Alterations of resting-state fMRI measurements in individuals with cervical dystonia
Zhihao Li1,2, CecĂ­lia N Prudente3,4, Randall Stilla3, Krish Sathian5,6, Hyder A Jinnah7, and Xiaoping Hu2

1Affective and Social Neuroscience, Shenzhen University, Shenzhen, China, People's Republic of, 2Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, GA, United States, 3Neurology, Emory University, Atlanta, GA, United States, 4Physical Medicine and Rehabilitation, University of Minnesota, Minneapolis, MN, United States, 5Neurology, Rehabilitation Medicine, Psychology, Emory University, Atlanta, GA, United States, 6Rehabilitation R&D Center for Visual & Neurocognitive Rehabilitation, Atlanta VA Medical Center, Decatur, GA, United States, 7Neurology, Human Genetics and Pediatrics, Emory University, Atlanta, GA, United States

Synopsis

Cervical dystonia (CD) is a neurological movement disorder where the pathophysiology remains to be characterized. The present rfMRI study explored CD-associated brain alterations of (i) functional connectivity (FC), (ii) fractional amplitude of low frequency fluctuation (fALFF), and (iii) regional homogeneity (ReHo). The results revealed 25 significant regional alterations that confirm and extend existing knowledge. Additionally, using these regional alterations as diagnostic features, a support vector machine classifier identified 8 features that together yielded a maximum classification accuracy of 97%.

PURPOSE

Cervical dystonia (CD) is a neurological disorder characterized by abnormal movements and postures of the head1. To date, there have been only 2 applications2,3 of resting-state functional MRI (rfMRI) in evaluating the associated pathophysiology; but these previous studies were not specifically focused on the brain network underlying head movement and rfMRI features other than functional connectivity (FC) were not explored. The present study examined CD-associated alterations of FC by capitalizing on newly identified brain regions underlying isometric head rotation4. In addition to FC, which only reflects inter-regional signal synchronization, local regional alterations were also explored using measurements of the fractional amplitude of low frequency fluctuation (fALFF)5 and of regional homogeneity (ReHo)6. Finally, with alterations of FC, fALFF, and ReHo identified, a support vector machine (SVM)7 learning algorithm was implemented to identify rfMRI features with the highest power in distinguishing the patient and control participants.

METHOD

Thirty-two participants (controls: N=16, 10F6M, Age=57.1±12.7; patients: N=16, 9F7M, Age=56.7±11.4) were scanned during rest (no specific task other than eye fixation) with a 3T Siemens Trio scanner (EPI-BOLD, TR/TE/FA/FOV=2000ms/30ms/90o/220cm, volume=240, 30 axial slices, thickness/gap=4mm/0mm, matrix=64×64). AFNI (http://afni.nimh.nih.gov) was used for the analysis with regular preprocessing steps of despiking, slice timing correction, volume registration, noise reduction8, band pass filtering (0.08Hz<f<0.009Hz), spatial smoothing (FWHM=5mm), and spatial normalization. For the analysis of FC, we used a compound seed that included 4 spheres (r=5mm) located in the precentral gyrus (Talairach coordinates: x=21/-15/30/-54, y=-22/-22/-13/-1, z=52/49/46/31, LPI orientation) reported in a recent study of isometric head movement4. The voxel-wise measurements of FC, fALFF, and ReHo were individually derived and then compared between groups (group t-test) with potential confound of head motion modeled by a covariate of the maximum pairwise displacement in the motion parameter. To identify the rfMRI features for best group classification, the SVM analysis (radius basis function kernel) involved the procedures of “recursive feature elimination” (RFE) and “leave-one-out cross-validation” (LOOCV) (Fig.1); thus the highest classification accuracy was achieved with the fewest but most important features.

RESULTS

At the corrected threshold of p<0.05 (p<0.01/voxel and 581mm3 cluster), the CD participants exhibited both reduced (e.g. bilateral postcentral gyrus) and enhanced (e.g. bilateral basal ganglia and thalamus) FC in different brain regions (Fig.2). For the measurements of fALFF (Fig.3) and ReHo (Fig.4), only reduced values were observed in CD individuals (e.g. bilateral postcentral gyrus). Combining FC, fALFF, and ReHo, we have identified 25 alterations (Table.1). Feeding these 25 features into the SVM classifier with RFE, 8 features survived (highlighted in Table 1) and a maximum group classification accuracy of 97% was achieved.

DISCUSSION

Seeding in brain regions activated by isometric head rotation4, the present study revealed FC alterations in a brain network specific to head movements. FC alterations were confirmed at both the cortical and subcortical level, which resolved limitations of the previous studies2,3. In addition to the inter-regional FC, cortical (e.g. postcentral gyrus) and subcortical (e.g. basal ganglia and thalamus) alterations were also identified in local regional measurements; thus insights of pathophysiology in CD extended to reduced power and regional homogeneity of the intrinsic neural fluctuation. The rfMRI features identified here may have potential in future applications for improving diagnostic and prognostic evaluations in CD.

Acknowledgements

This work was supported by: National Center for Advancing Translational Sciences and NIH (U54NS065701, U54NS06571-03S1), as well as the Emory University Research Committee (UL1 RR025008).

References

[1]. Jinnah et al. 2013. Mov Disord 28:926.

[2]. Delnooz et al. 2013. PLoS One 8:e62877.

[3]. Delnooz et al. 2015 Brain Struct Funct 220:513.

[4] Prudente et al. 2015. J Neurosci 35:9163.

[5]. Zou et al., 2008. J Neurosci Methods 172:137.

[6]. Zang et al., 2004. Neuroimage 22:394.

[7]. Chang et al., 2011. ACM Trans Intell Syst Technol 2:27.

[8]. Jo et al., 2010. Neuroimage 52:571.

Figures

Fig.1. Schematic diagrams of the “leave-one-out cross-validation” (left) and “recursive feature elimination” (right).

Fig.2. Comparison of FC between the control (CON) and patient (CD) group at different axial levels (Z indices).

Fig.3. Comparison of fALFF between the control (CON) and patient (CD) group at different axial levels (Z indices).

Fig.4. Comparison of ReHo between the control (CON) and patient (CD) group at different axial levels (Z indices).

Table 1. The 25 alterations of rfMRI measurements identified in the present study. The 8 features that survived the procedure of “RFE” with the highest classification accuracy are highlighted in bold face.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1329