Elsevier

NeuroImage

Volume 75, 15 July 2013, Pages 249-261
NeuroImage

Reliability analysis of the resting state can sensitively and specifically identify the presence of Parkinson disease

https://doi.org/10.1016/j.neuroimage.2011.06.056Get rights and content

Abstract

Parkinson disease (PD) is characterized by a number of motor and behavioral abnormalities that could be considered deficits of a “no task” or “resting” state, including resting motor findings and defects in emerging from a resting state (e.g., resting tremor, elevated resting tone, abulia, akinesia, apathy). PET imaging, and recently, the MRI technique of continuous arterial spin labeling (CASL) have shown evidence of changes in metabolic patterns in individuals with PD. The purpose of this study was to learn if the presence of PD could be “predicted” based on resting fluctuations of the BOLD signal. Participants were 15 healthy controls, 14 subjects with PD, and 1 subject who presented as a control but later developed PD. The amplitude of the low frequency fluctuation (ALFF) was used as an index of brain activity level in the resting state. Participants with PD using this index showed a reliable decrease in activity in a number of regions, including the supplementary motor cortex, the mesial prefrontal cortex, the right middle frontal gyrus, and the left cerebellum (lobule VII/VIII) as well as increased activity in the right cerebellum (lobule IV/V). Using a cross validation approach we term “Reliability Mapping of Regional Differences” (RMRD) to analyze our sample, we were able to reliably distinguish participants with PD from controls with 92% sensitivity and 87% specificity. Our “pre-diagnostic” subject segregated in our analysis with the PD group. These results suggest that resting fMRI should be considered for development as a biomarker and analytical tool for evaluation of PD.

Graphical abstract

Highlights

► Resting fMRI predicts presence of Parkinson disease (PD). ► Sensitivity of prediction was 92%, and specificity 87%. ► Resting state measured using amplitude of low frequency fluctuation. ► Introduce region of interested based cross-validation method.

Introduction

Recent reviews have focused on the need to develop sensitive biomarkers for Parkinson disease (PD) (Foulds et al., 2010, Jones, 2010). PD in clinical practice is usually diagnosed based on the presence or absence of characteristic clinical features, including resting tremor, bradykinesia, and rigidity. However, these clinical features can also be present in a number of other disorders, including multiple systems atrophy, progressive supranuclear palsy, diffuse Lewy body disease, corticobasal ganglionic degeneration, drug induced Parkinsonism, or vascular Parkinsonism. There are therefore many important clinical and scientific reasons to develop diagnostic biomarkers that correlate with clinical disease. Clinically, prognosis and diagnostic options differ among various causes of Parkinsonism. For example, medications have differential effects on various forms of Parkinsonism, and while deep brain stimulation is an effective treatment for appropriately selected patients with PD, it can be contraindicated in other forms of Parkinsonism (Shih and Tarsy, 2007). An effective biomarker might aid in early diagnosis or even in identifying preclinical disease, as it is well recognized that the pathological processes underlying PD predate the onset of clinical symptoms (Braak et al., 2004). Further, clinical trials typically rely on clinical diagnoses, occasionally informed by post-mortem pathological analysis; the lack of specificity of clinical diagnoses results in added noise in PD related clinic trials. In fact, Fahn et al. (2004) noted that up to 15% of individuals enrolled in PD related clinical trials were later found to have Parkinsonism on the basis of a disorder other than PD.

We hypothesized that resting fMRI could contain temporal information that may allow the technique to be developed as a biomarker for PD. Specifically, we approach development of an imaging PD biomarker from the perspective that PD is a disorder that is clinically recognized on the basis of motor and behavioral abnormalities (such as resting tremor, elevated resting tone, and freezing of gait or start hesitation, and apathy), that could be considered to be deficits related to the “resting” or “no-task” state. We therefore hypothesize that individuals with PD might have measurable changes in resting brain activity detectable by fMRI. Resting network activity as defined by a number of fMRI techniques has been shown to result in consistent, reliable patterns among and across healthy subjects (Beckman et al., 2005, Damoiseaux et al., 2006, Zuo et al., 2010). Positron Emission Tomography (PET) shows alterations in resting network activity in PD, often termed the PD-related covariance pattern (PDRP) (Lozza et al., 2004, Moeller et al., 1999). This approach, pioneered by Eidelberg et al., focuses on using scaled subprofile modeling principal component analysis (SSM/PCA). SSM/PCA is a multivariate method designed to identify significant spatial covariance patterns in combined samples of patients and control scans (Habeck et al., 2008, Ma et al., 2006, Spetsieris et al., 2009). SSM/PCA maps covariance patterns across multiple regions of the brain. Recently, this technique has been extended to show differences in the PDRP in MRI images using continuous arterial spin labeling (CASL) (Ma et al., 2010) — a technique that uses continuous magnetic labeling of blood flow to generate a “perfusion map.”

