The cerebellar glucose metabolism in moyamoya vasculopathy and its correlation with neurocognitive performance after cerebral revascularization surgery: a [¹⁸F]FDG PET study

This study was approved by the institutional review boards of Huashan Hospital of Fudan University (reference number: KY2014278). The study population was divided into two parts. Firstly, in order to determine the relationship between neurocognitive performance and cerebellar glucose metabolism, we retrospectively investigated the 132 patients with MMV who underwent neuropsychological assessment and [¹⁸F]FDG PET scan between 2019 January and 2022 July. Patients were excluded for the following reasons: severe brain structure damage (e,g., hydrocephalus, massive cerebral infarct history, intracerebral hematoma evacuation history), stroke history within 6 months, drugs and alcohol abuse, left-handedness, no other serious medical comorbidities (e.g. cancer, Alzheimer’s disease, etc.). Ultimately, ninety-three patients were deemed eligible and included in this study.

Secondly, between 2022 January and 2023 May, we prospectively followed up 49 MMV patients with VCI who underwent combined cerebral revascularization surgery. Inclusion criteria included: digital subtraction angiography (DSA) confirmed MMV patients, no stroke events within 6 months before surgery, modified Rankin Scale ≤ 3, age ranging from 18 to 60, no history of brainstem and cerebellum stroke confirmed by MRI scan, agreement to undergo the combined revascularization, and voluntary participation in this study. Subjects might be excluded as following reasons: left-handedness, severe perioperative complications (e,g, massive cerebral infarction, intracerebral hemorrhage, epidural hematoma, etc.), no suitable receptor vessels at middle cerebral artery (MCA) M4-5 segment, refusal to complete the PET scan and neuropsychological assessments, stroke events during follow-up period, and no any collateral circulation formation from external carotid artery confirmed by DSA when follow-up. Additionally, forty-seven age- and gender- matched health controls (HC) were recruited from PET center of Fudan University health pool.

Neuropsychological assessments were conducted by three professional neuropsychologists at baseline and follow-up. Global cognitive performance was assessed with Mini-Mental State Examination (MMSE), and Memory and Executive Screening (MES) [17]. Attention was tested by Symbol Digit Modalities Test (SDMT) [18], and Trail Making Test A (TMT-A) [19]. Memory was evaluated with Chinese auditory verbal learning test (AVLT) [20]. Animal/Vegetable Verbal Fluency Test (VFT) [21], and Boston Naming Test (BNT) [22] acted as measurements of Language. Trial Making Tests B (TMT-B) [19] and Alternation VFT was used to represent the executive function. Visuospatial function was estimated by Clock Drawing Test (CDT) [23] and Rey-Osterrieth Complex Figure Test (CFT) [24]. The diagnosis of VCI was based on the Guidelines from the Vascular Impairment of Cognition Classification Consensus Study (VICCCS) [25]. Patients were assigned cognitive impairment if any neuropsychological test was over 1.5 standard deviations below the normal Chinese population. Patients who lost independence in daily life were classified as vascular dementia (VaD).

The surgical method was revascularization using combined direct and indirect bypass [26, 27]. Briefly, one superficial temporal artery (STA) branch was anastomosed to MCA M4-5 segment. An encephalo-dura-myo-synangiosis (EDMS) was performed as the indirect bypass. Indocyanine green (ICG) fluorescence imaging was regularly used to check the patency of the STA-MCA bypass after microsurgical anastomosis. If there were no suitable receptor vessels at M4-5 segment, we only use EDMS as indirect bypass and these patients would be excluded.

After combined revascularization, these patients would go through the follow-up neuropsychological assessment and [¹⁸F]FDG PET scan at 6 months. The patency of graft vessels was confirmed by DSA. The determination of neurocognitive outcomes were comprehensively analysis by three neuropsychological were blinded to the clinical data of patients. To be specific, the postoperative changes of 6 cognitive domains were evaluated. Only the improved domains were more than those of declined would be assigned into improved group, or they would be defined as non-improved group. Whether a cognitive domain was improved will be compared by the baseline performance of this patients. Only the scores significantly higher than the baseline would be deemed as improvement. The concrete cut-off value for each scale can be seen in Table S1.

