Hemispheric imbalance in mild cognitive impairment: a graph-theoretical analysis of multimodal brain networks

Abstract

Background

Hemispheric lateralization of brain structure and function supports cognitive processing. However, its alterations across different cognitive impairment subtypes and severities remain unclear.

Methods

In this case-control study, 104 participants were enrolled, including 47 with amnestic mild cognitive impairment (aMCI), 27 with non-amnestic MCI (naMCI), and 30 healthy controls. Functional and white matter structural brain networks were constructed separately for the left and right hemispheres. Graph-theoretical metrics and structure–function coupling were computed to assess intergroup and interhemispheric differences. Correlation analyses examined associations between network alterations and cognitive performance.

Results

Functionally, both MCI groups showed a significant leftward reduction in small-world properties (p < 0.05), with aMCI patients exhibiting reduced clustering coefficient and local efficiency in the left hemisphere. Structurally, lateralization of left-hemispheric local efficiency was disrupted in patients. Altered structure-function coupling revealed loss of right-hemispheric dominance in patient groups. Abnormal interhemispheric parameters correlated with memory deficits in aMCI. Node-level analysis revealed aMCI abnormalities concentrated in limbic and multimodal networks, while naMCI displayed more diffuse alterations involving the executive control network.

Conclusions

aMCI and naMCI exhibit distinct patterns of lateralization abnormalities, with aMCI patients being characterized by the disintegration of specific local functional networks in the left hemisphere. Therefore, ameliorating abnormal patterns of hemispheric lateralization could serve as a purpose of the intervention.

Introduction

Mild cognitive impairment (MCI) represents a transitional state between normal aging and dementia, defined by objective cognitive decline without significant impairment in daily function [1]. It comprises amnestic MCI (aMCI), featuring episodic memory impairment, and non-amnestic MCI (naMCI), involving deficits in executive function, language, or attention [2]. With increasing life expectancy, MCI prevalence is rising globally [3]. Annually, 8–15% of individuals with MCI progress to dementia, with over 50% converting within 5 years [4,5]. Therefore, early detection and intervention are critical to delaying dementia onset.

Hemispheric asymmetry refers to the multidimensional differences between the left and right cerebral hemispheres in both structure and function, ranging from local brain regions to whole-brain connectivity architecture [6]. It represents one of the fundamental organizational principles of the human brain, serving as a critical foundation for maintaining higher-order cortical functions and potentially offering an important entry point for elucidating the central mechanisms underlying MCI [7–10]. Under physiological conditions, cerebral hemispheric asymmetry enhances cognitive efficiency and robustness through synergistic contributions at both structural and functional levels. Structurally, lateralized features—such as the larger cortical surface area of the left planum temporale [11], the leftward structural advantage of the arcuate fasciculus [12], and the precise topological organization of connecting pathways within the splenium of the corpus callosum [13]—serve to shorten communication pathways within key ipsilateral circuits and enable highly efficient interhemispheric integration when required. This establishes a solid anatomical foundation for high-capacity information exchange. At the functional level, hemispheric asymmetry optimizes cognition through the synergy of specialization and integration [14]. Left hemisphere advantages in language and sequential analysis, together with right hemisphere strengths in spatial attention, establish a lateralized division of labor that reduces redundant neural processing and supports parallel information processing [9,15]. At the same time, dynamic interactions between the hemispheres via the corpus callosum and other pathways enable efficient integration and support flexible cognitive behavior [11].

However, under pathological conditions, hemispheric asymmetry may become abnormally altered due to factors such as structural damage, neural functional decline, and abnormal protein deposition [16]. Therefore, lateralized characteristics may provide valuable insights into the pathogenesis and progression of neurodegenerative diseases. Specifically, studies have shown that patients with cognitive impairment exhibit asymmetric patterns of gray matter loss: MCI patients tend to show more pronounced right hippocampal atrophy, whereas Alzheimer’s disease (AD) patients exhibit more severe left hippocampal atrophy [17]. Moreover, a longitudinal study revealed that AD patients experience a faster decline in functional lateralization compared to normal aging, particularly in the left frontal and anterior temporal cortices, as well as in the right insula and lateral temporoparietal regions [18]. Additionally, the asymmetric distributions of tau protein deposition and extracellular amyloid-beta (Aβ) pathology have been demonstrated to be closely associated with cognitive decline [16,19]. Taken together, the normal maintenance of cognitive function depends on the balance and effective coordination between the two hemispheres. Once this lateralized division of labor is disrupted, the dynamic coordination of brain networks between segregation and integration may be impaired, ultimately leading to domain-specific or global cognitive deterioration [14,16]. However, most previous studies have been limited to single brain regions or specific modalities, focusing only on either structural or functional aspects of asymmetry. Such piecemeal research outcomes impede a comprehensive interpretation of the neural mechanisms underlying MCI.

Functional networks, derived from functional MRI (fMRI), capture information processing capacity, while structural networks, based on diffusion tensor imaging (DTI), reflect physical inter-regional connections [20,21]. Their integration enables a multi-level understanding of brain coordination and information transfer [22]. Graph theory provides a powerful means to quantify network efficiency, small-world properties, and topological complexity [23]. Additionally, structural-functional coupling, which assesses the extent to which white matter pathways support functional interactions, has been shown to be more sensitive than single-modality metrics in detecting subtle alterations in brain activity [24,25]. Investigating these properties in the context of hemispheric lateralization could fill gaps in previous studies and provide a more comprehensive understanding of how MCI disrupts brain organization across multiple levels.

In the present study, we integrated multimodal brain networks to establish a comprehensive analytical framework capable of jointly evaluating functional, structural, and structure–function coupling characteristics. This framework was designed to systematically reveal patterns of hemispheric lateralization alterations across different subtypes of MCI. This study has two objectives: first, to determine whether patients with aMCI and naMCI exhibit abnormal hemispheric lateralization at the brain network level; and second, to clarify whether these abnormal network parameters are associated with clinical assessments in MCI patients. We hypothesize that, compared to healthy controls, both aMCI and naMCI patients will exhibit varying degrees of lateralization abnormalities at the levels of functional networks, structural brain networks, and structural-functional coupling, and that these changes will be correlated with cognitive assessments. This study is expected to shed light on the central neural mechanisms underlying MCI and lay a theoretical foundation for the future development of neuromodulation interventions.

