Progressive reduction in basal ganglia explains and predicts cerebral structural alteration in type 2 diabetes

Abstract

Type 2 diabetes is consistently reported to be associated with reduced gray matter, mainly in the cortical-striatal-limbic networks. However, little is known about how the progression of diabetes affects cerebral gray matter. To investigate, we collected 543 age- and sex-matched participants of nondiabetes, prediabetes, and diabetes. Voxel-based morphometry using a linear trend model was performed to reveal brain regions associated with disease progression. The Granger causal network of structural covariance was used to assess the causal relationships of brain structural alterations according to disease progression. Multivariate pattern analysis was applied for the stage-specific predictions of hyperglycemia. We detected a linear trend of gray matter volume reduction in the basal ganglia with disease progression (P < 0.05, FWER corrected), which caused a reduction in bilateral temporal gyri, frontal pole, parahippocampus, and bilateral posterior cingulate/precuneus volumes. In addition, the gray matter pattern of the basal ganglia could predict patients with diabetes (accuracy 60.12%, p = 0.002). In conclusion, the basal ganglia is the brain area with progressive gray matter reduction as diabetes progress. The reduced volume in the basal ganglia causes widespread gray matter reductions throughout diabetes progression. These findings indicate that the basal ganglia play a key role in diabetes by affecting the cortical-striatal-limbic network.

Keywords

Type 2 diabetes structural MRI basal ganglia causal covariance network multivoxel pattern approach

Introduction

Type 2 diabetes is a common disease that impacts more than 360 million people worldwide.¹ Patients with type 2 diabetes show cognitive and mood impairment, indicating a potential for brain injury in regions that control these functions. A recent meta-analytic study using structural MRI revealed reduced gray matter volume in multiple areas of cortex-striatal-limbic regions—the hippocampus, middle temporal cortex, superior temporal cortex, medial frontal gyrus, and insula—indicating that multiregional brain abnormalities exist in type 2 diabetes.¹ However, the pathophysiological mechanisms underlying how hyperglycemia affects the entire brain remain poorly understood.

An important characteristic of type 2 diabetes is that it is a progressive disease. Prediabetes is an intermediate state of hyperglycemia with glycemic parameters above normal but below the diabetes threshold. Annually, approximately 5∼19% of prediabetes progresses to overt type 2 diabetes, in which the risks of myocardial infarction, stroke, microvascular events, and mortality are all strongly associated with hyperglycemia,² while any reduction in hyperglycemic level is likely to reduce the risk of complications.³ Given the progressive nature and multiregional brain abnormalities of type 2 diabetes, investigating how progressive brain damage begins and spreads to the entire brain is necessary to understand neurodegeneration in patients with type 2 diabetes.

Several recently developed analysis methods can be used to deepen our understanding of the cerebral effects of hyperglycemia. Causal structural covariance network (CaSCN) analysis has been introduced to quantify possible causal relationships between different brain regions.^4
–6 When morphometric data are ranked according to progression or disease severity, e.g., the level of hyperglycemia, Granger causality (GC) analysis can be applied to these sequenced data to assess the causal relationships of structural alterations among brain regions.⁶ In addition, multivariate pattern analysis (MVPA) methods for imaging data are a modern machine learning-based approach that takes into account interactions between regions, as opposed to mass-univariate methods. MVPA is ideally suited to make predictions for individual subjects based on brain imaging patterns.⁷

In our study, we aimed to investigate the progressive aspect of structural brain alterations according to the level of hyperglycemia and to further evaluate the causal effect and elucidate the predictive pattern using CaSCN and MPVA.

