Multi-Index Brain Function-Cognition Correlation in Amnestic Mild Cognitive Impairment: A Resting-State fMRI Study

Abstract

Background

This study aimed to investigate the correlation between multi-metric resting-state fMRI (rs-fMRI) changes and cognitive performance in amnestic mild cognitive impairment (aMCI), moving beyond single-metric analyses.

Methods

Forty aMCI patients and twenty healthy controls underwent 3.0T rs-fMRI and neurocognitive assessment (MoCA, DST). Multiple rs-fMRI metrics (ReHo, ALFF/fALFF, RSFC) were analyzed for their association with cognition.

Results

aMCI patients showed increased frontal ReHo, decreased ALFF in default mode network hubs, and reduced prefrontal-parietal-temporal RSFC. Frontal hyper-synchronization correlated negatively with cognitive scores, while DMN ALFF correlated positively.

Conclusions

Multi-metric fMRI detects characteristic functional alterations in aMCI, indicating regional brain damage or compensatory changes linked to cognitive performance.

1. Introduction

Mild cognitive impairment (MCI) represents a clinical continuum between normal cognitive aging and dementia,^1,2 serving as a critical transitional phase. ³ Amnesic mild cognitive impairment (aMCI) is a primary subtype of mild cognitive impairment that progresses to Alzheimer’s disease. Although aMCI is primarily defined by episodic memory decline, objective assessments frequently reveal concomitant deficits in working memory and executive function, which further accelerates conversion to dementia. ⁴ Without effective intervention, the annual progression rate of aMCI exceeds 10%. The global cost of dementia, already amounting to US$1.3 trillion in 2020, is projected to double by 2030. ⁵ However, diagnosis rates in China and other low- and middle-income countries (LMICs) lag significantly behind those in high-income nations, ⁵ which highlights an urgent need for accessible and reliable early diagnostic tools.

aMCI arises from heterogeneous neuropathological processes, including neurodegenerative changes, cerebrovascular insults, and systemic metabolic dysfunction. ⁶ Although aMCI initially causes only mild disruption to instrumental activities of daily living, affected individuals face a substantially elevated risk of progressing to dementia, with current therapeutic options offering limited symptomatic relief.^7,8 Therefore, identifying reliable, multi-dimensional biomarkers has emerged as a pressing research priority.

Neuropsychological testing remains the cornerstone of MCI diagnosis. While episodic memory deficits are central to aMCI, impairments in executive function, attention, and visuospatial abilities are also commonly observed. ⁹ MoCA is a widely adopted screening tool, demonstrating sensitivity exceeding 90% for detecting MCI, ¹⁰ MoCA performance has also been correlated with aberrant DMN connectivity. ¹¹ The DST provides complementary insights, specifically evaluating auditory attention and working memory capacity. Its forward (DS-F) and backward (DS-B) components tap into distinct cognitive processes: DS-F primarily indexes auditory attention and immediate recall, whereas DS-B is a more demanding test of working memory and executive control. ¹² Deficits in DST performance, particularly DS-B, have been documented in aMCI populations and are thought to reflect early frontal lobe dysfunction, predicting subsequent progression to AD. ¹³ MoCA and DST demonstrate sensitivity to early cognitive deficits, making them crucial tools for identifying prodromal AD.^10,14 Together, they underscore that aMCI is not solely a memory disorder but also involves early alterations in executive and working memory processes.

While cognitive screening provides essential clinical information, it does not directly elucidate underlying neural mechanisms. RS-fMRI offers a non-invasive window into spontaneous brain activity and functional organization.^15,16 Key rs-fMRI metrics—including ReHo, ALFF, fALFF, ¹⁷ and RSFC—provide complementary indices of local and large-scale brain function.^18,19 A major focus of aMCI research has been the DMN, a constellation of midline and parietal regions, including the posterior cingulate cortex (PCC), precuneus, and medial prefrontal cortex (mPFC) that is highly active during rest and internally directed cognition (e.g., episodic memory retrieval). ¹⁸ DMN connectivity disruption is one of the most consistent neuroimaging findings in MCI and AD, with hypoconnectivity between the PCC and mPFC strongly linked to early AD pathology. ²⁰ Recent graph-theoretical analyses of rs-fMRI data have further revealed declines in global network efficiency in aMCI. ²¹ Crucially, altered DMN activity may precede overt hippocampal atrophy, suggesting its potential as an early biomarker. ²² Conversely, prefrontal and sensorimotor regions may exhibit compensatory hyperactivation,^18,19 a mechanism thought to temporarily sustain cognitive performance. However, the efficiency of such compensation is finite, and excessive local synchronization may itself signify pathological network dysfunction. ²³

