Page 1

Journal of Cerebral Blood Flow and Metabolism

16:1288-1299 © 1996 The International Society of Cerebral Blood Flow and Metabolism

Published by Lippincott-Raven Publishers, Philadelphia

Kinetic Evaluation of [11 C]Dihydrotetrabenazine by Dynamic

PET: Measurement of Vesicular Monoamine Transporter

R. A. Koeppe, K. A. Frey, T. M. Vander Borght, A. Karlamangla, D. M. Jewett, L. C. Lee,

M. R. Kilbourn, and D. E. Kuhl

Division of Nuclear Medicine, Department of Internal Medicine, University of Michigan, Ann Arbor, Michigan, U.S.A.

Summary: (+)-et-C IC]Dihydrotetrabenazine (DTBZ) binds to

the vesicular monoamine transporter (VMAT2) located in pre

synaptic vesicles. The purpose of this work was to evaluate

various model configurations for analysis of [II C]DTBZ with

the aim of providing the optimal measure of monoamine ve

sicular transporter density obtainable from a single dynamic

PET study. PET studies on seven young normal volunteer sub

jects. ages 20-35. were pelformed following i.v. injection of

666 ± 37 MBq (18 ± I mCi) of (+)-et-C IC]DTBZ. Dynamic

acquisition consisted of a IS-frame sequence over I h. Analysis

methods included both creation of pixel-by-pixel functional

images of transport (KI) and binding (DVtot) and nonlinear

least-squares analysis of volume-of-interest data. Pixel-by

pixel calculations were performed for both two-compartment

weighted integral calculations and slope-intercept estimations

from Logan plots. Nonlinear least-squares analysis was per

formed applying model configurations with both two

compartments, estimating Kj and DV,ot> and three compart

ments, estimating KI-k4• For the more complex configuration,

The assessment of cerebral biochemistry in living per

sons via the use of radiolabeled tracers and positron

emission tomography (PET) has been a widely explored

field over the past decade. The purpose of this study is

to characterize the in vivo kinetic behavior of (+ )-a

e ICJdihydrotetrabenazine (DTBZ) binding to the ve

sicular monoamine transporter (VMA T2). The VMA T2

binding site is a specific protein located exclusively in

the membranes of presynaptic vesicles (Henry and

Scherman, 1989) and radioligands for VMAT2 have

been proposed as both in vitro (Scherman et aI., 1988;

Near, 1986; Masuo et aI., 1990) and noninvasive in vivo

Received December 18, 1995; final revision received March 14,

1996; accepted March 14, 1996.

Address correspondence and reprint requests to Robert A. Koeppe at

Division of Nuclear Medicine, University of Michigan Medical School,

3480 Kresge III, Box 0552, Ann Arbor MI 48109, U.S.A.

Abbreviations used: DTBZ, dehydrotetrabenazine; PET, positron

emission tomography; VMA T2, vesicular monoamine transporter

type 2.

1288

we examined the stability of various binding-related parameters

including k3 (konBmax'l, k31k4 (Bmax'IKd), DV,p [(K/k2)(k3Ik4)],

and DVtot [K1Ik2(1 + kik4)]. The three-compartment model

provided significantly improved goodness-of-fit compared to

the two-compartment model, yet did not increase the uncer

tainty in the estimate of the DVtot. Without constraining pa

rameters in the three-compartment model fits, DVtot was found

to provide a more stable estimate of binding density than either

k3, k31k4, or DVw The two-compartment least-squares analysis

yielded approximately 10% underestimations of the total dis

tribution. However, this bias was found to be very consistent

from region to region as well as across subjects as indicated by

the correlation between two- and three-compartment DV,ot es

timates of 0.997. We conclude that (+ )-et-C IC]DTBZ and PET

can provide excellent measures of VMAT2 density in the hu

man brain. Key Words: C IC]Dihydrotetrabenazine-Tracer

kinetics-VMA T2-Monoamine vesicular transporter

Positron emission tomography.

