XAI-based Advanced Computational Models for Predictive Analysis of Bone Fracture and Structural Repair Healing Prediction

Bone fractures are one of the most frequent musculoskeletal injuries, strongly impacting mobility and quality of life. Fracture healing evaluation is crucial for clinical decision-making, but the existing methods are not precise or interpretable. This paper suggests a new deep learning-based model for bone fracture healing quantification from X-ray images and measuring treatment efficacy. The method utilizes a four-stage architecture: ResNet50 for fracture versus nonfracture classification. Faster R-CNN is employed to localize and detect fracture areas through bounding box prediction, followed by Mask R-CNN for pixel-level segmentation of fracture cracks in the detected areas. Pixel-wise examination allows measurement of damaged and regenerated bone areas to estimate healing percentage. To increase model interpretability and clinical confidence, Gradient-weighted Class Activation Mapping (Grad-CAM) is integrated to highlight the important regions affecting model predictions. A comparative analysis of nine deep models suggests that the best one to use is ResNet-50, which classifies with a high accuracy of 83%.