Performance of a modified supervised cluster algorithm for extracting reference region input functions from (R)-[11C]PK11195 brain PET studies

(R)-[¹¹C]PK11195 is widely used as radiotracer for imaging activated microglia in the brain using positron emission tomography (PET). Recently, for quantification of specific binding, a supervised cluster analysis (SVCA) approach to extract the reference tissue input function has been reported. This method uses a database of six predefined kinetic classes (SVCA6). In the present study SVCA was modified to enhance performance of the algorithm. The modified SVCA first applies an anatomical mask to include brain tissue only. In this manner a smaller number of kinetic classes can be used (n = 4, SVCA4), potentially improving accuracy and precision. The purpose of this study was to evaluate the performance of SVCA4. To this end 60 min dynamic (R)-[¹¹C]PK11195 studies of 9 AD subjects and 9 healthy controls were acquired, including continuous arterial blood sampling. A reference tissue time activity curve (TAC) was extracted from these scans using SVCA6, SVCA4 and a cerebellar region of interest (ROI). All reference TACs were fitted to a 2 tissue compartment plasma input model, including blood volume fraction, with volume of distribution (VT ) as outcome parameter. VT obtained using SVCA4 and cerebellum showed better (i.e. less) intersubject and intergroup variability than when using SVCA6. Moreover, reference tissue input based quantification of (R)-[¹¹C]PK11195 binding in a target region (thalamus) showed better correlation with plasma input kinetic analysis when using SVCA4 rather than SVCA6 or cerebellum. However, use of SVCA4 results in a blood volume related upward bias in combination with SRTM. The latter may be resolved by using a reference tissue model with blood volume fraction corrections. It is concluded that SVCA4 is more accurate and precise than SVCA6. Use of cerebellar ROI may still be used in combination with SRTM, but provides a more conservative measure of binding.