Enhancement of Impaired Olfactory Neural Activation and Cognitive Capacity by Liraglutide, but Not Dapagliflozin or Acarbose, in Patients With Type 2 Diabetes: A 16-Week Randomized Parallel Comparative Study

Cognitive decline in diabetes has drawn increasing attention as a complication of diabetes (1,2) that mainly affects the subdomains of memory, attention, processing speed, and executive function (3,4) and seriously threatens patients’ quality of life in their senior years (5). Cognitive decrements can be grouped as mild cognitive decline and dementia, according to the disease progression (6). Current treatments cannot halt the dementia process (7), but up to 40% of dementias might be prevented or delayed if the contributing factors, such as diabetes and obesity, are modified (8).

Rapid progress has been made in the wide application of neuroimaging techniques in recent research (9). Functional MRI (fMRI) can now achieve real-time monitoring of the activation or inhibition of global and regional neurocytes in addition to evaluating cerebral parenchyma and cerebrovascular changes by structural MRI (10). Notably, activation abnormalities of the olfactory neural circuit occur before the onset of structural changes and clinically measurable cognitive decline in patients with diabetes (11). The multidisciplinary combination of an olfactory fMRI scan and comprehensive cognitive assessments helps in early evaluation and diagnosis of cognitive decline in diabetes.

Early interventions and management for modifiable risks are the main strategies for brain protection (8). Evidence is accumulating that several antidiabetes agents have protective effects on neural function and cognition (12,13). Among them, glucagon-like peptide 1 receptor agonists (GLP-1RAs) are the promising drugs for cognitive decline (12,13). Liraglutide has been shown to restore brain glucose metabolism in Alzheimer disease (AD) (14,15), and our previous study demonstrated that liraglutide or exenatide treatment could ameliorate delayed memory in patients with type 2 diabetes (16). A 5-year follow-up study indicated that dulaglutide could reduce the occurrence of accelerated cognitive impairment by 14% in patients with type 2 diabetes (17). The sodium–glucose cotransporter 2 inhibitors (SGLT-2i) have demonstrated neuroprotective effects in animal studies (18,19). However, the neuroprotective benefits of different antidiabetes drugs and their comparative effects have not been fully investigated in randomized controlled trials.

In this study, we conducted a prospective, randomized, parallel trial to evaluate the comparative effects of a 16-week treatment with liraglutide, dapagliflozin, or acarbose on odor-induced brain functional alterations and cognitive changes in patients with type 2 diabetes inadequately controlled with metformin monotherapy. The results are expected to provide new clinical guidance on antidiabetes drug selection for patients with a high risk of cognitive decline.

This randomized, open-label, parallel-group study included patients with type 2 diabetes diagnosed according to the American Diabetes Association criteria (20) and inadequately controlled with metformin monotherapy. The patients were randomized to receive liraglutide, dapagliflozin, or acarbose for 16 weeks between May 2019 and December 2020. All patients were aged 40 to 75 years, had hemoglobin A_1c (HbA_1c) of 7.0–10.0% and a BMI of ≥25 kg/m², and had been on a stable dose of metformin monotherapy (≥1,500 mg daily) for at least 90 days. Key exclusions included left handedness, <6 years of education, and a history or presence of neurological or psychiatric disorders. A full list of exclusion criteria is presented in Supplementary Table 1. Withdrawal criteria included intolerance or allergic reaction to the study drugs or unwillingness to follow the study. Written informed consent was obtained from all participants. The study was approved by the Ethics Committee of Nanjing Drum Tower Hospital Clinical College of Nanjing Medical University and was conducted according to the Declaration of Helsinki. The study was registered on ClinicalTrials.gov (NCT03961659).

Eligible patients were randomized 1:1:1 to receive subcutaneous liraglutide (Victoza), oral dapagliflozin (Farxiga; a SGLT-2i), or oral acarbose (Glucobay; an α-glucosidase inhibitor) in combination with metformin (Glucophage) for 16 weeks (Supplementary Fig. 1). Randomization was performed based on a block algorithm. Liraglutide was gradually titrated from 0.6 mg to 1.8 mg once daily, while dapagliflozin was administered at 10 mg once daily, and acarbose was gradually increased from 50 mg to 100 mg three times daily orally with meals. Visits were scheduled every 4 weeks. A dose reduction of metformin was permitted if hypoglycemia occurred.

