New Segmentation Models for the Radiologic Characterization of Polycystic Kidney Disease Patients from MR and CT Images

Recent advances in genomics have contributed to a better understanding of the pathogenesis of the polycystic kidney disease (PKD), suggesting new treatment strategies to inhibit or delay cyst formation and expansion. The efficacy of these therapies is evaluated by estimation of cystic burden measured by magnetic resonange imaging (MRI) as total kidney volume (TKV). In this Thesis, different imaging approaches are proposed for a correct characterization of the PKD patient by the estimation of renal and cyst volume from magnetic resonance and computed tomography (CT) images. TKV estimation method from MRI relies on a previously validated method developed for axial images that has been adapted and validated to work on coronal images. The results have been compared with the ones obtained from axial images and validated with volume estimation obtained from manual tracing. The performace of the semi-automated method in terms of misclassification of the PKD patient was also evaluated in comparison with other radiologic approaches currently usedfor TKV assessment such as the ellipsoid method and the mid-slice method. A novel method for TKV computation from CT images is proposed. This multi- step approach is completely automated and includes the use of a level set approach to identify the renal contour and so extrapolate the renal volume. The segmented kidneys obtained with the developed methods where used for the segmentation of the cysts. A similar strategy was used for cyst segmentation and counting from MR images. Every cyst agglomerate underwent a voting mechanism based on the curvature of the object interface to distinguish the single cysts. The results of this approach for TCV computation was validated through comparison with TCV obtained by manual segmentation. The last chapter is dedicated to the research activity conducted in the area of diffussion weighted imaging.