Evaluation of image resize and thresholding binary on Covid-19 detection using convolutional neural network

Abstract

Covid-19 is a disease caused by infection with the 2019 novel coronavirus with the rapid spread has resulted in the cause of a new pandemic in the world. Several things must be considered to suppress the spread of this virus, the most crucial strategy of which is through an effective patient tracking and diagnosis process. One way to diagnose Covid-19 is through radiological tests. In the conventional method, the radiologist performs observations of the chest x-ray manually concerning things that depend on the interests and judgment of the doctor. It is due to the inaccuracy of detecting Covid-19 patients. Therefore, it is necessary to have a system with high accuracy that can help the classification process of radiological test results. So, in this study, an analysis of the convolutional neural network was carried out to help diagnose this disease. By utilizing secondary data from images of Covid-19 thorax x-ray, viral pneumonia, and the normal with an amount at 1,300 for each class, the data is divided into 70% training data and 30% test data. The data set has gone through 3 preprocessing stages: resizing, threshold binary, and the original image without going through the preprocessing stage. The results showed that the accuracy value of the detection model with the CNN method is 91.11% for images without preprocessed, 93.68% for images that have been resized, and 89.91% for images subjected to images the threshold binary function. Applying the image resizing stage to the input image with a smaller resolution can increase the accuracy value and shorten the computation time required for the resulting Covid-19 detection model. At the same time, the application of the threshold binary stage on the input image can reduce the accuracy value and prolong the computational time required for the model. © 2022 American Institute of Physics Inc.. All rights reserved.