At screening, all patients underwent a brain fMRI scan, a battery of cognitive assessments, and anthropometric and biochemical measurements. Body weight, height, waist circumference, and blood pressure were measured by a qualified assessor. BMI was calculated as body weight in kilograms divided by the square of the height in meters. The body fat ratio was estimated using DEXA (Lunar iDXA software; GE Healthcare Technologies, Milwaukee, WI). HbA_1c, fasting plasma glucose, insulin, C-peptide, triglyceride (TG), total cholesterol (TC), HDL cholesterol (HDL-C), and LDL cholesterol (LDL-C), as well as liver and renal function, were analyzed using fasting blood samples in a central laboratory. Blood samples were collected after a standard meal tolerance test to detect 2-h postprandial plasma glucose, insulin, and C-peptide levels. β-Cell function was estimated by updated HOMA model (HOMA2)-β from fasting insulin using the HOMA2 Calculator (HOMA2 v2.2.3; Diabetes Trials Unit, University of Oxford, https://www.dtu.ox.ac.uk/homacalculator/). At week 16, all patients completing the study were reevaluated as at the screening.

All patients underwent brain structural MRI, odor-induced task fMRI, and resting-state fMRI on a 3.0T MR scanner (Achieva TX; Philips Medical Systems, Eindhoven, Netherlands) with an 8-channel head coil. Structural data were obtained by high-resolution T1-weighted three-dimensional fast field echo structural scans (repetition time, 9.7 ms; echo time, 4.6 ms; field of view, 256 mm × 256 mm × 192 mm; flip angle, 8°; voxel size, 1 mm × 1 mm × 1 mm). The fMRI data were collected by gradient-echo planar imaging sequence scans (repetition time, 2,000 ms; echo time, 30 ms; field of view, 192 mm × 192 mm × 140 mm; slice thickness, 4 mm; gap: 0 mm; flip angle, 90°; voxel size, 3 mm × 3 mm × 4 mm), with 222 volumes for task fMRI and 230 volumes for resting-state fMRI.

The odor-induced task consisted of “fresh air,” “rest,” and “scent,” as described previously (11). During the resting-state scan, patients were instructed to relax with their eyes closed, but without falling asleep.

Both task and resting-state fMRI data were preprocessed with Statistical Parametric Mapping 12 (SPM12) and Data Processing & Analysis for Brain Imaging (DPABI_V4.2_190919, https://rfmri.org/dpabi) by the following initial steps (21,22): 1) converting Digital Imaging and Communications in Medicine (DICOM) data into Neuroimaging Informatics Technology Initiative (NIfTI) format; 2) removing the first 6 images of the task scan and the first 10 images of the resting-state scan; 3) slice timing; 4) spatial realignment to remove head movements, >2.5 mm translation, or >2.5° rotation; 5) coregistration with T1-weighted high-resolution anatomical images and spatial normalization to the Montreal Neurological Institute (MNI) space template; 6) spatial smoothing with a Gaussian kernel of 6-mm full-width at half-maximum only for functional connectivity (FC) and amplitude of low-frequency fluctuations (ALFF) analyses; and 7) linear detrending and temporal band-pass filtering (0.01–0.08 Hz) only for FC and regional homogeneity (ReHo) analyses.

Odor-induced brain activation was assessed by a general linear model using Statistical Parametric Mapping 12 (SPM12) software. Following extraction of the three separate conditions “fresh air,” “scent,” and “rest” from the whole sequence, contrasts were made for each participant between “fresh air > rest” and “scent > rest.” Odor-induced fMRI data were analyzed in the mask of the olfactory network, including the regions of bilateral parahippocampus, amygdala, piriform cortex, insula, orbitofrontal cortex, hippocampus, and entorhinal cortices (23). The brain regions showing significant alterations after the 16-week treatment were selected as regions of interest (ROIs), and activations within the ROIs of each participant were estimated as β-values.

Resting-state fMRI data were processed as follows for FC, ALFF, and ReHo analyses: FC was computed using a seed-based correlation analysis method. Seed regions were selected as brain regions showing significant alterations after treatment. FC between the seed regions and the other regions of the whole brain was assessed voxel-by-voxel to generate an FC map. ALFF was calculated in the olfactory network as follows: 1) acquisition of the power spectrum by using fast Fourier transformation to switch each voxel’s time series to a frequency domain; 2) calculation of ALFF values based on the averaged squared root of the frequency of the power spectrum within the range of 0.01–0.1 Hz; and 3) conversion of the ALFF values to z-distribution standardization. ReHo was calculated in the olfactory network as the similarity of the time series in the functional cluster by calculating the similarity between a single voxel and its 27 surrounding voxels using the Kendall coefficient of concordance. We used the z-distribution standardized values of ALFF and ReHo (zALFF and zReHo) for the final analyses.