The [¹⁸F]FDG PET scan were conducted by using a PET/CT system (Biograph mCT Flow PET/CT, Siemens, Erlangen, Germany). To keep the blood glucose level under 6 mmol/L confirmed by finger-tip glucose test, all subjects fasted for at least 6 h before scan. After receiving an intravenous injection of 0.15 mCi/kg [¹⁸F]FDG, patients rested into a dim room for 50 min, following by a 10-min scan. The [¹⁸F]FDG PET images were analyzed by using SPM12 (https://​www.​fil.​ion.​ucl.​ac.​uk/​spm). The images were firstly transformed into standardized uptake values (SUV) corrected by body weight and then spatially normalized into Montreal Neurological Institute stereotactic space. The images would be smoothed by Gaussian filter with a kernel of [8 8 8]. Given that MMV mainly affects the anterior circulation, the brainstem was utilized as reference area to calculate the SUV ratio (SUVR) map. The SUVR of each region of interest (ROI) was extracted by Anatomical Automatic Labeling 2 (AAL2). The figures were generated by using MRIcron [28], SUIT-3.5 [29] and BrainNet Viewer [30].

The connection between two nodes \({v}_{j}\)​ and \({v}_{m}\) in the brain network is defined as the spatial similarity of the two cubes; the similarity of the two cubes is quantified by calculating the correlation coefficient. Considering the curvature of the cerebral cortex, two similar cubes may have a certain angle. Since the cubes are constructed from 3D discrete PET data, each cube \({v}_{j}\) is rotated by multiples of 45° along the three coordinate axes (the angle is defined as \(\theta\)), and the maximum correlation coefficient is found. Thus, the similarity operator \({r}_{jm}\) between \({v}_{j}\) and \({v}_{m}\)​ is defined as:where \({v}_{ji}\) represents the i-th voxel in node \({v}_{j}\), \(\overline{{v }_{j}}\) represents the mean of the voxels in node, \({v}_{mi}\) represents the i-th voxel in node \({v}_{m}\), \(\overline{{v }_{m}}\) represents the mean of the voxels in node \({v}_{m}\)​, and n is the number of voxels in the node, here n = 27. Cubes with variance close to 0 are excluded (mean < 0.01%) [31, 32].

At this point, we obtain the individual brain functional network based on similarity. However, due to different brain volumes (V) among individuals, the sizes of the brain networks vary, affecting the comparison between brain networks [33]. Therefore, we use the AAL atlas to standardize the above connection matrix to a uniform number of nodes (120 nodes): first, the edges of the obtained network are Bonferroni corrected, with edges passing the correction assigned a value of "1" and edges not passing the correction assigned a value of "0", resulting in a binary raw network (BRN). Then, each cube is assigned to different ROI in the AAL atlas based on its anatomical location and defined as nodes \(\left({V}^{norm}\right)\), thus obtaining a weighted normalized network (WNN). Next, the connection strength between nodes \({V}_{i}^{norm}\) and \({V}_{j}^{norm}\) in the WNN is defined as the ratio of actual connections to all possible connections:where if \({p}_{kl}^{bonf}<0.05\), \({p}_{kl}^{sig}=1\); otherwise, \({p}_{kl}^{sig}=0\); \({p}_{kl}^{bonf}\) is the Bonferroni corrected correlation coefficient between cubes V_k and V_l; \({N}_{i}\) and \({N}_{j}\) are the number of nodes in the BRN belonging to WNN nodes i and j; all self-connections are set to 0.

In the subsequent analysis of the correlation between changes in connection strength in various brain regions and cognitive function changes, we use the rate of change to represent the post-operative changes in connection strength, i.e.:where \({w}_{ij post}\) represents the connection strength between \({V}_{i}^{norm}\) and \({V}_{j}^{norm}\) after cerebral revascularization surgery, and \({w}_{ij pre}\) represents the connection strength between \({V}_{i}^{norm}\) and \({V}_{j}^{norm}\) before the surgery. Similarly, the changing rate in cognitive scales and SUVR after surgery is calculated to reflect the impact of cerebral revascularization surgery on connective strength and cognitive function.