Despite these advances, the literature on resting-state network alterations in aMCI remains somewhat limited and inconsistent. First, while the relationship between MoCA and DMN changes has been explored^11,24 association between DST performance and multi-parametric rs-fMRI metrics is underexplored, especially in populations where educational attainment may influence test scores. ²⁴ Second, previous fMRI studies have often adopted a unidimensional analytical approach, failing to integrate multiple parameters that collectively capture brain synchrony, activity strength, and rhythmic specificity. ²⁵ In contrast, emerging large-scale studies have begun to address this gap. A brain-wide risk score integrating structural MRI and resting-state functional network connectivity was developed in a cohort of over 37,000 participants, demonstrating that multimodal features effectively discriminate MCI from cognitively normal individuals. ²⁶ Xu et al further applied multimodal MRI analysis to characterize early-stage MCI, integrating structural and functional metrics. ²⁷ These studies highlight the promise of multi-metric and multimodal approaches for early MCI detection.

Moreover, emerging evidence from our group further supports the necessity of adopting a multi-network perspective in understanding aMCI-related brain dysfunction. Our preliminary work has demonstrated that: (1) Functional brain network alterations, characterized by compensatory increases in global efficiency and integration, occur as early as the subjective cognitive decline stage, particularly in carriers of the APOEε4 allele ²⁸ ; (2) Effective neuromodulation interventions, such as high-frequency rTMS, improve cognition in aMCI by simultaneously modulating spontaneous activity across widespread regions spanning the DMN, executive control network (ECN), and frontoparietal network (FPN) ²⁹ ; and (3) Non-cognitive symptoms in MCI, such as balance deficits, are linked to functional disconnection within specific networks like the corticocortical vestibular network (CVN). ³⁰ These findings collectively suggest that aMCI is a multi-network dysconnectivity syndrome. However, a systematic investigation that concurrently examines the integrity of multiple core neurocognitive networks (e.g., DMN, ECN, FPN) and their relationship with multi-domain cognitive performance (e.g., memory, executive function, working memory) in a well-defined aMCI cohort is still lacking.

Neuropsychological assessment serves as one of the fundamental screening tools for aMCI, while fMRI is now widely used for analyzing brain function in aMCI. We hypothesized that compared with healthy controls (HC), aMCI patients exhibit distinctive alterations in rs-fMRI metrics, reflecting a multi-network dysconnectivity pattern, which are correlated with cognitive performance scores.

This study therefore integrates sensitive neuropsychological assessments, including MoCA and DST, with multiple complementary rs-fMRI metrics, including ReHo, ALFF, fALFF, and RSFC, to characterize functional brain alterations in aMCI and clarify their associations with cognitive performance. These metrics were selected because they capture distinct but complementary dimensions of brain dysfunction. ReHo reflects local synchronization of spontaneous neural activity and may help identify regional functional abnormalities or compensatory hyper-synchronization. ALFF quantifies the intensity of spontaneous low-frequency neural activity and is useful for detecting reduced intrinsic activity in cognition-related regions, especially default mode network hubs. fALFF further improves the specificity of ALFF by reducing the influence of non-neuronal physiological noise, thereby providing a more reliable index of frequency-specific spontaneous activity. RSFC evaluates long-range functional communication between brain regions and can reveal disrupted network integration among the default mode, executive control, frontoparietal, and sensorimotor networks. By combining these indices, this study aims to provide a more comprehensive functional profile of aMCI than any single metric alone. Clinically, such a multi-index approach may improve the identification of early functional abnormalities, distinguish impaired neural activity from inefficient compensatory recruitment, ³¹ and provide imaging evidence that complements cognitive screening for early diagnosis and disease monitoring.

2. Materials and Methods

2.1. Participants

The present study recruited a total of 60 participants, including 40 patients diagnosed with aMCI and 20 HC. A schematic overview of the study design is presented in Figure 1.The Affiliated Baoan Hospital of Shenzhen, Southern Medical University’s Ethics Committee for Clinical Research approved this study. The aMCI and HC were recruited via advertisements posted in local communities in Fuzhou City, Fujian Province, China. All participants provided written informed consent after receiving a comprehensive explanation of the study’s purpose, procedures, potential risks, and benefits of the study’s, ensuring full compliance with ethical requirements. The study was registered on the Chinese Clinical Trial Registry (ChiCTR) website (https://www.chictr.org.cn/) under registration number: ChiCTR2300076140.

Figure 1.

