Validation of Reference Tissue Modelling for 11C FMZ_main document

Methods Following earlier studies, the pons was used as the reference region.PET scans were performed on 16 controls and 11 patients at least six months followinginjury, each of whom als

Objective [11C]flumazenil ([11C]FMZ) positron emission tomography (PET) can beused as a measure of neuronal loss The purpose of this study was to validate referencetissue kinetic modelling of [11C]FMZ PET within a group of patients with head injury

Methods Following earlier studies, the pons was used as the reference region.PET scans were performed on 16 controls and 11 patients at least six months followinginjury, each of whom also had arterial blood sampling to provide whole blood andmetabolite-corrected plasma input functions Regional non-displaceable binding

potentials (BP ND) were calculated from five reference tissue models and compared to

BP ND from arterial input models For the patients the regions included a peri-lesionalregion-of-interest (ROI)

Results Total distribution volume (V T ) of the pons was not significantly different between control and patient groups (P=0.24) BP ND from all the reference tissue

approaches correlated well with BP ND from the plasma input models for both controls(r2: 0.98-1.00; P<0.001) and patients (r2: 0.99-1.00; P<0.001) For the peri-lesionalregions (n=11 ROI values) the correlation was also high (r2=0.91)

Conclusions These results indicate that reference tissue modelling with the pons as the

reference region is valid for [11C]FMZ PET in head injured patients at six monthsfollowing injury within both normal appearing and peri-lesional brain regions

Key words: [11C]flumazenil, kinetic modelling, positron emission tomography (PET), reference tissue, traumatic brain injury (TBI)

[11C]flumazenil ([11C]FMZ) is a neuronal benzodiazepine / -aminobutyric acid (GABA)receptor ligand and a marker of neuronal integrity (1-3) Recent studies in stroke haveemployed [11C]FMZ positron emission tomography (PET) as a marker of selectiveneuronal loss (3-6) Like stroke, trauma can result in regions of pan necrosis, which areclearly visible, or selective neuronal loss which demonstrate little change on standardstructural magnetic resonance imaging (MRI) (7-9) However, unlike stroke, insultscaused by traumatic brain injury do not follow vascular or anatomical boundaries.Consequently, it is unsafe to assume that tissue regions that normally provide referenceareas are unaffected by pathology, and that such pathology does not impact on PETquantification Indeed, many common reference tissue sites (e.g pons, supratentorialwhite matter) are known to be involved in head injury (10) This issue has not beensatisfactorily addressed by previous publications and data which support the use ofreference tissue modelling in patients at late time points following head injury wouldfacilitate recruitment by removing the need for arterial cannulation In order to addressthese concerns, we have examined the validity of using a reference tissue input (11) inthis setting

Use of a reference tissue input for analysing [11C]FMZ is desirable as it obviatesarterial sampling and is computationally efficient when using methods such as Logangraphical analysis (12), multi-linear analysis (13) or basis function implementations ofthe simplified reference tissue model (SRTM (14)): receptor parametric mapping(RPM1) (15) and RPM2 (16) Reference tissue approaches produce values for

distribution volume ratio (DVR) and/or non-displaceable binding potential (BP ND) asmeasures of available receptor density (17)

The pons has been used as a reference tissue for [11C]FMZ and is validated forcontrol subjects (6, 18-22) and various neurological disease patients (20, 21).

In this study, [11C]FMZ BP ND values for head injury patients and controls determinedfrom various reference tissue methods (Logan plot with reference tissue input (12),reference tissue model (RTM (11)), SRTM, RPM1, RPM2) with pons as the reference

tissue are validated through comparison with BP ND from compartmental modelling with

a metabolite-corrected arterial plasma input function

Materials and Methods

Objective

To assess whether reference tissue modelling of [11C]FMZ PET is comparable with arterial input compartmental modelling within a group of patients recovering from head injury

Ethics statement

Full written approval was obtained from volunteers and patients prior to commencement

of any part of the study The study was conducted according to the Good ClinicalPractice guidelines, and the declaration of Helsinki The study protocol was approved

by the Cambridge Local Research Ethical Committee and the UK Administration ofRadioactive Substances Advisory Committee (ARSAC)

Imaging

MRI

Structural MRI was obtained in patients and control subjects The protocol includedhigh-resolution 3D volume T1-weighted spoiled-gradient (SPGR), T2-weighted andfluid-attenuation inversion-recovery (FLAIR) sequences, acquired on a 3T whole bodymagnet (Medspec s300, Bruker, Ettlingen, Germany) SPGR images were resized tovoxels of 1  1  1 mm3 and re-orientated to the AC-PC line

