ORIGINAL ARTICLE Annals of Nuclear Medicine Vol.9, No.2, 75-80, 1995 Error analysis of table look-up method for cerebral blood flow measurement by 123I-IMP brain SPECT: Comparison with conventional microsphere model method Hiroshi ITO,* Kiyoshi ISHII,** Hiroto ATSUMI** Toshifumi KINOSHITA,** Ryuta KAWASHIMA,* Shuichi ONO,* Seiro YOSHIOKA,* Hidehiro IIDA,*** Kazuo UEMURA*** and Hiroshi FUKUDA* *Department of Nuclear Medicine and Radiology, Division of Brain Sciences, Institute of Development, Aging and Cancer, Tohoku University **Department of Radiology, Sendai City Hospital ***Department of Rediology and Nuclear Medicine, Research Institute for Brain and Blood Vessels, Akita While N-isopropyl-p-[123l]iodoamphetamine (IMP) is commonly used as a flow tracer, significant clearance from the brain causes underestimation of CBF as compared with true CBF when conventional microsphere model analysis is applied. We previously reported a simple "table look-up" method for CBF measurement using IMP taking into account this clearance effect. The method is based on a two-compartment model, the K1 (corresponding to CBF) and k2 constants being obtained from a table from the ratio of the I st SPECT (40 min) to the 2nd SPECT (180 min) counts. Arterial input data used were obtained by one point blood sampling 10 min after IMP infusion against the standard input function. In the present study, this approach was compared with conventional microsphere model analysis. For 30 subjects, the latter method entailed 8 min continuous arterial blood sampling after IMP infusion and the use of SPECT data at the end of this period, calibrated by a count ratio of 8 minl40 min planar images of whole brains. A good correlation was observed between the two methods (r = 0.88), but an overestimation of table look-up method CBF as compared with microsphere model CBF was observed contrary to theoretical predictions. Limitations in the estimation of SPECT data at 8 min, obtained with SPECT data at 40 min for calibration of the count ratio of 8 min/40 min whole brain planar images, might be responsible for this. Key words: IMP, SPECT, cerebral blood flow, table look-up method INTRODUCTION IODINE-123 (123I) labeled N-isopropyl-p-iodoamphetamine (IMP) is used as a cerebral blood flow (CBF) tracer for single photon emission computed tomography (SPECT) due to its large extraction fraction and high affinity for the brain.1,2 But, significant clearance from the brain causes change in IMP distribution3,4 and underestimation of CBF when a conventional microsphere model analysis5 is applied to prolonged data acquisition.6-9 We previously reported a "table look-up method," a new simple approach to measurement of CBF with IMP, taking into account this clearance effect.10-12 The approach is based on a two-compartment model (influx: K1, efflux: k2), in which Kl (taken to represent CBF) and k2 are obtained from a table read with the count ratio of first SPECT scan (mid-scan time: 40 min) to second SPECT scan (mid-scan time: 180 min). Arterial input function is obtained by calibrating against the standard input function from one point arterial blood sampling at 10 min after intravenous infusion of IMP. The purpose of the present study was to compare this method with the conventional microsphere model method13 for a major patient series. MATERIALS AND METHODS Subjects SPECT studies were performed on 30 subjects including 19 patients suffering from cerebral contusion, 3 with cerebrovascular disease, 2 with hypoxic brain, 2 with carbon monoxide toxicosis and 4 normal volunteers. None of the patients had any heart or lung disease and informed consent was obtained from all subjects after proper explanation of the study being conducted. SPECT study Two SPECT scans were performed, at 40 min and 180 min of mid-scan time, after intravenous infusion of 222 MBq IMP lasting 1 min. Fifty sec planar brain images were obtained 8 min after IMP infusion for the conventional microsphere