Method

Study design and participants

This case–control study was conducted at Yueyang Hospital of Integrated Traditional Chinese and Western Medicine, affiliated with Shanghai University of Traditional Chinese Medicine, from 1 November 2021 to February 2022. Based on prior literature [26], we performed sample size estimation using G*Power (http://www.gpower.hhu.de/). For a one-way ANOVA comparing three groups with an assumed effect size of f = 0.4, α = 0.05, and power = 0.80, the required minimum sample size was 22 participants per group. The study protocol was approved by the Institutional Ethics Committee (approval No. 2021-103), and all procedures complied with the Declaration of Helsinki. All participants were fully informed about the study and provided written informed consent before enrollment. The study flowchart is shown in Figure 1.

Figure 1.

Study design flowchart. Participants recruited were initially evaluated based on inclusion/exclusion criteria, with those meeting the standards subsequently allocated to the aMCI, naMCI, and healthy controls groups. All participants underwent neuropsychological assessments and multimodal MRI scans. Structural and functional networks of bilateral cerebral hemispheres were constructed for each group, followed by between-group comparisons of graph-theoretical metrics and structure-function coupling coefficients, as well as intra-group interhemispheric comparisons. Additionally, correlation analyses were performed between neuroimaging indices demonstrating significant differences and clinical scale scores.

Abbreviation: aMCI: amnestic mild cognitive impairment; DTI: diffusion tensor imaging; fMRI: functional magnetic resonance imaging; MRI: magnetic resonance imaging; naMCI: non-amnestic mild cognitive impairment.

According to the Petersen/Winblad expert consensus, the diagnostic criteria for MCI are as follows [27]: Patients exhibit subjective cognitive decline and objective cognitive impairment (with at least one cognitive measure impaired by more than 1.5 standard deviations), but do not meet the criteria for dementia [28], while maintaining basic activities of daily living. Additionally, based on the classification criteria for MCI [29,30], patients are further divided into aMCI and naMCI for discussion. MCI patients were categorized as aMCI whenever there was a decrease in long-term delayed recall scores on the Auditory Verbal Learning Test (AVLT). Conversely, if long-term delayed recall in the AVLT is within the normal range and there is impairment in any of the other cognitive domains it is categorized as naMCI.

The inclusion criteria for aMCI in this study were as follows: (1) age 50–80 years; (2) right-handedness; (3) ≥6 years of education; (4) meeting the diagnostic criteria for MCI; and (5) impaired long-term delayed recall on the AVLT exceeding 1.5 standard deviations. The inclusion criteria for naMCI were: (1) age 50–80 years; (2) right-handedness; (3) ≥6 years of education; (4) meeting the diagnostic criteria for MCI; (5) AVLT scores within the normal range; and (6) impairment of more than 1.5 standard deviations on any of the cognitive indices of language (Boston Naming Test, BNT), attention (Symbol Digit Modalities Test, SDMT), and executive control (Shape Trajectory Test, Part B, STT-B) tests. In contrast, the healthy control group included right-handed participants aged 50–80 years with ≥6 years of education who never complained of cognitive decline and whose neuropsychological assessment showed no cognitive domain impairment. Exclusion criteria included (1) use of cognitive-enhancing medications; (2) comorbid tumors, severe cardiovascular, hepatic, renal, hematologic, or infectious diseases; (3) neurological disorders (e.g. dementia of various causes, cerebrovascular disease, Parkinson’s syndrome, epilepsy) or psychiatric disorders (e.g. anxiety, depression, schizophrenia, etc.); (4) inability to complete neuropsychological assessments, such as severe visual or auditory impairments; (5) drug or alcohol abuse and contraindications to MRI.

Clinical and neuropsychological assessments

Demographic data included sex, age, years of education, height, weight, smoking and drinking habits, as well as medical history. Neuropsychological data included daily functioning assessment (Functional Activities Questionnaire, FAQ) [31], mood status evaluations (Hamilton Depression Scale, HAMD [32]; Hamilton Anxiety Scale, HAMA [33]), and cognitive function assessments. The cognitive function evaluation included the Mini-Mental State Examination (MMSE) [34] and assessments of various cognitive domains using the AVLT [35], BNT [36], SDMT [37], STT-B [38]. Two senior neuropsychologists with years of experience conducted the neuropsychological assessments without knowledge of the clinical diagnosis, and the results were later reviewed by another senior neuropsychologist. Detailed information about these assessment scales is provided in Supporting Information.

Image acquisition and preprocessing

All participants entered a quiet and clean preparation room 30 min prior to the MRI scan. Vital signs such as heart rate, respiration, and blood pressure were measured and confirmed to be normal. After ensuring the participant met the scanning conditions, all metal items were removed, and earplugs were provided. During the scan, participants were positioned in a supine posture with their heads fixed using a coil. Participants were directed to keep their eyes shut, stay alert, maintain steady breathing, and limit any movement of the head. If any discomfort occurred, participants were instructed to raise their hands.

All neuroimaging data were acquired on a 3.0 Tesla MAGNETOM Verio system (Siemens Healthcare, Erlangen, Germany). The scanning protocol encompassed three modalities: fMRI, structural MRI (sMRI), and DTI. The detailed parameters for each sequence are as follows: Resting-state fMRI: Images were acquired using a gradient-echo echo-planar imaging (EPI) sequence. Key parameters were: repetition time (TR) = 3000 ms, echo time (TE) = 30 ms, flip angle (FA) = 90°, 43 slices, slice thickness = 3.0 mm, matrix size = 64 × 64, field of view (FOV) = 230 × 230 mm², voxel size = 3.6 × 3.6 × 3.0 mm³, and 200 volumes. Structural T1-weighted MRI: High-resolution anatomical images were obtained using a magnetization-prepared rapid gradient-echo (MPRAGE) sequence, configured with: TR = 1900 ms, TE = 2.93 ms, inversion time (TI) = 900 ms, FA = 9°, FOV = 256 × 256 mm², and slice thickness = 1 mm. Diffusion Tensor Imaging (DTI): Diffusion-weighted data were collected via a single-shot spin-echo EPI sequence in the axial plane with the following parameters: TR = 10,000 ms, TE = 89 ms, FA = 90°, slice thickness = 2.0 mm, in-plane resolution = 1.875 mm, 60 non-collinear diffusion directions (b = 1000s/mm²), and two b0 images. Details of the image preprocessing pipeline were provided in the Supplementary Materials.