Material and methods

Patient selection

This retrospective study was approved by the Institutional Review Board (IRB No. 2203-178-1313), and informed consent was waived. We retrospectively enrolled 1122 men and women aged 20 to 88 years who underwent a comprehensive annual or biennial physical examination between January 2016 and May 2018 at Seoul National University Hospital Health Promotion Center in the Republic of Korea. Those who had both glycated hemoglobin (HbA1c) and high-resolution structural magnetic resonance imaging (MRI) available were enrolled. MRI was acquired using a 3.0 T MR scanners (Discovery MR750w, GE Healthcare, Milwaukee, WI) with a 24-channel head coil. Volumetric T1-weighted images were obtained using a 3-dimensional, fast spoiled gradient-echo pulse sequence with the following parameters: repetition time, 8.5 ms; echo time, 3.2 ms; inversion time, 450 ms; flip angle, 12°; field-of-view, 256 × 256 mm; acquisition matrix, 256 × 256; number of excitations, 1; slice thickness, 1 mm; number of slices, 154-172 according to head size; sagittal plane; and voxel size, 1 mm³. Two board-certificated neuroradiologists (K.S.C and I.H.) evaluated the MRI and checked for motion artifacts and lesions. We excluded subjects as per the following criteria: previous brain defects including lacunar infarctions over 3 mm⁸ or reported previous history of type 1 diabetes. Then, we categorized the patients into three groups according to their HbA1c range: the nondiabetes group (less than 5.6% HbA1c), the prediabetes group (5.7%–6.4% HbA1c), and the diabetes group (6.5% or higher HbA1c) (Figure 1).⁹ Because we assumed that sufficient long-term deterioration from hyperglycemia is mandatory to cause a brain morphological change,¹⁰ HbA1c level, a marker of the average glucose levels spread over a two- to three-month period,⁹ was selected as a criterion.

Figure 1.

Patient selection procedure for brain effects of diabetes.

Propensity scores were matched by selecting the cases in the 3 groups, and age and sex were used as matching parameters using the ‘matchit’ R package (R core team, R foundation for statistical computing, Vienna, Austria).^11,12 Age and sex were chose following previous articles using CaSCN and voxel based morphometry (VBM).^4,13 We used optimal matching on the propensity score, which yielded adequate balance, as indicated in Supplementary figure 2. The propensity score was estimated using logistic regression. After matching, standardized mean differences for the covariates were below 0.001, indicating adequate balance. Ultimately, 543 subjects (181 per group) were enrolled. For between-group statistical analyses, ANOVA tests were used for continuous variables, and chi-square tests were used for categorical variables.

Voxel-based morphometric analysis: Stepwise reduction in diabetes

MRI was analyzed with FSL-VBM,¹⁴ (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLVBM), an optimized VBM protocol¹⁵ carried out with FSL tools.¹⁶ First, structural images were brain-extracted and gray matter-segmented before being registered to the MNI 152 standard space using nonlinear registration.¹⁷ The resulting images were averaged and flipped along the x-axis to create a left-right symmetric, study-specific gray matter template. Second, all native gray matter images were nonlinearly registered to this study-specific template and “modulated” to correct for local expansion (or contraction) due to the nonlinear component of the spatial transformation. The modulated gray matter images were then smoothed with an isotropic Gaussian kernel with a sigma of 3 mm. The output of each stage of the analysis was visually checked according to FSL-VBM guidelines, using ‘slicesdir’ command.

Finally, a voxelwise general linear model (GLM) was applied using permutation-based nonparametric testing, correcting for multiple comparisons across space. To estimate the stepwise GM volume alterations in the three groups – nondiabetes, prediabetes and diabetes – we modeled a linear trend of volume decrease (Figure 2, white panel). We modeled our linear trend explicitly as a regressor, i.e. the model had an intercept and a column of 1 = diabetes, 2 = prediabetes, 3 = nondiabetes. The testing was performed by the FSL randomize program, which uses 5000 random permutations using total intracranial volume (TIV) as a covariate. According to the advice of VBM, the TIV should be regressed out during morphological analysis.¹⁸ Threshold-free cluster enhancement (TFCE)¹⁹ was used to obtain the area of linear gray matter volume decrease according to diabetes progression after accounting for multiple comparisons by controlling for the familywise error (FWE) rate at p < 0.05.

Figure 2.