model analysis (Fig.l). The SPECT scanner used was a Neurocam (Yokogawa Medical Systems Corp., Tokyo, Japan),14 equipped with a three-head rotating gamma camera. In-plane resolution was 9 mm full width at half maximum (FWHM), and axial resolution was 10 mm FWHM. The SPECT scan protocol acquired 64 projections at 50 sec per projection with 120' rotation of the camera. Reconstruction was performed by filtered backprojection using a Butterworth filter (cutoff frequency 0.45 cycle/cm, power factor 10). Attenuation correction was made numerically by assuming the object shape to be circular or elliptical and the attenuation coefficient to be uniform. Image slices were set up parallel to the orbitomeatal (OM) line and obtained at 8 mm intervals through the whole brain. One point arterial blood sampling from the brachial artery was performed at 10 min after IMP infusion. Radioactivity of the whole blood was measured with a well counter and was used for calibration against the standard input function to provide an arterial input function for the table look-up method. 10-12 Continuous arterial blood sampling at a constant rate from the brachial artery was also performed during the first 8 min after IMP infusion and octanol extracted radioactivity was measured for the conventional microsphere model analysis (Fig. 1). A cross calibration scan was performed using an elliptic cylindrical uniform phantom (long axis: 19 cm, short axis: 14 cm inner diameter) for calibrating the relative sensitivities of the SPECT scanner and the well counter system. Image analysis Regions-of-interest in the cerebellum, pons, thalamus, putamen, centrum semiovale and cerebral cortex including frontal, temporal, parietal and occipital lobes were outlined on the 40 and 180 min SPECT images. The shape of regions-of-interest was circular with a 35 mm diameter for the cerebellum, and elliptic with a short axes of 16-25 mm and long axes of 25-50 mm for other region. Theory Table look-up method10-l2. In this method, a two-compartment model was employed in line with previous reports.6,7,15 dCb(t)/dt = K1Ca(t) - k2ECb(t) (1) where Cb(t): concentration of radioactivity in the brain Ca(t): arterial input function K1: influx rate constant (ml/ml/min) k2: efflux rate constant (l/min) In this study, we assumed the first-pass extraction fraction of IMP to be equal to 1,2,16 and therefore, Kl equals CBF. The ratio of Kl to k2 is called the distribution volume of IMP In the brain (Vd (ml/ml)). Solving Eq. I provides: Cb(t) = KlECa(t) ~ e-k2t (2) where ~ denotes the convolution integral. For this method, two SPECT scans are performed. The model equation (Eq. 2) can therefore be expressed for each scan. Cb(tc) = K1ECa(tc) ~ e-k2te (3a) Cb(td) = K1ECa(td) ~ e-k2td (3b) where te and td are mid-scan times at first and second scans, respectively. Calculating the ratio of Eq. 3a to Eq. 3b gives: Cb(te)/Cb(td) =Ca(tc) ~ e-k2te/Ca(td) ~ e-k2td (4) For a given input function, C*(t), the radioactivity ratio of the first to second scans (the right side of Eq. 4) can be considered to tabulate as a function of k2. For a given radioactivity ratio of first to second scans, the table look-up procedure then provides a corresponding k2 value. By inserting this k2 Value into Eq. 3a or 3b, a K1 value that corresponds to CBF can be calculated. The arterial input function, Ca(t) is obtained by calibration against the standard input function by using the arterial blood radioactivity gained from the one point sampling. Microsphere model method13. CBF values were also calculated by microsphere model analysis as follows: f = Cb/int 8min 0min Ca(t)dt = Cb R/Ca (5) where f: CBF (ml/ml/min) Cb: concentration of radioactivity in the brain at 8 min after IMP infusion Ca(t): arterial input function R: constant arterial blood sampling rate (ml/min) Ca: total octanol extracted radioactivity of the blood withdrawn over 8 min In this study, Cb was obtained as follows: Cb = Cb(Planar8)/Cb(Planar40) .Cb(SPECT40) (6) where