Automated Bone Fracture Detection in Radiographic Images Using a Fine-Tuned AlexNet Model

Bone fractures are major medical problems requiring correct diagnosis and fast treatment for effective patient care. This paper explores the implementation of a modified AlexNet model for automated bone fracture detection utilizing radiographic X-ray images. Deep learning and transfer learning let training on a dataset of 10,580 X-ray images split into test, validation, and training sets produce a model. The AlexNet model achieved an overall accuracy of 94% with precision and recall values of 92% and 95% for non-fractured cases and respectively 96% and 94%. These tests reveal the model's degree of discrimination between non-fractured and fractured contexts. Preprocessing techniques, including data augmentation, normalization, and contrast enhancement, improved the model’s robustness and adaptability to diverse imaging conditions. Despite a minor discrepancy between training and validation accuracies, the results indicate the model’s potential to support radiologists by enhancing diagnostic efficiency and accuracy in clinical workflows. Future work aims to address current limitations by expanding the dataset, exploring advanced architectures, and integrating the model into clinical systems for real-time diagnosis. This study highlights the potential of deep learning in reducing manual diagnostic workloads and improving patient outcomes through reliable and efficient automated fracture detection.