(DaSilva and Kilbourn, 1992) markers of nigrostriatal

neurons. Prior studies in our laboratories have demon

strated a regional in vivo brain binding of such radioli

gands, which correlate with the known distribution of

VMA T2 in rodent brain (Kilbourne, 1994; Kilbourn et

aI., 1995), and the linear relationship between the level of

in vitro striatal radioligand binding to VMAT2 and the

extent of nigral injury in 6-hydroxydopamine-Iesioned

rats (Vander Borght et aI., 1995). DTBZ exhibits high

affinity (low nanomolar) binding only to VMAT2, while

animal studies have shown a lack of regulation of this

transporter by repeated or chronic dopaminergic or cho

linergic drug treatments (Naudon et aI., 1994; Vander

Borght et aI., 1995; Wilson and Kish, 1996). This is in

contrast to other aspects of the dopaminergic nerve ter

minal, including dopamine synthesis and the neuronal

membrane dopamine transporter, where drug-induced

regulation of the enzyme (Zhu et aI., 1992; Hadjicon

stantinou et aI., 1993; Gjedde et aI., 1993) or transporter

(Ikegami and Prasad, 1990; Kilbourn et aI., 1992; Sharpe

Page 2

ANALYSIS OF [llCJDTBZ KINETICS

1289

et al., 1991; Wiener et aI., 1989; Wilson et aI., 1994) has

been clearly demonstrated. While DTBZ binding is not

specific to dopaminergic nerve terminals, as the radioli

gand binds to the vesicular transporter common for all

monoaminergic neurons, the PET signal measured in the

striatum largely represents storage vesicles in the pre

dominant (>95%) dopaminergic terminals (Kish et aI.,

1992). DTBZ binding to the VMAT2 is thus complimen

tary to but distinctly different from previous PET and

SPECT radiotracers that have been proposed for the

study of nigrostriatal pathology, including those that fol

low dopamine synthesis and storage, such as CSF]fluo

rodopa (Garnett et aI., 1983; Gjedde et aI., 1991; Huang

et aI., 1991) and CSF]fluoro-m-tyrosine (Dejesus et aI.,

1995), or those that bind to the neuronal membrane do

pamine transporter, such as [l1C]nomifensine (Aquilo

nius et aI., 1987, Salmon et aI., 1990), CIC]WIN 35,428

(Frost et aI., 1993), [IsF]GBR 12909 (Koeppe et aI.,

1990), and e31][3-CIT (Innis et aI., 1993; Laruelle et aI.,

1994).

In this work, we present an evaluation of kinetic analy

sis approaches for the estimation of a VMA T2 density

index following a single radioligand injection. Each ap

proach provides separate estimates for parameters repre

senting the tracer's blood-brain barrier (BBB) transport

rate and tissue distribution volume. We examined both

pixel-by-pixel estimation procedures capable of produc

ing functional images of the transport and binding pa

rameters and volume-of-interest (VOl) approaches using

two- and three-compartment model configurations esti

mating two, three, or four rate parameters as well as

accounting for the contribution of bloodborne radioac

tivity in the total PET signal. We compared the means

and coefficients of variation (COV

termine the appropriate trade-off between bias and pre

cision in the parameter estimates. Computer simulation

studies were performed in conjunction with the human

studies to examine the magnitude of bias in the param

eter estimates caused by various model simplifications or

assumptions.