Brain network construction and graph theory analysis

In this study, we used the Automated Anatomical Labeling (AAL) atlas to construct the brain network. This atlas divides the entire brain into 90 symmetrical regions (45 in the left hemisphere and 45 in the right hemisphere) [39]. Based on this, the 45 regions in the left hemisphere and 45 regions in the right hemisphere were segmented as independent templates for subsequent analysis. The structural and functional networks were constructed using the Graph-theoretical Network Analysis (GRETNA) toolbox (https://www.nitrc.org/projects/gretna/) [40] in Matlab R2013b (The Mathworks Inc., Natick, MA, USA), and the topological properties of the networks were calculated.

Functional brain network construction

For each participant, time series data for the 45 regions in the left and right hemispheres were extracted from the preprocessed fMRI data. Pearson correlation coefficients were calculated to assess functional connectivity between brain regions [41], resulting in two 45 × 45 correlation matrices for each participant, one for the left hemisphere and one for the right hemisphere. Network construction used a binary matrix for positive connections, and Fisher-Z transformation was applied to improve the normal distribution properties of the network. Based on the Gretna toolbox, the sparsity threshold for the left and right hemisphere brain networks was set between 0.09 and 0.48, with a step size of 0.01, ensuring that the networks maintained an appropriate density and avoided overly complex or sparse structures. Additionally, to minimize potential bias from a single threshold, statistical analysis was performed using the area under the curve (AUC) of global and nodal topological metrics across the sparsity range.

Structural brain network construction

The individual FA skeletons obtained from preprocessing were used to construct white matter structural brain networks for the left and right hemispheres. Whole-brain white matter tractography was performed using deterministic fiber tracking with the Fiber Assignment by Continuous Tracking (FACT) algorithm [42]. Tracking was terminated when FA <0.2 or the turning angle exceeded 45° [43]. If a fiber tract passed through or terminated in two brain regions, those regions were considered structurally connected. To ensure the reliability and stability of the connections, only those connections that were preserved in more than 80% of participants were retained as binary thresholds, forming binary matrices that represented the structural connectivity of white matter networks in the left and right hemispheres [44].

Graph theory analysis

Based on the previously constructed structural and functional network matrices of the left and right hemispheres, topological properties of the brain networks were calculated using the GRETNA toolbox. The global metrics included: clustering coefficient (Cp), characteristic path length (Lp), normalized clustering coefficient (γ), normalized characteristic path length (λ), small-worldness (σ), global efficiency (Eglob), and local efficiency (Eloc). In addition, two nodal metrics were also considered: nodal local efficiency [Eloc(i)] and nodal clustering coefficient (NCp). Detailed information on all network properties is provided in Supporting Information.

Structural-functional coupling analysis

To assess the global brain network structural-functional coupling (SFC), we first removed redundant data and retained only the upper triangular portion of each participant’s structural and functional brain networks. Then, the two brain networks were unfolded into vectors and converted into z-scores. The coupling between structure and function was quantified using Spearman’s correlation, which was calculated only for the non-zero components of the structural and functional connectivity features [45]. Finally, group-level analysis was performed using the global structural-functional coupling coefficients from all participants.

Statistical analysis

SPSS 26.0 software (IBM, Armonk, NY, USA) was used to analyze the data from the aMCI, naMCI, and healthy control (HC) groups, with a significance level set at 0.05 (two-sided test). First, demographic and clinical characteristics of the participants in each group were described, and differences between the groups were assessed. Continuous data were expressed as means (SD) and group differences were compared using one-way ANOVA or the Kruskal-Wallis test. Categorical variables were expressed as n (percentage, %), and differences were compared using the chi-square test. Second, for MRI network construction, Fisher-Z transformation was applied to enhance the normality of the data. Age, gender, and education years were considered as confounding variables. Covariance analysis (ANCOVA) was used to compare group differences in global brain network properties and structural-functional coupling coefficients, with post hoc testing performed using the least significant difference (LSD) method. Additionally, paired sample t-tests were used to compare hemispheric lateralization for different network parameters. Furthermore, Pearson partial correlation coefficients were calculated to explore the relationship between abnormal network parameters and clinical scores, with residual scatter plots plotted for visual representation. We also employed the Mahalanobis distance method to detect multivariate outliers in all variables involved in the correlation analysis, setting a statistical threshold (p < 0.001). The absence of any outliers indicates that the reported correlation results are robust. Lastly, to identify specific brain regions showing hemispheric lateralization, paired t-tests were conducted on nodal metrics for the left and right hemisphere brain networks using the GRETNA toolbox, with false discovery rate (FDR) correction applied to control for multiple comparisons.

Results

Demographic and clinical data

A total of 121 participants were recruited for this study. After excluding 4 participants with incomplete information and 13 participants with issues in their MRI images, 104 participants were ultimately included. This included 47 aMCI patients, 27 naMCI patients, and 30 healthy participants.

There were no significant differences in demographic data among the three groups, except for age and education years. The naMCI group was the oldest and had the shortest duration of education, while the HC group was the opposite. In light of this, the group comparisons in this study controlled for the confounding variables of age and education years. As shown in Table 1, significant statistical differences were observed among the three groups in memory function assessments (all trial scores on the AVLT), as well as in multiple cognitive domains such as language, attention, and executive function (p < 0.05). However, there were no significant differences among the three groups in terms of TIV, WM volume, GM volume, and volumes of the left and right hippocampus (p > 0.05).

