Analysis and measurement of tumour heterogeneity through multi-modality imaging

Cancer is among the leading causes of mortality worldwide. The need for more effective functional biomarkers as well as the awareness of the complexity of tumour biology, which reflects the tissue heterogeneity, have prompted the use of imaging modalities able to detect its biological aspects. Dynamic Contrast Enhanced-Computed Tomography (DCE-CT), Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) have shown promising results. However, there is still need for proper reliability analyses and development of more quantitative approaches, involving the evaluation of tumour heterogeneity. Part I presents topics and issues regarding the image-based biomarkers used till now, as well as the methodological contributions introduced; Part II is dedicated to the applications of the methodological approaches devised. A reliability analysis of the perfusion maps from DCE-CT has been performed. Downstream of this analysis, the ability of perfusion parameters to improve characterization of lung cancer subtypes has been investigated. Then, an automatic methodological approach has been developed to classify spatio-temporal heterogeneity of lung tumours, as visually performed by radiologists. A novel local-based method has been developed to evaluate the intra-tumoral heterogeneity. The ability of the features extracted to act as a prognostic image-based biomarker has been early assessed. To face the issue of the high variability characterising PET data, a robust approach was introduced to represent a high-uptake activity. To this purpose, a method to perform a 3D PET segmentation has been developed. Then, a multi-modal analysis of the intra-tumoural heterogeneity has been performed in the gastro-oesophageal junction (GOJ) cancer. GOJ heterogeneity has been analysed on FDG-PET/CT and FDG-PET/MRI series. Texture features have a better performance in prognosis compared to the commonly used parameters PET- and MRI-derived. Finally, an algorithm to detect sub-regions in tumour volume has been developed to combine multi-modal information. To this purpose, a 3D registration algorithm was implemented.