Authors: Abdulmajeed Alsufyani *
Affiliations:
Department of Computer Science, College of Computers and Information Technology, Taif University, Taif, Saudi Arabia
This study examines the performance of several deep learning models for detecting bone fractures from X-ray images. Traditional radiological methods depend on manual interpretation, which can lead to mistakes. Deep learning provides a useful alternative by automating the process of fracture detection. In this research, five models were tested: one custom Convolutional Neural Network (CNN) and four pre-trained models — AlexNet, DenseNet121, ResNet152, and EfficientNetB3. The models were trained on a dataset containing 10,581 X-ray images, which were labeled as either fractured or non-fractured. The models’ performance was measured using accuracy, precision, recall, and F1-score. Among these models, EfficientNetB3 achieved the best results, with 99.20% accuracy and perfect recall, showing its high potential for use in clinical practice. ResNet152 and the custom CNN also performed well, although with slightly lower accuracy. The findings of this study emphasize the value of using advanced deep learning architectures for medical image analysis.
Bone fractures, X-ray images, Deep learning, Fracture detection, Medical imaging
https://doi.org/10.21833/ijaas.2025.05.008
Alsufyani, A. (2025). Enhancing diagnostic accuracy in bone fracture detection: A comparative study of customized and pre-trained deep learning models on X-ray images. International Journal of Advanced and Applied Sciences, 12(5), 68–81. https://doi.org/10.21833/ijaas.2025.05.008