Degree Name

Master of Philosophy


School of Physics


Non-small cell lung cancer (NSCLC) patients in stage IV of the disease are commonly diagnosed with bone metastases, which is the leading cause of death in NSCLC patients. Once the cancer cells have metastasised, the disease is considered incurable and palliative treatment such as radiotherapy given as 8 Gy in a single fraction or 25 Gy in 5 fractions is one of the most common treatment options to provide relief from pain. Recent, clinical trials have shown that if a limited number of metastatic bone lesions, in their early stage of metastases (called, oligometastasis) treated with very high, stereotactic doses such as 30 Gy in 3 fractions or 16-20 Gy in single fraction could increase survival without the progression of the diseases as well as overall survival of these patients.

NSCLC patients undergo 18F-fluorodeoxyglucose (FDG) – positron emission tomography (PET)/ computer tomography (CT), in short, FDG-PET/CT imaging for identification of stage and progression of the disease as well as for the evaluation of the treatment response. More importantly, PET/CT imaging with FDG radiopharmaceutical has shown high sensitivity and specificity for the diagnosis of bone metastases. The CT-only images, taken before the radiotherapy treatment, for the generation of treatment plan and dose computation using a computerised radiotherapy treatment planning system inherently have inferior contrast for soft tissues, resulting in poor visualisation of malignant metastases compared to FDG-PET/CT images. Therefore, in recent years, radiation oncologists have increasingly used FDG-PET/CT images to detect and manually contour /delineate the extent of the bone lesion. The manual contours of the target volume (termed as Gross tumour volume or GTV) in bone are translated to planning-CT images for radiotherapy treatment planning. In radiotherapy, the accuracy of GTV delineation is crucial for the precise delivery of the prescribed radiation dose to the target volume to ensure an increased probability of tumour control and reduce the probability of normal tissue toxicity by ensuring that the manually contoured boundary of the target volume does not encompass the surrounding normal tissue. In comparison to conventional radiotherapy, it is prudent to have an accurately contoured GTV with reduced margins in Stereotactic ablative body radiotherapy, where high doses of radiation are delivered in small fractions, The manual contouring techniques, based on the clinical expertise of the radiation oncologist, are, by nature, subjected to inter-observer variability, which could potentially result in suboptimal treatment if the target volume is undercontoured or results in severe toxicities, if, over-contoured. Moreover, inconsistencies in contouring impact the outcome data from randomised trials or retrospective studies. Therefore, auto-segmentation of the target volumes is an essential tool in today’s technology-driven radiotherapy and is the main motivation of this thesis.

In this thesis, a software tool has been developed for the automatic search and segmentation of malignant lesions in bone using FDG-PET/CT images. The autosegmentation algorithm utilised the standardised uptake values (SUV) of FDG in PET images to identify and delineate the bone metastases and uses information from CT images taken during the PET/CT scan to localise the anatomical location of the bone metastasis. The software calculates metrics such as volume, SUVmean and SUVmax of the segmented lesions for quantification and validation. The software validates the auto-segmented lesions against manually drawn contours of a radiation oncologist using statistical analysis, making it a very useful tool for optimisation of SUV threshold that could be applied in scientific studies utilising large patient populations for investigation of the number of metastatic lesions.

The auto-segmented metastatic bone lesions are highly conformal biological tumour volumes (BTV) that could be used in radiotherapy treatment planning for dose escalation using stereotactic ablative body radiotherapy (SABR) for treatment of bone metastases.

Apart from radiotherapy applications, this tool can be used for diagnosing and staging diseases and monitoring the progression and treatment outcomes in a large patient population. The tool can be integrated into more sophisticated machine learning and deep learning algorithms, which can be instrumental in identifying biomarkers or radiomic signals within the lesions that can help clinicians provide more targeted and personalised treatment options.

FoR codes (2008)




Unless otherwise indicated, the views expressed in this thesis are those of the author and do not necessarily represent the views of the University of Wollongong.