Automatic covariance pattern analysis outperforms visual reading of 18 F-fluorodeoxyglucose-positron emission tomography (FDG-PET) in variant progressive supranuclear palsy

BACKGROUND

To date, studies on positron emission tomography (PET) with 18 F-fluorodeoxyglucose (FDG) in progressive supranuclear palsy (PSP) usually included PSP cohorts overrepresenting patients with Richardson's syndrome (PSP-RS).

OBJECTIVES

To evaluate FDG-PET in a patient sample representing the broad phenotypic PSP spectrum typically encountered in routine clinical practice.

METHODS

This retrospective, multicenter study included 41 PSP patients, 21 (51%) with RS and 20 (49%) with non-RS variants of PSP (vPSP), and 46 age-matched healthy controls. Two state-of-the art methods for the interpretation of FDG-PET were compared: visual analysis supported by voxel-based statistical testing (five readers) and automatic covariance pattern analysis using a predefined PSP-related pattern.

RESULTS

Sensitivity and specificity of the majority visual read for the detection of PSP in the whole cohort were 74% and 72%, respectively. The percentage of false-negative cases was 10% in the PSP-RS subsample and 43% in the vPSP subsample. Automatic covariance pattern analysis provided sensitivity and specificity of 93% and 83% in the whole cohort. The percentage of false-negative cases was 0% in the PSP-RS subsample and 15% in the vPSP subsample.

CONCLUSIONS

Visual interpretation of FDG-PET supported by voxel-based testing provides good accuracy for the detection of PSP-RS, but only fair sensitivity for vPSP. Automatic covariance pattern analysis outperforms visual interpretation in the detection of PSP-RS, provides clinically useful sensitivity for vPSP, and reduces the rate of false-positive findings. Thus, pattern expression analysis is clinically useful to complement visual reading and voxel-based testing of FDG-PET in suspected PSP. © 2023 The Authors. Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

Parkinson's disease progression: Increasing expression of an invariant common core subnetwork

Notable success has been achieved in the study of neurodegenerative conditions using reduction techniques such as principal component analysis (PCA) and sparse inverse covariance estimation (SICE) in positron emission tomography (PET) data despite their widely differing approach. In a recent study of SICE applied to metabolic scans from Parkinson's disease (PD) patients, we showed that by using PCA to prespecify disease-related partition layers, we were able to optimize maps of functional metabolic connectivity within the relevant networks. Here, we show the potential of SICE, enhanced by disease-specific subnetwork partitions, to identify key regional hubs and their connections, and track their associations in PD patients with increasing disease duration. This approach enabled the identification of a core zone that included elements of the striatum, pons, cerebellar vermis, and parietal cortex and provided a deeper understanding of progressive changes in their connectivity. This subnetwork constituted a robust invariant disease feature that was unrelated to phenotype. Mean expression levels for this subnetwork increased steadily in a group of 70 PD patients spanning a range of symptom durations between 1 and 21 years. The findings were confirmed in a validation sample of 69 patients with up to 32 years of symptoms. The common core elements represent possible targets for disease modification, while their connections to external regions may be better suited for symptomatic treatment.

Adaptive and pathological connectivity responses in Parkinson's disease brain networks

Functional imaging has been used extensively to identify and validate disease-specific networks as biomarkers in neurodegenerative disorders. It is not known, however, whether the connectivity patterns in these networks differ with disease progression compared to the beneficial adaptations that may also occur over time. To distinguish the 2 responses, we focused on assortativity, the tendency for network connections to link nodes with similar properties. High assortativity is associated with unstable, inefficient flow through the network. Low assortativity, by contrast, involves more diverse connections that are also more robust and efficient. We found that in Parkinson's disease (PD), network assortativity increased over time. Assoratitivty was high in clinically aggressive genetic variants but was low for genes associated with slow progression. Dopaminergic treatment increased assortativity despite improving motor symptoms, but subthalamic gene therapy, which remodels PD networks, reduced this measure compared to sham surgery. Stereotyped changes in connectivity patterns underlie disease progression and treatment responses in PD networks.

