Utilising a Deep Learning Approach for Bone Fracture Detection and Classification

Abstract

Bone is an essential part of every living thing. 
A person's mobility is facilitated by their 
bones. Humans often have bone fractures. In 
order to identify the broken bone, the 
physicians refer to the X-ray picture. The 
manual fracture diagnosis method is laborious 
and prone to mistakes. Consequently, the 
development of an automated method for 
diagnosing bone fractures is necessary. Power 
electrical device models often make use of 
Deep Neural Networks (DNNs). This research 
builds on previous work by developing a 
model for fracture and healthy bone 
classification using a deep neural network. 
Overfitting occurs in the deep learning model 
due to the limited data collection. So, to make 
the data set bigger, data augmentation methods 
were used. In order to assess the efficacy of 
the model, three experiments were conducted 
with the softmax and Adam optimiser. The 
suggested model achieves a classification 
accuracy of 92.44% for both healthy and 
broken bone when 5 fold cross validation is 
used. Ten percent of the test data has an 
accuracy of over 95%, while twenty percent 
has an accuracy of over 93%.