A battery of validated neuropsychological tests was used for comprehensive evaluation of individual cognition. General cognition was assessed by the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA) (Beijing version). Multiple cognitive subdomains, including immediate memory, visuospatial constructional, language, attention, and delay memory, were evaluated by the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS). Processing speed was evaluated by the Trail Making Test (parts A and B), and executive function was assessed by the completion time of Stroop Color-Word Test (parts I, II, and III). All of these tests were conducted in a fixed sequence by a well-trained examiner who was masked to the treatments and final analyses.

The primary end point was the change in odor-induced brain activation from baseline to week 16. Secondary end points were the changes in the following indexes: 1) cognitive assessments of MMSE, MoCA, immediate memory, visuospatial constructional, language, attention, delay memory, processing speed, and executive function; 2) resting-state fMRI, including FC, ALFF, and ReHo; and 3) metabolic parameters, including body weight, BMI, waist circumference, HbA_1c, fasting and 2-h postprandial plasma glucose, insulin, C-peptide, lipid profile (TC, TG, LDL-C, and HDL-C), and body fat ratio.

The change in odor-induced brain activation from baseline to follow-up was 0.666 ± 0.474 in our previous study (16); therefore, we computed that a sample size of 12 patients per treatment arm would provide 90% power and allow for a 20% potential dropout rate to detect the difference in odor-induced activation by the end of this study. The fMRI data at baseline were compared among the three groups using one-way ANOVA. The paired t test was used to compare fMRI data within groups between baseline and week 16. The significance threshold correction was conducted according to Gaussian random field (GRF) theory with a voxel level of P < 0.001 and a cluster level of P < 0.05. These analyses were performed using DPABI_V4.2_190919 software. Continuous variables at baseline were compared among the three groups using ANOVA, while categorical variables were compared using the χ² test. Continuous variables were compared between baseline and week 16 within groups using paired t tests, while after-treatment comparisons were made using ANCOVA with adjusted baseline values, and post hoc tests were done with the Bonferroni correction. Analyses were performed using SPSS 20.0 software (IBM, Armonk, NY).

To assess the contributing factors attributed to the olfactory brain functional alterations and cognitive improvements, we used the structural equation model and took the following indexes into account: the antidiabetes agents grouping, and the metabolic parameters either changed significantly after treatment in this study or showed significant association with cognition in our previous studies (11,16). The hypothesis model is shown in Supplementary Fig. 2. Furthermore, the olfactory brain functional alterations may perform a mediation role between the associations of antidiabetes agents grouping and metabolic parameters with cognitive improvements. The contributing factors were firstly screened by Pearson correlation test and finally determined based on a structural equation model performed using IBM SPSS Amos software (IBM) and including mediation analysis. A maximum likelihood estimation robust extraction method was used for parameter estimation. Bootstrapping with 2,000 resamplings and 95% CIs was used to analyze mediating effects. A P value of <0.05 was considered statistically significant.

Of the 55 patients screened, 36 were enrolled and randomly assigned to the three treatment groups (liraglutide, n = 12; dapagliflozin, n = 12; acarbose, n = 12). All patients completed the study and were included in the final analysis (Supplementary Fig. 1). At baseline, all general characteristics, including age, sex, education, alcohol habits, smoking habits, duration of diabetes, HbA_1c, and BMI, were matched in the three groups. After the 16-week treatment, HbA_1c fell to a significant and similar degree in each group (liraglutide, 8.5% ± 1.0% to 7.1% ± 0.7%, P < 0.001; dapagliflozin, 8.2% ± 1.0% to 7.0% ± 1.2%, P = 0.007; and acarbose, 7.9% ± 0.9% to 7.1% ± 0.7%, P = 0.011; P = 0.839 among groups). All three groups also showed a significant and similar reduction in BMI (liraglutide, 27.3 ± 1.6 kg/m² to 26.4 ± 1.6 kg/m², P = 0.012; dapagliflozin, 26.7 ± 1.2 kg/m² to 25.7 ± 1.3 kg/m², P < 0.001; and acarbose, 27.2 ± 1.9 kg/m² to 26.7 ± 1.8 kg/m², P = 0.049; P = 0.384 among groups), body weight (liraglutide, 73.7 ± 8.6 kg to 71.3 ± 9.1 kg, P = 0.010; dapagliflozin, 74.6 ± 8.1 kg to 71.8 ± 7.8 kg, P < 0.001; and acarbose, 73.4 ± 6.8 kg to 71.7 ± 5.9 kg, P = 0.049; P = 0.542 among groups), and body fat ratio (liraglutide, 33.7% ± 4.8% to 31.6% ± 4.8%, P < 0.001; dapagliflozin, 31.6% ± 4.5% to 29.6% ± 4.4%, P = 0.008; and acarbose, 33.8% ± 4.0% to 32.2% ± 4.0%, P = 0.005; P = 0.710 among groups) at week 16 compared with baseline. The waist circumference was significantly reduced in the liraglutide and dapagliflozin groups after treatment, but not in the acarbose group; however, the reduction did not differ significantly among the three groups (liraglutide, 94.8 ± 6.8 cm to 92.6 ± 6.3 cm, P = 0.020; dapagliflozin, 97.3 ± 5.4 cm to 94.4 ± 4.2 cm, P = 0.013; and acarbose, 93.0 ± 7.6 cm to 91.8 ± 8.3 cm, P = 0.375; P = 0.780 among groups) (Table 1).

