FractureNet: A Fusion Approach for Automated Hand Fracture Detection and Classification

Bone fractures are a prevalent health concern, and their precise and appropriate recognition is crucial for effective diagnosis and patient recovery. While conventional methods rely on radiologists’ interpretations of imaging studies, computer vision-based analysis offers enhanced fracture detection and classification. The current literature faces a lack of model generalization due imbalanced dataset and restrictions in several classes. The proposed study was undertaken to address these issues and classify hand fractures, with a specific emphasis on transverse and compound types. The proposed FractureNet model divides into two phases: Phase 1 utilizes the YOLOv5 model for fracture detection with CLAHE to enhance image quality, and Phase 2 classifies the fracture type by fusion approach using a HOG, LBP, and VGG16 model to improve the model performance. These combined features advance the model’s capability to distinguish, and the features are fed into various classifiers, among which SVM achieves 95.4% accuracy, 95.3% precision, 96.1% recall, and 95.6% f1-score for the collected fracture x-ray dataset.