Automated Diagnosis of COVID-19 Using Deep Learning and Data Augmentation on Chest CT

  1. Runwen Hu
  2. Guanqi Ruan
  3. Shijun Xiang
  4. Minghui Huang
  5. Qiaoyi Liang
  6. Jingxuan Li

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Background

Coronavirus disease 2019 (COVID-19) has surprised the world since the beginning of 2020, and the rapid growth of COVID-19 is beyond the capability of doctors and hospitals that could deal in many areas. The chest computed tomography (CT) could be served as an effective tool in detection of COVID-19. It is valuable to develop automatic detection of COVID-19.

Materials and Methods

The collected dataset consisted of 1042 chest CT images (including 521 COVID-19, 397 healthy, 76 bacterial pneumonia and 48 SARS) obtained by exhaustively searching available data on the Internet. Then, these data are divided into three sets, referred to training set, validation set and testing set. Sixteen data augmentation operations are designed to enrich the training set in deep learning training phase. Multiple experiments were conducted to analyze the performance of the model in the detection of COVID-19 both in case of no noisy labels and noisy labels. The performance was assessed by the area under the receiver operating characteristic (AUC), sensitivity, specificity and accuracy.

Results

The data augmentation operations on the training set are effective for improvement of the model performance. The area under the receiver operating characteristic curve is 0.9689 with (95% CI: 0.9308, 1) in case of no noisy labels for the classification of COVID-19 from heathy subject, while the per-exam sensitivity, specificity and accuracy for detecting COVID-19 in the independent testing set are 90.52%, 91.58% and 91.21%, respectively. In the classification of COVID-19 from other hybrid cases, the average AUC of the proposed model is 0.9222 with (95%CI: 0.8418, 1) if there are no noisy labels. The model is also robust when part of the training samples is marked incorrectly. The average AUC is 92.23% in the case of noisy labels of 10% in the training set.

Conclusion

A deep learning model with insufficient samples can be developed by using data augmentation in assisting medical workers in making quick and correct diagnosis of COVID-19.

Article activity feed

  1. ScreenIT

    SciScore for 10.1101/2020.04.24.20078998: (What is this?)

    Please note, not all rigor criteria are appropriate for all manuscripts.

    Table 1: Rigor

    NIH rigor criteria are not applicable to paper type.

    Table 2: Resources

    No key resources detected.

    Results from OddPub: Thank you for sharing your code and data.

    Results from LimitationRecognizer: An explicit section about the limitations of the techniques employed in this study was not found. We encourage authors to address study limitations.

    Results from TrialIdentifier: No clinical trial numbers were referenced.

    Results from Barzooka: We did not find any issues relating to the usage of bar graphs.

    Results from JetFighter: We did not find any issues relating to colormaps.

    Results from rtransparent:
    • Thank you for including a conflict of interest statement. Authors are encouraged to include this statement when submitting to a journal.
    • Thank you for including a funding statement. Authors are encouraged to include this statement when submitting to a journal.
    • No protocol registration statement was detected.

    About SciScore

    SciScore is an automated tool that is designed to assist expert reviewers by finding and presenting formulaic information scattered throughout a paper in a standard, easy to digest format. SciScore checks for the presence and correctness of RRIDs (research resource identifiers), and for rigor criteria such as sex and investigator blinding. For details on the theoretical underpinning of rigor criteria and the tools shown here, including references cited, please follow this link.

  2. Version published to 10.1101/2020.04.24.20078998v2 on medRxiv
  3. Version published to 10.1101/2020.04.24.20078998v1 on medRxiv