Trial flow chart

The diagnosis of aMCI was based on Petersen’s diagnostic criteria (Petersen, 2004). ³²

The inclusion criteria of the aMCI group include ³³ :(1) Male or female, ages 55 to 75;(2) memory complaint provided by the patient or family, Montreal Cognitive Assessment (MoCA) score<26 (with 1 point added to the raw score for participants with ≤12 years of education to account for educational background); (3) Global Deterioration Scale of 2-3; (4) Preserved activities of daily living (ADLs), defined by a score of ≤18 on the Lawton ADL scale(14-item); (5) Native Chinese, right-handed;(6) Willingness and ability to participate in all aspects of the intervention; (7) Availability of spouse or caregiver to provide collateral information and assist with study adherence. (8) All enrolled aMCI patients were treatment-naïve at the time of study enrollment. The inclusion criteria of the HC group include: (1) No subjective cognitive complaints (2) MoCA score≥26; (3) Normal activities of daily living, as indicated by a 14-item ADL scale score of ≤18; (4) Native Chinese, right-handed;(9) Provision of informed consent and voluntary participation. The exclusion criteria include:(1) History of neurological or psychiatric disorders (e.g. stroke, depression, epilepsy, Lewy Body disease, Parkinson’s disease); (2) Significant neurological deficits (e.g., severe visual or hearing impairment); (3) Use of medications known to affect cognitive function;(4) Moderate or severe dementia; (5) Substance abuse problems.

2.2. Data Acquisition

Demographic and clinical data were collected for all participants, encompassing age, gender, ethnicity, occupation, educational level, medical history, family medical background, and exercise frequency.

MRI data were acquired using a Siemens 3.0T MRI scanner equipped with a 64-channel head-neck combined coil. Before scanning, operators delivered individualized psychological support to participants: they explained scanning precautions and helped participants acclimate to the environment and procedures, with the goal of minimizing head movement and maintaining data quality. Specific scanning parameters were as follows:

• (1) T1-weighted imaging: Repetition time (TR) = 2100 ms, echo time (TE) = 30 ms, flip angle = 8°, slice thickness = 1.0 mm, field of view (FOV) = 250 mm × 250 mm, matrix = 256 × 256, voxel size = 0.98 × 0.98 × 1 mm³, number of slices = 160.

• (2) Resting-state fMRI: Repetition time (TR) = 2000 ms, echo time (TE) = 4.6 ms, flip angle = 70°, slice thickness = 1.0 mm, FOV = 230 mm × 230 mm, matrix = 256 × 256, voxel size = 3.6 × 3.6 × 3.9 mm³, number of slices = 30.

2.3. Cognitive Assessment

A comprehensive cognitive assessment battery was administered, which included the MoCA and the DST. All cognitive assessments were administered in a quiet, standardized environment by a trained neuropsychologist unaware of group allocation. Testing was performed between 9:00 AM and 12:00 PM to minimize circadian influences.

The MoCA was administered following the standard protocol. It evaluates seven cognitive domains: visuospatial/executive, naming, memory, attention, language, abstraction, and orientation, with total scores ranging from 0 to 30 (higher scores indicate better function). One point was added for participants with ≤12 years of education to adjust for educational bias. The MoCA demonstrates good test-retest reliability in older adult populations, with intraclass correlation coefficients (ICCs) ranging from 0.86 to 0.90. ³⁴ For Chinese elderly populations, the optimal cutoff for MCI detection is 24/25 for individuals with 7 or more years of education, with a sensitivity of 80.5% and specificity of 82.5%. ³⁵ In oldest-old Chinese populations (aged ≥80 years), the recommended cutoff is ≤24. ³⁶

The DST was administered as part of the Wechsler Memory Scale–Third Edition (WMS-III) protocol. It includes forward span (DS-F), assessing auditory attention and immediate recall, and backward span (DS-B), measuring working memory and executive control. The test terminates after two consecutive errors at a given span length. The DST shows acceptable split-half reliability, with ICCs ranging from 0.64 to 0.65 across repeated assessments in older adults. ³⁷ For Mandarin-speaking older adults, a recent large-scale cross-cultural study (N =3,681 across 14 languages) reported optimal cutoff values for identifying cognitive impairment of DS-F = 7 and DS-B = 5. ³⁸

To ensure stability of test results, we (1) read standardized instructions verbatim to all participants; (2) had the same examiner conduct all assessments; (3) completed testing within a single 40–50 minute session with short breaks as needed to avoid fatigue; and (4) double-checked all scores for accuracy.

2.4. Data Preprocessing

Preprocessing of resting-state fMRI data was performed using the Data Processing Assistant for Resting-State fMRI (DPARSF, version 7.0; https://rfmri.org/DPARSF), a toolbox based on MATLAB 2013b (The MathWorks, Inc., Natick, MA, USA). The initial ‘dicom’ files were transformed into '*.nif’ format using the dcm2niigui tool within the MRIcro software package. Preprocessing steps were executed in the following order:

• (1) Removal of Initial Time Points: The first 10 time points of each run were excluded to enable magnetic field stabilization and participant adaptation, resulting in 230 time points available for analysis;

• (2) Slice Timing Correction: Temporal alignment was carried out to address differences in acquisition time across slices, using slice 29 as the reference;