= SD/mean) to de

THEORY

Conceptually, the kinetic analysis of [llC]DTBZ

time-activity distribution begins with a model consisting

of one blood compartment representing free ligand in

arterial plasma (Cp) and three tissue compartments rep

resenting free ligand in tissue (Cf), nonspecifically bound

ligand (Cn,), and specifically bound ligand (C,p)' The

first simplifying assumption that has been made through

out this work is that the free and nonspecific binding

equilibrate rapidly and have been combined into a single

compartment (Cf+ns)' This three-compartment configura

tion has rate parameters defined as follows:

k2' = K/(DVr+ns) = (K/DVf)/(l + DVn)DVf)

k3' = konBmax'/(I + DVn)DVr)

k4 = koff

and thus,

k3' Ik4 = (Bmax'IKd)/(l + DVn/DVf)

(min-I)

(min-I)

(min-I)

(1)

wherefis mass specific blood flow (ml g-I min-I), Eo is

the single pass extraction fraction of ligand across the

BBB into brain, PS is the permeability surface area prod

uct (ml g-I min-I), DVf' DVns, DVr+ns are the tissue

distribution volumes of free ligand, nonspecifically

bound ligand, and their sum, respectively (ml g-l), kon is

the bimolecular association rate between ligand and re

ceptor (g pmol-1 min-I), Bmax' is the binding site density

or concentration of unoccupied transporter binding sites

(pmol g-I), koff is the dissociation rate of ligand from the

binding site complex (min-I), and Kd is the equilibrium

binding constant for the specific binding site (pmol g-l

or nM). The "prime" symbols on k2' and k3' are used to

differentiate these rate parameters from the true k2 and

k3, which describe rates of exchange from the compart

ment containing only free ligand. The ratio of binding

parameters Bmax'1 Kd has been previously referred to as

the "binding potential" by Mintun et al. (1984). The

term (l + DVn/DVf) reflects the apparent increase in

volume of the binding precursor pool when combining

free and nonspecific compartments. Thus, 1/(1 + DVn/

DVf) describes the fraction of radiolabel available either

for transport back to plasma or binding to transporter

sites and is equivalent to the termf used by Mintun et aI.

(1984) and others. The distribution volumes of free +

nonspecific, specific binding sites, and total are given by

the following equations:

DVf+ns = (K/k2')

DV,p

= (K/k2) X (kik4)

= DVf X (Bmax'IKd)

DVtot

= DVr+ns + DVsp

= DVr+ns + DVf X (Bmax'IKd)

= (K/k2') X (k3'lk4)

= (KIlk2') X 0+ k3' Ik4)

(2)

Because of the inherent difficulty in estimating indi

vidual rate parameters with high accuracy, particularly

when the k3 is large relative to k2 (Koeppe, 1990; Koeppe

et aI., 1991; Frey et aI., 1992) or when k4 is small

(Koeppe, 1990), and because of practical considerations

necessitating that studies must be able to be implemented

and performed routinely in human subjects, our group

has intentionally pursued radioligands that exhibit both

rapid binding and dissociation. The kinetic modeling of

such ligands can often be simplified from the model

described above to a model that combines all tissue com-

J Cereb Blood Flow Metab, Vol. 16, No.6, 1996

Page 3

1290

R. A. KOEPPE ET AL.

partments into a single compartment. This simplification

is valid only when the binding and release of ligand from

the specific binding sites is rapid compared to the trans

port parameters K] and k2'. The meaning of the transport

parameter K] is unchanged; however, the clearance pa

rameter k2" becomes:

k2" = k2' 1(1 + k3' Ik4)

= KJDVtot

and thus,

DVtot = (K/k2' ) X (1 + k3'lk4)

= DVp.ns+DVf X (Bmax'IKd)

(3)

where DVtot is the total distribution volume of the

summed tissue compartments and has the same theoret

ical definition as DVtot derived from the three-compart

ment approach. How closely the DVtot estimates from the

two different model configurations agree will depend

primarily on how rapidly the free and bound compart

ments approach equilibrium (i.e., how valid the simpli

fying assumption is).

The primary goal of this study is to provide a reliable

measure of vesicular monoamine transporter binding site

density. In the ideal, this would be a direct quantitative

measure of Bmax', the density of available binding sites;

however, for practical reasons this may not be possible.

