Source-based morphometry multivariate approach to analyze ¹²³I-FP-CIT SPECT imaging

Abstract

Background

¹²³I-FP-CIT (DaTSCAN®) SPECT imaging is widely used to study neurodegenerative parkinsonism, with the evaluation of presynaptic dopamine transporter (DAT) in striatal regions. Beyond DAT, ¹²³I-FP-CIT may be also considered for other monoaminergic systems, in particular serotonin transporter (SERT). Independent Component Analysis (ICA) implemented in Source-Based Morphometry (SBM), a multivariate approach, represents an alternative method to explore monoaminergic neurotransmission, studying the relationship among voxels, and grouping them into “metabolic” neural networks.

Methods

One-hundred forty-three subjects (84 with Parkinson’s disease (PD) and 59 control individuals (CG)) underwent DATSCAN® imaging. The ¹²³I-FP-CIT binding was evaluated by the multivariate SBM approach, as well as by region-of-interest (ROI) (caudate/putamen) (BRASS software) and whole-brain voxelwise univariate (Statistical Parametric Mapping, SPM) approaches.

Results

As compared to univariate approaches (BRASS and SPM), which only demonstrated striatal ¹²³I-FP-CIT binding reduction in the PD group, SBM identified six sources of non-artifactual origin, including basal ganglia and cortical regions as well as brainstem. Among them, three sources (basal ganglia and cortical regions) presented loading scores (as index of ¹²³I-FP-CIT binding) significantly different between PD and CG. Notably, even though not significantly different between PD and CG, the remaining three non-artifactual sources were characterized by a predominant frontal, brainstem and occipito-temporal involvement.

Discussion

The concept of source blind separation by the application of ICA (as implemented in SBM) represents a feasible approach to be considered in ¹²³I-FP-CIT (DaTSCAN®) SPECT imaging. Taking advantage of this multivariate analysis, specific patterns of variance can be identified (involving either striatal or extrastriatal regions) that could be useful in differentiating neurodegenerative parkinsonisms.

INTRODUCTION

In clinical practice, ¹²³I-FP-CIT (¹²³I-ioflupane, DaTSCAN®, GE Healthcare, Glattbrugg, Switzerland) SPECT imaging is widely used to study pre-synaptic dopaminergic changes ^{1, 2}. DaTSCAN® has been proved to be useful to differentiate between degenerative forms of parkinsonism, i.e. Parkinson’s disease (PD), Progressive Supranuclear Palsy (PSP), Dementia with Lewy Body (DLB) and Multiple System Atrophy (MSA), from non-degenerative parkinsonisms, namely drug-induced parkinsonism (DIP), psychogenic parkinsonisms (PP), or essential tremor^3-6. From a clinical point of view, image assessment has been focused on the evaluation of striatal regions, considering the high affinity of ¹²³I-FP-CIT for these regions ^{7, 8}, as index of presynaptic dopamine transporter (DAT) impairment ⁹. However, as previously suggested by studies in non-human primates ¹⁰, in healthy subjects ¹¹ and in pharmacological studies ^{1, 12}, ¹²³I-FP-CIT could be useful for the evaluation of other monoaminergic systems (i.e. serotonine transporter, SERT, norepinephrine transporter, NET) ^{13, 14}, in extrastriatal regions (cortical and midbrain)^15-17.

Striatal ¹²³I-FP-CIT binding (primarily related to nigrostriatal neuronal integrity) can be evaluated by a number of different methods. A region-of-interest (ROI) approach is based on a semi-quantitative analysis of the striatal region compared to a reference (occipital) region, after normalizing all images to a standardized ¹²³I-FP-CIT template derived from healthy subjects (BRASS software, Brain Registration & Analysis Software Suite, Hermes Medical Solutions, Stockholm, Sweden) ¹⁸. Furthermore, a whole-brain approach based on Statistical Parametric Mapping for SPECT ¹²³I-FP-CIT imaging has been proposed ^19-21, with the possibility to perform voxelwise studies on different groups of subjects with different statistical paradigms, and with the possibility to explore extrastriatal regions^22-25.

