Abstract

Effective categorization of bone fractures is crucial for accurate medical diagnosis and treatment planning. This paper presents a comprehensive approach using GradCAM imaging and Convolutional Neural Networks (CNN) to enhance bone fracture diagnosis accuracy. The dataset, "Bone Break Classifier Dataset," includes 1735 photos of 12 fracture categories, ensuring diverse representation. Data preprocessing optimizes model training, leading to a CNN architecture balanced for complexity and performance. Sophisticated optimization techniques, including learning rate scheduling, contribute to improved model performance. The integration of GradCAM visualization enhances interpretability, facilitating transparent decision-making. The proposed model achieves a training accuracy of 93.84%, validation accuracy of 90.62%, and test accuracy of 85.16%. Quantitative and qualitative results, along with comparative analysis, highlight the model's effectiveness. The study identifies areas for further improvement, offering a robust framework for fracture classification with implications for clinical decision-making and medical image analysis advancements.