CompDNet-512: Hybrid Deep Learning Architecture for Prediction of COVID-19

With a growing appearance of enormous cases of COVID-19 in people, computational solutions founded on deep learning can play a huge role in performing automated prediction of the corona virus disease. In this article a hybrid deep convolution neural network architecture named as CompDNet-512 is implemented to detect the presence of COVID disease in the chest radiographs of patients. Previously the researchers had only used deep transfer learning technique for train neural network model to detect the syndrome. They have succeeded to some extent, but none were able to developed generalized model which can be predicting various abnormalities present in clinical images. Herein, our model can be applied to range of medical application for abnormalities prediction. The proposed model is a combination of three techniques that is, lossless compression, dense network and progressive resizing, which when applied in a pipeline to the chest X-ray dataset consisting of positive corona virus patient studies and normal studies, outperformed the existing models in detecting COVID-19 and shown an accuracy of 98.3 %.