The brain olfactory-related regions in each group were activated in response to odor stimulation at baseline and at week 16. The liraglutide group showed significant activation of the left hippocampus after the 16-week treatment compared with baseline (with GRF correction, voxel level: P < 0.001, cluster level: P < 0.05, cluster size threshold: 18 voxels). However, neither the dapagliflozin nor the acarbose group showed any significant alteration in odor-induced activation between baseline and week 16 (Fig. 1A and B).

Comparison with the baseline data revealed no significant differences in FC between the seed region and the whole brain, ALFF, or ReHo after the 16-week intervention in any of the three groups (data not shown).

Delayed memory was significantly improved at week 16 compared with baseline in the liraglutide group (93.8 ± 11.4 to 111.6 ± 13.1, P < 0.001), but not in the dapagliflozin (91.1 ± 15.9 to 96.5 ± 12.8, P = 0.157) or acarbose (95.2 ± 14.4 to 97.1 ± 12.4, P = 0.567) groups, and was improved overall in the liraglutide group compared with dapagliflozin and acarbose groups. Attention (108.1 ± 9.6 to 112.9 ± 9.9, P = 0.001) and executive function (70.0 ± 12.7 to 62.4 ± 7.8, P = 0.012) were also markedly improved by liraglutide treatment compared with baseline, but not by dapagliflozin (103.4 ± 12.1 to 107.2 ± 12.9, P = 0.071 for attention; 85.2 ± 35.6 to 79.6 ± 21.3, P = 0.332 for executive function) or acarbose (105.5 ± 16.0 to 107.2 ± 13.2, P = 0.396 for attention; 74.9 ± 23.4 to 73.1 ± 22.6, P = 0.536 for executive function). The MMSE, MoCA, immediate memory, visuospatial constructional, language, and processing speed were not markedly changed by any of the three treatments between baseline and week 16 (Fig. 2).

The potential factors contributing to enhanced brain activation and cognitive performance were screened according to the significant correlationship showed in the Pearson correlation test (Supplementary Table 2) and determined in a structural equation model. Dapagliflozin treatment was not included in model construction, and acarbose treatment was excluded from the final model as no significant association was observed in model construction. The final model is presented in Fig. 3, with adequate indicators of the model fit as follows: χ²/degrees of freedom = 1.080 < 3, Tucker Lewis index = 0.939 > 0.9, comparative fit index = 0.956 > 0.9, and root mean square error of approximation = 0.048 < 0.08.

In the final model, liraglutide treatment improved delayed memory through a partial mediation role of left hippocampal activation (β = 0.011; 95% bootstrap CI 0.023, 0.255), with a mediating effect size of 21.31% (Fig. 3 and Supplementary Tables 3 and 4). Besides, the β-coefficients of the model demonstrated that a reduced waist circumference contributed to the enhanced left hippocampal activation (β = −0.401, P = 0.006), a decreased body fat ratio contributed to improved attention (β = −0.440, P = 0.003), and increased fasting insulin significantly contributed to reduced executive function test time (β = −0.386, P = 0.013) (Fig. 3 and Supplementary Table 3).