In this study, we evaluate if resting state BOLD fluctuations can be used to sensitively and specifically predict presence or absence of PD. Resting BOLD fluctuations do not formally create a perfusion map, but fluctuations in the BOLD signal are related to both brain activity and to blood flow. The magnitude of BOLD fluctuations can be measured, and in fact low frequency oscillations in the resting state can be relatively easily calculated and can provide a brain-wide voxel by voxel signal (often termed the analysis of low frequency fluctuation or “ALFF” signal) that can be compared between individuals with PD and controls (Zuo et al., 2010). Zou et al. (2009) showed in a recent work that the ALFF signal correlates at least modestly (R = 0.7) with CASL in healthy controls. The purpose of this study is to examine the concept that resting fMRI can be used as a diagnostic tool in PD.

Cross validation is a common statistical approach to measure reliability (Devroye et al., 1996). However, fMRI data sets have complex features including large dataset size, shifting sizes and shapes of regions of interest in repeated analyses, and moderate individual or “outlier” effects on the location and shape of regions of interest. While some groups have started to consider methods of fMRI cross validation (Chen and Herskovits, 2010, Mourao-Mirandmpel et al., 2005), proposed approaches have been complex and have not overlapped with current methods to anatomically localize regions of significant difference. Cross validation is therefore not commonly applied to fMRI datasets. We develop and present here the straightforward method we term “Reliability Mapping of Regional Differences (RMRD)”. We use this method to determine which regions are reliable in repeated analysis.

In our sample of participants, serendipity presented us with a “control” subject who developed symptoms of PD 6 months after our imaging scan (and was diagnosed with PD by a community physician 8 months after the resting state fMRI — later validated by our team). As an extension of our cross-validation protocol, we additionally evaluate if this participant's imaging could have pre-diagnostically “predicted” onset of PD.

Section snippets

Subjects

In this study we recruit 22 participants with PD and 19 matched normal controls (see Table 1 for subject demographics including age, gender, cognitive profile, and disease severity). All subjects are evaluated between 12 and 18 h after the last levodopa dose, in a practically-defined “off” state, as we wish to evaluate the patients in a state in which resting abnormalities in tone and mobility are maximized. Seven subjects with PD and 3 controls are excluded prior to any data processing due to

Demographics and neuropsychological testing of sample

In regards to our demographics of sample participants (Table 1), there are no significant difference in age is noted in the sample. The control population has more women, but this again is not significant (Fisher's Exact Test p = 0.25). Mini-Mental Status (MMSE — Folstein et al., 1975) and Montreal Cognitive Assessment (MoCA — Nasreddine et al., 2005) scores are not significantly different between the PD and control subjects. Individuals with PD perform significantly less well on the Trail Making

Discussion

The goal of our study is to investigate if ALFF analysis derived from resting fMRI can sensitively and specifically distinguish individuals with PD in the off medication state from healthy individuals. We show that individuals with PD have reliable alterations in ALFF signal that can be used to sensitively and specifically separate subjects with PD in the sample from controls; our results further suggest possible pre-diagnostic relevance for the technique. We use ALFF to create a full brain

Conclusion

In this work, we show that it is feasible to develop resting fMRI as a biomarker and diagnostic tool in PD. We were able to reliably distinguish subjects with PD from controls, and were able to identify a “preclinical” subject shortly before this participant developed motor signs and symptoms of PD. While sensitivity and specificity of our current methods are good, better reliability is needed to, for example, match reliability reported by Eidelberg's group using PET (Lozza et al., 2004,

In memoriam

Mark Yang passed away suddenly while gardening shortly after co-authoring this paper. This paper is therefore dedicated to the extraordinary life of Mark Yang, decorated statistician, magician, father, and gardener.

Author contributions

F.M.S. and Y.L. designed the study. Y.L. in addition provided critical expertise in functional imaging, without which this study could not have been performed. F.M.S. recruited the patients, performed the neurologic examination, analyzed the data, and wrote the manuscript. M.Y. assisted in statistical analysis and along with F.M.S. developed the reliability analyses. D.K. and A.S. stepped up to the plate after M.Y.'s tragic sudden death and reviewed the statistical analysis and approach, as

Conflict of interest

The authors declare no competing financial interests.

Acknowledgments

We thank Scott Higginbotham of Capitol City Harley Davidson, Gainesville Harley Davidson, and Hawg Wild for the Cure, without whose support this work could not have been performed. We also thank the Micanopy Doc Hollywood fund for support of this work. EPSRC grant EP/D066654/1 (JFC) also provided partial support for this work.

References (35)

  • Q. Zou et al.

    Static and dynamic characteristics of cerebral blood flow during the resting state

    Neuroimage

    (2009)
  • X.N. Zuo

    The oscillating brain: complex and reliable

    Neuroimage

    (2010)
  • C.F. Beckman et al.

    Investigations into resting-state connectivity using independent component analysis

    Philos. Trans. R. Soc. Lond. B Biol. Sci.

    (2005)
  • B. Biswal et al.

    Functional connectivity in the motor cortex of resting human brain using echo-planar MRI

    Magn. Reson. Med.

    (1995)
  • H. Braak et al.

    Stages in the development of Parkinson's disease-related pathology

    Cell Tissue Res.

    (2004)
  • J.S. Damoiseaux

    Consistent resting-state networks across healthy subjects

    PNAS.

    (2006)
  • L. Devroye et al.

    A Probability Theory of Pattern Recognition

    (1996)
  • Cited by (0)

    View full text