• Increase font size
  • Default font size
  • Decrease font size

PhD thesis

E-mail Print PDF
BY ››Simon DAVID
DIRIGEE PAR ››Roux C, Visvikis D, Hatt M
Title ››Image analysis for therapy response studies in Positron Emission Imaging
THESE DE ››SCIENCE
INSTITUTION ››Laboratoire de Traitement de l'Information Médicale
ECOLE DOCTORALE ››SICMA
DATE DEBUT ››2008-10-01
SOUTENUE LE ››2011-12-14
Keywords ››PET, oncology, follow-up studies, multi-tracer analyse, Bayesian classification, change detection, DS theory, DSm theory
RESUME ››Allowing the early detection of metabolic changes during treatment, Positron Emission Tomography (PET) is a promising tool for therapy response assessment. PET imaging is also currently used for the definition of a biological tumour volume in adaptive radiotherapy, with one or more PET scans measuring different tumour features in context of multi-tracers analysis. A therapeutic response is usually defined as variations of semi-quantitative parameters such as standardized uptake value (SUV) measured in PET scans performed during the treatment. The variability associated to the current segmentation approaches are usually affecting these quantitative measurements. Moreover, such measurements do not reflect overall tumour volume and radiotracer uptake variations. In context of adaptive radiotherapy, there is currently no gold standard to define a biological tumour volume according to several multi-tracers PET scans.
In this thesis, three fusion methods have been developed, based on different mathematical theories to combine several follow-up or multi-tracer PET scans at the voxel level. The first proposed approach is based on multi-observation image analysis for merging several PET acquisitions. Based on the Bayesian classification principle, the ASEM approach can merge either follow-up or multi-tracer PET acquisitions with the same fusion process. The second developed method is merging multi-tracer PET scans with the Dempster-Shafer theory (DS) in order to define a global tumour volume according to PET scans measuring different tumour features. The third approach is based on the change detection principle using the recent Dezert-Smarandache theory (DSm), which manages both imprecision, uncertainty and conflict between sources, applied simultaneously to follow-up PET scans.
The proposed methods were applied to both simulated and clinical datasets in followup and multi-tracer context. Their performances were compared to the use of fixed and adaptive thresholding and the Fuzzy-C-mean algorithm applied separately to the fused scans. Working with simulated data allows an accurate assessment of the fusion obtained for each approach. On simulated datasets, the fixed and adaptive thresholdsand the FCM algorithm applied independently on both images led to higher errors than the ASEM approach. If the change detection method is associated to contrasted results on simulated follow-up data, the multi-tracer fusion approach doesn’t seem to be adapted to define a biological tumour volume regarding different tumour features.
Applied to nine clinical follow-up cases, the fixed threshold and the FCM segmentation failed to provide coherent quantitative measurements regarding the partial responses considered. Applied separately by two clinicians,the adaptive threshold doesn’t seem to be robust to noise and tracer uptake heterogeneity leading to significant variability in the quantitative measurements. If the change detection method looks efficient to characterize a therapeutic response for seven patients, this approach failed for two patients responding partially to their treatment. The ASEM method is the one for which the variations of the measured quantitatives parameter are in line with the therapeutic response. This results are confirmed by the multi-tracer clinical cases. For the four considered patients, the threshold methods and the FCM segmentation were less robust to noise and the tumour contrast. The qualitative and quantitative analysis associated to the ASEM fusion emphasize the robustness of this approach. To conclude with this thesis, further development associated to the proposed methods will be described. A three follow-up scans fusion performed by the ASEM method will be present in order to illustrate its ability to identify the tumour changes occurring at different stages of the treatment. Regarding the multi-tracer fusion approach, a new model allowing the merging of FDG and FMISO scan will be present. With this new model, a dose map can be generated and applied in adaptive radiotherapy. Finally, concerning the change detection method, we will present a way to integrate to the current process, a measure of texture parameters by analyzing tumour heterogeneity.