Table 1.

Characteristics of the study population.

Characteristics	Total (N = 104)	Group			F/χ2	p-Value*	Effect size (η²)
		aMCI (n = 47)	naMCI (n = 27)	HC (n = 47)
Basic characteristics
Gender (% female)	73 (70.19)	31 (65.96)	22 (81.48)	20 (66.67)	2.226	0.329
Age, years	65.35 (6.86)	65.55 (6.23)	67.89 (6.73)	62.73 (7.20)	4.312	0.016	0.079
Education, years	10.45 (2.56)	10.06 (2.44)	9.85 (2.38)	11.60 (2.61)	4.594	0.012	0.083
Height, cm	163.21 (7.31)	163.19 (7.39)	161.78 (7.02)	164.53 (7.44)	1.010	0.368
Weight, kg	62.61 (9.55)	63.25 (9.89)	61.07 (8.73)	62.98 (9.86)	0.474	0.624
Smoking (%)	17 (16.35)	10 (21.28)	2 (7.41)	5 (16.67)	2.689	0.261
Drinking (%)	9 (8.65)	3 (6.38)	3 (11.11)	3 (10.00)	0.591	0.744
Hypertension(%)	59 (56.73)	28 (59.57)	15 (55.56)	16 (53.33)	0.311	0.856
Hypercholesterolemia (%)	21 (20.19)	13 (27.66)	4 (14.81)	4 (13.33)	2.987	0.225
Diabetes (%)	19 (18.27)	10 (21.28)	4 (14.81)	5 (16.67)	0.554	0.758
Daily life performance
FAQ	0.70 (2.70)	0.66 (2.16)	0.78 (3.66)	0.70 (2.54)	0.016	0.984
Emotional performance
HAMA	3.61 (3.18)	3.96 (3.89)	2.89 (2.99)	3.70 (3.01)	0.985	0.377
HAMD	1.89 (3.71)	2.17 (2.65)	1.19 (1.96)	2.1 (5.80)	0.666	0.516
Cognitive performance
MMSE	26.69 (2.10)	25.98 (2.30)	26.81 (1.52)	27.70 (1.80)	6.965	0.001	0.121
Memory function
AVLT_N1	3.29 (1.52)	2.55 (1.23)	3.26 (1.20)	4.47 (1.48)	19.894	<0.001	0.283
AVLT_N2	5.69 (1.83)	4.28 (1.30)	5.70 (1.56)	6.40 (1.98)	17.829	<0.001	0.261
AVLT_N3	6.35 (2.01)	5.23 (1.74)	6.85 (1.49)	7.63 (1.92)	19.136	<0.001	0.275
AVLT_N4	4.40 (2.39)	2.64 (1.54)	5.15 (1.81)	6.50 (1.85)	50.418	<0.001	0.500
AVLT_N5	3.87 (2.43)	1.89 (1.27)	4.81 (1.30)	6.10 (2.12)	72.308	<0.001	0.589
AVLT_N6	3.83 (2.59)	2.13 (1.50)	4.44 (2.29)	5.93 (2.45)	34.066	<0.001	0.403
AVLT_N7	20.13 (2.90)	18.74 (2.79)	20.74 (2.98)	21.77 (1.81)	13.346	<0.001	0.209
AVLT_IR	14.89 (4.57)	12.06 (3.19)	15.81 (3.20)	18.50 (4.66)	29.263	<0.001	0.367
AVLT	23.17 (8.88)	16.60 (5.32)	25.78 (5.64)	31.13 (8.04)	51.964	<0.001	0.507
Language function
BNT	22.31 (3.73)	21.49 (3.95)	21.11 (3.82)	24.67 (1.86)	10.018	<0.001	0.166
Attention function
SDMT_correct	30.24 (12.76)	25.94 (10.28)	23.22 (9.23)	43.30 (9.04)	39.112	<0.001	0.436
Execution function
STT_B (s)	169.43 (81.30)	186.43 (83.54)	196.50 (88.86)	118.43 (38.49)	9.882	<0.001	0.164
Brain volumes (cm3)
TIV	1369.67 (144.90)	1379.45 (146.60)	1371.98 (170.81)	1349.87 (111.18)	0.346	0.708
GM	572.70 (51.03)	577.70 (54.04)	572.78 (53.33)	563.78 (43.08)	0.614	0.543
Hippocampus_L volume	0.45 (0.05)	0.44 (0.06)	0.46 (0.04)	0.44 (0.05)	0.697	0.500
Hippocampus_R volume	0.41 (0.05)	0.41 (0.06)	0.42 (0.04)	0.40 (0.04)	1.362	0.261
WM	472.77 (52.36)	478.41 (53.27)	471.69 (59.35)	463.95 (42.48)	0.637	0.531

Data are mean (SD), unless otherwise indicated.

^*Comparisons among aMCI (n = 47), naMCI (n = 27) and HC (n = 30).

Abbreviations: aMCI: amnestic mild cognitive impairment; AVLT: Auditory Verbal Learning Test Total Recall (sum of AVLT_N1-N5); AVLT_IR: Auditory Verbal Learning Test Immediate Recall (sum of AVLT_N1-N3); AVLT_N1: Auditory Verbal Learning Test First Trial (first Immediate Recall); AVLT_N2: Auditory Verbal LearningTest SecondTrial (SecondImmediateRecall); AVLT_N3: AuditoryVerbal LearningTest ThirdTrial (ThirdImmediateRecall); AVLT_N4: AuditoryVerbal Learning Test Fourth Trial (short delay recall); AVLT_N5: Auditory Verbal Learning Test Fifth Trial (long delay recall); AVLT_N6: Auditory Verbal Learning Test Sixth Trial (category-cued recall); AVLT_N7: AuditoryVerbal LearningTest SeventhTrial (long delayed recognition); BNT: Boston Naming Test; FAQ: Functional Activities Questionnaire; GM: Gray Matter; HAMA: Hamilton Anxiety Scale; HAMD: Hamilton Depression Scale; HC: healthy control; L: left; MMSE: Mini-­Mental State Examination; naMCI: non-amnestic mild cognitive impairment; R: right; SDMT: Symbol Digit Modalities Test; STT_B: Shape Trail Test Part B; TIV: Total intracranial volume; WM: White Matter.

