From March 2024 to January 2025, we recruited 130 participants from the Endocrinology Department and Physical Examination Center of the First Affiliated Hospital of Guangzhou University of Chinese Medicine, comprising 73 cases with T2DM and 57 healthy controls (HCs). The inclusion criteria for all participants were: (1) aged 35–70 years; (2) right-handedness; and (3) ≥6 years of formal education. The exclusion criteria included: (1) organic central nervous system disorders (e.g., brain injury and tumors); (2) Parkinson’s disease, hyperthyroidism, or epilepsy; (3) history of alcoholism, substance abuse, or psychiatric disorders; (4) Fazekas scale score of 2 or 3; and (5) contraindication for magnetic resonance imaging (MRI). T2DM cases met the 2019 diagnostic criteria for diabetes with disease duration >1 year (World Health Organization). We excluded individuals with Montreal Cognitive Assessment (MoCA) scores ≤10, diabetic ketoacidosis, stage 4 diabetic nephropathy, diabetic foot, or other severe complications. All MRI scans were visually inspected by two experienced radiologists for motion artifacts and overall image quality. No participants were excluded due to excessive motion or poor image quality, resulting in a final cohort of 73 T2DM cases and 57 HCs. The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the First Affiliated Hospital of Guangzhou University of Chinese Medicine. Written informed consent was obtained from all participants.

All participants completed the following standardized tests: Clock Drawing Test (CDT), Montreal Cognitive Assessment (MoCA), Digit Symbol Substitution Test (DSST), Auditory Verbal Learning Test (AVLT), Mini-Mental State Examination (MMSE), Digit Span Test (DST), Stroop Color-Word Test (CWT), Grooved Pegboard Test (GPT), and Trail Making Test Part A (TMT-A).

The MMSE and MoCA evaluated global cognitive function. The AVLT assessed memory, and the DSST measured visuomotor coordination. The TMT-A assessed processing speed and attention, and the DST evaluated working memory and attention. The GPT measured motor coordination and cognitive flexibility, the CDT assessed visuospatial abilities and executive function, and the CWT evaluated executive function and selective attention.

Raw scores from all neuropsychological tests were used in the analyses. To minimize the influence of demographic variability, age, sex, and years of education were included as covariates in subsequent statistical models rather than applying test-specific normalized scores. To minimize fatigue effects and potential order-related bias, all neuropsychological tests were administered in a fixed sequence, with brief rest periods provided between tests. The sequence was designed to progress from less demanding tasks (e.g., MMSE and MoCA) to more cognitively demanding or time-sensitive tasks (e.g., TMT-A, DSST, and CWT), ensuring consistency across participants. All cognitive tests were administered within 1 week of the MRI scan to minimize variability related to timing. The neuropsychological evaluations were conducted by trained examiners who were blinded to participants’ group allocation (T2DM or HC) to reduce potential assessment bias.

Neuroimaging was performed using a Siemens 3.0 T Prisma MRI scanner with a 64-channel head coil. Initial structural scans (T1-weighted, T2-weighted, and T2-FLAIR sequences) excluded participants with brain abnormalities. High-resolution T1-weighted images were acquired using sagittal 3D T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) sequences with the following parameters: echo time (TE) = 2.98 ms, repetition time (TR) = 2,530 ms, slice thickness = 1.0 mm, 192 slices; field of view (FOV) = 256 mm × 256 mm, flip angle = 7°, isotropic voxel size = 1 mm × 1 mm × 1 mm, and total acquisition time = 5 min 58 s.

