Efficient Transfer Learning for COVID-19 Diagnosis using X-Ray and Computed Tomography Images

Abstract

In this work, we aim to find an effective model to diagnose COVID-19 by using a Transfer Learning (TL) model. The purpose is to classify COVID-19 infected persons from chest X-Ray (XR) and Computed Tomography (CT) images. Several Transfer Learning models have been studied to find the most efficient and effective among them. The proposed approach is based on Tensorflow and the architecture uses the MobileNet_V2 model. The datasets that are used in this study are publicly available. In order to train and evaluate our proposed model, we collected the CT scans dataset of 8000 images with two classes of infected and normal lungs, and the XR dataset contains 616 images. Two experiments are conducted with samples of different sizes to evaluate the model using google colab. The results revealed that the performance of our model MobileNet_V2 is highest with validation accuracy for XR and CT scans images: Val_Accuracy <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">XR</inf> =96.77% and Val_Accuracy <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CT</inf> =99.67%, and test time for XR and CT scans images: t <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">XR</inf> =0.18s, t <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">CT</inf> =0.03s respectively.