Bone fracture diagnosis using radiographic imaging is a critical yet challenging task due to subtle fracture patterns, image quality variations, and increasing clinical workload on radiologists. Although deep learning–based methods have demonstrated promising performance in automated fracture detection, their clinical adoption remains limited due to black-box decision making, lack of explainability, and reliance on unimodal imaging data. This paper presents an explainable multimodal deep learning framework for automated bone fracture detection, severity assessment, and clinical decision support. The proposed framework integrates radiographic images with structured clinical metadata using an attention-based feature fusion strategy to enhance diagnostic accuracy and contextual understanding. Convolutional Neural Networks and Vision Transformer architectures are employed for image feature extraction, while clinical parameters are encoded through a multilayer perceptron. Model interpretability is achieved using Gradient-weighted Class Activation Mapping (Grad-CAM), enabling visual localization of fracture-relevant regions. Extensive analysis using publicly available musculoskeletal datasets demonstrates that the multimodal explainable approach outperforms conventional unimodal models in terms of accuracy, robustness, and clinical reliability. The results highlight the importance of explainable and context-aware AI systems in musculoskeletal imaging and support their potential integration into real-world clinical workflows for improved fracture diagnosis and decision support.