Stepwise gray matter volume reduction in the nondiabetes, prediabetes and diabetes groups. The result of FSL-VBM using linear trend modeling (white panel on the left). Reduced gray matter volume in bilateral basal ganglia (red) images are overlaid on an axial template. The color bar represents p values <0.05, corrected by TFCE (raw image available at https://neurovault.org/images/791029/).

Mapping the causal effect of stepwise basal ganglia atrophy on diabetes progression using seed-based CaSCN

The GM volume data of all 181 diabetes patients were sequenced according to the ranking of HbA1c levels from low to high. This data sequencing was analogous to characterizing the progressive property of hyperglycemia on the basis of cross-sectional data. Subsequently, pseudoprogression was used to construct seed-based CaSCNs using the Brain Covariance Connectivity Toolkit.¹⁸ The seed region was selected from the previously mentioned VBM analysis (shown in Figure 2). Residual-based GC analysis was performed in a voxelwise manner for all the voxels in the mask of the seed region (the formal details are shown in Appendix 1). In detail, the Granger causal effect of the seed region (i.e., X, Figure 2 in our case) on the other regions (i.e. Y) were is determined by two linear regressions. After sorting the brain images according to the HbA1c ranking, if the combination of the lower rank X and Y can better predict (i.e. small residual in regression) the present Y than the lower rank Y alone, then the X have a causal effect on Y.

The CaSCN could allow assessment of the causal effect of GM volume alteration of the seed region (i.e., shown in Figure 2) on the other regions (i.e., entire gray matter other than the seed region). As the seed exhibited a stepwise reduction in GM volume in diabetes, a positive GC value indicated that the same GM volume alteration (reduced) in the regions lagged behind the seed atrophy, which may suggest that the reduction is driven by the seed.⁴ Sex, age, total intracranial volume, and increases in HbA1c between subjects were regressed as covariates in conducting CaSCN analysis.⁴

Multiclass gaussian process classification using basal ganglia gray matter volume

We used a 3-class (GPC) classification that was aimed at simultaneously discriminating each group (nondiabetes, prediabetes and diabetes) from one another as implemented in the Pattern Recognition for Neuroimaging Toolbox (PRoNTo) software (http://www.mlnl.cs.ucl.ac.uk/pronto).^20,21 Whole-brain GM density images with or without basal ganglia mask were used as prediction features. We employed a 10-fold cross-validation approach to assess classifier generalizability, whereby we excluded a matched pair of subjects to comprise the test set and inferred all parameters from the remaining data (training set) before applying this classifier to predict the labels for the test samples. Sex, age, and total intracranial volume were selected as regressed-out covariates. Finally, we computed the balanced accuracy as the mean of sensitivity and specificity, which quantifies the overall categorical classification performance of the classifier in a way that accommodates potential class imbalance in the data. Permutation testing, which uses 1000 random permutations, was used to derive a p value to determine if the balanced accuracy exceeded chance levels (33.33%).

A pattern of predictive weights for each of the classes was obtained. A high positive score in the weight vector for a given group denotes a strong positive contribution to a prediction in favor of that group, while a high negative score for the same group denotes a strong negative contribution.

Results

Table 1 summarizes the characteristics of the study population. There were no significant differences with respect to age or sex in those matched with propensity score.

Table 1.

Clinical characteristics of the study population.

	Normal(n = 181)	Prediabetes(n = 181)	Diabetes(n = 181)	P value^a
Age (years)	63 ± 6	63 ± 7	63 ± 7	.775
Sex (Male: Female)	50:131	52:129	53:128	.939
HbA1c level	5.4 ± 0.2	5.9 ± 0.2	7.4 ± 1.0	.000
Total intracranial volume (cm³)	1,965 ± 40	1,964 ± 34	1,973 ± 42	.081

Data are the mean ± standard deviation.

By ANOVA test for continuous variables and chi-square test for categorical variables.

Stepwise gray matter volume reduction in the basal ganglia

As illustrated in Figure 2, the substantial area of the bilateral basal ganglia showed a linear trend of smaller gray matter volume as the disease progressed from nondiabetes to prediabetes to diabetes. Two clusters were detected, both located in the basal ganglia with peak voxel locations of −14, 14, −8 and 14, 8, −10 in MNI coordinates (mm).