PET

FMZ-PET was acquired in 3D mode on a GE Advance PET Scanner (General ElectricMedical Systems, Milwaukee, WI, USA) Prior to FMZ injection a 15 min transmissionscan using rotating 68Ge rod sources was acquired to correct for photon attenuation.FMZ was labelled with 11C using a methylation process (25), providing high specificactivities (370–550 GBq/mol) FMZ was injected intravenously as a bolus (418 ± 21MBq) and data was acquired for 75min post-injection (55 times frames: 18  5s, 6 15s, 10  30s, 7  60s, 4  150s and 10  300s) Images were reconstructed using thePROMIS 3D filtered back projection algorithm (26) into 128  128  35 arrays with a

voxel size of 2.34  2.34  4.25mm Corrections were applied for randoms, dead time,normalisation, scatter, attenuation and sensitivity.

The arterial input function was determined from 18 blood samples via a radialartery cannula (6  20s, 3min, 4min, 5min, 7min, 10min, 15min, every 10minthereafter) Radioactivity in both whole blood and plasma was measured in a gammawell counter Seven blood samples were used to determine the radio-labelled metabolitefraction in plasma (3min, 6min, 10min, 20min, 45min, 60min, and 75min afterinjection) Concentrations of hydrophilic metabolites and [11C]FMZ within plasma weredetermined using a validated technique (27) The metabolite fraction data were fittedwith the Hill function (28) and hence a metabolite-corrected plasma time-activity curve(TAC) was obtained

Image realignment, co-registration and spatial normalisation

SPM5 (www.fil.ion.ucl.ac.uk/spm/software/spm5/) was used to realign the dynamicPET images, co-register PET to the re-orientated SPGR MR using normalised mutualinformation and spatially normalize MR and hence PET images to the MontrealNeurological Institute (MNI) standard space (MNI152 template)

Regions of interest

Following review of MR data to exclude evidence of injury, the pons region of interest(ROI) was drawn using Analyze 7.0 (Biomedical Imaging Resource, Mayo Clinic,Rochester, MN, USA) on 10 contiguous transverse planes of the re-orientated SPGR

MR, as previously described (20) In addition, for each patient, a peri-lesional ROI wascreated Regions of brain displaying evidence of a lesion were delineated on FLAIR and

SPGR images and dilated by 7mm to limit partial volume error In order to createperilesional ROIs these lesion ROIs were dilated by a further 10mm to include a margin

of potentially injured brain and the original lesion volume subtracted These ROIs wereprojected onto co-registered PET images to generate a TAC for each region, which wasthen subject to arterial input and reference tissue kinetic modelling

ROIs were also taken from the WFU Pick atlas (Wake Forest University,Winston-Salem, NC) toolbox in SPM5 The following 12 ROIs were selected: anteriorcommissure, cerebellum Brodmann area (BA) 7b, corpus callosum, frontal BA 10,hippocampus, motor cortex BA 4, occipital BA 18, parietal BA 40, temporal BA 20,caudate, putamen and thalamus These MNI space ROIs were projected onto spatiallynormalised PET images to generate a TAC for each region, which was then subject toarterial input and reference tissue kinetic modelling

Kinetic modelling

Arterial input function methods

Rate constants and blood volume were estimated from compartmental modelling of ROITAC data, which can be parameterised as

)()

()1()

1 1

1

k

k k

V T values from both 1TCM and 2TCM were converted to BP ND using (17)

V

V

V T pons is used as an estimate of the non-displaceable distribution volume V ND

Hence correlation with BP ND from arterial input modelling was used to verify BP ND fromthe reference tissue models

As an illustration of the impact of head injury, a BP ND map (derived from V T 1TCM)was produced for one patient using voxel-wise modelling with the same methodology asdescribed above for regional modelling

Reference tissue methods

ROI BP ND was obtained for RTM (11) and SRTM (14) using the pons TAC andweighted non-linear least squares fitting optimised with the Levenberg-Marquardt

method Determination of BP ND with RPM1 (15) used 100 basis functions with 0.001 

3  0.01 sec-1 A parametric map of k 2 was also produced using RPM1 in order to

produce a map of k 2  (k 2 in the reference tissue, i.e pons), where k 2  = k 2 /R I The median

value of k 2  in voxels with BP ND  0.5  BP NDmax was used in RPM2 (16) RPM2 used

100 basis functions with 0.001  k2  0.01 sec-1 DVR and hence BP ND (DVR–1) were

determined with reference tissue Logan graphical analysis (12) using data acquired 17.5min post-injection, where the fitting time was chosen through observation of Loganplots of high binding cortical TACs Using the same methodology at the voxel level,

BP ND maps were generated using RPM1, RPM2 and reference tissue Logan graphical

analysis for the same patient as the aforementioned arterial input BP ND map

between arterial input and reference tissue BP ND values were assessed using one-way