Cb(Planar8): the whole brain radioactivity of the planar image at 8 min after IMP infusion Cb(Planar40): the whole brain radioactivity of the planar image at 40 min which is one of the projections of SPECT scans Cb(SPECT40): the brain radioactivity concentration of the SPECT scan with the mid-scan time of 40 min after IMP infusion Simulation of CBF correlation between the two compartment model and the microsphere model For prediction of systematic underestimation of CBF by microsphere model analysis, a simulation of the correlation between CBF values evaluated by the two-compartment model analysis and those from microsphere model analysis was performed. CBF values from the microsphere model were calculated as follows: Firstly, the brain radioactivity curve, Cb(t) was generated for a CBF range of 0 to 100 ml/100 ml/min according to the two-compartment model equation (Eq. 2) where the Vd Values were 20, 30, 40 or 50. The standard input function used in the table look-up method was employed for the arterial input function, Ca(t). Secondly, for each calculated Cb(t), the microsphere model CBF values were calculated using Cb(t) at 8 min and integrated with Ca(t) for the time period [0,8 min] (Eq. 5). The resultant microsphere model CBF values were compared with those generated by two-compartment model analysis. Simulation of the effects ofgray-white matter mixture The limited spatial resolution of SPECT scanners causes gray-white matter mixture in regions-of-interest. The effects of gray-white matter mixing on CBF values calculated by the table look-up method were evaluated.n The heterogeneous tissue radioactivities at first and second SPECT scans were generated as mixtures of gray and white matter. CBF values of the gray and white matter were assumed to be 80 and 20 ml/100 ml/min, respectively. The Vd Value of gray and white matter was assumed to be the same as 30, 35, 40, 45 or 50 ml/ml. The difference between true CBF values (= 80 ml/100 ml/ min x gray matter fraction + 20 ml/100 ml/min x white matter fraction) and CBF values calculated by table look-up method with the generated heterogeneous tissue radioactivity were estimated where the fraction of gray matter per given region-of-interest varied from 0 to 100%. In this simulation, the arterial input function was the standard input function used for the table look-up method. Simulation of CBF correlation between the two-compartment model and the microsphere model method in which 8 min SPECTdata were obtained from the 40 min SPECT and 8 min/40 min whole brain ratio In this study, in the microsphere model method, the SPECT data at 8 min were obtained from count ratios of 8 minf40 min whole brain planar images and 40 min SPECT data. But the radioactivities of 40 min SPECT scans are non-1inear for CBF due to a significant clearance of IMP, and this could cause error. For this reason, a simulation of the CBF correlation between the twocompartment model analysis and the microsphere model method with count ratios of 8 minl40 min was also performed. Firstly, the brain radioactivity at 40 min, Cb(40 min) was generated for each CBF according to the two-compartment model equation (Eq. 2) with the standard input function as an arterial input function with the Vd value assumed to be 50 ml/ml. Secondly, for each count ratio of 8 minf40 min planar images, i.e., 0.6. 0.7, 0.8, or 0.9, CBF values were calculated by the microsphere model method (Eq. 5 and 6). The resultant microsphere model method CBF values were compared with those generated by two-compartment model analysis. RESULTS Figure 2 shows the simulation of the CBF correlation between the two-compartment model and the microsphere model. This indicated systematic underestimation of CBF values evaluated by the microsphere model analysis as compared with those from the two-compartment model analysis. The magnitude of CBF underestimation was thus expected to be 5.0% for two-compartment model CBF of 50 