• (3) Head Motion Correction: Realignment was conducted to mitigate head movement. Participants exhibiting excessive motion (translation > 2.0 mm or rotation > 2.0° in any direction) were excluded from the study;

• (4) Image registration: T1-weighted structural images were aligned with their corresponding functional data. A template was created using all participants to spatially normalize the structural images—this step supports standardized group analysis and reduces the impact of inter-subject variations in brain structure;

• (5) Spatial Normalization: Individual T1-weighted structural images were first co-registered to the mean functional image and subsequently normalized to the Montreal Neurological Institute (MNI) standard space using a study-specific template. The normalized functional images generated from this process were resampled to a voxel size of 3 mm × 3 mm × 3 mm;

• (6) Spatial Smoothing: Normalized functional images were smoothed using a 6 mm full-width at half-maximum (FWHM) Gaussian kernel, which improved the signal-to-noise ratio (SNR) and addressed residual anatomical discrepancies;

• (7) Detrending: Linear trends were eliminated to remove signal drifts linked to scanner instability or participant fatigue;

• (8) Temporal Filtering: A band-pass filter (0.01–0.08 Hz) was applied to the time series of each voxel, aiming to reduce low-frequency drift and high-frequency physiological noise.

2.5. Data Analysis

2.5.1. ReHo Analysis

ReHo was calculated and computed via the DPARSF toolkit. For each voxel, the Kendall’s coefficient of concordance (KCC) was calculated between the voxel’s time series and those of its 26 nearest neighboring voxels. The resulting ReHo maps were then standardized by dividing by the global mean ReHo value and spatially smoothed. Group differences in ReHo were assessed using a two-sample t-test, with age and gender included as covariates (voxel-wise p < 0.01, Gaussian Random Field (GRF) corrected, cluster-level p < 0.05). Regions exhibiting significant group differences were extracted as regions of interest (ROIs), and their mean ReHo values were correlated with participants’ cognitive scores.

2.5.2. ALFF Analysis

The ALFF was calculated to quantify the intensity of spontaneous neural activity. For each voxel, the time series was transformed into the frequency domain; the square root of the power spectrum was then computed and averaged across the 0.01–0.08 Hz frequency range—this averaged value was defined as the ALFF. Individual ALFF maps were standardized by dividing by the global mean ALFF. Group comparisons were performed using a two-sample t-test, with age and gender as covariates (cluster-level False Discovery Rate correction, p < 0.01, cluster size > 100). Mean ALFF values from significant clusters were extracted for correlation analysis with cognitive measures.

2.5.3. fALFF Analysis

fALFF was determined as the ratio of power in the low-frequency range (0.01∼0.08 Hz) to power across the entire frequency range (0∼0.25 Hz). This metric serves to enhance the specificity of ALFF to neural activity by reducing the influence of physiological noise. Standardized fALFF maps were generated, and group differences were analyzed using a two-sample t-test (controlling for age and gender; voxel-wise p < 0.01, GRF corrected, cluster-level p < 0.05). ROI-based mean fALFF values were correlated with cognitive scores.

2.5.4. RSFC Analysis

To explore whole-brain RSFC patterns, RSFC analysis was conducted using seed regions derived from the ReHo, ALFF, and fALFF analyses—specifically, regions that showed significant group differences. The mean time series was extracted from each seed region, and voxel-wise Pearson correlation coefficients were computed between the seed time series and the time series of every other voxel in the brain. Correlation coefficients were converted to z-scores using Fisher’s r-to-z transformation to improve normality. Group differences in RSFC maps for each seed were assessed using two-sample t-tests (p < 0.05, corrected).

2.5.5. Statistical Analysis

Statistical analyses of demographic, clinical, and behavioral data were performed using IBM SPSS Statistics 26.0. Personnel responsible for data processing were blinded to participants’ group assignments. The normality of continuous variables was assessed using histograms, normal probability plots, and the Shapiro-Wilk test. Normally distributed data are presented as mean ± standard deviation (SD) and were compared using two-sample t-tests. Non-normally distributed data are reported as median (interquartile range) and were analyzed using the Mann-Whitney U test. Categorical data are summarized as frequency (percentage) and were compared using the chi-square test. All statistical tests were two-tailed, with a significance level set at α = 0.05.

2.5.6. Correlation Analysis

Regions exhibiting significant group differences were extracted as regions of interest (ROIs), and their mean ReHo values, mean ALFF values and mean fALFF values were correlated with participants’ cognitive scores.Age, sex, and educational level were included as control variables. To address the issue of multiple comparisons and control the family-wise error rate, Bonferroni correction was applied. The significance level was adjusted from α = 0.05 to α = 0.05/14 ≈ 0.0036, based on a total of 14 statistical tests (7 ROIs × 2 cognitive scales).