Dynamic 18F-FPCIT PET: Quantification of Parkinson's disease metabolic networks and nigrostriatal dopaminergic dysfunction in a single imaging session

Previous multi-center imaging studies with 18F-FDG PET have established the presence of Parkinson's disease motor- and cognition-related metabolic patterns termed PDRP and PDCP in patients with this disorder. Given that in PD cerebral perfusion and glucose metabolism are typically coupled in the absence of medication, we determined whether subject expression of these disease networks can be quantified in early-phase images from dynamic 18F-FPCIT PET scans acquired to assess striatal dopamine transporter (DAT) binding. Methods: We studied a cohort of early-stage PD patients and age-matched healthy control subjects who underwent 18F-FPCIT at baseline; scans were repeated 4 years later in a smaller subset of patients. The early 18F-FPCIT frames, which reflect cerebral perfusion, were used to compute PDRP and PDCP expression (subject scores) in each subject, and compared to analogous measures computed based on 18F-FDG PET scan when additionally available. The late 18F-FPCIT frames were used to measure caudate and putamen DAT binding in the same individuals. Results: PDRP subject scores from early-phase 18F-FPCIT and 18F-FDG scans were elevated and striatal DAT binding reduced in PD versus healthy subjects. The PDRP scores from 18F-FPCIT correlated with clinical motor ratings, disease duration, and with corresponding measures from 18F-FDG PET. In addition to correlating with disease duration and analogous 18F-FDG PET values, PDCP scores correlated with DAT binding in the caudate/anterior putamen. PDRP and PDCP subject scores using either method rose over 4 years whereas striatal DAT binding declined over the same time period. Conclusion: Early-phase images obtained with 18F-FPCIT PET can provide an alternative to 18F-FDG PET for PD network quantification. This technique therefore allows PDRP/PDCP expression and caudate/putamen DAT binding to be evaluated with a single tracer in one scanning session.

Spectral guided sparse inverse covariance estimation of metabolic networks in Parkinson's disease

In neurodegenerative disorders, a clearer understanding of the underlying aberrant networks facilitates the search for effective therapeutic targets and potential cures. [¹⁸F]-fluorodeoxyglucose (FDG) positron emission tomography (PET) imaging data of brain metabolism reflects the distribution of glucose consumption known to be directly related to neural activity. In FDG PET resting-state metabolic data, characteristic disease-related patterns have been identified in group analysis of various neurodegenerative conditions using principal component analysis of multivariate spatial covariance. Notably, among several parkinsonian syndromes, the identified Parkinson’s disease-related pattern (PDRP) has been repeatedly validated as an imaging biomarker of PD in independent groups worldwide. Although the primary nodal associations of this network are known, its connectivity is not fully understood. Here, we describe a novel approach to elucidate functional principal component (PC) network connections by performing graph theoretical sparse network derivation directly within the disease relevant PC partition layer of the whole brain data rather than by searching for associations retrospectively in whole brain sparse representations. Using sparse inverse covariance estimation of each overlapping PC partition layer separately, a single coherent network is detected for each layer in contrast to more spatially modular segmentation in whole brain data analysis. Using this approach, the major nodal hubs of the PD disease network are identified and their characteristic functional pathways are clearly distinguished within the basal ganglia, midbrain and parietal areas. Network associations are further clarified using Laplacian spectral analysis of the adjacency matrices. In addition, the innate discriminative capacity of the eigenvector centrality of the graph derived networks in differentiating PD versus healthy external data provides evidence of their validity.

