Altered Neurofluid Dynamics Markers in Middle‐Aged and Older Women With Insomnia: A Multiparametric MR Neuroimaging Study

ABSTRACT

Background

Despite known brain alterations in insomnia—particularly prevalent in older females—how it affects sleep‐dependent glymphatic clearance remains understudied due to in vivo human measurement challenges.

Purpose

To investigate altered neurofluid dynamics in women with insomnia using indirect neuroimaging markers.

Study Type

Prospective.

Population

46 healthy controls (HC; 56.3 ± 5.5 years) and 28 females with insomnia disorder (ID; 58.2 ± 4.9 years).

Field Strength/Sequence

3.0 T, 3D T1‐weighted magnetization‐prepared rapid acquisition gradient echo, resting‐state fMRI using gradient‐echo echo‐planar imaging, and multi‐shell diffusion MRI using spin‐echo‐planar sequence.

Assessment

(1) Blood oxygen level dependent cerebral spinal fluid (BOLD‐CSF) coupling measuring temporal coordination between neurovascular and CSF oscillations, (2) diffusion tensor image analysis along the perivascular space (DTI‐ALPS) quantifying directional diffusivity in periventricular white matter, (3) choroid plexus (ChP) volume reflecting morphology of primary CSF‐producing structures, and (4) nucleus basalis of Meynert (NBM) volume evaluating cholinergic system integrity potentially relevant to vascular regulation; (5) all participants completed self‐reported sleep measures, including the Pittsburgh Sleep Quality Index (PSQI), Insomnia Severity Index (ISI), and Fatigue Severity Scale (FSS), and also underwent cognitive function testing.

Statistical Tests

Analysis of covariance evaluated between‐group differences controlling for demographic and clinical covariates. Relationships with cognitive and sleep scores were assessed using partial correlations, stratified by group only when significant interaction effects were detected. Multiple comparisons were false discovery rate corrected (p < 0.05). Classification model performance was evaluated using the area under the receiver operating characteristic curve (AUC). Model comparisons were performed using DeLong's tests (ΔAUC) and stepwise likelihood ratio tests (LRT) to assess classification gain and the independent incremental contribution of each biomarker; all tests were two‐sided with α = 0.05.

Results

Compared to controls, insomnia patients showed significantly reduced BOLD‐CSF coupling (−0.18 ± 0.20 vs. −0.32 ± 0.17), indicating altered temporal coordination between neurovascular and CSF dynamics. ChP volume was significantly enlarged in the insomnia group (1.72% ± 0.42% vs. 1.55% ± 0.39% of total intracranial volume), potentially reflecting compensatory CSF production upregulation, inflammatory changes, or vascular remodeling. NBM volume was significantly reduced in insomnia patients (201.75 ± 17.66 vs. 217.47 ± 21.04 mm³), suggesting cholinergic system alterations. In contrast, the DTI‐ALPS index did not differ between groups (p = 0.85). BOLD‐CSF coupling positively correlated with PSQI (r = 0.34), ISI (r = 0.41), and FSS (r = 0.40); ChP volume positively correlated with ISI (r = 0.32) and FSS (r = 0.35) (all FDR‐corrected). A dataset consists of 74 participants (46 HC and 28 ID) were included, the four‐marker classification model achieved moderate performance (AUC = 0.785, accuracy = 71.9%).

Data Conclusion

Multiple indirect neuroimaging markers potentially related to neurofluid dynamics were altered in middle‐aged and older women with chronic insomnia, except for DTI‐ALPS. These findings include altered neurovascular‐CSF coordination, ChP enlargement, and cholinergic system volume reduction.

Evidence Level

2.

Technical Efficacy

Stage 2.

Plain Language Summary

Insomnia is common in middle‐aged and older women, but its effect on the brain's waste‐clearance system, known as glymphatic system, remains unclear. Using several magnetic resonance imaging (MRI) to compare 28 women with insomnia and 46 women who slept well. Three measures differed between the groups. Insomnia group showed weaker coordination between brain activity and cerebrospinal fluid flow, a larger choroid plexus (the tissue that makes this fluid), and a smaller nucleus basalis of Meynert (a region that helps control brain blood flow). A fourth, diffusion‐based measure did not differ. These results link insomnia to measurable changes in how the brain manages its fluids.