3. Results

3.1. Subject Characteristics

We screened 259 participants (196 with and 63 without memory impairment). A total of 199 individuals were excluded, as 162 did not meet the inclusion criteria and 37 declined to participate. Consequently, 60 participants (40 with aMCI and 20 healthy controls, HC) provided informed consent and underwent assessments and rs-fMRI. After excluding 3 aMCI participants due to excessive head motion, the final analysis included 57 participants (37 aMCI and 20 HC). The clinical and demographic characteristics of the participants are summarized in Table 1. The two groups were well-matched for age, sex, and education level (all p > 0.05). As expected, the aMCI group scored significantly lower on the MoCA and DST compared to the HC group (p < 0.001 and p < 0.05, respectively).

Table 1.

Characteristics of the MCI Patients and HC( $\overline{\mathrm{x}}$ ±SD)

Group	HC (n=20)	MCI (n=37)	F/χ2/z value	P-value
Subjects	20	37	-	-
Age(years)	69.50±5.12	67.08±5.53	0.237	0.629
Sex(M/F)	6/14	8/29	-	0.348
Education level(years)	12.30±1.922	11.27±2.755	2.085	0.1456
MoCA(scores)	27.30±0.923	22.27±1.94	16.262	<0.001
DST(scores)	13.00±1.806	11.54±2.049	-2.595	<0.05

*MoCA, Montreal Cognitive Assessment; DST, Digit Span Test.

3.2. ReHo Findings

Significant group-level differences in ReHo were identified in two brain regions: the right precentral gyrus (preCG.R) and the left dorsolateral superior frontal gyrus (SFGdor.L) (Figure 2A, Table 2). A two-sample t-test—with age and gender included as covariates—revealed that the aMCI group exhibited significantly higher ReHo values in these two regions relative to the HC group (Figure 2B).

Figure 2.

(A) Brain regions showing significant ReHo differences between the HC and aMCI groups (preCG.R and SFGdor.L).; (B) The small multiples plot illustrating the increased ReHo in the aMCI group compared to the HC group in these regions. *p < 0.05

Table 2.

ReHo Differences Between MCI and HC Groups

Cluster	Brain regions	Peak MNI coordinate			Cluster size	Peak intensity
		x	y	z
Cluster 1	preCG.R	24	-20	70	826	-4.434
Cluster 2	SFGdor.L	-20	6	64	770	-4.5536

Note. *HC, health control; MCI, mild cognitive impairment; GRF-corrected voxel p < 0.01, cluster p < 0.05. RF-corrected voxel p < 0.01, cluster p < 0.05.

3.3. ALFF Results

Significant alterations in ALFF were detected in several brain regions (Figure 3A, Table 3). After controlling for age and gender, the aMCI group exhibited significantly lower ALFF in the right dorsal anterior cingulate cortex (ACCsup.R), left angular gyrus (ANG.L), and right medial superior frontal gyrus (SFGmed.R) compared to the HC group (Figure 3B).

Figure 3.

(A)Brain regions showing significant ALFF differences between groups (ACCsup.R, ANG.L, SFGmed.R);(B) The small multiples plot illustrating the decreased ALFF in the aMCI group relative to the HC group in these regions. *p < 0.05

Table 3.

ALFF Differences Between MCI and HC Groups

Cluster	Brain regions	Peak MNI coordinate			Cluster size	Peak intensity
		x	y	z
Cluster 1	ACCsup.R	4	36	6	113	-3.781
Cluster 2	ANG.L	-46	-76	26	100	-3.786
Cluster 3	SFGmed.R	8	52	32	121	-3.420

Note. *Cluster-level FDR correction p < 0.01, cluster size = 100.

3.4. fALFF Results

Group comparisons of fALFF revealed significant differences in the left medial superior frontal gyrus (SFGmed.L) and the left postcentral gyrus (poCG.L) (Figure 4A, Table 4). The aMCI group exhibited distinct fALFF patterns relative to the HC group: significantly lower fALFF in the SFGmed.L and significantly higher fALFF in the poCG.L (Figure 4B).

Figure 4.

(A) Brain regions showing significant fALFF differences between groups (SFGmed.L and poCG.L). (B) The small multiples plot illustrating the decreased fALFF in SFGmed.L and increased fALFF in poCG.L in the aMCI group compared to the HC group. *p < 0.05

Table 4.

fALFF Differences Between MCI and HC Groups

Cluster	Brain regions	Peak MNI coordinate			Cluster size	Peak intensity
		x	y	z
Cluster 1	SFGmed.L	2	54	16	585	-4.9569
Cluster 2	poCG.L	-40	-14	38	722	4.2418

*GRF-corrected voxel p < 0.01, cluster p < 0.05.