The data analysis was conducted by SPSS 16.0 and Matlab R2022b. The network difference was calculated by GRETNA [34].The data were presented as mean ± standard deviation or frequencies (percentage). The voxel-wise SUVR comparison in two groups or multiple groups would utilize a one-way ANOVA design, and the comparison of surgical changes would use the repeated-measures ANOVA and paired t-test design by SPM12. The multiple regression model in SPM12 was used to calculate the voxel-wise correlation between neuropsychological scores and SUVR. The uncorrected outcomes will be mentioned at main body of text. The ROI-wise differences among groups were calculated by using one-way ANOVA followed by least significant difference (LSD) post hoc analysis. The Person’s correlation analysis between ROI SUVR and neuropsychological score was utilized. Postoperative SUVR and connective strength changes was represented by the ratio of difference value between baseline and follow-up value and baseline value, while the difference value was used to represent the changes of each neuropsychological scale. The statistical significance level was set at 0.05. The multiple comparison of voxel-wise analysis was corrected by using cluster level false discovery rate (FDR) and family-wise error (FWE) < 0.05. While the network comparison was corrected by network-based statistic (NBS) < 0.05.

Finally, ninety-three eligible MMV patients were enrolled in retrospective cohort, who underwent both [¹⁸F]FDG PET scan and neuropsychological assessment, including 48 male and 45 female with the mean age of 42.3 ± 11.2 years. Twenty-one (22.6%) patients did not show any evidence of VCI, 61 (65.6%) patients manifested mild VCI and 11 (11.8%) were diagnosed as VaD. The distribution of cognitive scores of each scale can be seen in Fig. S1a. The clinical characteristics of these patients and 47 age and gender matched HC were summarized in Table 1.

A total of 49 patients were enrolled in prospective cohort. Seven patients were excluded as following reasons: four patients without appropriated receptor vessels at middle cerebral artery M4-5 segment and transferred to indirect revascularization; two patients suffered from severe perioperative complication and could not complete the follow-up assessments; one patient was lost. Finally, we prospectively observed 42 patients who underwent combined revascularization surgery with a mean follow up interval of 183.4 days (range: 162–201 days). The patency of STA graft at follow-up time was confirmed by DSA. There were 20 (47.6%) patients with Matsushima Grade A, 13 (31.0%) Grade B, and 9 (21.4%) Grade C. Twenty-two (52.4%) patients showed significant neurocognitive improvement and the rest 20 (47.6%) patients were not. There was no statistical difference in gender, age, Suzuki Stage, initial clinical manifestation and follow-up interval between two groups. The distribution of cognitive scores of each scale can be seen in Fig. S1b. The clinical characteristics of this cohort were summarized in Table 2.

The voxel-wise analysis revealed metabolic differences in cerebellum between MMV and HC. Hypometabolic regions of MMV mainly located at cerebellum posterior lobe, including bilateral Lobule VI, Crus I, and partial right Lobule IX. While hypermetabolic regions manifested at the anterior part of right Lobule VIII, IX, X, and Vermis IV & V (P < 0.05, FWE corrected, Fig. 1a). To determine whether this kind of cerebellar glucose metabolism is associated with neuropsychological performance, the cerebellar mean SUVR map was generated (Fig. 1b). With the progression of neurocognitive performance, the SUVR of right Lobule Crus I and VI (Fig. 1c and d) also showed decreased tendency, while Vermis X (Fig. 1e) presented significantly higher SUVR in VaD group (P < 0.05, one-way ANOVA with LSD post hoc analysis). Other lobule located at posterior part also manifested this decreased tendency, such as VIIb and Crus II, but there was no statistical significance (Table S2).

Further ROI-based analysis demonstrated the correlation between the SUVR of 26 cerebellum regions and neuropsychological score (Fig. 2g). The SUVR of right Lobule Crus I and VI were significantly correlated with the majority of neuropsychological score, while Vermis X presented negatively correlated with them (P>0.05). Additionally, we found that these cerebellar regions can be mainly divided into two parts by using hierarchical cluster analysis: the SUVR of most cerebellar anterior lobe and vermis have a negatively correlated trend with neurocognitive performance while the posterior lobe manifests positively correlated trend. The voxel-wise analysis also indicated the cerebellar posterior lobe were positively correlated with MMSE, MES, AVLT N1-3 and TMT-B score (P < 0.001, FWE corrected at cluster level, Fig. 2a-d), including Lobule VI, VIIb, VIII, CrusI, and CrusII. While the vermis and anterior lobe showed significant negative association with TMT-A and TMT-B score ((P < 0.001, FWE corrected at cluster level, Fig. 2e-f). The details of the cluster information were shown in Table S3.