As defined in Eqs. 1-3, there exist other parameters that

relate to binding density that might provide more reliable

measures of binding density. These include k3

( = konBrnax'), kik4 ( = Brnax'IKd), DV,p [= DVf x (Bmax' 1

Kct)), and DVtot [DVf+ns + DVf x (Brnax'IKd)]. The distri

bution volume estimates, particularly DVtol, are functions

of more than just binding site density and therefore con

tain additional intrinsic bias compared to either k3 or

kik4. However, the precision of distribution volume es

timates is typically much better than those of individual

rate constants. It is the focus of this work to determine

which measure yields the best trade-off between bias and

precision and the greatest sensitivity to actual variations

in binding site density.

MATERIALS AND METHODS

Seven young nonnal volunteers, 20-35 years of age, were

studied following administration of 666 ± 37 MBq (18 ± I

mCi) of (+)-cY-[IIClDTBZ. No-carrier-added C IClDTBZ

(500-2,000 CiJmmol) was prepared as reported by Kilbourn

(1995). A sequence of 15 PET scans (4 x 30 s, 3 x I min, 2 x

2.5 min, 2 x 5 min, 4 x 10 min) was acquired on a Siemens/CTI

ECAT EXACT-47 scanner (Knoxville, TN, U.S.A.) for 60 min

following i.v. injection. Data were reconstructed using a Han

ning filter with 0.5 (cycles/projection ray) cutoff. Calculated

attenuation correction was perfonned on all images.

Blood samples were withdrawn via a radial artery catheter as

rapidly as possible for the first 2 min of the scan and then at

progressively longer intervals for the remainder of the study.

Plasma was separated from red cells by centrifugation, and

J Cereb Blood Flow Metab, Vol. 16, No.6, 1996

counted in a Nal well-counter. The plasma radioactivity time

course was corrected for radiolabeled metabolites using a rapid

Sep-Pak CIS cartridge chromatographic technique similar to

that previously reported for scopolamine and tlumazenil (Frey

et aI., 1995). The samples at I, 2, and 3 min and at all subse

quent samples (13 in all) were analyzed for metabolites. Me

tabolite fractions of other samples prior to 2 min after injection

were estimated by linear interpolation. Due to the relatively

long scan duration, we employed a method using radioactive

fiducial markers to correct for any patient motion occurring

throughout the study. Molecular sieve beads (1-2 mm diam

eter) were placed at three points on the patient's scalp prior to

the study. Approximately I IJ.I of the ligand preparation was

pi petted onto each bead at an activity of -2 IJ.Ci/bead. Follow

ing reconstruction of the dynamic PET sequence, the beads

were defined on a single base frame (frame 10; 10-15 min after

injection). Details of the reorientation method are described in

Koeppe et al. (1991).

Volumes-of-interest (VOls) were created on the base frame

and applied to all other frames generating time-activity curves

for the ditl'erent brain structures. Regions analyzed included

caudate nucleus, putamen, frontal cortex, thalamus, and cer

ebellar hemispheres. Left and right hemisphere regions were

analyzed separately then averaged within each subject prior to

calculation of group means and standard deviations. No sig

nificant hemispheric differences were detected.

Nonlinear least-squares analysis was performed on the VOI

generated time-activity data. Parameters were estimated for

both two- and three-compartment configurations using the

Marquardt algorithm (Bevington, 1969) with constraints re

stricting parameters (except time shift) to positive values. Each

model configuration tested was implemented to account for the

contribution from activity in the cerebral blood volume (CBV)

