Effect of CT-Scan Image Resizing, Enhancement and Normalization on Accuracy of Covid-19 Detection

Abstract

Covid-19 continues to be a global health problem with an impact on at least 70 million people exposed and more than 1.5 thousand people died in December 2020. Detection by RT-PCR as the gold standard of WHO is still difficult to reach in some areas and has a low sensitivity issue. Many studies have focused on the detection of Covid-19 using computer vision, especially deep learning methods. However, it is necessary to evaluate the preprocessing stage before carrying out the classification to increase the accuracy of its detection. Therefore, the objective of this study was to compare the choice of the CT-Scan image pre-processing method and its effect on the results of covid-19 classification accuracy. The benefit of this study is that it can be used as a recommendation when considering the choice of a CT-Scan image preprocessing method to improve the accuracy of Covid-19 detection through more comprehensive deep learning. This study uses a CT scan image because it is considered to be of better quality than an X-ray image, although the price is relatively more expensive. The various methods used were resizing (deformed and non-deformed), enhancement(HE, CLAHE, EFF), and normalization ranges ([–1 1] and [0 1]). Meanwhile, the deep learning method used is the VGG-16 classifier. The results showed that there was an influence generated by the variations in the preprocessing methods on the precision of the Covid-19 classification. The highest accuracy results were obtained with 88.54% using the deformation ranges of size change, CLAHE improvement and normalization [0 1] and [–1 1]. This result quite competitive compared to other studies.