Performance Analysis of Convolutional Neural Network Architecture in Pneumonia Detection From X-Ray Images

Pneumonia is a pulmonary infectious disease and one of the leading causes of global mortality. Radiologists typically perform the diagnosis of pneumonia through the analysis of chest X-ray images, which is a process that is time-consuming and prone to diagnostic errors. Deep learning technology, particularly Convolutional Neural Networks (CNNs), offers an automatic diagnostic solution with high accuracy. This study compares the performance of various pre-trained CNN architectures, including ResNet (50, 101, 152), InceptionV3, VGG-16, DenseNet (121, 169), MobileNet (V1, V2), and EfficientNet (B1, B7) in detecting pneumonia from chest X-ray images. The dataset used consists of 441 pneumonia images obtained from Pertamina Central Hospital and 234 normal images from the Mendeley Data repository. The data were divided into training (541 images), validation (134 images), and testing (675 images) sets, with preprocessing including rescaling and data augmentation. Model evaluation was conducted using accuracy, precision, recall, F1-score, and confusion matrix analysis. Experimental results indicate that DenseNet-169 achieved the best performance with an accuracy of 99.40 % and an F1-score of 0.9954, whereas EfficientNet-B7 demonstrated the lowest performance with an accuracy of 65.33 %. This study highlights the importance of selecting an appropriate CNN architecture for clinical applications.