Causal analysis of the structural covariance network – Stepwise reduction in the basal ganglia explains the cerebral effects of diabetes

To elucidate the causal effect of stepwise volume reduction due to diabetes, the CaSCN using the resulting voxels in Figure 2 was constructed. The result of CaSCN is illustrated in Figure 3. CaSCN results demonstrated positive GC from the seed in the basal ganglia to the bilateral temporal gyri, bilateral frontal pole, left parahippocampus, and bilateral posterior cingulate/precuneus. This indicates that the seed regions may be the epicenter and exhibit positive causal effects on other regions, thus potentially exerting a damaging effect on other regions. Interestingly, as demonstrated in Figure 3 (white box), the distribution of regions revealed in the CaSCN resembles a previous meta-analysis investigating structural alterations in type 2 diabetes.

Figure 3.

The results of causal analysis of the structural covariance network show causal effects of gray matter atrophy patterns in patients with diabetes. Causal networks were constructed by applying Granger causality (GC) analysis to sequenced morphometric data according to the ranks of HbA1c from low to high. The area of reduced volume in Figure 2 was used as the seed region on the basis of voxel-based morphometric analysis. Areas in red indicate that the gray matter volume reduction lagged behind seed region atrophy, which may suggest that it is driven by the seed region (p < 0.001, (raw image available at https://neurovault.org/images/790995/)). The areas resemble previous meta-analyses investigating gray matter reduction in type 2 diabetes (image in white rectangle, adapted from Liu et al. Front. Aging Neurosci. 9:161¹).

Multiclass gaussian process classification: Stepwise reduction in basal ganglia predicts cerebral effects of diabetes

The three-class classifier trained to discriminate among nondiabetes, prediabetes, and diabetes subjects using basal ganglia mask achieved a class accuracy of 60.12% for diabetes (permutation p = 0.002). The balanced accuracy was 37.7% (permutation p = 0.049, see Figure 4 for the 3-class weight maps and Figure S3 for confusion matrix). When the whole-brain gray matter was used as a prediction feature, the balanced accuracy was 38.6% (permutation p = 0.013, see Figure S2 for atlas-based weight maps and Figure S3 for confusion matrix). For both classifications, the accuracy significantly exceeded the 33.33% that would be predicted for a 3-class classifier by chance. We emphasize that multiclass classification is a more difficult problem than binary classification.²² In MVPA analysis, we are generally asking whether the decoding classification accuracies are significantly greater than what could be expected by chance, and significant classification accuracy suggests sufficient information contained in the basal ganglia to indicate which hyperglycemic state the subject is in.²³

Figure 4.

Nonthresholded three-class multivariate discrimination weight maps. Multivariate discrimination weight map for diabetes (upper row, raw image: https://neurovault.org/images/790996/) vs. prediabetes (middle row, raw image: https://neurovault.org/images/791007/) vs. nondiabetes (bottom row, raw image: https://neurovault.org/images/791018/). The intensity values of the multivariate discrimination weight maps illustrate the relative positive weight distributions (orange) and negative weight distributions (cyan). Note that the gray matter volume in the caudate nucleus (red arrow) negatively predicts DM, similar to the mass-univariate approach.

Discussion

In this study, we aimed to discover which brain region was affected according to the progression of hyperglycemia and to demonstrate the region’s causal relationship to the entire brain. The results of VBM analysis showed progressive gray matter reduction in the bilateral basal ganglia (with FWER <0.05, Figure 2), and the CaSCN analysis showed that gray matter volume reduction in the basal ganglia explains the cerebral effects of diabetes. Furthermore, MPVA revealed that the gray matter pattern of the basal ganglia is predictive of type 2 diabetes (class accuracy = 60.12%, permutation p = 0.002). Based on the results, we propose a model explaining how hyperglycemia affects the brain (Figure 5).

Figure 5.

