Ensemble Model based Osteoporosis Detection in Musculoskeletal Radiographs

In both adult and paediatric imaging, musculoskeletal radiographs provide a significant depth of knowledge in treating illnesses or injuries of the joints, bones, muscles, and spine. Because musculoskeletal disorders affect over 1.7 billion people, detecting anomalies in musculoskeletal investigations is difficult. In this project, artificial intelligence will be utilised to categorise images from X-ray equipment of the elbow, finger, humerus, shoulder, and wrist as fracture or nonfracture. The performance of five deep learning-based pretrained models in identifying these fractures was evaluated using the musculoskeletal radiographs (MURA) dataset, and an ensemble learning model for each category was created. DenseNet-169, MobileNetV2 ResNet-50, ResNeXt-50, and VGG16 are the pre-trained models employed. In the ensemble model, test accuracy, AUC, and Cohen's kappa values were produced as a result of 5 separate classifications in total, with certain categories performing better in ensemble and others in individual models.