3.5. RSFC Findings

Voxel-wise analysis revealed that aMCI patients had significantly decreased RSFC compared to HCs between several key region pairs: between preCG.R and the left inferior parietal lobule (IPL.L), between SFGdor.L and the left middle temporal gyrus (MTG.L), and between the SFGmed.R and the right middle frontal gyrus (MFG.R) (Figure 5A). The mean RSFC strength for these connections was significantly lower in the aMCI group (Figure 5B).

Figure 5.

(A) Brain maps showing significantly reduced RSFC in the aMCI group compared to the HC group. (B) Comparison of mean RSFC (z-scores) between groups for the identified connections. ***p < 0.0005

3.6. Correlation Analysis

After controlling for age, gender, and the level of education, Bonferroni correction was performed. The results of correlation analyses indicated that:

• (1) A significant negative correlation was observed between mean ReHo in the preCG.R and SFGdor.L and MoCA scores.

• (2) Mean ALFF in the ACCsup.R showed significant positive correlations with MoCA scores Whereas the mean ALFF values in the ANG.L and SFGmed.R was no statistically significant correlation with MoCA scores.

• (3) Mean fALFF in the SFGmed.L was positively correlated with MoCA scores, whereas mean fALFF in the poCG.L was no statistically significant correlation with MoCA scores.

• (4) Mean ReHo in the SFGdor.L was also significantly negatively correlated with DST scores.

• (5) No other statistically significant correlations with DST scores were found in other regions.

4. Discussion

4.1. This Work

This study aims to integrate sensitive neuropsychological assessments (MoCA, DST) with multi-metric rs-fMRI (ReHo, ALFF, fALFF, RSFC) to explore characteristic functional brain changes in aMCI and analyze their relationship with cognitive function. We discovered that our ALFF and fALFF analyses reveal consistent hypoactivation across DMN hubs. Reduced fALFF in SFGmed.L and reduced ALFF in ACCsup.R demonstrated a positive correlation with MoCA scores, indicating that resting-state activity intensity in these regions is critical for maintaining overall cognitive function (Figure 6A and B).

Figure 6.

Scatter plots showing significant correlations between mean ReHo, ALFF, and fALFF values in specific brain regions and MoCA scores across the participant groups

The RSFC reductions between SFGmed.R and MFG.R reflect compromised integrity within the DMN’s anterior subsystem and the RSFC reductions between preCG.R and IPL.L, and between SFGdor.L and MTG.L, demonstrate impaired integration between the DMN, frontoparietal control network (FPN), and sensorimotor systems, ³⁹ We also observed increases in ReHo in SFGdor.L and preCG.R, and increase of fALFF in poCG.L represent the most direct evidence of compensatory mechanisms and likely reflect compensatory recruitment aimed at stabilizing performance when integrative hubs are compromised. Increased ReHo in SFGdor.L exhibited significant negative correlations with both MoCA and DST scores (r = 0.43–0.56), and increased ReHo in preCG.R exhibited a significant negative correlation with MoCA scores (r = 0.31), indicating that compensatory activation of prefrontal and sensorimotor networks fails to effectively support complex cognitive tasks (Figure 6C–E).

The significantly reduced ALFF observed in ACCsup.R, ANG.L, and SFGmed.R collectively indicates a broad attenuation of spontaneous neural activity within the DMN framework. Our multi-metric approach provides evidence for this “impairment” phase, affecting the DMN and its interconnected systems (Figure 6F and G). Critically, as fALFF specifically suppresses non-neuronal noise, the fALFF reduction in SFGmed.L indicates genuine neuronal dysfunction rather than vascular or other confounds, strengthening the evidence for anterior DMN impairment (Figure 7A). This finding suggests that aMCI patients may experience impaired higher cognitive function, which aligns with meta-analytic evidence of disrupted ALFF in DMN hubs. ⁴⁰ This ALFF reduction aligns with multimodal studies linking spontaneous neural activity to executive control and language processing, ⁴¹ further validating its contribution to global cognitive status. Our RSFC analysis extends the impairment paradigm from regional activity to systems-level integration (Figure 7B–G). The disrupted connectivity between SFGmed.R and MFG.R reflects compromised integrity within the DMN’s anterior subsystem. More broadly, these cross-network RSFC reductions indicate a failure of systems-level integration extending beyond the DMN to encompass the FPN and sensorimotor systems. ³⁹ More importantly, the weakened connections within SFGdor-MTG and SFGmed-MFG suggest the failure of collaborative function between the executive control network and the DMN/semantic processing network. ⁴² These interactive impairments between networks, combined with abnormalities in the regions indicated by ReHo, ALFF, and fALFF, collectively constitute the neural basis of cognitive impairment in aMCI, consistent with recent dynamic functional connectivity studies, ⁴³ likely reflecting the pathophysiological mechanism of coexisting local functional disorders and global network disintegration. ⁴⁴ This large-scale network “decoupling” represents a fundamental characteristic of brain aging and neurodegeneration, ⁴⁵ where the delicate balance between network integration and segregation becomes disrupted. Decreased ALFF in ACCsup.R, ANG.L, and SFGmed.R, together with reduced RSFC among prefrontal, parietal, and temporal pairs, indicates loss of intrinsic activity and long range integration in the DMN and adjacent association cortex, consistent with recent work showing that default mode disconnection emerges early in the disease trajectory. RSFC declines were notably selective, affecting connectivity between the anterior DMN and the frontoparietal network, and between the sensorimotor network and parietal/temporal regions. This indicates a failure of local-to-system integration and corroborates the meta-analytic evidence in MCI, supporting ‘early default network decoupling’ as a core pathophysiology. ⁴⁶