We further constructed the MCN for each participant. The mean connective strength matrix for each group with different cognitive diagnosis was generated (Fig. 3a). The 120 cerebral regions based on AAL2 were presents as numbers (Table S4). The connective strength gradually decreased as cognitive impairment growth. Significant difference was found between HC and VCI (including mild VCI and VaD) (P < 0.05, NBS corrected, Fig. 3b and supplementary material 2). However, no statistical difference was found between non-VCI and VCI. Meanwhile, the degree of right Lobule VI, Crus II, and VIII also showed a significant decreasing trend as cognitive impairment progressed (P < 0.05, one-way ANOVA with LSD post hoc analysis, Fig. 3c-e). The degrees of other cerebellar regions were shown in Table S5.

Repeated-measures ANOVA revealed that, following combined cerebral revascularization surgery, posterior cerebellar glucose metabolism showed significant recovery in cognition improved group. The regions with significantly improvement were located in right Lobule Crus I, Crus II, VI, VIIb and VIII (P < 0.001, T > 3.2, uncorrected, Fig. 4a and Table S6). In contrast, the non-improved group exhibited post-operative metabolic improvement primarily in the anterior and left hemisphere of the cerebellum, adjacent to the vermis, including left Lobule IV & V, III L, and VI (P < 0.001, T > 3.2, uncorrected, Fig. 4a and Table S6). Comparison between the improved and non-improved group (main effect) demonstrated significantly higher metabolism in right Lobule Crus I, Crus II, VI, VIIb and VIII in improved group, while the metabolism of Vermis X and Lobule IX was higher in non-improved group (P < 0.001, T > 3.2, uncorrected, Fig. 4a and Table S6). The whole brain postoperative SUVR changes were also compared by using paired T test in improved group. The SUVR of the dominant cerebral hemisphere showed significant improvement, with larger clusters than the right hemisphere, while the significant improvement in the cerebellum was located at the right side. Voxel-wise analysis demonstrated that the SUVR of left thalamus and midbrain red nucleus, the critical components connecting the cerebro-cerebellar circuit, also significantly enhanced (P < 0.001, uncorrected, T > 3.5, Fig. 4b). We also analyzed the correlation between the postoperative cognitive performance and the cerebellar SUVR changes. The hierarchical cluster analysis also revealed that the SUVR of posterior lobe were positively associated with improved cognitive performance, while the anterior lobe presented opposite tendency (Fig. 4c).

In order to study the relationship between cerebro-cerebellar network changes and postoperative neurocognitive performance, we identified 502 significantly increased connectivities after surgery in improved group (P < 0.05, NBS corrected, Fig. 5a and supplementary material 2). Several cerebro-cerebellar connections displayed significant improvement, including the connection with right Lobule VI, left Lobule IV & V, IX, and X. However, none of connection survived in non-improved group. We therefore choose these cerebellar regions as seed points to identify the postoperative cerebro-cerebellar connectivity changes and its relationship with cognitive performance. When right Lobule VI was set as seed point, the improved connective strength in several cerebral regions was significantly positively related to the enhancement of MES performance (FDR q < 0.05, Fig. 5b and Table S7). These cerebral regions include right middle frontal gyrus (MFG), left medial superior frontal gyrus (SFG), right anterior and posterior orbital frontal gyrus (OFG), right superior occipital gyrus (SOG), left inferior and middle occipital gyrus (IOG and MOG), left fusiform gyrus (FFG), right precuneus (PCUN), right pallidum (PAL), right superior temporal gyrus (STG), and right temporal pole (TPO): middle temporal gyrus. Other neurocognitive performance also showed significant correlation with cerebro-cerebellum connective strength change. The increasing connective strength of left lobule IV & V (anterior lobe) and left Lobule VIII and left posterior cingulate cortex (PCC) were negatively associated with AVLT short delayed recall (r = − 0.530, FDR q < 0.05, Fig. 5c and Table S7) and VFT-animal (r = − 0.521, FDR q < 0.05, Fig. 5d and Table S7). While left Lobule IX (the posterior lobe) were positively correlated with the TMT-B performance as its connection with left medial SFG boosted (r = 0.577, FDR q < 0.05, Fig. 5e and Table S7).