ANOVA P values < 0.05 were considered to be statistically significant

Results

Participants

The controls had a median (range) age of 42 (24 – 71) years, which was not

significantly different from the patients (35 (21 – 67) years, P = 0.18) In addition, the

proportion of males in the control group (69%) was not significantly different from the

patients with head injury (73%, P = 0.83) The median (range) Glasgow Outcome Score

of patients at the time of PET imaging was 4 (3 – 5) The patient characteristics are

detailed in Table 1 On review of early CT and follow up MR data no patientsdemonstrated evidence of injury within the pons Figure 1 shows a FLAIR MR of a

representative patient together with a co-registered arterial input BP ND map Corresponding reference tissue BP ND maps are given in Figure 2

Pons distribution volume values for controls and patients

Pons V T following head injury was not significantly different from that of controls for

either V T 1TCM (0.90 vs 0.84; P=0.24) or V T 2TCM (0.90 vs 0.84; P=0.28)

Arterial input regional BP ND values

Mean ROI BP ND values obtained from arterial input modelling are given in Table 2 Forboth controls and patients the two arterial input models (1TCM and 2TCM) were highlycorrelated (r2: 0.97 and 0.99; P < 0.001).

Comparison of arterial input and reference tissue model fits to regional data

As an example of the fits achieved with both arterial input and reference tissue models,Figure 3 shows 1TCM, 2TCM, RPM1 and RPM2 fits to a peri-lesional ROI TAC Allmethods provide good fits, although the noise propagating from the reference tissueinput can clearly be seen with RPM1 and RPM2

Comparison between arterial input and reference tissue BP ND values

For both controls and patients, there were strong correlations between ROI BP ND from

the reference tissue methods and corresponding BP ND values derived from arterial input

modelling (Table 3) In addition, for both controls and patients there were no differences

in BP ND values between the reference tissue methods (P > 0.99) As an illustration, the

correlations for RPM2 are shown graphically for the WFU Pick atlas ROIs (Figure 4)and peri-lesional ROIs (Figure 5)

In all cases, mean slopes of less than unity were found for the correlations Onedifference between the arterial input and reference tissue methods is that blood volume

is included in the former (equation 1) but neglected in the latter If blood volume isneglected in the arterial input model, the correlations with reference tissue modellinghave slopes much closer to unity This is illustrated for RPM2 through comparison ofFigures 6 and 7

Discussion

This study investigated the validity of reference tissue modelling of [11C]FMZ inpatients following head injury with the pons as a reference region Although, it has beenreported that there is some specific binding in pons (31), the use of the pons as areference tissue for [11C]FMZ has been validated in controls (6, 18-22) and otherneuronal conditions (20, 21) Head injury is known to result in damage across the brain,even in regions which appear structurally normal (7, 32) Previous studies have failed

to assess the impact this damage may have on PET quantification using reference tissuemodelling This study demonstrates that reference tissue modelling of [11C]FMZ PET isvalid for patients recovering from head injury

As there is not uniform agreement in the literature with respect to the optimalcompartment model to use for FMZ, we tested the reference tissue approaches against

both one-tissue (1TCM) and two-tissue (2TCM) arterial input models For the data inthis study the two arterial input models were highly correlated (r2: 0.97-0.99).

The correlations found between non-displaceable binding potential (BP ND) fromreference tissue modelling and arterial input modelling were as high for patients (n=11)

as for the control group (n=16) The high degree of correlation was maintained across awide range of [11C]FMZ binding (BP ND: 0.0-6.0) in a set of cortical and sub-corticalregions (n=12) For the patient group, the correlation was also high in peri-lesional

ROIs, whose BP ND values were in the range 2.0-3.5 These results suggest that referencetissue modelling is valid for late [11C]FMZ PET in head injured patients within normalappearing and perilesional brain regions Although reference tissue modelling using thepons as a reference has been validated in healthy controls and patients with variousforms of neurological disease, these are the first data to address this issue followinghead injury

The slope of the correlations between arterial and reference tissue methodstended to be less than unity when blood volume was included in the arterial inputmodel However, blood volume is neglected in the reference tissue model (11, 14) andthe reference tissue Logan plot (12) Once blood volume was also neglected in thearterial input models, the slope of the correlations neared unity, providing furthervalidation of the reference tissue methods

There was no difference between the BP ND values calculated using the variousreference tissue models Therefore, these data do not suggest that any of the referencetissue models are superior in this setting However, the basis function methods (RPM1and RPM2), together with the reference tissue Logan plot are better suited to modelling

of noisy TACs than the free-fitting full or simplified reference tissue models