Source blind separation through Independent Component Analysis (ICA) has been widely used in functional neuroimaging, in particular for functional Magnetic Resonance Imaging (fMRI) ²⁶. This statistical method efficiently separates multivariate signals into non-overlapping spatial and time components and is able to capture functional/metabolic brain areas that are “working together”, thus representing functional brain neural networks ²⁷. In a data-driven fashion, ICA allows the removal of a great proportion of data noise, improving the signal-to-noise ratio ²⁸. More recently, ICA approach has been successfully applied to structural data (named Source Based Morphometry, SBM) ^{29, 30}, relying on the assumption that functionally correlated brain regions show greater concordance on structural measures as a result of mutually trophic influences or common experience-related plasticity ^{31, 32}. In contrast with the univariate approach of SPM (detecting the mean difference between two groups), SBM is based on a multivariate paradigm and considers the relationship among voxels, grouping them into “metabolic” neural networks. From this point of view, the SBM approach represents an interesting alternative method to explore monoaminergic neurotransmission by the application of ¹²³I-FP-CIT SPECT imaging. Taking advantage of the source blind separation of the variance in ¹²³I-FP-CIT by SBM, different sources could be identified, with potential clinical and neurobiological relevance. These considerations prompted the present study, aimed at evaluating multivariate (SBM) analysis application on ¹²³I-FP-CIT SPECT imaging in PD patients compared to a control group (CG) of patients with a clinical diagnosis of Essential Tremor (ET).

METHODS

Subjects

Consecutive patients with PD diagnosis according to UK Parkinson’s disease Society Brain Bank clinical criteria ³³ and supported by at least one-year follow-up were recruited from the Neurology Unit, Department of Clinical and Experimental Sciences, University of Brescia, Italy. All individuals underwent routine laboratory analyses and brain structural Magnetic Resonance Imaging (MRI). A standardized neurological examination was carried out, including the Unified Parkinson Disease Rating Scale (UPDRS-III) and the Mini Mental State Examination (MMSE).

Patients with a history of alcohol or drug abuse were excluded from the study. As control group (CG), a series of patients with clinical diagnosis of Essential Tremor (ET) ^{34, 35} (all with a normal ¹²³I-FP-CIT imaging), were considered. Written informed consent from the subject was obtained for each procedure. The research protocol has been approved by the Ethics Committee of the Brescia Hospital, Brescia, Italy.

SPECT imaging

Intravenous administration of 110–185 MBq of ¹²³I-FP-CIT was performed 30 min after thyroid blockade (800 mg of KClO4) in all subjects. Taking into account that SERT blockers (like Selective Serotonine Reuptake Inhibitors, SSRI) may influence ¹²³I-FP-CIT binding ratios (especially for sertraline) leading to increases of striatal to occipital ratios of approximately 10%, antidepressant therapy, if present, was suspended three weeks before the assessment. Brain SPECT was performed using a dual-head gamma-camera (Infinia Hawkey, GE Healthcare Ltd) calibrated with 159 KeV photopeak and ±10% energy window. Rotational radius was minimized (<16 cm). 120 projections over 360° (40 s/view) were acquired using a step-and-shoot protocol at 3° intervals with the camera heads following a circular orbit (matrix size 128×128; zoom 1.1) resulting in 43 min of total scan time (Total counts 1.8-3 × 10⁶). Data were reconstructed by filtered backprojection, with a Butterworth 3-dimensional (3D) post-filter (order = 10.0; cut-off 0.50 cycle/cm) and corrected for attenuation (Chang’s method, 0.12/cm^-1) ²². Three different methods of analysis of ¹²³I-FP-CIT imaging were considered:

1. BRASS™ software (BRASS, Brain Registration & Analysis Software Suite, Hermes Medical Solutions, Stockholm, Sweden; http://www.hermesmedical.com/products/hybrid-nm-processing/brass.html) was used for region-of–interest (ROI) analysis of striatal regions. A single ROI over the occipital lobe was set as reference background uptake area, and specific ratios for each caudate/putamen were calculated. Between-groups (PD vs CG) differences were evaluated by univariate General Linear Model, considering age and gender as nuisance variables. Statistical threshold was set at p<0.05 ^{18, 22}.