ml/100 ml/min and Vd of 50 ml/ml. Figure 3 shows the effects of gray-white matter mixing for CBF values calculated by the table look-up method. When the Vd Value for gray and white matter was 30 ml/ ml, calculated CBF values were systematically underestimated when the fraction of gray matter varied from 0 to 100%. But, when the Vd Value for gray and white matter was more than 40 ml/ml, calculated CBF values were systematically overestimated (9.8% for gray matter fraction of 50% and Vd of 50 ml/ml). Figure 4 shows the simulation of the CBF correlation between the two-compartment model and the microsphere model method in which 8 min SPECT data were obtained from the 40 min SPECT and 8 min/40 min whole brain count ratio. In this simulation, comparative overestimation of CBF values by the microsphere model method was observed in CBF ranges for the two-compartment model of 0.0-30.0, 0.0-45.0, 0.0-70.0 and 0.0-95.0 ml/100 ml/ min with 8 min/40 min planar image count ratio of 0.6, 0.7, 0.8 and 0.9, respectively. In this study, the actual mean count ratio of the 8 min/40 min planar image was 0.762 +- 0.072 (+- S.D.). A good correlation was obtained between CBF values evaluated by the table look-up method and those from the microsphere model method (Y = 0.92X + 7.34, X: table look-up method, r = 0.88) (Fig. 5), but overestimation of CBF values was observed with the microsphere model method as compared with those from the table look-up method. Mean CBF values evaluated by the table look-up method were 10.6% lower than those from the microsphere model method (table look-up method: mean CBF +- S.D. = 37.4 +- 8.09 ml/100 ml/min, microsphere model method: mean CBF +- S.D. = 41.8 +- 8.46 ml/100 ml/min). Vd Values were not uniform in the brain, especially low Vd Values being observed in lesions, i.e., cerebral infarctions and contusions. The mean Vd Value in X-ray CTnormal densityregions was 48.7 +- 9.15 ml/ml(+- S.D.). There was no significant difference between gray and white matter in Vd Values. DISCUSSION The microsphere model analysis has been routinely used as a method for measuring CBF using IMP, but under-estimation of CBF is caused by significant clearance of IMP from the brain, especially when data acquisition is prolonged.6-9 In this study, the simulation study similarly indicates systematic underestimation of CBF values with evaluation by microsphere model analysis as compared with those from the two-compartment model analysis even when data acquisition is limited to the early phase i,e., within 8 min after IMP infusion (5.0% underestimation for a two-compartment model CBF of 50 ml/lOOml/ min and Vd Of 50 ml/ml) (Fig. 2). The simulation of the effects of gray-white matter mixture also indicated differences between true CBF values (= 80 ml/100 ml/min x gray matter fraction + 20 ml/100 ml/min x white matter fraction) and table look-up method CBF values. In this study, the mean Vd Value was 48.7 +- 9.15 ml/ml (+- S.D.) for normal regions on X-ray CT. When the Vd Value of gray and white matter was more than 40 ml/ml, the table look-up method CBF was systematically overestimated (9.8% for a gray matter fraction of 50% and a Vd Of 50 ml/ml) (Fig. 3). On the other hand, there were no effects of gray-white matter mixing on CBF values calculated by the microsphere model analysis, because the correlation between the brain radioactivity and CBF value is linear in the microsphere model (Eq. 5),17 A good correlation was obtained between CBF values evaluated by the table look-up method and those from the conventional microsphere model method (Fig. 5), suggesting equivalent applicability, but while the two simulation studies (Figs. 2 and 3) indicated that CBF values obtained from the table look-up method would be higher than those from microsphere model analysis, in fact the opposite was the case. The microsphere model method values were thus actually 10.6% higher than the table look-up method CBF values (Fig. 5). As reasons for this, a