DETECTION OF COVID-19 USING TRANSFER LEARNING

Abstract

Coronavirus disease-19 spread has grown to almost every corner of the globe. As a result, it is necessary to take steps toward a much earlier diagnosis of its infection. Coronavirus Illness (COVID-19) is a recently identified coronavirus infectious disease. In most COVID-19 infected persons, respiratory illnesses are mild to severe. without particular treatment they recover. The risk of serious sickness is increased among elderly and those with underlying medical conditions, such as cardiovascular disease, diabetes, chronic respiratory disease and cancer. The greatest method to avoid and delay transmission is to know the COVID-19 virus, the sickness it produces and how it is spreading well. The only protection from it’s illness via washing your hands or regular use of alcohol-based rubber without touching your face. Chest X-Rays, Computed Tomography, and RT-PCR are early-stage diagnostic techniques.Visually detecting and inspecting these clinical images for any hidden anomalies is a time-consuming task. Serology is used to identify anticorps in clinical settings and population surveillance. Due to the restricted availability of test kits every person afflicted by the virus is difficult to detect.In addition, these tests take from several hours to a day to generate the result, which in the current situation of urgency becomes excessively tedious, time-consuming and mostly mistake prone. A quicker and more accurate screening approach is therefore urgently needed, which may also be validated by the PCR test. Transfer Learning in medical imaging has a lot of research potential. The method proposed here is a transfer learning-based binary classification model which predicts whether a Lung CT image has SARS-CoV-2 infection. It has a three-stage procedure for fine-tuning various pre-trained architectures. It uses progressive resizing an optimization technique in which our approach is to resize

the input images to 128×128×3, 150×150×3, and 224×224×3 pixels and fine- tuning the neural network at each stage. As a result, CT transfer learning with

progressive resizing outperforms various published models in the recent research work with improved accuracy of 97.4% with only 22 epochs. This technique may help to diagnose COVID-19 patients at an early stage and reduce the pressure on medical systems.