Indian Journal of Science and Technology

Abstract

Objectives: The objective of this research is to enhance accuracy on the COVID-19 case identification using X-ray imagery by addressing the drawbacks of utilising a single deep learning model, such as overfitting, high variance, and generalisation errors, by generating predictions with numerous frameworks as opposed to one model. Methods: In this study, secondary data sets from a group of experts from Qatar University in Doha, Qatar, and the University of Dhaka in Bangladesh, together with partners from Pakistan and Malaysia, have produced a dataset of 21,135 CXR pictures for COVID-19 patients, as well as pictures of normal and viral pneumonia. The performance of proposed strategy is, EnDL-COVID-19 is compared with three parameters Accuracy, Sensitivity, PPV Assessment. Findings: ENDL-COVID-19 gives good results for COVID-19, instances identification with a performance of 95%, higher than COVID-Net at 93.3%, are according to research observations ENDL-COVID-19 outperformed by a significant margin in a series of experiments using QU&UD test data consisting of 1592 CXR images. It was able to achieve a sensitivity of 96% and a PPV of 94.1% in determining whether or not COVID-19 occurrences were present. Novelty: The proposed weighted averaging ensemble technique, which is aware of the various sensitivities of deep-learning frameworks on various category types, is used to combine multiple snapshot frameworks of COVIDNet, which made a breakthrough in an open sourced COVID-19 case identification approach using chest X-ray pictures analyzed by deep neural networks.

Keywords: COVID-19, Ensembling Learning, Deep-Learning, EnDL-COVID-19, X-Ray Pictures