Leveraging transfer learning in computer vision for AI-powered orthopaedic assistance: a sustainable approach for Healthcare 4.0

Abstract

Bones are considered to be the most important support system of human body. Generally human bones are made up of protein, collagen, and minerals, especially calcium. Injury in human bone generally results as fracture. A bone fracture is an incident that leads to a crack or break in the bone structure, typically occurring due to accidents when sudden pressure is applied to any part of the bone. To detect bone fractures and provide appropriate treatment, traditional healthcare diagnostics utilise various imaging techniques such as X-rays. Orthopaedic specialists typically rely on their expertise and experience to accurately determine the presence of fractures. Diagnosis based on a doctor's expertise can sometimes result in inaccuracies. In this research work, a deep learning based system using transfer learning is developed for classifying bone fractures using X-ray images. For enhancing the performance, Adam optimiser is used in this work. To transform the image pixels, the raw X-ray images are pre-processed using BGR method. To prevent the model from overfitting issue, dropout regularisation method is applied in this case. The proposed VGG16 model has shown the highest validation accuracy of 96.62% during the model validation.

Ähnliche Arbeiten