2. Statistical Parametric Mapping (SPM8 Wellcome Department of Imaging Neuroscience; http://www.fil.ion.ucl.ac.uk/spm/); ¹²³I-FP-CIT template from 30 healthy controls ³⁶ was used for normalization process. Individual ¹²³I-FP-CIT data were realigned, spatially normalized to ¹²³I-FP-CIT template in Montreal Neurological Institute space and smoothed (3D Gaussian filter with 8 mm Full Width at Half Maximum). Each voxel value of each scan was normalized to the mean value within the occipital cortex. An univariate General Linear Model, considering age and gender as nuisance variables was applied, with statistical threshold set at p<0.05 with a whole-brain false-discovery rate (FDR) correction for multiple comparisons ³⁷.

3. Source Based Morphometry (SBM): SBM was initially described to study co-varying patterns of alterations in structural Magnetic Resonance Imaging (MRI) (i.e. grey matter density ²⁹, cortical thickness ³⁸, fractional anisotropy ³⁰) in different conditions like healthy aging ³⁹, schizophrenia ^{29, 40} and Parkinson’s Disease ⁴¹. In the case of structural data (i.e. MRI), SBM can be used to analyze images which have only a single image per subject (i.e. grey matter but also functional techniques like positron emission tomography (PET) or single-photon emission computed tomography (SPECT)). Preprocessed images used for SPM analysis (realigned, spatially normalized to ¹²³I-FP-CIT template in Montreal Neurological Institute space and smoothed) were considered for SBM analysis. Briefly, SBM used spatial independent component analysis (ICA) to decompose local ¹²³I-FP-CIT binding variation across subjects into sources of common variance, considering a subjects-by-voxels data matrix. ICA was calculated by GIFT toolbox (GroupICAT v4.0a; http://mialab.mrn.org/software/gift) ⁴², using a neural network algorithm (Infomax) that attempts to minimize the mutual information of the network outputs⁴³; the number of sources was set to thirty, and to ensure reliability of the sources the decomposition was tested by using the ICASSO toolbox ⁴⁴ by running Infomax 10 times with different initial conditions and bootstrapped data sets and selected the best (most central) run. Individual source maps were converted to Z-scores before entering group statistics, to obtain voxel values comparable across subjects. As previously described for SBM ²⁹, the mixing matrix (containing the loading parameters for each subject and for each source) was used for statistical analysis, testing the difference between PD and CG groups, considering age and gender as nuisance variables. The statistical threshold was set at p<0.05 with whole-brain false-discovery rate (FDR) correction for multiple comparisons ³⁷. As previously described ²⁹, source matrix was used for visualization, scaling each map to unit standard deviation (SBM Z map) and threshold at |Z|>3.0. The maps of significant sources were then superimposed onto the MNI-normalized template brain.

Statistical analysis

Comparisons of demographic and clinical characteristics between groups (PD vs CG) were assessed via a Student’s t-test for continuous variables and a χ² test for categorical variables; significance level was set at p< 0.05. The data were analyzed by IBM SPSS Statistics 22.0 for Windows.

RESULTS

Demographic and clinical characteristics of PD and CG were reported in Table 1. Age and gender were significantly different among PD and CG, and they were considered as nuisance variables in the analyses. BRASS analysis demonstrated significant (p<0.001) striatal (putamen and caudate) ¹²³I-FP-CIT binding reduction in PD compared to CG, as expected (Table 1). SPM analysis showed a single cluster (109880 voxels, p<0.05 FDR whole-brain) of bilateral basal ganglia reduction in PD (x,y,z: −26, −6, 2; T=17.23, left putamen; x,y,z: 26, −6, 4; T=14.57, right putamen; x,y,z: −16, 16, 0; T=9.35, left caudate; x,y,z: 14, 16, 2; T=9.20, right caudate), with no evidence of extrastriatal differences between groups (Figure 1).

Table 1.

Demographic and clinical characteristics of the studied groups.

