features of all sets are extracted and pooled into a global feature vector for the whole CT-scan volume. A simple multiple-layer perceptron (MLP) network is used to further classify the aggregated feature vector. The models are trained and evaluated on the provided training and validation datasets. On the validation dataset, the accuracy is 0.9278 and the F1 score is 0.9261.

Predictive Factors of Severe Coronavirus Disease 2019 in Previously Healthy Young Adults: A Single-Center, Retrospective Study (preprint)

Background: Several previously healthy young adults have developed Coronavirus Disease 2019 (COVID-19), and a few of them progressed to the severe stage. However, the factors are not yet determined. Method: We retrospectively analyzed 123 previously healthy young adults diagnosed with COVID-19 from January to March 2020 in a tertiary hospital in Wuhan. Patients were classified as having mild or severe COVID-19 based on their respiratory rate, SpO2, and PaO2/FiO2 levels. Patients’ symptoms, computer tomography (CT) images, preadmission drugs received, and the serum biochemical examination on admission were compared between the mild and severe groups. Significant variables were enrolled into logistic regression model to predict the factors affecting disease severity. A receiver operating characteristic (ROC) curve was applied to validate the predictive value of predictors. Result: Age; temperature; anorexia; and white blood cell count, neutrophil percentage, platelet count, lymphocyte count, C-reactive protein, aspartate transaminase, creatine kinase, albumin, and fibrinogen values were significantly different between patients with mild and severe COVID-19 (P<0.05). Logistic regression analysis confirmed that lymphopenia (P=0.010) indicated severe prognosis in previously healthy young adults with COVID-19, with the area under the curve (AUC) was 0.791(95% Confidence Interval (CI) 0.704–0.877)(P<0.001). Conclusion: For previously healthy young adults with COVID-19, lymphopenia on admission can predict severe prognosis.

Temporal radiographic changes in COVID-19 patients: relationship to disease severity and viral clearance (preprint)

Background: COVID-19 is “public enemy number one” and has placed an enormous burden on health authorities across the world. Given the wide clinical spectrum of COVID-19, understanding the factors that can predict disease severity will be essential since this will help frontline clinical staff to stratify patients with increased confidence. Purpose: To investigate the diagnostic value of the temporal radiographic changes, and the relationship to disease severity and viral clearance in COVID-19 patients. Methods: : In this retrospective cohort study, we included 99 patients admitted to the Renmin Hospital of Wuhan University, with laboratory confirmed moderate or severe COVID-19. Temporal radiographic changes and viral clearance were explored using appropriate statistical methods. Results: : Radiographic features from HRCT scans included ground-glass opacity, consolidation, air bronchogram, nodular opacities and pleural effusion. The HRCT scores (peak) during disease course in COVID-19 patients with severe pneumonia (median: 24.5) were higher compared to those with pneumonia (median: 10) (p=3.56×10 -12 ), with more frequency of consolidation (p=0.025) and air bronchogram (p=7.50×10 -6 ). The median values of days when the peak HRCT scores were reached in pneumonia or severe pneumonia patients were 12 vs . 14, respectively (p=0.048). Log-rank test and Spearman's Rank-Order correlation suggested temporal radiographic changes as a valuable predictor for viral clearance. In addition, follow up CT scans from 11 pneumonia patients showed full recovery. Conclusion: Given the values of HRCT scores for both disease severity and viral clearance, a standardised HRCT score system for COVID-19 is highly demanded.

Association of radiologic findings with mortality of patients infected with 2019 novel coronavirus in Wuhan, China (preprint)

Radiologic characteristics of 2019 novel coronavirus (2019-nCoV) infected pneumonia (NCIP) which had not been fully understood are especially important for diagnosing and predicting prognosis. We retrospective studied 27 consecutive patients who were confirmed NCIP, the clinical characteristics and CT image findings were collected, and the association of radiologic findings with mortality of patients was evaluated. 27 patients included 12 men and 15 women, with median age of 60 years (IQR 47-69). 17 patients discharged in recovered condition and 10 patients died in hospital. The median age of mortality group was higher compared to survival group (68 (IQR 63-73) vs 55 (IQR 35-60), P = 0.003). The comorbidity rate in mortality group was significantly higher than in survival group (80% vs 29%, P = 0.018). The predominant CT characteristics consisted of ground glass opacity (67%), bilateral sides involved (86%), both peripheral and central distribution (74%), and lower zone involvement (96%). The median CT score of mortality group was higher compared to survival group (30 (IQR 7-13) vs 12 (IQR 11-43), P = 0.021), with more frequency of consolidation (40% vs 6%, P = 0.047) and air bronchogram (60% vs 12%, P = 0.025). An optimal cutoff value of a CT score of 24.5 had a sensitivity of 85.6% and a specificity of 84.5% for the prediction of mortality. 2019-nCoV was more likely to infect elderly people with chronic comorbidities. CT findings of NCIP were featured by predominant ground glass opacities mixed with consolidations, mainly peripheral or combined peripheral and central distributions, bilateral and lower lung zones being mostly involved. A simple CT scoring method was capable to predict mortality.