Evaluating the Effectiveness of Deep Learning Models for Chest X-Ray Image Classification

Abstract

A chest X-ray (CXR) examination is one of the radiological examinations used to help a doctor diagnose a disease in patients safely, quickly, and inexpensively. The development of Computer-Aided Diagnosis (CAD) systems has prompted numerous researchers to explore methods for detecting diseases using X-ray imaging. By implementing this research, it is hoped that it will enable medical personnel to accurately and quickly diagnose patients' diseases. This study utilizes three datasets of aortic enlargement, cardiomegaly, and COVID-19. A Convolutional Neural Network (CNN) is one method that researchers widely use to build Computer-Aided Design (CAD) systems. This study aims to compare the performance of seven CNN models with different architectures to determine which one produces the highest accuracy. The seven CNN models used include DenseNet121, DenseNet169, DenseNet201, InceptionV3, MobileNet, ResNet50, and Xception. The test results show that the DenseNet201 model, with an input size of 224 × 224 pixels, achieves the highest accuracy value for all datasets, reaching over 90% accuracy.