Bone Fracture Identification Using Convolutional Neural Network and Fine-Tuned DenseNet Models

Abstract

In the recent past, there have been various musculoskeletal disorders (MSK) that are caused by bone illness. As per the report provided by the WHO, it has been estimated that 1.71 billion people have been suffering from musculoskeletal disorders. Apart from this, the rate of increase in the femoral neck injuries and fractures is also expected to increase 4 times in the next 30 years. Therefore, the timely detection and treatment of patients with bone fractures is essential to avoid emergency care failure. In this work, two deep learning-based (DL) models, namely, convolutional neural networks (CNN) and DenseNet, have been used for the early detection of bone fractures. These models were employed with varied hyperparameters, where the outperforming results have been achieved with epoch 10. The accuracy achieved by the CNN model has been resulted as 96.25%, while the DenseNet model has resulted the accuracy of 62.29%. These outcomes identify that the CNN model is more proficient in terms of achieving improved accuracy for bone fracture identification when compared to the fine-tuned DenseNet model. The proposed DLbased CNN model has shown improved accuracy that may assist radiologists in identifying different fractures in the radiographic images.

Ähnliche Arbeiten