A Deep Learning Classifier for Accurate Detection of the Novel Corona Virus

Abstract

Although India witnessed the second slowest 100 to 1000 jump in COVID-19 cases, according to WHO, the number may be inaccurate because of the lack of rapid and large-scale testing facilities. According to reports, India is yet to face the gruesome effects of this pandemic as it moves closer to stage 4 of the community spread. Though standardized tests used in detecting coronaviruses, such as RT- PCR or transcriptase-polymerase chain reaction, take a minimum of 24 hours to generate useful results, they are also prone to high false negatives. Consequently, multiple periodic tests are required to arrive at a firm confirmation. Owing to this gap in the Indian Coronavirus testing scenario, this study focuses on a comparatively rapid and accurate method of testing employing AI-based image analysis of X-Ray and CT scans of the Lungs. Artificial intelligence based deep learning methodologies involving Convolutional Neural Networks with a sharp eye on accuracy of results and practical usage could be used for image analysis. Pre-trained and well-known convolutional neural networks along with a standard dataset for training and testing the same have been selected for the process. The performance of the model is also analyzed using standardized convolutional neural network analysis techniques to infer the best model for the particular use-case. The main objective of the study is to evaluate whether deep learning has the potential to provide accurate results and could provide aid to the existing X-ray methodology.