COVIZONE: A Deep Transfer Learning-Based System for Automatic Detection of Covid-19 using Chest X-Ray

With increasing rise of COVID-19 infected patients in India and worldwide, examining and detecting COVID-19 among such large number of populations is becoming a humongous task for the medical practitioners and civic authorities. RT-PCR, real time reverse transcription-polymerase chain reaction technique is widely accepted and one of the reliable methods for detection of novel COVID-19.However, being a time consuming, laborious and expensive method for declaring results for the patients in over 6-8 hours to even 3 days in remote places, this technique is not being widely used. The high and very fast spread rate of COVID-19 and low availability of RT-PCR kit, is making the use of computer assisted technologies an inevitable and a potentially faster response mechanism catering to a large population with least human error and a cost-effective solution. Therefore, an intelligent system COVIZONE has been presented, in the proposed work, designed using state of the art pre-trained CNN model to analyze and detect COVID-19 presence in the lungs using Chest X-Ray and CT-Scan Images. In the proposed work, a multi-class classification (Normal, Pneumonic and COVID-19) of patients using ResNet and ResNext CNN model has been done. Both the models show similar performance with high accuracy of 96% and 97% respectively on public dataset of COVID-19, Pneumonia and Normal CXR and CT-Scans. To avoid skewness due to lesser number of COVID-19 CXR images, dataset used has limited Pneumonia and Normal CXR images to train the system and achieved noticeable high accuracy. The proposed COVID-19 detection model i.e. COVIZONE, even if not used as a primary Covid testing and detection tool, can still be a very helpful tool for screening potentially infected persons and help the physicians who are yet not trained for this pandemic diagnosis.