Analysis of Dynamic SPECT/CT Measurements of the Arterial Input Function in Human Subjects Celeste D. Winant, Yuval R. Zelnik, Bryan W. Reutter, Senior Member, IEEE, Arkadiusz Sitek, Member, IEEE, Steven L. Bacharach, Grant T. Gullberg, Fellow, IEEE, Carina Mari Aparici Abstract– Measurement of the arterial input function (AIF) is essential to deriving quantitative estimates of regional myocardial blood flow using kinetic models. Accurate measurements have been possible with a wide range of radiotracers in both research and clinical PET/CT imaging. However, accurate measurements of the AIF with dynamic SPECT or SPECT/CT have posed various challenges; foremost being that imaging a rapidly evolving radiotracer distribution with a slowly-rotating single- or dual-head SPECT scanner yields temporally inconsistent projection data. A method is developed for quantifying the AIF in human subjects from dynamic SPECT/CT measurements of 99m Tc-tetrofosmin concentration in the left atrium imaged with a Philips Precedence SPECT/CT scanner. A 2-minute infusion is imaged in a series of eight back- to-back 180-degree continuous-mode acquisitions (or rotations), with the dual camera heads. In each acquisition a set of 36 projections (128 × 128 pixels of dimension 3.19 mm × 3.19 mm) is acquired in each rotation each over a time span of 54 seconds yielding a total acquisition time of 432 seconds. The AIF is computed using both traditional image-based analysis and full spatiotemporal image reconstruction methods (referred to as 4D recon). The errors induced by data inconsistency are evaluated by two approaches. The first method derives SPECT-like dynamic (inconsistent) projection data from selected forward projections, chosen from modeled SPECT acquisitions, of existing dynamic 94 Tc-MIBI PET images. The second validation method uses a database of SPECT measurements of an anthropomorphic phantom to generate SPECT-like projections. The first validation study using a two minute infusion showed very little bias in the time-activity curves estimated from the simulated dynamic cardiac SPECT patient study; whereas, the second validation study using a one minute infusion showed considerable more bias in the estimated time-activity curves and parametric parameters. We believe that this is the result of selecting non optimum basis functions. I. INTRODUCTION EASUREMENT of the arterial input function (AIF) is essential to deriving quantitative estimates of regional Manuscript received November 13, 2009. This work was supported in part by the University of California Discovery Grant Program (Contract DIG06- 2110-UC) and by Philips Medical Systems and by the Director, Office of Science, Office of Biological and Environmental Research, Medical Science Division of the U.S. Department of Energy under Contract No. DE-AC02- 05CH11231. C. D. Winant, S. L. Bacharach and C. Mari Aparici are with the University of California, San Francisco Department of Radiology and Biomedical Imaging, San Francisco, CA 94107 USA (telephone: 415-353-9441, e-mail: cwinant@radiology.ucsf.edu). B. W. Reutter and G. T. Gullberg are with the Lawrence Berkeley National Laboratory (LBNL), Department of Radiotracer Development & Imaging Technology, Berkeley, CA 94720 USA. Y. R. Zelnik was with LBNL and is now with the Hebrew University, Edmond J. Safra, Jerusalem 91904, Israel. A. Sitek is with the Department of Radiology at the Harvard Medical School and with the Brigham and Women’s Hospital, Boston, MA 02115 USA. myocardial blood flow using kinetic models. Accurate measurements have been possible with a wide range of radiotracers in both research and clinical PET/CT imaging. However, accurate measurements of the AIF with dynamic SPECT have posed various challenges; foremost being that imaging a rapidly evolving radiotracer distribution with a slowly-rotating single- or dual-head SPECT scanner yields temporally inconsistent projection data. Thesis: The implementation of spatiotemporal reconstruction methods that are able to reconstruct both spatial and temporal basis functions from projection data can better improve the bias due to inconsistent projection measurements. Most image reconstruction algorithms operate on the assumption of temporally consistent projection data. Non- negligible artifacts arise from relatively low count rates, increased attenuation, and greater collimation-induced blurring in SPECT images compared to PET. In addition, the reconstruction of inconsistent data due to temporal variations in the uptake of the radiotracer can also present significant artifacts if the time variation in the data is not taken into consideration. In the following, we estimate the AIF in human subjects from dynamic SPECT/CT acquisitions of 99m Tc-tetrofosmin concentration in the left atrium imaged with a commercial SPECT/CT scanner (Precedence, Philips Healthcare, Andover, MA). We compute the AIF with both traditional image-based analysis and with spatiotemporal reconstruction methods (4D reconstruction methods) that take the temporal variation of the projections into account. Spatiotemporal reconstruction methods [1-3] fit the time-varying activity in each of the image voxels to a linear combination of three quadratic B- spline basis functions that have the capability to model the expected range of physiologic kinetics of the tracer distribution in the subject. II. METHODS A. Patient Study Subjects were imaged at rest with a Philips Precedence SPECT/CT dual-headed scanner. An injection of 10 to 25- mCi injection of 99m Tc-tetrofosmin was administered while the patient was in the scanner, and was delivered in a slow 2- minute infusion using a Harvard ADS 975 pump. Imaging began at the start of infusion, using a series of eight to ten back-to-back 180-degree continuous-mode acquisitions (54 seconds for each rotation), with the heads at 90 degrees. A set of 36 projections (128 × 128 pixels of dimension 3.19 mm × 3.19 mm pixels) was acquired during each rotation yielding a total acquisition time of 432 - 540 seconds, for the dynamic portion of the study. A set of early projection images is shown M 2009 IEEE Nuclear Science Symposium Conference Record M09-338 U.S. Government work not protected by U.S. copyright 3404