All software procedures were carried out in the MATLAB computational environment (version R2022b; MathWorks, Natick, MA, United States). Three-dimensional T1-weighted images were processed using the SUIT toolbox¹ implemented within the Statistical Parametric Mapping software package (SPM12; https://www.fil.ion.ucl.ac.uk/spm). The preprocessing pipeline comprised the following sequential steps: (1) format conversion, in which DICOM image data were converted to NIfTI format; (2) spatial normalization, with anatomical origin aligned with the anterior commissure-posterior commissure (AC-PC) line, orientation verified, and resolution corrected; (3) tissue segmentation, including cerebellar isolation and gray matter segmentation; (4) spatial transformation, with normalization to SUIT standard space using the DARTEL algorithm; (5) atlas registration, achieved by reslicing into the SUIT cerebellar atlas space; and (6) spatial smoothing, with application of a 3-mm full width at half-maximum (FWHM) Gaussian kernel. Statistical analysis was performed using the DPABI 3.0 toolbox.² Specifically, the VBM module was applied to examine group-level differences in regional GMV, and the seed-based SCN analysis module was used to assess covariance patterns of cerebellar subregions. In addition to cerebellar-focused preprocessing, whole-brain VBM analyses were performed using the Computational Anatomy Toolbox (CAT12) in SPM12 to explore potential regional cerebral gray matter alterations. The same preprocessing pipeline (segmentation, normalization to MNI space, modulation, and smoothing with an 8-mm FWHM Gaussian kernel) was applied for cerebral GMV estimation. In this study, a 3-mm FWHM Gaussian smoothing kernel was applied during structural MRI preprocessing using the SUIT toolbox. This choice follows recommendations from Diedrichsen (19) and Diedrichsen et al. (20), who highlighted that a 3-mm kernel preserves anatomical precision in cerebellar analyses. Specifically, Diedrichsen (19) noted that smoothing the final probability map using a 3-mm Gaussian kernel produces smooth segmentation edges. Moreover, the SUIT toolbox, developed by Diedrichsen and colleagues, facilitates accurate normalization of cerebellar structures into atlas space using the DARTEL algorithm and has been widely used with a 3-mm smoothing kernel in prior neuroimaging studies. All MR images were visually inspected by two experienced radiologists for motion artifacts, image distortions, and other anomalies prior to inclusion in analyses. Participants whose scans exhibited significant artifacts or failed quality control checks were excluded from further processing.

Between-group differences in GMV between T2DM participants and HCs were evaluated using two-sample t-tests. Cerebellar subregions demonstrating statistically significant GMV alterations were then defined as regions of interest (ROIs) for subsequent seed-to-voxel SCN analyses. To account for interindividual neuroanatomical variability, global tissue volume and total intracranial volume (TIV) were estimated in native T1 space using the estimate TIV and global tissue volume module within the Computational Anatomy Toolbox (CAT12).

Tables 1, 2 present the demographic, clinical, and neuropsychological characteristics of the two groups. No significant differences were observed between T2DM participants and HCs with respect to age, body mass index (BMI), sex distribution, years of education, low-density lipoprotein (LDL) levels, systolic blood pressure, smoking status, or total cholesterol levels (all p > 0.05). In contrast, the T2DM group showed significantly lower high-density lipoprotein (HDL) levels and higher fasting blood glucose, glycated hemoglobin (HbA1c), and triglyceride levels compared to HCs. Regarding cognitive performance, T2DM participants demonstrated significantly poorer outcomes on the MMSE, MoCA, AVLT, GPT, DSST, TMT-A, DST, and CWT (all p < 0.05), whereas no significant group difference was observed in CDT scores.

The VBM analysis revealed that, compared to HCs, participants with T2DM exhibited significant GMV reductions in the bilateral cerebellar crus I, left crus II, left lobules I–IV, left lobule IX, and right lobule VIIb. No cerebellar subregions demonstrated increased GMV in the T2DM group (Table 3 and Figure 1). Seed-to-voxel SCN analyses were then performed using these six GMV-reduced regions as ROIs. This analysis identified two cerebellar-based structural networks showing significant between-group differences. Specifically, reduced morphological covariance was observed between the left crus I and the left middle temporal gyrus, middle occipital gyrus, and angular gyrus, whereas increased morphological covariance was detected between the left lobules I–IV and the right caudate nucleus (Table 3 and Figure 2). Notably, none of the cerebral regions that covaried with cerebellar atrophy (e.g., middle temporal gyrus, angular gyrus, middle occipital gyrus, and caudate nucleus) exhibited significant GMV reduction in the between-group whole-brain VBM analysis, indicating that these areas were structurally preserved at the gross morphological level.