Variable	PD (n=84)	CG (n=59)	p
Age at evaluation, y	67.8±10.4	60.4±12.7	<0.001#
Gender, M % (n)	64.3% (54)	42.4% (25)	0.009^
MMSE	28.2±2.0	29.3±0.6	0.126#
UPDRS-III	15.1±7.5	-	-
L Caudate binding	1.58±0.44	2.61±0.43	<0.001°
R Caudate binding	1.69±0.50	2.79±0.42	<0.001°
L Putamen binding	1.14±0.51	2.71±0.40	<0.001°
R Putamen binding	1.10±0.52	2.62±0.45	<0.001°

Abbreviations and symbols: M, male; MMSE, Mini Mental State Examination; PD, Parkinson’s Disease; CG, control group (DAT negative subjects); y, years; UPDRS-III, Unified Parkinson Disease motor score

^#t student test;

^{^}chi square test

^°Univariate General Linear Model considering age and gender as nuisance variables. Caudate and Putamen bindings are expressed as ratio to occipital region of interest (ROI).

Figure 1. ¹²³I-FP-CIT (DaTscan®) imaging alterations in PD compared to CG with SPM approach.

PD<CG contrast is presented (p<0.05 FDR whole-brain), showing the reduced DAT binding in basal ganglia bilaterally. Significant clusters were superimposed on a standardized MRI T1 3D template. Montreal Neurological Institute of standardized space are shown below of each slices. R: right. PD: Parkinson’s Disease. CG: control group. SPM: Statistical Parametric Mapping.

For the SBM results, visual inspection of the 30 estimated sources highlighted obvious artifacts (i.e., signal near the external boundary of the brain or appearing primarily in grey matter-related regions) of 24 sources. Six sources of non-artifactual origin were then considered (see Figure 2) including basal ganglia and cortical regions (frontal, temporal and occipito-parietal) as well as brainstem (pons-midbrain). Among them, three sources presented loading scores (as index of ¹²³I-FP-CIT binding) significantly different between PD and CG (Table 2). In particular, source 8 and 23 included basal ganglia (putamen, caudate, thalamus) bilaterally, as well as cortical areas such as anterior cingulate, insula and frontal regions, whereas source 21 was characterized by a left-predominant involvement of basal ganglia/cortical areas. Notably, even if not significantly different between PD and CG, the remaining three non-artifactual involved frontal (source 10), brainstem (source 17) and occipito-temporal (source 18) regions.

Figure 2. ¹²³I-FP-CIT (DaTscan®) imaging alterations in PD compared to CG with SBM approach.

Each sources represented the absolute spatial pattern of the non-artifactual ¹²³I-FP-CIT binding derived from ICA analysis. Source 8: caudate and lentiform nuclei, insula anterior cingulate gyri; Source 10: bilateral inferior, middle, superior frontal gyri, right middle temporal and occipital gyri, caudate and lentiform nuclei, cingulate gyri; Source 17: pons, midbrain, bilateral posterior cingulate, thalami; Source 18: bilateral middle occipital, fusiform, inferior temporal, middle temporal, inferior occipital gyri; Source 21: left putamen, left caudate and left insula; Source 23: lentiform nuclei, caudate, insula, pons, right medial frontal gyrus and left inferior temporal gyrus. See Table 2 for details about the components significantly different between PD and CG. All the spatial components were thresholded at Z=3.0. Spatial patterns were superimposed on a standardized MRI T1 3D template. R: right. Montreal Neurological Institute of standardized space are shown below of each slices. R: right. PD: Parkinson’s Disease. CG: control group. SBM: source-based morphometry.

Table 2.

Sources derived from SBM analysis significantly different among the studied groups (PD vs CG).

Component	Direction	Cluster	Region	MNI	mm3
8	PD<CG	1	Putamen L	−24, 0, +2	8300
		2	Putamen R	+22, +4, +2	9400
		3	Caudate L	−12, +14, +2	2400
		4	Caudate R	+12, +12, +2	3700
		5	Insula L	−36, 0, +2	200
		6	Insula R	+36, 0, 0	500
		7	Thalamus R	+14, −6, +6	400
		8	Anterior Cingulate L	−12,+22, −4	100
		9	Anterior Cingulate R	+12, +22, −6	300
21	PD<CG	1	Putamen L	−20, +8, +2	8100
		2	Caudate L	−14, +16, +2	3600
		3	Insula L	−36, −8, +8	1300
23	PD<CG	1	Putamen L	−28, −6, +2	4700
		2	Putamen R	+26, −6, +4	3800
		3	Insula L	−36, −2, +2	1300
		4	Insula R	+36, −2, −2	200
		5	Pons L	−16, −14, −18	200
	PD>CG	1	Pallidum L	−12, +2, −4	600
		2	Pallidum R	+14, +6, 0	1900
		3	Thalamus L	−12, −10, +16	1000
		4	Thalamus R	+14, −8, +14	800
		5	Caudate L	−8, +4, 0	300
		6	Caudate R	+12, −2, +18	1300
		7	Medial Frontal gyrus R	+8, +26, −20	300
		8	Inferior Temporal gyrus L	−40, −12, −34	100