number of factors must be considered. One possibility is error in estimating the SPECT brain counts at 8 min in the microsphere model method with count ratios 8 min/40 min whole brain planar images and 40 min SPECT data.18 This error would be caused by non-linearity of 40 min SPECT radioactivities due to a significant clearance of IMP. The simulation study (Fig. 4) revealed comparative overestimation of CBF values by the microsphere model method in CBF ranges for the two-compartment model of 0.0-30.0, 0.0-45.0, 0.0-70.0 and 0.0-95.0 ml/100 ml/min with 8 min/40 min planar image count ratios of 0.6, 0.7, 0.8 and 0.9, respectively. In this study, the actual mean count ratio of the 8 min/40 min planar image was 0.762 +- 0.072 (+- S.D.), and therefore this could have been responsible for the overestimation of CBF values determined by the microsphere model method. In addition, other unknown errors due to radioactivities from extracerebral arteries included in 8 and 40 min planar images could have played roles. Another potential source of error is in the determination of the arterial input function. For accurate CBF measurement, accurate determination of this function including corrections for time delay and dispersion of input is required. It has been shown in the H215O PET studies that no correction for these is associated with greater overestimation of CBF when the scan duration is shorter.19-21 The standard input function used in table look-up method does not feature these corrections, because errors from time delay and dispersion would not be significant due to the sufficient delay until the mid-scan time of the two SPECT scans, i.e., 40 and 180 min,10-12 but these errors in the microsphere model analysis case will be more significant, because the scan time was very early at 8 min. This could have directly caused the overestimation of CBF values determined by the microsphere model analysis. With the table look-up method, on the other hand, there might have been unknown errors due to difference in the arterial input curve shape for each subject. In conclusion, a relatively good correlation was obtained between CBF values gained by table look-up method and those from the conventional microsphere model method. Since the table look-up method is simple, and does not require a continuous arterial blood sampling, it can be recommended for routine application. Possible reasons for the contrast to expectations from theoretical considerations, higher CBF values from the microsphere model method than the table look-up method are : 1 . Errors in estimation of SPECT data at 8 min by calibration of SPECT scan at 40 min with count ratios of 8 min/40 min whole brain planar images in the microsphere model method. 2. Errors in determination of arterial input function with both methods. ACKNOWLEDGMENTS We are greatly indebted to the staff of Sendai City Hospital and the Institute of Development, Aging and Cancer. Tohoku University, particularly Messrs. Yoshimasa Inukai, Shigeto Abe, Masami Sato for operating the SPECT scanner. This study was supported by a Grant-in-Aid No. 05454297 for Scientific Research from the Japanese Ministry of Education, Science and Culture. REFERENCES l. Winchell HS, Baldwin RM, Lin TH. Development of I-123labeled amines for brain studies: localization of I- 1 23 iodophenylalkyl amines in rat brain. J Nucl Med 21 : 940-946, 1980. 2. Winchell HS, Horst WD, Braun L, OldendorfWH, Hattner R, Parker H. N-isopropyl-[123I]p-iodoamphetamine: single-pass brain uptake and washout; binding to brain synaptosomes; and localization in dog and monkey brain. J Nucl Med 21: 947-952, 1980. 3. Creutzig H, Schober O, Gielow P, Friedrich R, Becker H, Dietz H, et al. Cerebral dynamics of N-isopropyl-(1231)piodoamphetamine. J Nucl Med 27: 178-183, 1986. 4. Nishizawa S, Tanada S, Yonekura Y, Fujita T, Mukai T, Saji H, et al. Regional dynamics of N-isopropyl-(123I)piodoamphetamine in human brain. J Nucl Med 30: 150-156, 1989. 