A Novel Framework for Automated Detection of Bone Fracture using Deep Learning

Abstract

Better patient outcomes and prompt medical intervention depend on the early and precise bone fractures detection. However, manually interpreting X-ray images is time-consuming and frequently produces inconsistent results, particularly when comparing different bone areas. Using the Dataset of Bone Fracture, this study suggests an automated framework for categorizing bone fractures. 10,580 radiographic pictures of both broken and unbroken bones from a variety of anatomical locations, including the upper and lower limbs, the lumbar region, the hip, and the knee, are included in this dataset. The proposed approach integrates the VGG16 with a Convolutional Block Attention Module (CBAM) to enhance feature learning by adaptively focusing on salient spatial and channel information. Preprocessing steps included image normalization, resizing, and augmentation to mitigate overfitting and address class imbalance. The experimental results demonstrate that the VGG16 + CBAM model achieved 99% accuracy, 99% precision, 98.5% recall, and 99% F1-score, outperforming standard CNN, RESNET50 and plain VGG16 models. The addition of CBAM significantly improved interpretability and reduced false negatives, highlighting the potential of attention-based architectures in computer-aided diagnostic systems for multi-region bone fracture detection.