Abstract--This study investigates different detector normalization procedures for a small animal scanner, specifically the Concorde microPET ® R4. The procedures were compared in terms of: (i) image uniformity, (ii) performance as a function of count rate, and (iii) impact on the estimate of the binding potential (BP) in brain rat studies. Image uniformity studies were performed on two cylindrical phantoms of different size filled with an aqueous concentration of 11 C (38 kBq/mL and 615 kBq/mL). BP was estimated with the Logan graphical approach on 12 11 C-Methylphenidate and 9 11 C- Dihydrotetrabenazine rat studies processed with all the normalization procedures. Results/Conclusion: We found that: 1) the combination, the geometry normalized combination, the component-point, and the geometry normalized component procedures significantly improve radial image uniformity compared to the direct-point and direct-cylinder procedures, 2) the geometry normalized combination procedure seems to provide the best radial and axial uniformity, 3) a mismatch between the count rates at which the normalization and the emission scans are acquired degrades the axial uniformity by 47% to 98% whereas this effect was not observed for radial uniformity, and 4) the difference in BP values obtained from data corrected with different normalization procedures is as high as 15% for normal striatum and 75% for lesion striatum. I. INTRODUCTION Detector normalization factors in PET are required to correct for non-uniform detector efficiency response. The first normalization method developed for 2D PET scanners was the direct inversion method. In direct inversion the elements of the normalization correction factors are obtained by scanning a uniform source (typically rod sources or a uniform cylinder) centrally placed in the Manuscript received October 4, 2004. This work was supported in part by NSERC, MSFHR and the Triumf Life science grant. M.-L. Camborde, V. Sossi, A. Rhamim and E. Vandervoort are with the University of British Columbia, Vancouver, BC V6T 1Z1, Canada (email: marie@pet.ubc.ca; vesna@physics.ubc.ca; rhamim@physics.ubc.ca; evander@physics.ubc.ca) D. F. Newport and S. Siegel are with CTI Concorde Microsystems LLC, Knoxville, TN 37932, USA (email: dnewport@cms-asic.com ; ssiegel@cms-asic.com) K. R. Buckley and T. Ruth are with TRIUMF, Vancouver, BC V6T 2A3, Canada (email: truth@triumf.ca, buckley@triumf.ca) scanner field of view (FOV) and normalizing the number of events in each line of response (LOR) to a global average value. With the development of 3D PET scanners, the requirement to include oblique LORs increased the number of LORs by an order of magnitude and in order to achieve data of the statistical quality necessary to obtain accurate normalization factors very long acquisition times were required [1]. The next normalization methods to be introduced were indirect or component-based methods. These methods entail expressing the normalization correction factors as a product of a geometric and an intrinsic efficiency component. These basic components were first proposed by Hoffman et al. [2]. The efficiency factors account for the non-uniformity of individual detector response (which may vary over time) while the set of fixed correction factors is related to the geometry of the scanner. These factors are then combined to form the overall normalization factors that are applied to the data typically prior to reconstruction [3]. The performance of the normalization methods used in our study has been thoroughly studied on human scanners [4]. In general, approaches adopted for 3D human PET scanners are either (i) direct inversion methods with a 3D scan of a uniform cylinder or a rotating line source [4-5] or (ii) component- based methods using a 2D [6] or a 3D rotating rod sources scan to calculate the geometrical factors and a 3D scan of a uniform cylinder, a rotating uniform plane, or moving line source to calculate the efficiency factors [4,7-9]. Furthermore 3D PET suffers high frequency image artifacts when there is a count rate mismatch between the normalization and the emission acquisitions. This effect is due to event mispositioning when using detector blocks as a consequence of pulse pile up as shown by Germano et al. [10]. Possible solutions to this problem have been addressed by Badawi and Marsden [11,12]. The goal of the present study is to evaluate normalization procedures for an animal scanner, specifically the Concorde microPET ® R4. Characteristics and performance evaluations of this scanner have been previously described [8,13], but no optimal normalization procedures have yet been investigated. Briefly, the system consists of 24 detector modules arranged Effect of normalization method on image uniformity and binding potential estimates on microPET ® Marie-Laure Camborde, Arman Rhamim, Danny F. Newport, Stefan Siegel, Ken R. Buckley, Eric Vandervoort, Thomas J. Ruth and Vesna Sossi, Member, IEEE 0-7803-8701-5/04/$20.00 (C) 2004 IEEE 0-7803-8700-7/04/$20.00 (C) 2004 IEEE 3467