Use of intensity- and spatial-based image descriptors to characterise and quantify neoplastic lesions in positron emission tomography

• Main Author: Willaime, Julien Michel Yvon
• Other Authors: Aboagye, Eric : Turkheimer, Federico
• Published: Imperial College London 2013
• Subjects: 610
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.634079

Intra-tumour biological heterogeneity is a characteristic shared by all cancers and is thought to contribute to treatment failure. Within-lesion spatial heterogeneity can be qualitatively visualised in Positron Emission Tomography (PET) imaging. Quantifying the variability of the biological processes and the complexity of the signal being measured in PET oncology is essential. The aim of this thesis was to develop and validate intensity- and spatial-based metrics to quantitatively account for the complexity of radiotracer uptake and to annotate intra-tumour PET heterogeneity. Texture analysis was employed to characterise the in vivo tumour heterogeneity of cell proliferation in breast tumours using 18F-fluorothymidine (18F-FLT) PET. The repeatability of the feature measurements was assessed in patients who had two PET scans prior to therapy. Associations between features at baseline and clinical response measured after three cycles of chemotherapy were explored. Associations between feature changes at one week after the start of chemotherapy and clinical response were also explored. Furthermore, the influence of analysis parameters and imaging protocols were studied. A subset of textural features produced reliable measurements and were associated with treatment response. A technique based on multifractal analysis was also developed for characterising the space-filling properties of an object of interest in PET imaging. The derived spatial index was further combined with intensity metrics and the technique was shown to correct for partial volume effects. The method was illustrated on mathematical objects, validated on test-retest 18F-FLT PET clinical data and applied to realistic PET simulations. This work contributes to the demonstration that intensity- and spatial-based image analysis methods can supplement existing methods in PET quantification studies. These techniques provide some improvements on existing methods to derive classical quantitative PET indices and permit extraction of additional information to further characterise patient populations in the clinical setting and in relation to therapy.