Network Structure and Function in Parkinson's Disease

Little is known of the structural and functional properties of abnormal brain networks associated with neurological disorders. We used a social network approach to characterize the properties of the Parkinson's disease (PD) metabolic topography in 4 independent patient samples and in an experimental non-human primate model. The PD network exhibited distinct features. Dense, mutually facilitating functional connections linked the putamen, globus pallidus, and thalamus to form a metabolically active core. The periphery was formed by weaker connections linking less active cortical regions. Notably, the network contained a separate module defined by interconnected, metabolically active nodes in the cerebellum, pons, frontal cortex, and limbic regions. Exaggeration of the small-world property was a consistent feature of disease networks in parkinsonian humans and in the non-human primate model; this abnormality was only partly corrected by dopaminergic treatment. The findings point to disease-related alterations in network structure and function as the basis for faulty information processing in this disorder.

A full-brain, bootstrapped analysis of diffusion tensor imaging robustly differentiates Parkinson disease from healthy controls

There is a compelling need for early, accurate diagnosis of Parkinson's disease (PD). Various magnetic resonance imaging modalities are being explored as an adjunct to diagnosis. A significant challenge in using MR imaging for diagnosis is developing appropriate algorithms for extracting diagnostically relevant information from brain images. In previous work, we have demonstrated that individual subject variability can have a substantial effect on identifying and determining the borders of regions of analysis, and that this variability may impact on prediction accuracy. In this paper we evaluate a new statistical algorithm to determine if we can improve accuracy of prediction using a subjects left-out validation of a DTI analysis. Twenty subjects with PD and 22 healthy controls were imaged to evaluate if a full brain diffusion tensor imaging-fractional anisotropy (DTI-FA) map might be capable of segregating PD from controls. In this paper, we present a new statistical algorithm based on bootstrapping. We compare the capacity of this algorithm to classify the identity of subjects left out of the analysis with the accuracy of other statistical techniques, including standard cluster-thresholding. The bootstrapped analysis approach was able to correctly discriminate the 20 subjects with PD from the 22 healthy controls (area under the receiver operator curve or AUROC 0.90); however the sensitivity and specificity of standard cluster-thresholding techniques at various voxel-specific thresholds were less effective (AUROC 0.72-0.75). Based on these results sufficient information to generate diagnostically relevant statistical maps may already be collected by current MRI scanners. We present one statistical technique that might be used to extract diagnostically relevant information from a full brain analysis.

Characterization of disease-related covariance topographies with SSMPCA toolbox: effects of spatial normalization and PET scanners

To generate imaging biomarkers from disease-specific brain networks, we have implemented a general toolbox to rapidly perform scaled subprofile modeling (SSM) based on principal component analysis (PCA) on brain images of patients and normals. This SSMPCA toolbox can define spatial covariance patterns whose expression in individual subjects can discriminate patients from controls or predict behavioral measures. The technique may depend on differences in spatial normalization algorithms and brain imaging systems. We have evaluated the reproducibility of characteristic metabolic patterns generated by SSMPCA in patients with Parkinson's disease (PD). We used [(18) F]fluorodeoxyglucose PET scans from patients with PD and normal controls. Motor-related (PDRP) and cognition-related (PDCP) metabolic patterns were derived from images spatially normalized using four versions of SPM software (spm99, spm2, spm5, and spm8). Differences between these patterns and subject scores were compared across multiple independent groups of patients and control subjects. These patterns and subject scores were highly reproducible with different normalization programs in terms of disease discrimination and cognitive correlation. Subject scores were also comparable in patients with PD imaged across multiple PET scanners. Our findings confirm a very high degree of consistency among brain networks and their clinical correlates in PD using images normalized in four different SPM platforms. SSMPCA toolbox can be used reliably for generating disease-specific imaging biomarkers despite the continued evolution of image preprocessing software in the neuroimaging community. Network expressions can be quantified in individual patients independent of different physical characteristics of PET cameras.