Figure 7.

Scatter plots showing significant negative correlations between mean ReHo values in the preCG.R and SFGdor.L and DST scores

The significantly elevated ReHo in SFGdor.L and preCG.R represents the most direct evidence of compensatory mechanisms. Increased local synchrony typically indicates enhanced neural resource recruitment. The SFGdor.L hyper-synchronization exemplifies the classic “posterior-anterior shift in aging” phenomenon, where prefrontal regions compensate for posterior network impairment. ⁴⁷ As hippocampal and DMN functions decline, the dorsolateral prefrontal cortex increases its computational load to maintain cognitive control. ⁴⁸ These hyperactivation patterns likely reflect compensatory neural resource mobilization in response to integrative hub failure. ⁴⁹ The preCG.R functional enhancement suggests compensatory recruitment extends beyond traditional cognitive networks to include sensorimotor regions, indicating a broader neural resource mobilization. However, the negative correlations between ReHo in these prefrontal and sensorimotor regions and cognitive performance fundamentally challenge a purely beneficial interpretation of these changes. Specifically, increased ReHo in SFGdor.L was negatively correlated with both MoCA and DST scores, while increased ReHo in preCG.R was negatively correlated with MoCA scores, suggesting neural inefficiency, where increased metabolic and computational effort yields diminishing cognitive returns. ²³ Similar mechanisms have been reported in early AD studies, but compensatory mechanisms in the aMCI stage are more dynamic and complex, ⁵⁰ as supported by recent findings of task-based hyperactivation in prefrontal networks. ⁵¹ Regarding maladaptive sensorimotor recruitment, the significantly increased fALFF in poCG.L provides evidence for network-level dedifferentiation. The aberrant engagement of primary somatosensory cortex for cognitive operations represents a failure of functional specialization, ⁵² although the absence of a statistically significant correlation between poCG.L fALFF and cognitive scores after Bonferroni correction suggests that this maladaptive recruitment may not directly translate to measurable cognitive decrements at the individual level, or that its effect size is insufficient to survive stringent multiple comparison adjustment. ⁵³ The DST is the primary tool used for assessing an individual’s working memory capacity. ⁵⁴ Working memory is defined as the ability to temporarily store and simultaneously process information, playing a key role in complex cognitive activities such as learning, reasoning, and understanding. Executive/working memory relates to FPN–DMN coupling: weaker segregation or abnormal coupling is linked to poorer executive performance. This is consistent with the significant negative correlation we observed between SFGdor.L ReHo and DST scores, further supporting the role of prefrontal dysregulation in working memory decline in aMCI.

Taken together, our multi-metric neuroimaging indices (ALFF, fALFF, ReHo, RSFC) revealed a profile in aMCI of compromised DMN activity and connectivity alongside prefrontal and sensorimotor hyper-synchronization. This aligns with the model of disrupted network balance in MCI, specifically DMN decoupling. We hypothesize that the local enhancements in aMCI signify inefficient recruitment and reduced neural efficiency, not stable functional compensation. ⁴⁶ This compensatory mechanism appears to be inefficient and may evolve into maladaptive reorganization as the disease progresses. The characteristic functional alterations in the some brain are significantly correlated with cognitive performance, underscoring the importance of implementing cognitive screening in clinical practice.

4.2. Contribution

The major contribution of this study lies in the integration of multiple complementary rs-fMRI indices to characterize aMCI-related functional brain alterations from both regional and network-level perspectives. Our findings demonstrate that each metric captured a distinct aspect of aMCI-related dysfunction: ReHo and ALFF revealed regional abnormalities (increased frontal synchrony and reduced DMN activity, respectively), fALFF confirmed frequency-specific alterations in sensorimotor regions, and RSFC uncovered widespread disruptions in interregional connectivity. From a clinical perspective, this multi-index framework may provide a more comprehensive imaging signature for aMCI than single-metric analyses.