This is the first randomized, prospective, head-to-head study to compare the therapeutic effects of three antidiabetes agents on cognitive function in patients with type 2 diabetes. After the 16-week treatment, liraglutide markedly enhanced odor-induced left hippocampal activation and improved cognitive performance of delayed memory, attention, and executive function, while dapagliflozin and acarbose had no similar effects. The structural equation model indicated that the improvements in brain health and cognitive function could be partly ascribed to a direct effect of liraglutide treatment as well as to metabolic amelioration.

Functional neuroimaging techniques, such as fMRI, are now the most important noninvasive methods for evaluating brain neural activation abnormalities that occur even before pathological changes (24). Disruption of olfactory neural circuits has typically been found in patients with mild cognitive decline and AD and has been confirmed as an early and sensitive indicator of cognitive decline in patients with type 2 diabetes (11,25,26). Our previous single-arm study found that GLP-1RAs treatment for 3 months markedly increased olfactory brain activation in obese patients with type 2 diabetes (16). However, the possibility that GLP-1RAs show ubiquitous neuroprotective effects has not been investigated nor has their effectiveness been compared with other antidiabetes drugs. The current study demonstrated that liraglutide treatment for 16 weeks markedly enhanced activation of the left hippocampus, a crucial center for memory formation and learning in the dominant hemisphere (27,28), whereas dapagliflozin and acarbose did not. These results suggest that liraglutide has a special effect not shared by dapagliflozin or acarbose for restoring the early impaired brain regions involving cognitive function in diabetes.

Subtle impairment of one or more cognitive subdomains often occurs before they can be detected by general cognitive assessments such as MMSE and MoCA (29,30). Delayed memory, attention, and executive function are recognized as early damaged domains in diabetes (3,4). One study showed that liraglutide improved memory and attention in patients with type 2 diabetes compared with the control group, without unifying their combined antidiabetes medication (31). Another study showed that liraglutide improved memory in patients with obese prediabetes and type 2 diabetes compared with a lifestyle intervention group, with comparable weight loss (32). A recent longitudinal study with a median follow-up of 5.4 years reported that dulaglutide reduced cognitive decline, as assessed by two general cognitive tests, in participants with type 2 diabetes (17). No prospective studies have been conducted on dapagliflozin or acarbose effects on cognition in diabetes. Only a few retrospective studies have explored the effect of SGLT-2i on cognition in diabetes, but the results were inconsistent (33,34). Similarly, a few studies have reported that acarbose is ineffective in reducing the risk of dementia in type 2 diabetes (35,36). Our study provided comprehensive assessments of general cognition as well as multiple cognitive subdomains and verified that patients in the liraglutide group had significantly improved cognitive scores for delayed memory, attention, and executive subdomains after the 16-week treatment, while those in the dapagliflozin and acarbose groups showed no improvement. Our head-to-head study also compared the therapeutic effects of different antidiabetes agents on cognitive function and confirmed a beneficial effect on the early affected cognitive domains of type 2 diabetes following liraglutide, but not dapagliflozin or acarbose, treatment.

Metabolic disorders are considered major modifiable risk factors for cognitive decline (8). Notably, all three antidiabetes treatments showed comparable metabolic benefits, but significantly improved brain activation and cognitive performance were only observed with liraglutide. This finding suggests a unique neuroprotective role of liraglutide on brain health and cognitive function. Our structural equation model showed that the relationship between liraglutide treatment and delayed memory was partially mediated by left hippocampal activation, with a mediating effect of 21.31%, suggesting that the cognitive benefits of liraglutide may involve restoration of hippocampal neuron function. Indeed, GLP-1 receptors are widely distributed in the brain regions related to memory and learning (37), and GLP-1RAs can cross the blood-brain barrier and improve neural function (38). Animal studies of AD and diabetes confirmed that liraglutide could alleviate hippocampal neuroinflammation and reduce synapse loss, thereby delaying or partially halting the progressive decline in memory and learning (39,40).

Meanwhile, our model demonstrated that metabolic ameliorations, including elevated fasting insulin, reduced waist circumference, and decreased body fat ratio, contributed to the improvement of brain activation and cognition. Our previous study confirmed the beneficial effect of elevated insulin level on cognition (16). Emerging evidence demonstrates insulin has a multifactorial role in the brain, whereas insulin signaling dysregulation contributes to neural injury and neurodegeneration (41). By binding to its receptor, insulin modulates synaptic plasticity and vascular function through direct effects on neurotransmission, energy metabolism, and inflammation (41). Insulin signaling dysregulation also affects AD pathology through the clearance of the amyloid β peptide and phosphorylation of tau (42). In addition, a meta-analysis study indicated that high waist circumference, a convenient and effective measure of central adiposity, was associated with a greater risk of cognitive impairment and dementia (43). The significant correlation of the reduced waist circumference with the enhanced brain activation in our study supported such neuroprotective effect of reduced waist circumference. However, with the comparable reduction of waist circumference between liraglutide and dapagliflozin groups, such neuroprotective effect was only observed in the liraglutide group. One possible explanation might be that waist circumference is a rough proxy for visceral obesity compared with imaging techniques, while liraglutide has been shown to be more effective in reducing visceral adipose than lifestyle intervention with comparable weight loss (44).