A proposed model of hyperglycemia affecting the brain. We propose a model for hyperglycemia affecting the brain. Hyperglycemia insults the cerebral small vessels (e.g., lenticulostriate artery, white arrow) supplying the basal ganglia, thus resulting in volume reduction. The insult on the basal ganglia causes additional networkwise volume reduction in large areas within the cortex-striatal-limbic network (yellow curved arrows).

Stepwise gray matter volume reduction in the basal ganglia

Our VBM analysis focusing on progressive reduction according to disease progression revealed gray matter reduction in the bilateral basal ganglia (Figure 2). The basal ganglia, which are primarily supplied by the small branches of the middle cerebral artery (in particular, lenticulostriate perforating arteries, Figure 5), are affected by small vessel pathologies caused by hyperglycemia.^24
–26 Specifically, by reducing endothelial nitric oxide synthase activity and increasing adhesion molecule expression in the endothelium, the hyperglycemic state causes cerebral small vessel disease (SVD).²⁵ Imaging evidence of SVD, including lacunar infarctions and enlarged perivascular spaces, involving the basal ganglia is typically found, indicating that the basal ganglia is prone to SVD.²⁴ In addition, long-standing hyperglycemia often results in diabetic striatopathy,²⁷ supporting our result of decreased volume in the striatal area (Figure 2).

To support the mass-univariate VBM result, we additionally conducted a MVPA using basal ganglia gray matter as a prediction feature of hyperglycemic status. The above chance level prediction result of MVPA indicates that there is sufficient information contained in the gray matter pattern of the basal ganglia; in other words, the gray matter pattern predicts the hyperglycemic status. The pattern of brain regions that discriminated diabetes, prediabetes and nondiabetes is largely in areas that have been found to be progressively reduced, e.g., the left anterior basal ganglia (Figure 4, red arrow). When predicted with whole-brain, the left putamen was remained highly important in predicting diabetes patients (Figure S2).

Stepwise gray matter volume reduction in the basal ganglia explains the cerebral effects of diabetes

We subsequently evaluated the causal relationship between the basal ganglia and progressive morphometric alterations according to hyperglycemia using the CaSCN. A directional network demonstrated that changes in the basal ganglia seed region were potentially causally related to multiple brain areas of the bilateral temporal gyri, frontal pole, left parahippocampus, and bilateral posterior cingulate/precuneus (Figure 3). Interestingly, the results of the CaSCN analysis resemble and overlap with the results of the previous meta-analysis investigating brain morphologic alterations in type 2 diabetes (Figure 3, image in white rectangle, adapted from Liu et al. Front. Aging Neurosci. 9:161¹) In a previous meta-analysis investigating the cerebral effect of type 2 diabetes, Liu et al. suggested that most of the brain regions with reduced gray matter volume were located in the default mode network (DMN), which may eventually cause poor cognitive performance.¹ Additionally, it has been suggested that gray matter decreases in the limbic system (e.g., insula and parahippocampus) are important components of various functions, including emotion, behavior, motivation and long-term memory. These functions are prone to be impaired in type 2 diabetes patients.

A proposed model of hyperglycemia affecting the brain

To combine previously known facts and our discovery, we propose a model of hyperglycemia affecting the brain (Figure 5). Hyperglycemia causes small vessel endothelial damage, and the basal ganglia, which are supplied by small vessels, are prone to endothelial damage.²⁸ The existence of small vessel damage due to hyperglycemia is indirectly supported by the T1 hypointense signal intensities in white matter, which was also observed in our study (see supplementary figure S4 for more details).²⁹ Based on the VBM and MVPA results, we suggest the basal ganglia as the region where brain alterations begin due to hyperglycemia.