5. Kuhl DE, Barrio JR, Huang SC, Selin C, Ackermann RF, Lear JL, et al. Quantifying local cerebral blood flow by Nisopropyl-p-[123I]iodoamphetamine (IMP) tomography. J Nucl Med 23: 196-203, 1982. 6. Greenberg JH, Kushner M, Rango M, Alavi A, Reivich M. Validation studies of iodine-123-iodoamphetamine as a cerebral blood flow tracer using emission tomography. J Nucl Med 31 : 1364-1369, 1990. 7. Murase K, Tanada S, Mogami H, Kawamura M, Miyagawa M, Yamada M, et al. Validation of microsphere model in cerebral blood flow measurement using N-isopropyl-p(123I)iodoamphetamine. Med Phys 17: 79-83, 1990. 8. Yokoi T, Iida H, Itoh H, Kanno I. A new graphic plot analysis for cerebral blood flow and partition coefficient with iodine-123-Iodoamphetamine and dynamic SPECT validation studies using oxygen-15-water and PET. J Nucl Med 34: 498-505, 1993. 9. Yonekura Y, Nishizawa S, Mukai T. Iwasaki Y, Fukuyama H, Ishikawa M, et al. Functional mapping of flow and backdiffusion rate of N-isopropyl-p-iodoamphetamine in human brain. J Nucl Med 34: 839-844, 1993. 10. Iida H, Itoh H, Bloomfield PM, Munaka M, Higano S, Murakami M, et al. A methcfd to quantitate cerebral blood flow using a rotating gamma camera and iodine-123-iodoamphetamine with one blood sampling. Eur J Nucl Med 21 : 1072-1084, 1994. 11. Iida H, Itoh H, MunakaM, Murakami M, Higano S, Uemura K. A clinical method to quantitate CBF using a rotating gamma camera and I-123-amphetamine (IMP) with one blood sampling. J Nucl Med 33: 963P, 1992. 12. Itoh H, Iida H, Murakami M, Bloomfield PM, Miura S, Okudera T, et al. A method for measurement of regional cerebral blood flow using N-isopropyl-p-[123I]iodoamphetamine (123I-IMP) SPECT; two scans with one point blood sampling technique. KAKU I GAKU ( Jpn J Nucl Med) 29: I 193-1200, 1992. 13. Matsuda H, Seki H, Sumiya H, Tsuji S, Tonami N, Hisada K, et al. Quantitative cerebral blood flow measurements using N-isopropyl-(iodine 123)p-iodoamphetamine and single photon emission computed tomography with rotating gamma camera. Am J Physiol Imag I : 186-194, 1986. 14. Kouris K, Jarritt PH, Costa DC, Ell PJ. Physical assessment of the GE/CGR Neurocam and comparison with a single rotating gamma-camera. Eur J Nucl Med 19: 236-242, 1992. 15. Itoh H, Iida H, Bloomfield PM, Murakami M, Higano S, Munaka M, et al. A technique for rapid imaging of regional CBF and partition coefficient using dynamic SPECT and I-123-amphetamine (IMP). J Nucl Med 33: P911, 1992. 16. Murase K, Tanada S, Inoue T. Ochi K, Fujita H, Sakaki S, et al. Measurement of the blood-brain barrier permeability of I-123 IMP, Tc-99m HMPAO and Tc-99m ECD in the human brain using compartment model analysis and dynamic SPECT. J Nucl Med 32: P911, 1991. 17. Huang SC, Mahoney DK, Phelps ME. Quantitation in positron emission tomography: 8. Effects of nonlinear parameter estimation on functional images. J Comput Assist Tomogr 11 : 314-325, 1987. l8. Takahashi N, Ohkubo M, Odano I, Sakai K. A problem of quantitative measurement of regional cerebral blood flow using microsphere model and N-isopropyl-p-[123I]iodoamphetamine (IMP): Comparison with 133Xe SPECT and sequential dynamic 123I-IMP SPECT. KAKUIGAKU (Jpn J Nucl Med) 31 : 319-326, 1994. 19. Iida H, Kanno I, Miura S, Murakami M, Takahashi K, Uemura K. Error analysis of a quantitative cerebral blood flow measurement using H215O autoradiography and positron emission tomography, with respect to the dispersion of the input function. J Cereb Blood Flow Metab 6: 536-545, 1986. 20. Iida H, Higano S, Tomura N, Shishido F, Kanno I, Miura S, et al. Evaluation of regional differences of tracer appearance time in cerebral tissues using [150]water and dynamic positron emission tomography. J Cereb Blood Flow Metab 8: 285-288, 1988. 21. Iida H, Kanno I, Miura S, Murakami M, Takahashi K, Uemura K. A determination of the regional brain/blood partition coefficient of water using dynamic positron emission tomography. J Cereb Blood Flow Metab 9: 874-885, 1989.