TMI-2013-0201.R1 1 Abstract— In the standard software for the Siemens high resolution research tomograph (HRRT) positron emission tomography (PET) scanner the most commonly used segmentation in the μ-map reconstruction for human brain scans is maximum a posteriori for transmission (MAP-TR). Bias in the lower cerebellum and pons in HRRT brain images have been reported. The two main sources of the problem with MAP-TR are poor bone / soft tissue segmentation below the brain and overestimation of bone mass in the skull. Method: We developed the new transmission processing with total variation (TXTV) method that introduces scatter correction in the μ-map reconstruction and total variation filtering to the transmission processing. Results: Comparing MAP-TR and the new TXTV with gold standard CT-based attenuation correction, we found that TXTV has less bias as compared to MAP-TR. We also compared images acquired at the HRRT scanner using TXTV to the GE Advance scanner images and found high quantitative correspondence. TXTV has been used to reconstruct more than 4000 HRRT scans at 7 different sites with no reports of biases. Conclusion: TXTV-based reconstruction is recommended for human brain scans on the HRRT. Index Terms— HRRT, attenuation correction, transmission scatter correction, PET, total variation I. INTRODUCTION HE high resolution research tomograph (HRRT, CTI/Siemens) is a positron emission tomography (PET) scanner dedicated to human brain scanning [1], [2], [3]. It has resolution of 1.4 mm using point spread function modeling (PSF) [4], which is 2-3 times better than other human PET scanners (typically PET/CT) [5] (see Fig. 1A-B). This high resolution makes it ideal for neurology studies. In the image reconstruction of PET emission scans (EM) one has to do attenuation correction (AC) to account for the Copyright (c) 2010 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to pubs-permissions@ieee.org. S. H. Keller is with the Department of Clinical Physiology, Nuclear Medicine and PET, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark (sune@pet.rh.dk ). C. Svarer is with the Neurobiology Research Unit, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark (csvarer@nru.dk ). M. Sibomana was with the Department of Clinical Physiology, Nuclear Medicine and PET, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark. He is now with Sibomana Consulting, Emines, Belgium (sibomana@gmail.com ). PET signal loss in matter using a so-called μ-map. Today most PET scanners in use are PET/CTs, and the μ-map is derived from CT scans by a mapping from the CT-energy range to the 511 keV of the photons detected in PET. For PET scanners without a CT such as the HRRT, a transmission scan (TX) is performed by rotating a radioactive transmission source with an energy of, or close to, 511keV around the object in the scanner and from that a μ-map for attenuation correction is created. The HRRT scanner software offers 3 choices for transmission scan processing (image reconstruction and possible post processing) of which maximum a posteriori for transmission (MAP-TR) [6], [7], [8] is most widely used. MAP-TR was shown by van Velden et al. [9] to provide the most accurate μ-maps for human brains amongst the methods available for the HRRT, but the paper also showed MAP-TR to have shortcomings as it was also the case with the works by Anton-Rodriguez et al. [10] and Son et al. [11]. The major issues with MAP-TR are overestimation of the skull attenuation, inaccurate segmentation of bone structures at the base of the skull, and the use of the water attenuation correction factor (ACF or μ-value) for brain tissue (0.096 cm -1 instead of 0.099 cm -1 [11], [12]). These inaccuracies in the μ- maps will lead to incorrect quantification in the PET emission images. Such a bias was reported for the cerebellum, pons and brain stem by several HRRT users as documented in May 2007 by Koeppe [13], suspecting scatter or attenuation correction to be the source of the bias (12% lower pons/global mean ratios for the HRRT as compared to all other scanners included in the project). The suspicion was later directed at the attenuation correction alone through discussion and investigations amongst HRRT users, but with potential problems with using MAP-TR in scatter correction as documented in [10]. Quantitatively correct images are essential, especially when used for research purposes, e.g. in neuroreceptor imaging with quantification of binding potentials. The objective of our work as presented in this paper was to improve the quantification of HRRT emission images by providing better μ-maps than what was available in the standard HRRT software. Thus we developed the transmission processing with total variation (TXTV) method. Its evolution is two-fold: Firstly, it introduces the use of scatter correction in the iterative reconstruction of the transmission image, and Attenuation correction for the HRRT PET- scanner using transmission scatter correction and total variation regularization Sune H. Keller, Claus Svarer and Merence Sibomana T