Global topology property analysis of functional brain networks

Between-group comparisons of global topological properties for the aMCI, naMCI, and HC groups

The functional brain networks of the left and right hemispheres in the aMCI, naMCI, and HC groups all exhibited small-world properties. In the left hemisphere functional brain network, after controlling for age, gender, and years of education, the results of the covariance analysis showed significant group differences in two topological metrics, Eloc (F = 3.228, p = 0.044, partial η² = 0.063) and Cp (F = 3.456, p = 0.036, partial η² = 0.067). No significant differences were found for the other metrics (p > 0.05). Pairwise comparisons showed that the Eloc and Cp metrics were lower in the aMCI group compared to the HC group, while no significant differences were observed between the naMCI group and the other two groups (Figure 2). Additionally, no significant group differences were found in all global topological metrics for the right hemisphere brain network.

Figure 2.

Comparison of global topological properties in left hemisphere functional brain networks among the aMCI, naMCI, and HC groups under sparsity thresholds (0.09–0.48). The central line in the box indicates the median, the box boundaries represent the 25th and 75th percentiles (IQR), and the whiskers show the minimum and maximum values within 1.5 IQR. Individual data points from all subjects are superimposed. *Compared with HC group, p < 0.05

Abbreviation: aMCI: amnestic mild cognitive impairment; AUC: area under curve; Cp: clustering coefficient; Eglob: global efficiency; Eloc: local efficiency; HC: healthy control; L: left; Lp: characteristic path length; naMCI: non-amnestic mild cognitive impairment; R: right; γ: normalized clustering coefficient; λ: normalized path length; σ: small worldness.

Comparison of differences in interhemispheric topological organization of functionally connected networks

Figure 3 displayed the differences in global topological properties between the left and right hemispheres in the aMCI, naMCI, and HC groups. The results indicated that both the aMCI and naMCI groups exhibited significantly lower γ and σ values in the left hemisphere compared to the right hemisphere (aMCI-γ: t = 2.311, p = 0.025, Cohen’s d = 0.337; aMCI-σ: t = 2.059, p = 0.045, Cohen’s d = 0.301; naMCI-γ: t = 2.127, p = 0.043, Cohen’s d = 0.409; naMCI-σ: t = 2.165, p = 0.040, Cohen’s d = 0.417), while no such differences were observed in the HC group. Additionally, the aMCI group showed a significant right hemisphere advantage in Cp (t = 2.288, p = 0.027, Cohen’s d = 0.334) and Eloc (t = 2.243, p = 0.030, Cohen’s d = 0.327), which was not found in either the HC or naMCI groups.

Figure 3.

Differences in hemispheric lateralization of global topological properties in bilateral functional brain networks among the aMCI, naMCI, and HC groups. The white dot indicates the median, the thick black bar represents the interquartile range (IQR), and the thin black lines (whiskers) extend to the minimum and maximum values within 1.5 × IQR from the lower and upper quartiles, respectively. *Statistical difference in topological properties between left and right hemispheres, p < 0.05

Abbreviation: aMCI: amnestic mild cognitive impairment; Cp: clustering coefficient; Eglob: global efficiency; Eloc: local efficiency; HC: healthy control; Lp: characteristic path length; naMCI: non-amnestic mild cognitive impairment; γ: normalized clustering coefficient; λ: normalized path length; σ: small worldness.

Analysis of global topological properties of FA white matter structured networks

In the group comparisons of global topological organization between the aMCI, naMCI, and HC groups, no significant differences were found in any of the indicators of the left and right hemisphere FA white matter structural networks.

When comparing hemispheric lateralization in the FA white matter structural networks, we observed that the HC group exhibited a significant left hemisphere advantage in Eglob, Eloc, and Cp, while showing a significant right hemisphere advantage in Lp, γ, λ, and σ (p < 0.05). Interestingly, no clear hemispheric lateralization was observed in the Eloc metric in the aMCI and naMCI groups, and the lateralization patterns for the other metrics were consistent with those observed in the healthy control group (Figure 4). The detailed statistical analysis results were provided in Table S1.

Figure 4.

Comparison of hemispheric lateralization in FA white matter structural networks among the aMCI, naMCI, and HC groups. The white dot indicates the median, the thick black bar represents the interquartile range (IQR), and the thin black lines (whiskers) extend to the minimum and maximum values within 1.5 × IQR from the lower and upper quartiles, respectively. *Statistical difference in topological properties between left and right hemispheres, p < 0.05.

Abbreviation: aMCI: amnestic mild cognitive impairment; Cp: clustering coefficient; Eglob: global efficiency; Eloc: local efficiency; HC: healthy control; Lp: characteristic path length; naMCI: non-amnestic mild cognitive impairment; γ: normalized clustering coefficient; λ: normalized path length; σ: small worldness.

Global structure-function coupling coefficient analysis

In the group comparison of global structural-functional coupling coefficients(S-FCC) between the aMCI, naMCI, and HC groups, no significant differences were found (p > 0.05). However, it is noteworthy that in the within-group comparisons for hemispheric lateralization, the HC group exhibited a significant right hemisphere advantage over the left hemisphere in structural-functional coupling (t = 2.215, p = 0.035, Cohen’s d = 0.404). In contrast, this hemispheric lateralization phenomenon disappeared in both the aMCI and naMCI groups (Figure 5).

Figure 5.

Comparison of hemispheric lateralization in structure-function coupling coefficients among the aMCI, naMCI, and HC groups. The black dots represent the median, and the lines indicate ±1 standard deviation. *Significant differences exist between the left and right hemispheres, p < 0.05.

Abbreviation: aMCI: amnestic mild cognitive impairment; HC: healthy control; naMCI: non-amnestic mild cognitive impairment; S-FCC: structural-functional coupling coefficient.

