Comparison of Hybrid Models with Multi-Feature Fusion Using Covid-19 Radiography Database

COVID-19, which emerged in 2019 and was subsequently classified as a pandemic, has affected millions of individuals worldwide. Different variations of the illness continue to persist, even though it may seem to have subsided at the moment. Hence, it remains essential to promptly and precisely diagnose COVID-19. Chest imaging has been proven to clearly demonstrate COVID-19 infection even in the early stages of the disease, assisting physicians and radiologists in making quicker and more accurate judgements. This study proposes a hybrid model with feature fusion based on Convolutional Neural Network based models and classifiers to accurately distinguish infected patients from healthy people. The extracted features from two different Convolutional Neural Network based models are concatenated, or added before feature selection. On a publicly accessible radiography database containing 21168 images of the four classes (Covid, Lung_Opacity, Normal, and Viral Pneumonia), extensive tests utilizing five fold cross-validation have been conducted. According to the tests, an accuracy rate of about 96% has been obtained. The findings also demonstrate that the proposed approach can contribute significantly to the rapidly expanding workload in health-care systems.