and for the time offset or shift between the plasma input func

tion measured from the radial artery and the actual arterial input

to the brain. The time offset was estimated by fitting the entire

slice average from a single mid-brain level to a three

compartment four-parameter model using the entire 60-min

data set and assuming a whole-slice average CBV of 0.035 ml

g-I (3.5%). In all subsequent fits for the subject, the time offset

was fixed to this estimated value. Thus, each subject had a

single value for the time offset (used for all regions) while CBV

was estimated separately for each region. Even though rate

constant estimates proved to be fairly insensitive to CBV and

time shift, inclusion of both parameters in the model reduced

the bias in KI caused by their effects on the early data. For the

two-compartment configuration, Kl, DVto, (K/k2")' and CBV

were estimated for each Val. For the three-compartment con

figuration, KI, DVf+ns (K/k2'), k3', k4, and CBV were esti

mated, Two other sets of fits were perfonned, both assuming a

three-compartment configuration but with one parameter fixed

to the mean value averaged across subjects. Estimates of K1,

DVf+m, k3', and CBV were obtained using the group mean

value for k4 for average across all regions (0,077 min-I). Es

timates of KI, k3', k4, and CBV were obtained using the group

mean value averaged across all regions excluding basal ganglia

for DVl+ns (2.4 ml g-I). Basal ganglia regions were excluded

due to difficulties in differentiating free + nonspecific and spe

cific compartments due to the higher rate of binding in these

regions. The binding parameters k3/k4, DV,,,, and DV,ot were

then calculated for each of these model configurations.

Two pixel-by-pixel analyses were also perfonned on the dy

namic data sets. First, a two-parameter weighted integral analy

sis (Alpert et aI., 1984) was implemented as previously used for

[IIClflumazenil (Koeppe et aI., 1991) producing functional im-

Page 4

ANALYSIS OF 11lCjDTBZ KINETICS

1291

ages of ligand delivery (K1) and binding (DVtot)' The first 30 s

of scan data was omitted from the calculations in order to

reduce CBV effects on the estimated parameters. Second, a

graphical analysis for reversible ligands was used to calculate

DVto, and an approximation for K1• As discussed in their paper,

the slope of the linear portion of the Logan plot yields an

estimate of the total distribution volume of the ligand, DV,o"

independent of the number of compartments in the model. For

ligand kinetics described by a two-compartment model, the plot

is linear at all times, while for ligand kinetics requiring three (or

more) compartments, linearity is not reached until some time

later. The meaning of the intercept changes with model

complexity and is easily interpretable only when a two

compartment model is assumed. The intercept is given by -1/

[kz"(1 + CBVIDV,ot)]· Ignoring blood volume contributions,

this reduces to -lIkz" and thus, Kl can be approximated by the

slope/-intercept. Pixel-by-pixel functional images of DV,o'

were obtained from slope estimates using data beginning 10

min after injection, while images of Kl were obtained from

separate estimates of the slope and intercept made using the

entire data set. Simulation and human studies were performed

and the above time intervals were selected based on minimizing

parameter bias caused by failure of the DTBZ kinetics to be

described by a two-compartment model (see Discussion). After

creation of Kl and DVto, images for both the weighted integral

and Logan plot methods, the VOIs previously defined for ki

netic analysis were applied to the functional images to obtain

regional estimates of the model parameters.

RESULTS

The fraction of authentic e IC]DTBZ in plasma de

creased quite rapidly following administration, typically

reaching 65-70% by 10 min, 50-60% by 30 min, then

becoming relatively constant at 40-50% by 60 min after

injection.

Results from least-squares fits to the DTBZ time

activity curves for two- and three-compartment model

configurations are shown in Table 1. The entire 60-min

data sequence was used in these fits. Table I values give

the mean and percent coefficient of variation (COV; 100

x standard deviation/mean) of each parameter for the

seven volunteers. The mean reduced chi-squared values

(X2; sum of the squared discrepancies between data and

model predictions divided by the number of degrees of

freedom) for fits to the regional time activity curves are

reported for each model configuration.

Figure I depicts the goodness-of-fit using a two

compartment model estimating KI, DVtot, and CBV (2C

Fit) and a three-compartment model estimating Kj, k2',

k3', k4, and CBV (3C-Fit). [IIC]DTBZ time-activity

curves in putamen (Put) and frontal cortex (FCtx) with

2C (solid lines) and 3C (dashed lines) fits are shown in

the top panel. The bottom panel is a plot of the residuals

from these fits demonstrating the significantly greater

discrepancy between the 2C-Fit and the measured data

(solid lines, filled symbols) than the 3C-Fit and the mea

sured data (dashed lines, open symbols).

