Automated Fracture Detection and Classification Using Siamese Neural Networks

Abstract

This study aims to address the critical need for accurate and efficient fracture detection and classification in medical radiography, by leveraging recent developments in deep learning techniques. A two-stage pipeline is proposed combining a cutting-edge convolutional neural network (CNN), You only look once (YOLOv9), for wrist fracture detection using a Siamese neural network (SNN) for fracture type classification. The proposed methods aim to overcome the limitations of a single-stage approach by optimizing detection and classification stages independently. The results obtained demonstrate the efficacy of the two-stage pipeline in improving accuracy and reducing class imbalance issues, highlighting the potential of artificial intelligence (AI) driven solutions in enhancing patient care and reducing diagnosis time.