Hybrid AI Models for Improved Bone Fracture Diagnosis

Abstract

Fractures are the most common musculoskeletal injury, the identification of which is essential for the treatment and rehabilitation of injured patients. The present work proposes an improved method for automatic bone fracture detection via Deep Learning (DL) and Machine Learning (ML) with the assistance of medical imaging diagnostics such as X-ray images. The proposed system was trained and evaluated on a set of 10000 labelled X-ray images of normal and fracture cases. The proposed approach is based on transfer learning, wherein the main feature of the proposed architecture involves using pre-trained CNNs, including ResNet50 and DenseNet121. The extracted features are utilized by ML classification models, including SVM and Random Forest. Regularization is used to decrease the amount of data from certain classes, and data augmentation is used to minimize imbalanced classes. Regularization is also used to detect minor fractures that could be difficult to detect using other techniques. The performances of the designed models were assessed based on the accuracy, precision, recall, and F1-score. The hybrid system resulted in 95% accuracy for standard and fractured bone images, with an overall accuracy of 96.5%.

Ähnliche Arbeiten