Comparison of biased lateralization of node properties

Nodal metrics analysis of functional brain networks

In the aMCI group, paired t-tests for hemispheric comparison revealed that NCp in the Precentral gyrus (PreCG, t = 4.138, p < 0.001, Cohen’s d = 0.604), Calcarine fissure and surrounding cortex (CAL, t = 4.229, p < 0.001, Cohen’s d = 0.617), and Temporal pole and superior temporal gyrus (TPOsup, t = 3.954, p < 0.001, Cohen’s d = 0.577) regions exhibited a significant left hemisphere deficit compared to the right hemisphere (Figure 6(A,B)). Additionally, as shown in Figure 6(C,D), the Eloc(i) metric in the CAL (t = 3.700, p = 0.001, Cohen’s d = 0.540) and TPOsup (t = 3.890, p < 0.001, Cohen’s d = 0.567) regions of the left hemisphere was also significantly lower than that in the right hemisphere. All the difference results were corrected for FDR. However, no significant hemispheric differences in NCp or Eloc(i) were observed in any brain regions in the HC or naMCI groups.

Figure 6.

Lateralization comparison of the nodal clustering coefficient (A,B) and nodal local efficiency (C,D) in functional networks between the left and right hemispheres. The central line in the box indicates the median, the box boundaries represent the 25th and 75th percentiles (IQR), and the whiskers show the minimum and maximum values within 1.5 IQR. Individual data points from all subjects are superimposed. *Significant differences exist between the left and right hemispheres, p < 0.05.

Abbreviation: aMCI: amnestic mild cognitive impairment; CAL: Calcarine fissure and surrounding cortex; HC: healthy control; L: left; PreCG: precentral gyrus; naMCI: non-amnestic mild cognitive impairment; R: right; TPOsup: temporal pole and superior temporal gyrus.

Nodal metrics analysis of FA white matter structural brain networks

There are nodes with different lateralization trends in the paired t-test comparisons between the left and right hemispheres in the aMCI group, naMCI group, and HC group (Figures S1 and S2). First, for the NCp metric, compared to the HC group, the aMCI patients showed a loss of right hemisphere dominance in the dorsolateral superior frontal gyrus (SFGdor) and left hemisphere dominance in the middle frontal gyrus (MFG). However, right hemisphere dominance was observed in the olfactory cortex (OLF) and insula (INS), while left hemisphere dominance appeared in the hippocampus (HIP). Besides, naMCI patients showed the disappearance of right hemisphere dominance in the SFGdor and left hemisphere dominance in the MFG, recurrent sulcus (REC), and paracentral lobule (PCL). However, right hemisphere dominance was found in CAL and INS, and left hemisphere dominance was observed in the caudate nucleus(CAU) (Figure 7(A,B)). In addition, for the Eloc(i) metric, both the aMCI and naMCI groups, compared to the HC group, exhibited the loss of left hemisphere dominance in the superior orbital frontal gyrus (ORBsup), MFG, rostral part of the central sulcus (ROL), supramarginal gyrus (SMG), and TPOsup. At the same time, left hemisphere dominance was observed in the parahippocampal gyrus (PHG) and right hemisphere dominance in the REC. However, the aMCI group showed specific right hemisphere dominance in the OLF and thalamus (THA), while the naMCI group showed specific right hemisphere dominance in the posterior cingulate gyrus (PCG) (Figure 7(C,D)). All results were corrected for FDR, as detailed in Tables S2 and S3.

Figure 7.

Lateralization comparison of the nodal clustering coefficient(A,B) and nodal local efficiency(C,D) in the white matter structural brain network between the left and right hemispheres. Bar heights represent the mean, and error bars indicate the standard deviation. *Significant differences exist between the left and right hemispheres, p < 0.05.

Abbreviation: aMCI: amnestic mild cognitive impairment; CAL: calcarine fissure and surrounding cortex; CAU: caudate nucleus; HC: healthy control; HIP: hippocampus; INS: insula; L: left; MFG: middle frontal gyrus; naMCI: non-amnestic mild cognitive impairment; OLF: olfactory cortex; ORBsup: superior orbital frontal gyrus; PCG: posterior cingulate gyrus; PCL: paracentral lobule; PHG: parahippocampal gyrus; R: right; REC: recurrent sulcus; ROL: rostral part of the central sulcus; SFGdor: dorsolateral superior frontal gyrus; SMG: supramarginal gyrus; THA: thalamus; TPOsup: Temporal pole and superior temporal gyrus.

Correlation with clinical scores

Based on Pearson partial correlation analysis, we evaluated the relationship between abnormal hemispheric parameters and clinical scores in the aMCI group, naMCI group, and the overall cohort of 104 participants. The results revealed significant positive correlations between AVLT_N7 and the Cp metric of the left-hemispheric functional brain network, as well as the Eloc metric of the FA-based white matter structural network, in the overall population (Cp: r = 0.232, p = 0.018; Eloc: r = 0.307, p = 0.002). Furthermore, this correlational pattern was also observed in the aMCI subgroup, with even stronger associations (Cp: r = 0.332, p = 0.023; Eloc: r = 0.409, p = 0.004). Additionally, a positive correlation was identified between the structure-function coupling coefficient of the right hemisphere and AVLT_N5 in the aMCI group (r = 0.340, p = 0.019), whereas no significant correlation was found in the overall population (r = 0.013, p = 0.898) (Figure 8). No significant correlations were found in the naMCI group.

Figure 8.

Correlation between abnormal hemispheric parameters and clinical scores.

Abbreviation: aMCI: amnestic mild cognitive impairment; AVLT_N5: Auditory Verbal Learning Test Fifth Trial (long delay recall); AVLT_N7: AuditoryVerbal LearningTest SeventhTrial (long delayed recognition); Cp: clustering coefficient; Eloc: local efficiency; FN: functional network; L: left; R: right; S-FCC: structural-functional coupling coefficient; SN: structural network.

