Skip to Main Content
Kinetic analysis of brain PET data with [11C]-(R)-PK11195 frequently uses data partitioning techniques for the extraction of a reference tissue kinetic class. To date, these unsupervised or supervised clustering methods have not yet addressed the question of the optimal number of clusters to extract in total. Here, results from k-means clustering into 2 to 10 classes of a cohort of 12 non-diseased subjects are presented. To characterise the separation, the Mahalanobis distance is used to measure the distance between the centroids and the other clusters. The cluster maps suggest the presence of about 3 distinguishable clusters in brain tissue and a further 2 to 3 extracerebral clusters. The maximum mean Mahalanobis distance was observed for 7 clusters.