PD: Parkinson’s Disease; CG: control group; MNI: Montreal Neurological Institute coordinates system; L: left; R: right; mm³: cluster dimension expressed in cubic millimeter.

DISCUSSION

In the present work, we explored the potential application of the multivariate approach of SBM to ¹²³I-FP-CIT SPECT imaging, taking advantage of the source blind separation procedure of ICA ^{26, 29}. After ICA estimation, six non-artifactual sources were identified. In particular, three sources (8, 21, and 23) were mainly characterized by basal ganglia areas in association with frontal and temporal regions. In line with the clinical indication for ¹²³I-FP-CIT SPECT imaging and previous literature data ^{1, 2}, ¹²³I-FP-CIT binding was significantly impaired in PD patients in comparison to control group. Univariate approaches were able to detect only striatal ¹²³I-FP-CIT binding impairment in PD patients, however the SBM approach demonstrated significant involvement of cortical areas. Conversely, such cortical (extrastriatal) involvement was not recognized by univariate approaches, even if the large cluster reported by SPM analysis partially included extrastriatal regions (but without peak activations).

Interestingly, the identification of three different sources centered on basal ganglia suggested that ¹²³I-FP-CIT biding might serve to answer to additional clinical questions. First of all, ¹²³I-FP-CIT imaging was able to explore molecular fronto-striatal connectivity, as previously reported ^{22, 23}. As already mentioned, voxels carrying similar information were grouped by ICA in sources that can be considered as “metabolic” pattern, reflecting a similar ¹²³I-FP-CIT binding in the different ²⁹. Beyond basal ganglia, sources (10, 17 and 18) identified by ICA were characterized by the involvement of cortical regions (10: frontal; 18: temporo-occipital) as well as brainstem (source 17). Considering the affinity of ¹²³I-FP-CIT for different neurotransmitters (DAT and SERT in particular) and literature data ^{3, 6, 25, 45-47}, ¹²³I-FP-CIT binding in brainstem (pons) could be more related to SERT than to DAT neurotransmission. Nevertheless, taking into account that ICA maximized the statistical independence of the estimated sources, the variance of the signal (namely ¹²³I-FP-CIT) considered in one source (i.e. source 17) is highly “specific” for that spatial pattern. On the contrary, in all other types of analysis (ROI-based, SPM, spatial covariance analysis) ¹²³I-FP-CIT binding signal was considered “as a whole”, including noise, white matter but also the influence of other brain regions. In other words, considering that in ¹²³I-FP-CIT imaging the highest signal-to-noise ratio was present in striatal regions, the evaluation of other brain regions could be hampered by striatal regions. In this sense, ICA approach implemented in SBM analysis, separating the multivariate signal of ¹²³I-FP-CIT binding in maximally independent sources, could be potentially able “to move the lens” from striatal to extrastriatal brain regions. Altogether, obtained sources from SBM analysis on ¹²³I-FP-CIT imaging (and in particular sources 10, 17 and 18) supported the idea that this approach could be very useful in differentiating Parkinson’s Disease from atypical parkinsonisms ^{19, 46-51}. For instance, source 17, which maps brainstem, has the potential to be useful to PSP diagnosis. In addition, the evaluation of source 18, mapping occipital regions, has the potential to be helpful in DLB diagnosis.

Future studies considering atypical neurodegenerative parkinsonisms need to be performed to further evaluate the extra potential applications of ¹²³I-FP-CIT imaging as diagnostic tool in the spectrum of extrapyramidal syndromes.