Indeed, adipokines disorders associated with excess adiposity, especially central adiposity, often accompanied by oxidative stress, mitochondrial dysfunction, and increased inflammation, have been linked with cognitive impairment (45,46). Leptin facilitates presynaptic and postsynaptic transmitter release and sensitivity in hippocampal neurons and its dysregulation causes impairment of spatial learning and memory formation (46). Proinflammatory factor interleukin 6 is shown to increase amyloid β deposition and promote cognitive impairments (46). Additionally, although no significant difference of HbA_1c reduction was observed among the three groups, the liraglutide group showed a tendency of greater reduction of HbA_1c. However, in the structural equation model, no significant association was observed of HbA_1c reduction with the improved olfactory activation and cognition, which is consisted with the results from our previous study (11). Also, a systematic review showed the negative but weak relationship between HbA_1c and cognition in type 2 diabetes (47). Actually, a growing number of studies have shown that dysregulation of glucose variability, glucose peaks, and hypoglycemia are reported to be associated with a higher risk of dementia in diabetes (48–50). Assessed together, the results of the model indicated a beneficial role of liraglutide in brain health and cognitive function arising both through its neuroprotective effects and from metabolic amelioration, and we conducted a preliminary exploration of the underlying mechanism.

This study has two major strengths. One is that it is the first head-to-head study to evaluate the comparative effects of three antidiabetes agents on brain function and cognition and to explore the potential contributing factors. The second is that we used a combination of olfactory task fMRI, resting-state fMRI, and a battery of cognitive tests to provide a comprehensive evaluation of the early and subtle alterations in cognitive function.

The limitations of this study include its small sample size and the relatively short intervention duration. Thus, we cannot rule out interference due to a practice effect arising from repeated cognitive assessments conducted at relatively short intervals. Further longitudinal studies in larger sample populations are needed. Moreover, the olfactory bulbs were not scanned in our study, and the activation of olfactory bulbs is difficult to detect in the human brain. The relationships between glycemic control with olfactory behaviors and olfactory bulb volumes remains unclear (51). Further long-term investigations of effects of glycemic control on olfaction in diabetes are warranted.

In conclusion, our prospective head-to-head comparative study suggested that liraglutide, but not dapagliflozin or acarbose, could improve brain health and restore early impaired cognitive domains in type 2 diabetes. Our findings validated the unique beneficial effect of liraglutide on cognition and provided new clinical evidence for optimal treatment of individuals with diabetes and a high risk of cognitive decline.

Acknowledgments. The authors would like to thank all volunteers for their participation in this study and thank medical personnel from Department of Endocrinology and Department of Radiology, Nanjing Drum Tower Hospital Clinical College of Nanjing Medical University, for their valuable assistance.

Funding. This work was supported by the National Natural Science Foundation of China Grant Awards (82030026, 81770819, 82000775, 81970689, 82070837, 81970704, 81800752, 81900787, 82000735, 82100868, and 81800719), the Key Project of Nanjing Clinical Medical Science, the Natural Science Foundation of Jiangsu Province of China (BK20200114, BK20201115, BK20200118, and BK20200136), the Six Talent Peaks Project of Jiangsu Province of China (YY-086), the China Postdoctoral Science Foundation (2020M671449), the Doctor of Entrepreneurship and Innovation Program of Jiangsu Province, and the Nanjing Health Science and Technology Development Project (JQX19004 and YKK18067).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. H.C. contributed to data collection, statistical analyses, and wrote the manuscript. Z.Z. designed the protocol, contributed to statistical analyses, and reviewed the manuscript. B.Z. designed the protocol and reviewed the manuscript. W.Z. contributed to MRI data collection and analysis and wrote the manuscript. J.W. contributed to writing the manuscript. W.N. and Y.M. contributed to data collection. J.L. contributed to MRI data collection. Y.B. designed the study and oversaw all clinical aspects of study conduct and manuscript preparation. Y.B. is the guarantor of this work, and as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.