T2DM patients exhibited significant GMV reductions in the bilateral cerebellar crus I, left crus II, left lobules I–IV, left lobule IX, and right lobule VIIb. These findings largely align with previous reports of GMV reductions in the right crus I and bilateral lobules I–V/I–IV in T2DM (21). Notably, previous studies did not systematically assess the relationships between these cerebellar structural alterations and cognitive function.

The anterior lobe of the cerebellum plays a central role in regulating muscle tone, maintaining postural balance, and coordinating voluntary movements (22, 23). Atrophy within this region in T2DM may therefore contribute to impaired motor coordination and reduced fine motor control. Consistent with this, the present study demonstrated that participants with T2DM exhibited significantly poorer motor coordination and diminished neurocognitive flexibility compared to HCs.

Beyond its motor functions, the cerebellum also contributes to higher-order cognition, particularly through the posterolateral hemisphere. Cognitive tasks typically engage three functionally distinct, non-motor cerebellar regions: (1) lobule VI–crus I, (2) crus II–VIIB, and (3) lobules IX–X (24–26). Our neuroimaging results showed that GMV reductions in these posterolateral regions were significantly associated with multidomain cognitive impairments, including visuospatial deficits, executive dysfunction, impaired memory consolidation, disrupted limb coordination, and reduced neurocognitive flexibility. Notably, bilateral crus I/II are key nodes within networks supporting executive function, working and episodic memory, navigation, and attentional control (27–30), while lobules VI and VII are typically engaged during reading tasks (31). These converging findings confirm the present results and indicate that cerebellar degeneration constitutes a critical neuroanatomical substrate underlying T2DM-related cognitive impairment.

Notably, structural covariance (SC) is understood to reflect both structural and functional connectivity (32), and cerebro-cerebellar covariance patterns are thought to mirror the underlying anatomical coupling between sensorimotor and cognitive networks (33). In the present study, SCN analysis revealed decreased connectivity between the left crus I and both the DMN and the VN. Importantly, reduced SCN integrity in these pathways was negatively correlated with memory and executive performance in T2DM, suggesting that disrupted cerebro-cerebellar structural connectivity constitutes a potential neural mechanism underlying cognitive impairment.

Crus I/II has been identified as the principal cerebellar hub interfacing with the DMN (24, 29). Consistent with our findings, previous resting-state fMRI studies have reported disrupted and reorganized functional connectivity between the cerebellum and the DMN in T2DM (34). The present results, therefore, not only highlight the role of the cerebellum as a critical neuroanatomical substrate for T2DM-related cognitive decline but also emphasize that aberrant morphological connectivity within cerebro-cerebellar DMN/VN circuits contributes to the pathophysiology of this dysfunction. Taken together, converging evidence from structural and functional network studies strengthens the reliability of these findings. It may be hypothesized that the left crus I represents a neuroanatomically vulnerable locus, or an early-affected region, mediating cognitive decline in T2DM.

The caudate nucleus is critically involved in both motor and cognitive domains, contributing to motor planning/execution, spatial working memory, reward processing, motivation, and emotion (35–37). Aberrations in caudate networks have been recognized as predictors of cognitive decline in aging and neurodegenerative conditions (38, 39). Notably, the present study revealed enhanced structural connectivity between the left lobules I–IV and the right caudate nucleus. This connectivity was negatively correlated with short-term memory performance (AVLT-3), while it was positively correlated with processing speed and attention (TMT-A).

Previous studies have reported increased caudate-cortical connectivity in elderly populations exhibiting dual cognitive and motor decline (11). This finding may thus represent a unique neuroanatomical feature of T2DM-related cognitive dysfunction. Whether this connectivity enhancement reflects a maladaptive change or a compensatory mechanism remains uncertain and warrants further mechanistic investigation.

Several methodological limitations warrant consideration: First, the cross-sectional design with a relatively small sample size limits causal inference. Second, individual variability in hypoglycemic medications precluded assessment of their effects on cerebellar subregional GMV. Third, the potential confounding effects of mild microvascular complications in some participants remain unaddressed. Future studies should examine cerebellar structural changes across different T2DM progression stages to better characterize neuroimaging biomarkers of cerebellar impairment.