Metastatic bone disease (MBD) is a common secondary feature of cancer that can cause significant complications, including severe pain and death. Current methods of diagnosis require a highly trained radiologist capable of interpreting medical images and recognising the sites of MBD. These medical images are often noisy, two dimensional, greyscale and usually have a poor resolution.

In order to help assist with these issues, several studies have shown that computer aided methods can locate MBD within medical images. However these methods are limited in scope, accuracy, sensitivity, explainability and do not improve upon the poor visualisations of the underlying medical imaging data.

To address these limitations, I have developed a novel method of automatic MBD assessment and visualisation using computed tomography (CT) imaging data as the input. The method is fully automated and does not require any human interaction -- although users can interact with a viewer that visualises the results. This method has been tested on CT data from prostate cancer patients as prostate cancer is one of the most common sources of MBD.

The method described in this thesis has a sensitivity of 0.871 when detecting sclerotic and lytic lesions within a single data set. This sensitivity is comparable to existing methods, however the scope in detecting these lesions was limited to the vertebrae in previous studies. My method significantly expands this scope to include the ribs, vertebrae, pelvis and proximal femurs.

The work in this thesis also provides novel visualisations of the disease and does not suffer from explainability issues that plague modern machine learning algorithms.

In addition, I developed a novel method of tracking the spread of MBD at multiple time points using longitudinal CT data. This method is capable of calculating the change in lesion volume size across multiple time points, providing a novel numerical assessment.