Abnormal metabolic brain networks in a nonhuman primate model of parkinsonism

Parkinson's disease (PD) is associated with a characteristic regional metabolic covariance pattern that is modulated by treatment. To determine whether a homologous metabolic pattern is also present in nonhuman primate models of parkinsonism, 11 adult macaque monkeys with parkinsonism secondary to chronic systemic 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) and 12 age-matched healthy animals were scanned with [(18)F]fluorodeoxyglucose (FDG) positron emission tomography (PET). A subgroup comprising five parkinsonian and six control animals was used to identify a parkinsonism-related pattern (PRP). For validation, analogous topographies were derived from other subsets of parkinsonian and control animals. The PRP topography was characterized by metabolic increases in putamen/pallidum, thalamus, pons, and sensorimotor cortex, as well as reductions in the posterior parietal-occipital region. Pattern expression was significantly elevated in parkinsonian relative to healthy animals (P<0.00001). Parkinsonism-related topographies identified in the other derivation sets were very similar, with significant pairwise correlations of region weights (r>0.88; P<0.0001) and subject scores (r>0.74; P<0.01). Moreover, pattern expression in parkinsonian animals correlated with motor ratings (r>0.71; P<0.05). Thus, homologous parkinsonism-related metabolic networks are demonstrable in PD patients and in monkeys with experimental parkinsonism. Network quantification may provide a useful biomarker for the evaluation of new therapeutic agents in preclinical models of PD.

Three-fold cross-validation of parkinsonian brain patterns

Abnormal physiological networks of brain areas in disease can be identified by applying specialized multivariate computational algorithms based on principal component analysis to functional image data. Here we demonstrate the reproducibility of network patterns derived using positron emission tomography (PET) data in independent populations of parkinsonian patients for a large, clinically validated data set comprised of subjects with idiopathic Parkinson's disease (iPD), multiple system atrophy (MSA) and progressive supranuclear palsy (PSP). Correlation of voxel values of network patterns derived for the same condition in different data sets was high. To further illustrate the validity of these networks, we performed single subject differential diagnosis of prospective test subjects to determine the most probable case based on a subject's network scores expressed for each of these distinct parkinsonian syndromes. Three-fold cross-validation was performed to determine accuracy and positive predictive rates based on networks derived in separate folds of the composite data set. A logistic regression based classification algorithm was used to train in each fold and test in the remaining two folds. Combined accuracy for each of the three folds ranged from 82% to 93% in the training set and was approximately 81% for prospective test subjects.

Differential diagnosis of parkinsonian syndromes using PCA-based functional imaging features

In the current paper, we describe methodologies for single subject differential diagnosis of degenerative brain disorders using multivariate principal component analysis (PCA) of functional imaging scans. An automated routine utilizing these methods is applied to positron emission tomography (PET) brain data to distinguish several discrete parkinsonian movement disorders with similar clinical manifestations. Disease specific expressions of voxel-based spatial covariance patterns are predetermined using the Scaled Subprofile Model (SSM/PCA) and a scalar measure of the manifestation of each pattern in prospective subject images is subsequently derived. Scores are automatically compared to reference values generated for each pathological condition in a corresponding set of patient and control scans. Diagnostic outcome is optimized using strategies such as the derivation of patterns in a voxel subspace that reflects contrasting image characteristics between conditions, or by using an independent patient population as controls. The prediction models for two, three and four way classification problems using direct scalar comparison as well as classical discriminant analysis are assessed in a composite training population comprised of three different patient classes and normal controls, and validated in a similar independent test population. Results illustrate that highly accurate diagnosis can often be achieved by simple comparison of scores utilizing optimized patterns.

New strategies for automated differential diagnosis of degenerative brain disorders

New strategies are considered for automated, single-subject differential diagnosis of independent degenerative brain disorders characterized by similar clinical symptoms using functional imaging. The methodology of these strategies is described and its application in parkinsonian movement disorders is illustrated for PET data. Using an automated diagnostic Topographic Profile Rating (TPR) technique based on the Scaled Subprofile Model (SSM-PCA), single-subject score values for different conditions are compared with reference values to predict diagnosis. The discriminatory parameters of reference score sets associated with significant SSM principal components referred to as group invariant subprofiles (GIS networks) are examined. It is shown that the extraction of exclusive sub-networks that stem from contrasting image features between conditions can be an effective tool for optimization that does not require expert knowledge.

