BoneX — Multi Bone Fracture Detection System
An AI system for detecting and classifying bone fractures across seven anatomical regions from X ray images, deployed as a usable web application.
Problem
Fracture detection across multiple bone types is challenging due to variability in anatomy and imaging conditions. Baseline models can underperform without careful fine tuning and evaluation.
Solution
BoneX applies a Keras-based deep CNN workflow to classify fractures across seven regions (elbow, finger, forearm, hand, humerus, shoulder, wrist) and returns both fracture presence and type via a simple upload and predict web interface.
System Architecture
End to end fracture classification pipeline using fine tuned MobileNet with frozen feature extraction layers and task specific adaptation for multi bone fracture detection.
Key Design Decisions
- Use fine tuning/transfer learning to improve diagnostic performance over baseline models.
- Treat the work as an end to end system (upload → inference → result) instead of a notebook.
- Evaluate using precision/recall to reflect real diagnostic needs beyond simple accuracy.
Outcomes
Achieved 84% precision and recall for fracture detection/classification, improving upon a previous 79% benchmark. Supported seven fracture categories and packaged the workflow into a web application for interactive use.