Deep Learning-based Analysis and Classification of COVID Patients Through CT Images

COVID 19 disease also known as SARS-CoV-2 has spread rapidly all over the world, crippling industries all around the world, caused many death and affected life in all aspects. SARS-CoV-2 has become a global pandemic within three months. Even though some test facilities has been available, but they are not useful because of varying symptoms. In severe cases most patients have been diagnosed by lung infection, most of the patients with lungs infection go unnoticed or have been confused with pneumonia which caused the rise in mortality rate. Modern facilities, such as artificial intelligence and machine learning, and neural network-based technologies can be used to resolve these issues. In this research, we present a technique for lung CT-scan images analysis to classify the infected patients, for this we have used the lightweight neural network-based EfficientNet using a publicly available dataset and achieved an accuracy of 98.00 %. Other datasets have also been tested on trained weights of EfficientNet classification architecture and accuracy of 86.62 % and 88.98% is achieved. We also used specialized pre-processing techniques on the dataset, which gives the accuracy of 99.90%, and fine-tuned the trained weights on two other datasets and achieved an accuracy of 99.89% and 99.18% respectively. Also, it has been proved that training weights of the neural network on one dataset, could detect infected patients and give good accuracy on any other CT scan datasets.