Published Paper Details:

Bone fracture and bone tumor diagnosis are among the most important issues in orthopedic and oncological practice, especially in areas with low expertise in radiological diagnosis. This paper presents a deep learning based framework for the automated detection of fractures and tumors in radiographic and MRI images using the YOLOv11n object detection model. The system was trained on a combination of two curated sets of 1,500 tumor specimens including osteosarcoma and benign lesions and 3,000 fracture samples including transverse fractures, oblique fractures, and comminuted fractures. Grazing conversion, normalising, and worsening strategies were applied to improve dataset variety and robohardness. The model showed good performance with an accuracy of 92.8%, precision of 92.4%, recall of 90.1%, and mean average precision (mAP) at detection threshold of 0.5 for tumor detection (90.8% mAP), and 94.3% accuracy, 93.2% precision and 94.9% recall for fracture detection. To enable real-world applicability, the framework was deployed through a Streamlit-based web application offering role-specific access for patients and doctors, real-time image uploads, annotated visualizations, heatmap-based interpretability, and multi-format report exports (CSV, PDF, JSON). By offering strong detection accuracy while keeping the interface approach easy to use, the proposed system offers reliable assistance to minimize delay in diagnosis, error rates and enhance patient outcome in both orthopaedic and oncological treatment.