ABSTRACT
The coronavirus infection (COVID-19) is indeed one of the pandemic diseases that caused millions of casualties and also caused infection in millions of individuals globally. The fast diagnostic testing tool has become a necessity in the medical field. The understanding and identifying of chest radiograph and/or computed tomography exams are crucial for COVID-19 diagnosis. Research has been reported on COVID-19 case identification, which faces several limitations, including data asymmetry, limited generalizability, lack of comparison studies, and so on. To address these problems, this study propounds custom deep convolutional neural network architecture for classifying COVID-19 CXR images. For instance, a convolutional neural net remediation approach is aimed at tackling the gradient vanishing issue and enhancing classification performance by dynamically integrating information in various layers of CNN using CXR scans. We propose a convolutional neural network classifier in order to detect COVID-19, pneumonia, lung capacity, and normal from a chest radiograph examination. The method makes use of conglomeration of state-of-the-art CNNs by applying transfer learning. To solve the problem of the data imbalance, we used oversampling. The proposed methods with the MobileNet V3 implementation achieved 95.91% of the test accuracy of four class predictions and 96% recall. According to the obtained results, the model also achieves 97% precision and a 98% F1 score in coronavirus detection task; the other implementations also show good results, with an F1 score of 93–96%. These findings reveal that the suggested techniques exceed the comparison models in classification accuracy, recall, precision, and F1 score, which illustrates their promise in computer-aided diagnosis and smart healthcare.
