Deep learning-driven diagnosis of humerus1 fractures from radiographic data

Bone fractures are among the most frequently treated conditions in hospitals, arising from causes such as sports injuries, traffic accidents, and other trauma. X-ray imaging, computed tomography (CT), and, in more severe cases, magnetic resonance imaging (MRI) are the most commonly employed diagnostic methods. While these techniques offer high diagnostic accuracy, they are time-consuming due to the need for expert interpretation and the volume of cases requiring evaluation. This research aims to develop a machine learning-driven algorithm for the classification and diagnosis of humerus fractures, highlighting the potential of artificial intelligence in medical diagnostics. The proposed algorithm is based on Artificial Neural Networks (ANN) and trained using a dataset of X-ray images depicting both fractured and healthy bones. To optimize performance, the model was trained with rescaled input images, identifying the ideal resolution to preserve critical features necessary for accurate classification. The algorithm achieved over 90% accuracy across different image scales, and its performance was validated using metrics such as confusion matrix, sensitivity, F1-score, and Matthews correlation coefficient (MCC), demonstrating its effectiveness as a diagnostic tool for fracture detection.