Explainable AI-Based Humerus Fracture Detection and Classification from X-Ray Images

The human skeletal framework relies heavily on bones, and one such crucial component is the “Humerus.” Positioned in the upper arm, extending from the shoulder to the elbow junction, the Humerus provides essential structural support for muscles and facilitates upper-body movement, particularly in the arms and hands. Consequently, Humerus fractures significantly impact daily life, causing disruptions and limitations. This paper presents a thorough exploration of an Explainable AI-based Humerus Fracture Detection and Classification system, employing various deep learning models. Leveraging a dataset of 1266 X-ray images, encompassing fractured and non-fractured humerus bones from the publicly available “MURA” dataset, our research evaluates the effectiveness of Convolutional Neural Networks (CNN), VGG16, VGG19, DenseNet121, and DenseNet169 in detecting fractures. After 30 epochs of training, we assessed their performance using critical metrics: accuracy, precision, recall, and F1 score. Notably, DenseNet121 and DenseNet169 exhibited superior accuracy, precision, and recall, laying a robust foundation for automated humerus fracture diagnosis. We also introduced two ensemble models, "Ensemble-1 (VGG16 and VGG19)" and "Ensemble-2 (DenseNet121 and DenseNet169)," which delivered substantial improvements in accuracy, precision, recall, and F1 score, showcasing the potential of ensemble techniques in clinical settings. Furthermore, we enhanced model interpretability and transparency by incorporating Saliency Maps and GRAD-CAM (Gradient-weighted Class Activation Mapping) for Explainable AI (XAI). This visualization allowed us to identify regions of interest in X-ray images contributing to the model’s predictions, providing valuable insights for medical practitioners.

Перевод пока недоступен