BONE FRACTURE CLASSIFICATION USING CNN

Abstract

Bone fractures are a serious medical problem, caused by traumatic injury, sports injuries, and high-energy collisions. Precise fracture detection and characterization are mandatory for effective clinical management and successful patient outcomes. Conventional fracture diagnosis involves manual radiographic interpretation by radiologists, which is slow, vulnerable to inter-observer variance, and susceptible to diagnosis delays in resource-scarce environments. Recent developments in computer vision and deep learning have made it possible to perform automated medical image analysis with clinically useful accuracy. This paper suggests an automated bone fracture detection system based on Convolutional Neural Networks (CNNs) for X-ray image binary classification as fractured or normal. The system uses a bespoke CNN architecture that is trained on 10,247 X-ray images (6,643 fractured, 3,604 normal) from publicly available datasets. Our model is 94.2% accurate, 93.8% sensitive, 95.1% specific, and 0.968 AUC-ROC on the held-out test set. A Flask web application offers real-time fracture classification with 15 milliseconds average inference time per image. The light weight (4.8 million parameters) facilitates deployment in resource-constrained clinical settings. This system has promise as a clinical decision support tool for screening for fractures, triage, and medical education, but external validation and prospective clinical studies will be necessary prior to clinical implementation.