Detection of COVID19 from X-ray images using multiscale Deep Convolutional Neural Network

Neha Muralidharan; Shaurya Gupta; Manas Ranjan Prusty; Rajesh Kumar Tripathy

doi:10.1016/j.asoc.2022.108610

Detection of COVID19 from X-ray images using multiscale Deep Convolutional Neural Network

Appl Soft Comput. 2022 Apr:119:108610. doi: 10.1016/j.asoc.2022.108610. Epub 2022 Feb 14.

Authors

Neha Muralidharan¹, Shaurya Gupta², Manas Ranjan Prusty^{1

3}, Rajesh Kumar Tripathy⁴

Affiliations

¹ School of Computer Science and Engineering, Vellore Institute of Technology (VIT), Chennai, India.
² School of Electronics Engineering, Vellore Institute of Technology (VIT), Chennai, India.
³ Centre for Cyber Physical Systems, Vellore Institute of Technology (VIT), Chennai, India.
⁴ Department of EEE, Birla Institute of Technology and Science (BITS) Pilani, Hyderabad, India.

Abstract

The Coronavirus disease 2019 (COVID19) pandemic has led to a dramatic loss of human life worldwide and caused a tremendous challenge to public health. Immediate detection and diagnosis of COVID19 have lifesaving importance for both patients and doctors. The availability of COVID19 tests increased significantly in many countries, thereby provisioning a limited availability of laboratory test kits Additionally, the Reverse Transcription-Polymerase Chain Reaction (RT-PCR) test for the diagnosis of COVID 19 is costly and time-consuming. X-ray imaging is widely used for the diagnosis of COVID19. The detection of COVID19 based on the manual investigation of X-ray images is a tedious process. Therefore, computer-aided diagnosis (CAD) systems are needed for the automated detection of COVID19 disease. This paper proposes a novel approach for the automated detection of COVID19 using chest X-ray images. The Fixed Boundary-based Two-Dimensional Empirical Wavelet Transform (FB2DEWT) is used to extract modes from the X-ray images. In our study, a single X-ray image is decomposed into seven modes. The evaluated modes are used as input to the multiscale deep Convolutional Neural Network (CNN) to classify X-ray images into no-finding, pneumonia, and COVID19 classes. The proposed deep learning model is evaluated using the X-ray images from two different publicly available databases, where database A consists of 1225 images and database B consists of 9000 images. The results show that the proposed approach has obtained a maximum accuracy of 96% and 100% for the multiclass and binary classification schemes using X-ray images from dataset A with 5-fold cross-validation (CV) strategy. For dataset B, the accuracy values of 97.17% and 96.06% are achieved using multiscale deep CNN for multiclass and binary classification schemes with 5-fold CV. The proposed multiscale deep learning model has demonstrated a higher classification performance than the existing approaches for detecting COVID19 using X-ray images.

Keywords: COVID19; FB2DEWT; Multiscale deep CNN; No-findings; Pneumonia; X-ray.