A Residual Neural Network with a Novel Orthogonal Regularization for Covid-19 Diagnosis using X-ray images

Abstract

Covid-19 is a viral infection that affects the respiratory tract and causes serious health problems on a global scale. Due to the high contagiousness of the disease, early detection and accurate classification are of great importance. In this study, a novel orthogonal regularization method is proposed to improve the detection accuracy of Covid-19 disease from X-ray images. The proposed regularization method, evaluated using ResNet110, improves the classification accuracy compared to traditional Orthogonal regularization approaches. In the experimental studies, the proposed method is compared with various regularization techniques and the highest classification success rate is achieved by increasing the test accuracy rate to 96.52%. In addition, it is observed that the proposed method optimizes the learning curve of the model, especially in the later stages of the training process, and increasing the test accuracy. In addition, compared to the existing orthogonal regularization methods for Covid-19 detection, the proposed approach improved the test classification performance by approximately 1% in accuracy, F1-score, sensitivity, recall and specificity metrics.