Figure 2 contrasts the group mean ± 1 SD for four

different possible measures of binding density, k3', k3' /

k4, DV,p' and DVtot' The values for k3' have been mul

tiplied by 10 for better viewing. Note that the variability

in the measures for k3' and k3' /k4 are prohibitively high

in regions of high binding density, with coefficients of

variation exceeding 75%, while variability in regions

with lower binding density is much lower, particularly

for k3' /k4 with coefficients of variation of only 20-30%.

Variability in DV estimates is quite consistent across

regions with coefficients of variation ranging from 10%

to 20%, except of DVsp in the putamen, which was more

highly variable.

Table 2 gives results when constraining either D�+ns

or k4 in the three-compartment analysis. Results are

shown for parameter estimates fitting all rate constants,

fixing the free + nonspecific distribution volume to 2.40

ml g-J, and fixing the dissociation rate constant to 0.077

min-I. Fits are reported only for putamen and frontal

cortex. Results from fits of the caudate nucleus data were

similar to those of the putamen, while results from fits of

TABLE 1. Least-squares estimates o{ model parameters f or tlVo- and three-compartment model conf igurations

2-compartment fits

3-compartment fits

Kl

DV,o,

(ml g-')

K,

DVJ+ns

(ml g-l)

k'

3

k

DVsp

(ml g-l)

DV,o,

(m! g-l)

"

4

Region

(m! g-' min-')

X-

(m! g-l min-')

(min-I)

(min-I)

k' Ik4

X2

Caudate

nucleus

0.378

(16)

11.3

(20)

0.97

0.483

(17)

2.46

(87)

0.92

(64)

0.113

(37)

6.4

(76)

9.2

(19)

1 1.7

(20)

0.39

Putamen

0.401

(17)

11.4

(20)

2.10

0.476

(17)

4.53

(51 )

0.23

(94)

0.078

(50)

2.42

(67)

7.92

(34)

12.5

(16)

0.30

Thalamus

0.362

(14)

4.32

(15)

5.83

0.458

( 12)

2.66

(20)

0.051

(38)

0.063

(31 )

0.81

(21)

2.09

(10)

4.75

(12)

0.61

Frontal

cortex

0.336

(15)

3.71

(22)

7.08

0.450

(12)

2.23

(30)

0.054

(48)

0.058

(36)

0.91

(24)

1.93

(13)

4.16

(19)

0.33

Cerebellum

0.329

(18)

3.75

(17)

4.45

0.418

(17)

2.29

(26)

0.066

(63)

0.075

(41)

0.83

(32)

1.77

(18)

4.06

(13)

0.32

Values shown are the group mean and coefficient of variation (expressed as percent of the mean).

J Cereb Blood Flow Metab. Vol. 16. No.6, 1996

Page 5

1292

R. A. KOEPPE ET AL.

TABLE 2. Effect of constraining DVf+n, or k4 on three-compartment model estimations

Kl

DV!+"-,

(ml g�l)

k'

k4

DV,,,

(ml g�l)

DVtot

(ml g�l)

3

Region

Model

(ml g�l min�l)

(min�l)

(min�l)

k'/k4

X2

Putamen

Fit all

0.476

(17)

4.53

(51)

0.23

(94)

0.078

(50)

2.42

(67)

7.92

(34)

12.5

(16)

0.30

Fixed

DV'

0.496

(17)

2.40

Fixed

0.77

(142)

0.25

(161)

3.34

(24)

8.18

(24)

11.7

(18)

0.46

Fixed k4

0.454

(12)

6.19

(42)

0.087

(78)

0.077

Fixed

1.13

(78)

