Classification of X-Ray Images Using CNN Models

Authors : Havva Ersöz, Burhanettin Durmuş, Mehmet Ali Gedik

Pages : 203-209

Doi:10.47897/bilmes.1814345

View : 64 | Download : 142

Publication Date : 2025-12-29

Article Type : Research Paper

Abstract :Among medical imaging systems that play a crucial role in modern medical diagnosis and treatment processes, X-ray imaging stands out as an essential diagnostic tool due to its low cost and wide accessibility. This study focuses on developing a model based on a Convolutional Neural Network (CNN) architecture to automatically identify and classify anatomical regions in X-ray images. Using the MURA dataset and the UNIFESP X-Ray Body Part Classification dataset obtained from Kaggle, detailed anatomical region and projection view classification was performed on 7,487 multi-view musculoskeletal radiographs. The classification process utilized the AlexNet and ResNet50 architectures. To enhance the transparency and interpretability of the decision mechanisms, visual analysis was conducted using the Grad-CAM technique on misclassified samples. The obtained results showed that the AlexNet model achieved a validation accuracy of 91.52%, while the ResNet50 model achieved 94.20%. These findings demonstrate that detailed anatomical and directional classification can be performed with high accuracy, suggesting that this method could serve as an effective approach to improving labelling accuracy in hospital information systems.
Keywords : AlexNet, ResNet50, datasets, pacs, Grad-CAM

ORIGINAL ARTICLE URL