Detection of COVID 19 using X-ray Images with Fine-tuned Transfer Learning

Abstract

Recently, COVID-19 infection has been spread to a wider human population worldwide and deemed a pandemic for its rapidity. The absence of medicine or immunization for the “COVID-19” illness, along with the requirement for early discovery and isolation of affected persons, is critical in reducing the risk of infection in healthy population. Blood specimens, or “RT-PCR” are primary screening technique for “COVID-19”. However, average positive “RT-PCR” is expected as 30 to 60%, leading to undiscovered infections and potentially endangering a broad population of healthy persons with infectious symptoms. With the quick examination approach, chest radiography as a common approach for identifying respiratory disorders is straightforward to execute. A board-certified radiologist indicated the presence of disease in these radiographs. Four transfer learning techniques to COVID-19 illness identification were trained using 2,000 X-rays: VGG-16, GoogleNet, ResNet, and SqueezeNet. The result of the experimental assessment shows that the VGG-16 network fine-tuned with Keras achieved sensitivity of 100% with specificity of 98.5% and accuracy of approximately 99.3%.