4.85

(38)

12.2

(21)

0.49

Frontal

cortex

Fit all

0.450

(12)

2.23

(30)

0.054

(48)

0.058

(36)

0.91

(24)

1.93

(13)

4.16

(19)

0.33

Fixed

DV'

0.421

(19)

2.40

Fixed

0.042

(90)

0.050

(58)

0.79

(29)

1.94

(29)

4.39

(14)

1.32

Fixed k4

0.456

(II)

2.13

(30)

0.072

(23)

0.077

Fixed

0.93

(23)

1.87

(16)

4.00

(21)

0.58

Values shown are the group mean and coefficient of variation (expressed as percent of the mean).

the thalamus and cerebellar data were similar to those of

the frontal cortex.

Table 3 shows K1 and DVtot estimates obtained from

both the two-compartment weighted integral and the Lo

gan plot methods using the VOIs defined for the least

squares analysis. Weighted integral estimates excluded

data from the first 30 s to reduce CBV effects. Data from

10 min and later were used for the Logan DV101 estimate,

while the entire data set was used for the determination

of K1• Reported values give the mean and percent coef

ficient of variation for the group. Figure 3 shows func

tional images of K[ (top) and DVtot (bottom) at three

brain levels from a typical subject for the weighted in

tegral (left) and Logan (right) methods of analysis.

Figures 4 and 5 present results directly comparing

two- and three-compartment nonlinear least-squares VOl

analysis with pixel-by-pixel weighted integral and Logan

TABLE 3. Pixel-by-pixel estimates of model parameters f or

weighted integral and Logan plot analysis

Weighted Integral

Logan Plots

K,

DVtot

(m! g-l)

Kl

DVtot

(ml g-l)

Region

(m! g�l min�l)

(m! g�l min-1)

Caudate

nucleus

0.433

(14)

12.3

(16)

0.410

(15)

11.7

(14)

Putamen

0.472

(17)

12.6

(17)

0.442

(IS)

12.2

(14)

Thalamus

0.480

(12)

4.39

(12)

O.4lO

(13)

4.62

(II)

Frontal

cortex

0.450

(12)

3.77

(16)

0.370

(12)

4.03

(14)

Cerebellum

0.407

(17)

3.70

(15)

0.344

(17)

3.86

(13)

Values shown are the group mean and coefficient of variation.

J Cereb Blood Flow Metab, Vol. 16, No.6. 1996

analysis. Figure 4 shows DTBZ binding estimates of

DV101 for each method. Note the extremely similar vari

ability in all DV measurements. Figure 5 shows DTBZ

transport rate constant estimates for the four analysis

methods. The two-compartment least-squares approach

yields the lowest DV101 estimates, as might be expected,

while the weighted integral approach tends to show

slightly higher contrast between regions of high and low

binding density. The three-compartment least-squares

approach and the Logan method yielded the most similar

results. The two-compartment least-squares and the Lo

gan analyses yield consistent underestimation of K1 com

pared to the three-compartment analysis, primarily due to

the insufficiency of the two-compartment model for de

scribing the measured PET time course. Surprisingly, the

weighted integral approach, also based on a two

compartment model, yields less biased estimates than the

other approaches making a two-compartment assump

tion.

Besides comparing means and standard deviations for

the different measures of binding density, it is important

to know the degree of correspondence between the vari

ous measures. If two measures are very highly corre

lated, then they contain nearly same degree of informa

tion for distinguishing between levels of binding both

from region to region and from subject to subject. Table

4 shows results from linear regression analysis between

the various binding or transport parameters for the dif

ferent analysis methods. In each section of the table cor

relations are reported to the lower left of the diagonal

identity line. Regression slopes and intercepts (in paren

theses) are reported to the upper right of the diagonal.

For each regression, the parameter in the row heading

corresponds to the ordinate, while the parameter from the

column heading corresponds to the abscissa. The top

third of the table gives results between the various bind-