Discussion

To the best of our knowledge, this study is the first to investigate hemispheric lateralization abnormalities in MCI at the functional, structural, and structure–function coupling levels. We constructed the functional and structural brain networks of the left and right hemispheres of aMCI, naMCI and HC, and compared intergroup differences and lateralization using graph theory metrics and structure–function coupling analysis. The results revealed that both aMCI and naMCI showed abnormal hemispheric lateralization, especially in the left hemisphere, at both structural and functional levels, and that abnormal hemispheric asymmetry in brain network properties was significantly associated with memory performance. Notably, patients with aMCI showed reductions in local efficiency in the left hemisphere, predominantly in the limbic system and in integrated motor, visual, and semantic networks, and these alterations may represent the most promising potential biomarkers. These findings offer new insights into the central nervous mechanisms of MCI and provide a theoretical basis and direction for future neuroregulatory interventions.

There is no doubt that hemispheric lateralization is a core architectural principle of human brain organization, which is the result of a combination of evolutionary pressures, genetic programming, environmental interactions, and energetic optimization [10,46]. Both at the structural and functional levels, the human brain is characterized by varying degrees of asymmetry. Many brain phenotypes have been shown to have left-right hemispheric asymmetry, including gray matter volume, cortical thickness, and white matter integrity, which are closely related to brain function [47–49]. A study on the functional and structural brain connectivity network topological properties in community-dwelling older adults indicated that, in healthy elderly individuals, the functional network was generally in a state of relative balance between the left and right hemispheres, while the structural network exhibited leftward asymmetry [50]. Brandon T. Craig et al. [51] also conducted a detailed analysis of the structural connectivity patterns of the left and right hemispheres of the brain during development. The results indicated that global efficiency, hierarchical complexity, and clustering coefficient exhibited leftward lateralization, whereas betweenness centrality showed rightward lateralization. These findings are consistent with our results. Furthermore, abnormal hemispheric asymmetry measured by brain networks may have potential as a biomarker of cognitive impairment. Previous studies have observed abnormal right-sided dominance in cognitively impaired patients possibly related to left hemisphere damage and the initiation of compensatory mechanisms in the right hemisphere [52]. The results of this study in terms of function, structure, and structure-function coupling are described in detail below.

First, from the perspective of functional brain network analysis, we observed that hemispheric lateralization changes at both global and nodal levels in the aMCI and naMCI patients primarily occurred in the left hemisphere. This aligns with previous studies suggesting that aging-related cognitive impairments preferentially affect left hemisphere memory circuits [53,54]. At the global level, compared to healthy individuals with symmetrical topological properties, both aMCI and naMCI patients showed a significant left hemisphere deficit in σ, with lower values in the left hemisphere. Small-world properties, critical for maintaining a balance between functional segregation and integration, were reduced in the left hemisphere of MCI patients, suggesting degenerative changes in information transmission efficiency [55–57]. Notably, in the aMCI group, both Cp and Eloc in the left hemisphere were reduced and positively correlated with delayed recognition memory, indicating a decline in local information transmission efficiency within the left hemisphere network [23]. Previous studies have also revealed changes in local brain function in aMCI patients, though the regions of local damage have varied, including the left precuneus, left frontal cortex, left parietal cortex, and others [58–60].

In view of the decreased efficiency of biased local brain regions at the global level, this study further evaluated the lateralization phenomenon at the local level using Eloc(i) and NCp metrics. We discovered that only in aMCI patients did the left hemisphere show lower Eloc(i) and NCp values compared to the right hemisphere. The significantly affected brain regions were located in the primary motor cortex, primary visual cortex, and temporoparietal junction, with specific regions including PreCG, CAL, and TPOsup. These brain areas are involved in the integration of multimodal information such as motor, visual, auditory, and semantic processing. Such lateralized network damage may indicate the early decline of source memory in episodic memory in aMCI patients. It is well known that episodic memory includes both item memory and source memory. Source memory, in particular, is essential for the reliability of memories and cognitive integrity, as it facilitates the accurate integration of temporal, spatial, and other contextual information [61,62]. PreCG is involved in the planning and execution of action sequences, and disruption of its modular structure may affect the encoding and extraction of action-related source memories [63,64]. Previous studies have also analyzed resting-state fMRI and found that the sensory-motor network, including the PreCG, shows significantly reduced functional activity in MCI patients compared to healthy controls [65,66]. CAL, as a key brain region of the visual network, could provide important cues to visual sources in source memory [67]. Its local network inefficiency may result in encoding deficits of visual sources, disrupting the synchronous processing of visual features with other modalities, leading to the loss of sensory details in source memory. Last but not least, the TPOsup, located at the junction of the temporal, parietal, and occipital lobes, is a critical node in both the default network and semantic network. It is also considered a shared foundation for auditory memory and semantic comprehension [68,69]. A decline in its connectivity efficiency may disrupt the integration of cross-modal information, such as auditory and semantic processing, preventing effective associations between item memory and context, leading to source memory confusion and impairments in memory function. Previous studies have also shown that the left TPOsup exhibits reduced neural activity in aMCI patients [70,71], which is consistent with our findings. In summary, we hypothesize that the memory impairments in aMCI patients are likely due to the damage to source memory, which results from the disruption of multi-information integration networks.

Secondly, in the white matter structural brain network analysis, hemispheric asymmetry in healthy individuals is a widely recognized phenomenon [72]. This lateralization arises from evolution-driven functional differentiation, with the left hemisphere white matter primarily supporting high-density short-range connections, while the right hemisphere is more involved in long-range integration. Specifically, the left hemisphere, as the dominant hemisphere for language, logical analysis, and fine motor control, typically exhibits higher Eloc and Cp, supporting rapid and dense local information processing. The advantage of Eglob may reflect the optimization of cross-brain region integration by central hubs in the left hemisphere. The right hemisphere, which plays a dominant role in spatial navigation and overall context integration, may rely on longer Lp and stronger small-world properties to support the dynamic balance of widely distributed networks, adapting to the multitasking demands of complex environments [73,74]. Notably, we observed that the left hemisphere advantage in Eloc disappeared in both the aMCI and naMCI patients, which may indicate a decline in local information processing in the left hemisphere or compensatory functional reorganization in the right hemisphere.