Abnormal metabolic network activity in Parkinson's disease: test-retest reproducibility

Parkinson's disease (PD) is associated with an abnormal pattern of regional brain function. The expression of this PD-related covariance pattern (PDRP) has been used to assess disease progression and the response to treatment. In this study, we validated the PDRP network as a measure of parkinsonism by prospectively computing its expression (PDRP scores) in (15)O-water (H(2)(15)O) and (18)F-fluorodeoxyglucose (FDG) positron emission tomography (PET) scans from PD patients and healthy volunteers. The reliability of this measure was also assessed within subjects using a test-retest design in mildly affected and advanced PD patients scanned at baseline and during treatment with levodopa or deep brain stimulation (DBS). We found that PDRP expression was significantly elevated in PD patients (P<0.001) relative to controls in a prospective analysis of brain scans obtained with either H(2)(15)O or FDG PET. A significant correlation (R(2)=0.61; P<0.001) was evident between PDRP scores computed from H(2)(15)O and FDG images in PD subjects scanned with both tracers. Test-retest reproducibility was very high (intraclass correlation coefficient (ICC)>0.92) for PDRP scores measured both within PET session and between sessions separated by up to 2 months. This high reproducibility was observed in both early stage and advanced PD patients scanned at baseline and during treatment. The within-subject variability of this measure was less than 10% for both unmedicated and treated conditions. These findings suggest that the PDRP network is a reproducible and stable descriptor of regional functional abnormalities in parkinsonism. The quantification of PDRP expression in PD patients can serve as a potential biomarker in PET intervention studies for this disorder.

Parametric mapping of [18F]FPCIT binding in early stage Parkinson's disease: a PET study

We have shown that fluorinated N-3-fluoropropyl-2-beta-carboxymethoxy-3-beta-(4-iodophenyl) nortropane ([(18)F]FPCIT) and PET offer a valuable means of quantifying regional abnormality in dopamine transporter (DAT) imaging associated with Parkinson's disease (PD). The objective of this study was to delineate the topographic distribution of DAT binding in early stage idiopathic PD using statistical parametric analysis of [(18)F]FPCIT PET data. We performed dynamic PET studies in 15 hemi-parkinsonian (Hoehn & Yahr I) patients and 10 age-matched normal volunteers over 100 min and calculated images of [(18)F]FPCIT binding ratios on a pixel-by-pixel basis. Statistical parametric mapping (SPM) was then used to localize binding reductions in PD and to compute the absolute change relative to normal. [(18)F]FPCIT binding decreased significantly in the contralateral posterior putamen of the PD group (P < 0.001, corrected). A significant reduction was also seen in the ipsilateral putamen, which was smaller in extent but localized more posteriorly. A quantitative comparison of DAT binding in the two clusters showed that the onset of motor symptoms in PD was associated with an approximate 70% loss relative to the normal mean in the contralateral posterior putamen. These results suggest that SPM analysis of [(18)F]FPCIT PET data can be used to quantify and map abnormalities in DAT activity within the human striatum. This method provides a useful tool to track the onset and progression of PD at its earliest stages.

Visualizing the evolution of abnormal metabolic networks in the brain using PET

By applying novel statistical methods and visualization techniques to PET data obtained from combined groups of patients and normals, we are able to illustrate topographic covariance profiles unique to neurodegenerative disorders such as Parkinson's Disease at various stages of progression. Each profile represents a neuroanatomical network of metabolically covarying regions. The expression of the profile in each patient is characterized by a subject score which can correlate with independent clinical disease severity measures. To visualize these profiles, a semi-automatic routine is used (3D) animation of the metabolic topography as it evolves from initial to final stages of the disease.