Comparison of Parameter Optimizers on COVID-19 Detection From A New Lung CT Scan Dataset

The aim of this study is to determine the success of different parameter optimization methods in a classification done by using a new Convolutional Neural network (CNN) proposed in this study on a new original dataset consisting of lung CT images collected by the researchers from ill/healthy people in order to diagnose COVID-19. To achieve this purpose, the researcher prepared a new data set involving computerized tomography images collected from Erzurum City Hospital Emergency Department for the diagnosis of COVID-19 after the necessary permissions were taken. This new data set consists of 1081 lung CT images, 568 of which belong to 313 people infected with COVID-19. The results obtained for the proposed CNN model through different optimization methods were compared. Precision, recall and f1-score parameters were used while evaluating the results. In our dataset, the highest performance values for the proposed CNN model were achieved by AdaMax optimizer as follows: %89.86 accuracy, %86.51, precision, %95.61 recall and %87.33 f1-score. The classification accuracy values achieved by other optimizers were %89.40 (RMSProp), %87.10 (Adam), %81.11 (SGD).

Journal

2022 MEDICAL TECHNOLOGIES CONGRESS (TIPTEKNO'22)

Year

2022

DOI

10.1109/TIPTEKNO56568.2022.9960199

WoS UID

WOS:000903709700054