Interestingly, brain regions are not segregated, but they are wired. There are attempts to presume the brain as a multiscale networked system and understand the pathogenesis of brain disorders from a network perspective.³⁰ The structural covariance connectivity, constructed by morphological images, delineates the synchronous gray matter atrophy among regions and maps the topological pattern of network-reorganized regions,¹⁸ suggesting that the morphologic properties of brain regions are related to one another. Taking this network perspective, an alteration in one brain region can propagate to other regions that is linked in a network. Thus, our model suggests that progressive reduction in the basal ganglia propagates with diabetes progression and results in overt cortical-striatal-limbic network alterations in previous studies.¹

Our study had several limitations. First, the data organized according to hyperglycemia ranges do not directly reflect the real temporal sequence of illness progression. Thus, longitudinal studies must be performed to further clarify the causal relationship between structural deficits. Second, other disease etiologies that can result in small vessel disease, such as hypertension, dyslipidemia, and obesity were not matched in this study. A history of these diseases may confound the results. Third, this is a single-center study that may restrict the generalizability of the results. Fourth, the relatively low prediction accuracy of MVPA may also limit the use of brain imaging as a biomarker for hyperglycemia.

In conclusion, we demonstrated that the basal ganglia is the area of stepwise gray matter reduction according to diabetes progression. The reduced volume in the basal ganglia causes widespread, networkwise gray matter reduction over diabetes progression. To interpret the result, we propose a basal ganglia initiation and networkwise propagation model of hyperglycemia affecting the brain. These findings indicate that the basal ganglia play a key role in type 2 diabetes by affecting the cortical-striatal-limbic network.

Supplemental Material

sj-pdf-1-jcb-10.1177_0271678X231197273 - Supplemental material for Progressive reduction in basal ganglia explains and predicts cerebral structural alteration in type 2 diabetes

Supplemental material, sj-pdf-1-jcb-10.1177_0271678X231197273 for Progressive reduction in basal ganglia explains and predicts cerebral structural alteration in type 2 diabetes by Kyu Sung Choi, Inpyeong Hwang, Joon Ho Moon and Minchul Kim in Journal of Cerebral Blood Flow & Metabolism

Footnotes

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Data and resource availability

The raw result of neuroimaging analysis can be seen and downloaded at . The de-identified data are available upon reasonable request, subject to approval from the institutional IRB.

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: This work has been supported by the Bio & Medical Technology Development Program of National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. 2021M3E5D2A01022493) and the Phase III (Postdoctoral fellowship) grant of the SPST (SNU-SNUH Physician Scientist Training) Program.

Authors’ contributions

Study Design, KSC, IH, MK; Data collection, KSC, IH; Data analysis, interpretation, KSC, IH, JHM, MK; Figures, MK; Manuscript Writing, KSC, IH, JHM, MK. All authors revised and approved the final version of the manuscript. KSC, IH, and MK are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

ORCID iD

Minchul Kim

Supplemental material

Supplemental material for this article is available online.

References

Liu

Wang

, et al. Reduced gray matter volume in patients with type 2 diabetes mellitus. Front Aging Neurosci 2017; 9: 161.

Fonseca

VA.

Defining and characterizing the progression of type 2 diabetes. Diabetes Care 2009; 32 Suppl 2: S151–S156.

Stratton

Adler

Neil

HAW

, et al. Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. BMJ 2000; 321: 405–412.

Jiang

Luo

, et al. Progressive reduction in gray matter in patients with schizophrenia assessed with MR imaging by using causal network analysis. Radiology 2018; 287: 633–642.

Zhang

Liao

, et al. Hippocampus‐associated causal network of structural covariance measuring structural damage progression in temporal lobe epilepsy. Hum Brain Mapp 2017; 38: 753–766.

Chen

Cui

Fan

Y-S

, et al. Progressive brain structural alterations assessed via causal analysis in patients with generalized anxiety disorder. Neuropsychopharmacology 2020; 45: 1689–1697.

Cohen

Daw

Engelhardt

, et al. Computational approaches to fMRI analysis. Nat Neurosci 2017; 20: 304–313.

Loos

Makin

Staals

, et al. Long-term morphological changes of symptomatic lacunar infarcts and surrounding white matter on structural magnetic resonance imaging. Stroke 2018; 49: 1183–1188.

