Wrist Fracture Detection using Deep Learning

Wrist fractures are among the most frequently missed injuries in X-ray interpretation, particularly when the fracture line is fine or subtle. This project presents a deep learning-based solution that automatically detects wrist fractures from radiological images. Two custom detection algorithms, FissureNet‑WD and OssiFrac‑CBAX26, are developed using an enhanced YOLOv11 architecture. FissureNet‑WD emphasises thin fracture line identification through specialised edge-focused modules, while OssiFrac‑CBAX26 uses multi‑scale bone texture analysis to locate fractures across varying image densities. The models are trained and evaluated on a Roboflow dataset containing 10,374 training, 984 validation, and 514 test images belonging to a single fracture class. A web application built with the Flask framework and an SQLite database provides an interface for user registration, login, X‑ray upload, automated detection, and scan history management. All front‑end styling and scripts are embedded within HTML templates. The system outputs annotated images with bounding boxes and confidence scores for each detected fracture. This work demonstrates that domain‑specific architectural modifications can improve fracture detection accuracy and offers an accessible platform for experimentation with automated bone imaging analysis.

Keywords: wrist fracture detection, deep learning, YOLOv11, FissureNet‑WD, OssiFrac‑CBAX26, X‑ray imaging, bounding box, convolutional neural network, web application, medical imaging.