City Research Online

Abstract

This thesis studies wrist fractures seen on radiographs. Wrist radiographs are anal ysed by two different approaches; first by traditional image processing to extract geometric measurements, then by deep learning to classify risks as normal or abnormal (i.e. fractures or implants). Two data sets are used. The first data set includes wrist radiographs obtained from the Department of Radiology at the University of Exeter. The second data set corresponds to MURA X-ray images (MUsculoskeletal RAdiographs) obtained by the Stanford Machine Learning Team. The MURA data set provides more X-ray images to explore than the first data set.

In the first task, a semi-automated geometric image analysis algorithm is proposed to analyse and compare the radiographs of healthy controls and patients with wrist fractures treated by Manipulation under Anaesthesia (MuA). The first dataset was used in this task. Thirty-two geometric and texture measurements were created. Image texture emerged as a metric of the most distinct geometric features from wrist X-rays associated with fractures.

In the second task, eleven pre-trained convolutional neural network (CNN) architectures were used. CNN classified the MURA data set into normal and abnormal categories. Transfer learning technique applied to all eleven pre-trained CNNs to deal with wrist X-ray datasets. ResNet-50 and Inception-ResNet-V2 were then explored further using data augmentation strategies. Transfer learning techniques and data augmentation strategies greatly enhance CNN’s ability to classify wrist X-ray images.

Class activation mapping (CAM) explores the convolutional neural network’s activation associated with the abnormality within the wrist X-ray image. It shows that CAM can indicate the abnormality area in the wrist’s X-ray image. The graphical heatmap of CAM overlaid on the wrist X-ray image marks the visual point of the area that triggers the CNN’s decision.

Publication Type:	Thesis (Doctoral)
Subjects:	R Medicine > R Medicine (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering
Departments:	School of Science & Technology > Engineering > Electrical & Electronic Engineering School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads