AutoCov22: A Customized Deep Learning Framework for COVID-19 Detection

Abstract

The novel coronavirus disease 2019 (COVID-19) spill has spread rapidly and appeared as a pandemic affecting global public health. Due to the severe challenges faced with the increase of suspected cases, more testing methods are explored. These methods, however, have several disadvantages, such as test complexity and associated problems—sensitivity, reproducibility, and specificity. Hence, many of them need help to achieve satisfactory performance. Motivated by these shortcomings, this work proposes a custom deep neural network framework named “AutoCov22” that detects COVID-19 by exploiting medical images—chest X-ray and CT-scan. First, multiple neural networks extract deep features from the input medical images, including popularly used VGG16, ResNet50, DenseNet121, and InceptionResNetV2. Then, the extracted features are fed to different machine-learning techniques to identify COVID-19 cases. One objective of this work is to quicken COVID-19 detection. Another goal is to reduce the number of falsely detected cases by a significant margin. Comprehensive simulation results achieve a classification accuracy of 99.74%, a precision of 99.69%, and a recall of 98.80% on exercising chest X-ray images. Extended experiment results in accuracy, precision, and recall up to 87.18%, 84.98%, and 85.66%, respectively, in processing CT-scan images. Thus, the AutoCov22 approach demonstrates a promising and plausible best solution over several methods in the state-of-the-art. © 2023, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.