Evaluation of the automatic three-dimensional delineation of caudate and putamen for PET receptor occupancy studies

Abstract

PET receptor occupancy studies with a baseline study and an intervention study are increasingly used as an aid in dose-finding procedures for central nervous system drug development. The aim of this study was to evaluate, and confirm the feasibility of two automatic, paired, three-dimensional delineation methods of striatal structures (caudate and putamen) for the purposes of PET receptor occupancy studies. The automatic extraction was done with the deformable surface models from PET binding potential images. The segmentation result of the baseline study was utilized for segmenting the intervention study. The methods were evaluated with Monte Carlo simulated phantom images, a test-retest dataset with 11C-raclopride, and two receptor occupancy datasets (11C-raclopride and 18F-SPA-RQ). With the test-retest dataset, the reproducibility (normalized absolute differences (NAD)) and reliability (intraclass correlation coefficient (ICC)) of binding potential values were assessed with the proposed methods designed specifically for receptor occupancy and compared with the manual segmentation. A similar comparison was also performed for the receptor occupancy estimates. In the test-retest dataset, the two automatic three-dimensional segmentation methods had mean NADs ranging from 2.44 to 5.11% and ICCs from 0.88 to 0.98. The corresponding values for the manual two-dimensional segmentation were 3.45 to 6.65% and 0.82 to 0.96 (NAD differences non-significant). The automatic methods also yielded similar receptor occupancy values to the manual segmentation method in a range of high and low occupancy values. We believe that the automated three-dimensional segmentation will be useful in enhancing the analysis of large-scale PET receptor occupancy studies with high-resolution scanners.