LONG BONES X-RAY FRACTURE CLASSIFICATION USING MACHINE LEARNING
Soaad Nasser Eldin Ali, Hala Maghraby Sherif, Sabry Mohammed Hassan, Ashraf Abd El Rahman El Marakby
Abstract
Accurate long bone fracture diagnosis is essential to prevent permanent deformities resulting from misdiagnosis. This study uses machine learning to introduce a multi-class classification and detection system for long bone fractures. In this study, two image classifications are applied Binary classification and Multi-class classification, and an image detection model. Binary classification to distinguish normal and fractured bone X-ray images. Three models are used for this classification, Model A and Model B are used for grayscale images, and a ResNet50 pertained model for RGB images. Multi-class classification to identify fracture type using ResNet50 fine-tuned model And a Faster RCNN detection model to classify and detect the fracture type and its location in the X-ray images. The dataset was collected from various resources and labeled and annotated following Müller AO classification for bone fracture types. Binary classification achieved a 90.2% accuracy rate for Model A, 90.85% for Model B, and 96.5% for ResNet50, While the multi-class classification model achieved 87.7% accuracy in identifying fracture types for ResNet50 and 80% for Faster RCNN in fracture detection. Special Issue of AEIC 2024 (Electrical and System & Computer Engineering Session)