The combination of DMN hypoactivity, disrupted interregional connectivity, and frontal or sensorimotor hyper-synchronization reveals a pattern in which posterior default-mode regions show reduced activity while anterior and sensorimotor regions exhibit increased local synchronization, suggesting a possible redistribution of neural resources in aMCI. Moreover, significant associations were observed between several of these imaging alterations and MoCA or DST performance; specifically, reduced ALFF in ACCsup.R, reduced fALFF in SFGmed.L, and increased ReHo in SFGdor.L and preCG.R showed significant correlations with MoCA scores, while increased ReHo in SFGdor.L was additionally correlated with DST scores, indicating that these rs-fMRI metrics are not merely descriptive neuroimaging findings but are functionally relevant to cognitive impairment. Therefore, the integrated use of ReHo, ALFF, fALFF, and RSFC may complement neuropsychological assessment, support early diagnosis, and provide potential imaging markers for monitoring disease progression or intervention effects in aMCI.

4.3. Limitation

This research is subject to several methodological limitations that warrant consideration. First, the cross-sectional design precludes direct observation of the dynamic progression from compensatory mechanisms to systemic decompensation within individuals over time. Second, although the sample size is consistent with similar neuroimaging investigations in aMCI research, a larger cohort would enhance statistical power and enable more nuanced subgroup analyses, such as stratifying participants based on genetic risk factors (e.g. APOE ε4 carrier status) or cerebrospinal fluid biomarker profiles. Given the limited sample size and the exploratory positioning of this research, we primarily focused on characterizing the independent alteration pattern and cognitive relevance of each individual metric; for this reason, we have not yet systematically explored the potential synergistic or antagonistic interactions across these metrics, nor attempted to establish a multi-metric combined prediction model in the present work. In addition, the absence of task-based fMRI data limits our ability to evaluate how compensatory brain activity is modulated under specific cognitive demands. Finally, the lack of inclusion of Alzheimer’s disease (AD) patients prevents us from capturing the full trajectory of neural changes as aMCI progresses to AD, thereby limiting insights into which compensatory mechanisms ultimately fail and which brain regions are most vulnerable during this transition.

4.4. Future Work

To address these limitations and extend current findings, several future research directions are proposed. Multi-metric interaction analysis and integrated predictive modeling, which require a well-powered sample and rigorous validation beyond the scope of this study, should be explored in future multicenter studies with enlarged cohorts to clarify inter-metric relationships and construct multimarker predictive models, thereby refining the multi-metric integrative framework. Longitudinal studies tracking participants from aMCI to AD are essential to validate functional decline trajectories and capture temporal changes in network compensation and failure. Integrating multi-modal neuroimaging data (e.g., structural MRI, diffusion tensor imaging, amyloid/tau PET) will facilitate the establishment of a comprehensive model linking neuropathology, brain structure, function, and cognition. Additionally, task-based fMRI should be incorporated to clarify the recruitment and breakdown of compensatory activation under controlled cognitive challenges, while including AD patients will enable direct comparison with aMCI to identify critical neural changes and the point of compensatory failure during disease progression.

5. Conclusions

This study demonstrates that multi-parametric fMRI can detect characteristic alterations in aMCI, indicating the presence of both regional brain damage and compensatory mechanisms. These alterations likely reflect a cascade of pathological events, beginning with impairment of the core DMN, which then triggers maladaptive or ineffective compensatory responses from the prefrontal and sensorimotor networks. Significant correlations between specific imaging alterations and cognitive scores, particularly between frontal ReHo and both MoCA and DST performance, and between DMN activity metrics and MoCA scores, highlightingg the distinct roles of different brain regions: demonstrating that while the hubs of the DMN are crucial for maintaining cognitive performance, the compensatory involvement of the executive control and sensorimotor networks is characterized by maladaptive or inefficient functional compensation.

Appendix.

Abbreviations

aMCI

Amnestic mild cognitive impairment

AD

Alzheimer’s disease

fMRI

Functional magnetic resonance imaging

ReHo

Regional homogeneity

ALFF

Amplitude of low-frequency fluctuation

fALFF

Fractional amplitude of low-frequency fluctuation

RSFC

resting-state functional connectivity

MoCA

Montreal Cognitive Assessment

DST

Digit Span Test

preCG.R

the right precentral gyrus

SFGdor.L

the left dorsolateral superior frontal gyrus

ACCsup.R

the right dorsal anterior cingulate cortex

ANG.L

the left angular gyrus

SFGmed.R

The right medial superior frontal gyrus

SFGmed.L

the left medial superior frontal gyrus

poCG.L

the left postcentral gyrus.

ORCID iDs

Haoyu Huang https://orcid.org/0009-0001-2367-1603

Qin Wang https://orcid.org/0009-0005-1688-5506

Consent for Publication

Informed consent was obtained from all subjects involved in the study. Written informed consent has been obtained from the patients to publish this paper.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.*