CovidMulti-Net: A Parallel-Dilated Multi Scale Feature Fusion Architecture for the Identification of COVID-19 Cases from Chest X-ray Images

摘要: The COVID-19 pandemic is an emerging respiratory infectious disease, having a significant impact on the health and life of many people around world. Therefore, early identification patients fastest way to restrain spread pandemic. However, as number cases grows at alarming pace, most developing countries are now facing shortage medical resources testing kits. Besides, using kits detect time-consuming, expensive, cumbersome procedure. Faced with these obstacles, physicians, researchers, engineers have advocated for advancement computer-aided deep learning models assist healthcare professionals in quickly inexpensively recognize from chest X-ray (CXR) images. With this motivation, paper proposes "CovidMulti-Net" architecture based transfer concept classify normal other pneumonia three publicly available datasets that include 1341, 446 CXR images healthy samples 902, 1564, 1193 infected Viral Pneumonia, Bacterial diseases. In proposed framework, features extracted well-known pre-trained models, including DenseNet-169, ResNet-50, VGG-19. then fed into concatenate layer, making robust hybrid model. framework achieved classification accuracy 99.4%, 95.2%, 94.8% 2-Class, 3-Class, 4-Class datasets, exceeding all state-of-the-art models. These results suggest frameworks ability discriminate individuals infection ones provides opportunity be used diagnostic model clinics hospitals. We also made materials accessible research community at: https://github.com/saikat15010/CovidMulti-Net-Architecture.git.