Characterization of tumor tissue using multi-parametric MRI data of a preclinical tumor xenograft modelPusovnik, Matic (Avtor) Miklavčič, Damijan (Mentor) multi-parametric MRIin vivo image registrationtumor segmentationIntroduction: In small animal studies multiple imaging modalities can be combined to complement each other in providing information on anatomical structure and function. Non-invasive imaging studies on animal models are used to monitor progressive tumor development. This helps to better understand the efficacy of new medicines and prediction of the clinical outcome. The aim was to construct a framework based on longitudinal multi-modal parametric in vivo imaging approach to perform tumor tissue characterization in mice. Materials and Methods: Multi-parametric in vivo MRI dataset consisted of T1-, T2-, diffusion and perfusion weighted images. Image set of mice (n=3) imaged weekly for 6 weeks was used. Multimodal image registration was performed based on maximizing mutual information. Tumor region of interested was delineated in weeks 2 to 6. These regions were stacked together, and all modalities combined were used in unsupervised segmentation. Clustering methods, such as K-means and Fuzzy C-means together with blind source separation technique of non-negative matrix factorization were tested. Results were visually compared with histopathological findings. Results: Clusters obtained with K-means and Fuzzy C-means algorithm coincided with T2 and ADC maps per levels of intensity observed. Fuzzy C-means clusters and NMF abundance maps reported most promising results compared to histological findings and seem as a complementary way to asses tumor microenvironment. Conclusions: A workflow for multimodal MR parametric map generation, image registration and unsupervised tumor segmentation was constructed. Good segmentation results were achieved, but need further extensive histological validation.[M. Pusovnik]20212021-03-03 18:55:37Magistrsko delo/naloga125057sl