CoviXNet: A novel and efficient deep learning model for detection of COVID-19 using chest X-Ray images. (September 2022)

• Record Type: Journal Article
• Authors: Srivastava, Gaurav; Chauhan, Aninditaa; Jangid, Mahesh; Chaurasia, Sandeep
• Is Part Of: Biomedical signal processing and control. Volume 78(2022)
• Publication Date: 2022-09
• Subjects: COVID-19 -- Chest X-Ray images -- Deep transfer learning -- Convolutional neural networks -- CoviXNet; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103848
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: The Coronavirus (COVID-19) pandemic has created havoc on humanity by causing millions of deaths and adverse physical and mental health effects. To prepare humankind for the fast and efficient detection of the virus and its variants shortly, COVID-19 detection using Artificial Intelligence and Computer-Aided Diagnosis has been the subject of several studies. To detect COVID-19, there are numerous publicly accessible datasets of Chest X-rays that the researchers have combined to solve the problem of inadequate data. The cause for concern here is that in combining two or more datasets, some of the images might be duplicates, so a curated dataset has been used in this study, taken from an author's paper. This dataset consists of 1281 COVID-19, 3270 Normal X-rays, and 1656 viral-pneumonia infected Chest X-ray images. Dataset has been pre-processed and divided carefully to ensure that there are no duplicate images. A comparative study on many traditional pre-trained models was performed, analyzing top-performing models. Fine-tuned InceptionV3, Modified EfficientNet B0&B1 produced an accuracy of 99.78% on binary classification, i.e., covid-19 infected and normal Chest X-ray image. ResNetV2 had a classification accuracy of 97.90% for 3-class classification i.e., covid-19 infected, normal, and pneumonia. Furthermore, a trailblazing custom CNN-based model, CoviXNet, has been proposed consisting of 15 layers that take efficiency into account. The proposed model CoviXNet exhibited a 10-fold accuracy of 99.47% on binary classification and 96.61% on 3-class. CoviXNet has shown phenomenal performance with exceptional accuracy and minimum computational cost. We anticipate that this comparative study, along with the proposed model CoviXNet, can assist medical centers with the efficient real-life detection of Coronavirus. Highlights: Efficient detection of COVID-19 using lightweight CNN architecture—CoviXNet. Deep feature extraction using various popular pre-trained networks. The curated dataset used has no duplicate images or data leakage. Top-performing models are investigated in a comparative analysis of pre-trained models. Proposed Model CoviXNet achieved 99.47% 10-fold accuracy on binary and 96.61% on 3-class classification.