Performance of True Transfer Learning using CNN DenseNet121 for COVID-19 Detection from Chest X-Ray Images

Abstract

Confirmed cases of the COVID-19 virus are still growing to date since the discovery of the first case in Wuhan, China at the end of 2019. Fast and accurate early identification of people infected with the COVID-19 virus can help control the rate of spread. One of the Deep Learning methods, Convolutional Neural Networks (CNN), is used to identify diseases in image data. The study proposes the CNN method using Chest X-ray images in the detection of COVID-19 cases with a transfer learning model that can improve higher accuracy. The Transfer Learning model used in this research is DenseNet121. In this study, the transfer learning model is used to see the accuracy obtained and compare it with the accuracy results of the traditional learning method. We used a larges open-source dataset of 1500 Chest X-ray images. System performance is measured from the level of accuracy by the use of transfer learning and the selection numbers of epoch and batch size parameters. That show that pairing parameters with epochs 100, batch size 64 produces a 98% accuracy rate which is the best pair parameter using transfer learning methods. Our transfer learning methods show better classification accuracy than traditional learning methods.