Positron emission tomography image texture analysis

Alsyed, Emad 2023. Positron emission tomography image texture analysis. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (18MB) | Preview
	PDF (Cardiff University Electronic Publication form) - Supplemental Material Restricted to Repository staff only Download (269kB)

Abstract

According to the World Health Organisation (WHO), cancer is the second leading cause of death around the world. Cancer is responsible for about 18 percent of deaths worldwide and about 10 million people die from it each year. Multidisciplinary teams participate to manage and deliver effective diagnosis and treatments for cancer patients. Currently, PET images and other medical images are interpreted visually by radiologists and clinicians. However, medical images contain more information than what can be assessed visually. The rapid development of medical image analysis has revolutionised the ability to recognise complex patterns in imaging data and provide a depth of quantitative analysis previously unachievable. Radiomics is defined as extracting quantitative features from medical images which cannot be seen by the naked eye. It is now accepted that further data extraction has the potential to enhance the prognostic and diagnostic power of the radiologist or oncologist. However, despite the promising aspect of radiomics, several challenges remain in the field of radiomics. The major challenges that need to be addressed before radiomics can be applied in the clinic are reproducibility, repeatability, and stability of radiomic features. Several studies have reported that some of PET radiomic features are very sensitive to different sources such as segmentation method, image acquisition and reconstruction protocols. Thus, multiple variables, parameters and condition may cause a variation on radiomic features. For increased confidence in the utilisation of texture features as imaging biomarkers, this thesis intends to determine whether v different confounding factors have an effect on PET image radiomic analysis. In this thesis, preclinical, homogeneous phantom and heterogeneous phantom studies were conducted to assess the impact of different reconstruction settings (TOF, number of iteration, number of subsets, FWHM of the gaussian filter) on PET image radiomic features. In addition, the self organising map (a type of artificial neural network algorithm) were applied to cluster and visualise the resulting data. The results presented in this body of work, indicate that different reconstruction settings have an influence on PET radiomic features and some of the robust features were able to distinguish between regions (phantom inserts). Furthermore, the findings of this thesis showed evidence that suggests self-organising map (SOM) has ability to identify emergent properties that effect their variability, in this case contour size. In addition, the SOM can be used with outcome data to serve as a predictive tool for dependent variables (e.g therapy response, prognosis). In so doing the learnt representations of self-organised features serve as the attributes for prediction which will take into consideration the statistical variability in the underlying dataset.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Engineering
Uncontrolled Keywords:	radiomic, reconstruction, PET, texture, analysis, cancer
Publisher:	Cardiff University
Date of First Compliant Deposit:	27 February 2023
Last Modified:	02 Mar 2023 15:31
URI:	https://orca.cardiff.ac.uk/id/eprint/157379

Actions (repository staff only)

Edit Item