At the nodal level, integrating the results of NCp and Eloc(i), we observed a significant rightward asymmetry in numerous brain regions in both the aMCI and naMCI patients. The affected regions primarily involve the salience network, cingulo-opercular network, and fronto-parietal control network, which are well-established networks commonly associated with cognitive impairments. These networks are responsible for the integration and execution of cognitive information [75–77]. Furthermore, we observed specific changes in the limbic system in the aMCI patients. Several studies have confirmed that aMCI patients exhibit early atrophy in the left medial temporal lobe [78], with the entorhinal cortex being one of the first areas to show atrophy, often at a faster rate than the hippocampus [79,80]. Atrophy in these brain regions results in a decline in information processing abilities, potentially leading to the disuse or degeneration of white matter connections, thereby affecting local efficiency. In this context, the local connectivity efficiency of the HIP and PHG may increase compensatorily to maintain memory function. Additionally, we observed that the white matter network damage in naMCI patients was more extensive, involving the sensorimotor network, default network, executive control network and others. These findings were consistent with previous reports [81].

Although functional brain network analysis has revealed abnormal patterns of dynamic information integration in patients with cognitive impairments, and white matter structural network research has further clarified the degenerative characteristics of their anatomical basis, the relationship between these two mechanisms remains inadequately investigated. To address this gap, this study employed SFC to examine the connection between structural connectivity and functional activity [25]. We found that in the healthy population, the S-FCC exhibited a right hemisphere dominance, whereas this lateralization phenomenon disappeared in both aMCI and naMCI patients. The underlying neural mechanism of this may be related to functional compensation between the left and right hemispheres. Under physiological conditions, the left hemisphere, due to the complexity of its higher cognitive functions, often exhibits a lower degree of coupling. This characteristic may facilitate dynamic reorganization of functional connections by reducing the rigid constraints imposed by structural connections, thus supporting flexible problem-solving and adaptive behaviors [82,83]. Through the analysis of structural and functional networks, we also observed that during the pathological process of cognitive impairment, the left hemisphere often became an early target of damage. This damage may force the right hemisphere to compensate by enhancing functional connectivity. However, this compensation might lead to a decrease in S-FCC due to insufficient structural support.

Building on the above analyses, we further consider neuromodulatory strategies for MCI, particularly the potential of transcranial magnetic stimulation (TMS) and transcranial direct current stimulation (tDCS) to modulate hemispheric lateralization. Evidence has shown that non-invasive brain stimulation techniques could restore interhemispheric balance and improve cognitive function by targeting specific brain regions or networks. For example, repetitive TMS applied to the left temporoparietal association cortex or to key nodes of the default mode network has been reported to enhance local neural activity and interhemispheric connectivity, thereby improving memory encoding and retrieval [84,85]. Furthermore, bilateral tDCS montages involving anodal stimulation of the left dorsolateral prefrontal cortex combined with cathodal stimulation of the right homologous region have also been shown to effectively modulate cortical activity and enhance cognitive performance in MCI patients [86]. In summary, neuromodulation approaches that specifically target the observed decreased network efficiency in the left hemisphere hold promise as precise therapeutic strategies for addressing memory decline in aMCI patients, thereby providing a potential translational pathway for the network impairment mechanisms revealed in this study.

Undoubtedly, this study has certain limitations. First, this study enrolled only right-handed participants. Although this design effectively controlled the potential confounding effect of handedness on hemispheric lateralization and improved internal validity, it unavoidably limits the generalizability of our conclusions to left-handed populations. Because left-handed individuals may exhibit distinct patterns of hemispheric lateralization, future studies will include them to more comprehensively delineate similarities and differences in the neural mechanisms of MCI across subgroups. Second, although we divided MCI patients into aMCI and naMCI groups, we did not further refine the subgroups based on specific cognitive domain impairments (such as memory, language, or executive function subtypes). The existing classification may mask the heterogeneity within the MCI population. Finally, and most critically, the design and nature of this study must be considered. As an exploratory cross-sectional investigation, its limitations—including a finite sample size, the lack of an independent validation cohort, and the inherent constraints of its design—determine that the present findings should be interpreted primarily as revealing phenomena and generating hypotheses. Although the observed intergroup differences were statistically significant, they cannot be directly inferred as indicative of a causal relationship with the disease, nor do they demonstrate diagnostic classification accuracy at an individual level. Therefore, the clinical translational potential of these lateralization indicators requires further validation in prospective, large-sample longitudinal studies.

Conclusion

In summary, this study confirmed that MCI patients exhibited significant hemispheric lateralization differences in both structural and functional brain networks, as well as alterations in structural-functional coupling. The mechanisms underlying these changes may be related to functional impairments, structural reorganization, and hemispheric compensation. Notably, aMCI patients showed specific breakdowns in local functional networks, which may highlight the importance of source memory impairment in aMCI. This study expands the exploration of the central mechanisms of mild cognitive impairment from a multimodal brain network analysis perspective.

Additional supporting information could be found online in the Supporting Information section at the end of this article.

Funding Statement

This work was supported by the National Natural Science Foundation of China [Grant Nos.: 82302870, 82472613, 82472589, 82402991]; Shanghai Municipal Special Fund for Promoting High-Quality Industrial Development [240404]; Shanghai Innovative Medical Device Application Demonstration Project [23SHS05600]; Shanghai Rising-Star Program [Grant No.: 23QA1409200, 24QA2709300];High-level Chinese Medicine Key Discipline Construction Project (Integrative Chinese and Western Medicine Clinic) of National Administration of TCM [zyyzdxk-2023065]; Shanghai Hospital Development Center Foundation-Shanghai Municipal Hospital Rehabilitation Medicine Specialty Alliance [SHDC22023304].

Ethical approval

This study was approved by the Institutional Ethics Committee of Yueyang Hospital of Integrated Traditional Chinese and Western Medicine (Approval No. 2021-103). All participants provided written informed consent prior to their participation in accordance with the Declaration of Helsinki.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.