Sherwani

Khan

Ekhzaimy

, et al. Significance of HbA1c test in diagnosis and prognosis of diabetic patients. Biomark Insights 2016; 11: 95–104. BMI.

10.

Rezaeyan

Asadi

Kamrava

, et al. Reorganizing brain structure through olfactory training in post-traumatic smell impairment: an MRI study. J Neuroradiol 2022; 49: 333–342.

11.

Stuart

King

Imai

, et al. MatchIt: nonparametric preprocessing for parametric causal inference. J Stat Softw 2011; 8.

12.

Team

RC.

R. A language and environment for statistical computing. Vienna, Austria, 2013.

13.

Douaud

Mackay

Andersson

, et al. Schizophrenia delays and alters maturation of the brain in adolescence. Brain 2009; 132: 2437–2448.

14.

Douaud

Smith

Jenkinson

, et al. Anatomically related grey and white matter abnormalities in adolescent-onset schizophrenia. Brain 2007; 130: 2375–2386.

15.

Good

Johnsrude

Ashburner

, et al. A voxel-based morphometric study of ageing in 465 normal adult human brains. Neuroimage 2001; 14: 21–36.

16.

Smith

Jenkinson

Woolrich

, et al. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 2004; 23 Suppl 1: S208–S219.

17.

Andersson

Jenkinson

Smith

Non-linear optimisation FMRIB technical report TR07JA1. Oxford, UK: University of oxford FMRIB Centre, 2007.

18.

Zhang

Liu

, et al. BCCT: a GUI toolkit for brain structural covariance connectivity analysis on MATLAB. Front Hum Neurosci 2021; 15: 641961.

19.

Smith

Nichols

, Threshold-free cluster enhancement: addressing problems of smoothing, threshold dependence and localisation in cluster inference. Neuroimage 2009; 44: 83–98.

20.

Marquand

Howard

Brammer

, et al. Quantitative prediction of subjective pain intensity from whole-brain fMRI data using gaussian processes. Neuroimage 2010; 49: 2178–2189.

21.

Schrouff

Rosa

Rondina

, et al. PRoNTo: pattern recognition for neuroimaging toolbox. Neuroinformatics 2013; 11: 319–337.

22.

Lim

Marquand

Cubillo

, et al. Disorder-specific predictive classification of adolescents with attention deficit hyperactivity disorder (ADHD) relative to autism using structural magnetic resonance imaging. PloS One 2013; 8: e63660.

23.

Etzel

JA.

MVPA significance testing when just above chance, and related properties of permutation tests. In: 2017 international workshop on pattern recognition in neuroimaging (PRNI), pp. 1–4. Piscataway, NJ: IEEE, 2017.

24.

Mahammedi

Wang

Williamson

, et al. Small vessel disease, a marker of brain health: what the radiologist needs to know. Am J Neuroradiol 2022; 43: 650–660.

25.

Umemura

Kawamura

Hotta

, Pathogenesis and neuroimaging of cerebral large and small vessel disease in type 2 diabetes: a possible link between cerebral and retinal microvascular abnormalities. J Diabetes Investig 2017; 8: 134–148.

26.

Liu

Rutten-Jacobs

Liu

, et al. Causal impact of type 2 diabetes mellitus on cerebral small vessel disease: a mendelian randomization analysis. Stroke 2018; 49: 1325–1331.

27.

Chua

C-B

Sun

C-K

Hsu

C-W

, et al. “Diabetic striatopathy”: clinical presentations, controversy, pathogenesis, treatments, and outcomes. Sci Rep 2020; 10: 1594–11.

28.

Shibuya

Leite

C. d C

Lucato

, Neuroimaging in cerebral small vessel disease: update and new concepts. Dement Neuropsychol 2017; 11: 336–342.

29.

Dadar

Maranzano

Ducharme

, et al. Validation of T 1w‐based segmentations of white matter hyperintensity volumes in large‐scale datasets of aging. Hum Brain Mapp 2018; 39: 1093–1107.

30.

Bullmore

Sporns

, Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 2009; 10: 186–198.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.28 MB