1. Introduction

Adequate sleep is fundamental for human health and physiological homeostasis [1]. Insomnia disorder (ID), a major public health challenge affecting 17%–28% of the population [2], is characterized by persistent sleep difficulties causing significant distress or functional impairment [3]. Beyond impaired daily functioning, ID increases the risk of psychiatric, cardiovascular, and metabolic diseases [4, 5]. Prevalence patterns show striking sex differences, with ID occurring 1.5–2 times more frequently in females and peaking in middle‐aged and older women [2, 6]. This disparity likely reflects interacting factors such as hormonal transitions and sex differences in sleep architecture; nevertheless, this high‐risk demographic remains relatively understudied.

With recent meta‐analytic evidence delineating its synaptic and metabolic molecular substrates [7], neuroimaging studies have established ID as a brain disorder involving functional and structural alterations [4, 8], they have predominantly focused on neurons and glia [9]. Consequently, potential alterations in specialized brain clearance systems for fluid circulation and waste removal remain largely unexplored, despite their potential relevance to sleep‐related brain dysfunction.

The glymphatic system is a complex, heterogeneous network of perivascular pathways facilitating cerebrospinal fluid (CSF) circulation and brain waste clearance [10, 11]. Its morphology and function vary markedly across vessel types, brain regions, and individuals [11, 12]. This system is dynamically regulated by factors including cerebrovascular pulsatility, aquaporin‐4 (AQP4) water channel polarization on astrocytic endfeet, and vascular tone [13]. Specifically, CSF enters the brain parenchyma via periarterial spaces, exchanges with interstitial fluid through AQP4 channels, and exits through perivenous pathways to remove metabolic waste products such as amyloid‐β and tau proteins [11, 13].

Animal studies have demonstrated robust sleep–wake regulation of glymphatic function, with clearance efficiency peaking during slow‐wave sleep [13, 14]. Conversely, sleep deprivation impairs clearance and leads to the accumulation of waste products. However, direct noninvasive measurement of human glymphatic function is methodologically challenging [12]. Available neuroimaging methods provide only indirect inferences about neurofluid dynamics, with uncertain specificity for glymphatic processes versus other physiological phenomena [12, 15].

To infer potential alterations in neurofluid dynamics, studies have employed multiple complementary indirect markers. The analysis along the perivascular space (DTI‐ALPS) is a widely adopted diffusion‐based marker in periventricular white matter [12, 16], yet its specificity is frequently questioned as it may reflect white matter microstructure, myelin content, or vascular pathology rather than perivascular fluid dynamics per se [16]. Similarly, blood oxygen level dependent BOLD‐CSF coupling measures temporal synchronization between hemodynamic fluctuations and CSF oscillations [17, 18], potentially reflecting coordination driven by sleep slow waves [14], although reduced coupling might also indicate primary vascular pathology or measurement artifacts [12]. Structural assessments of choroid plexus (ChP) volume reflect the morphology of primary CSF‐producing structures, where enlargement has been linked to poor sleep quality and may indicate neuroinflammation or compensatory upregulation [19, 20]. Furthermore, the nucleus basalis of Meynert (NBM) volume provides insights into a cholinergic nucleus regulating both sleep–wake states and cerebrovascular tone, with a recent study suggesting that its activity modulates CSF flux [18] despite its inclusion remaining exploratory [21]. Ultimately, these markers serve as indirect proxies for complex physiological processes rather than direct measures of clearance [12], and observed alterations in insomnia may reflect primary clearance dysfunction, compensatory adaptations, or unrelated processes such as neuroinflammation or vascular pathology [22].

Preliminary studies have suggested associations between sleep disorders and neurofluid dynamics alterations. Obstructive sleep apnea patients exhibit enlarged perivascular spaces, altered ALPS indices, and decreased BOLD‐CSF coupling [23, 24, 25], while narcolepsy involves ALPS and ChP volume changes [23]. Furthermore, acute sleep deprivation increases BOLD‐CSF coupling, whereas chronic insomnia shows reduced coupling [26]. However, these studies are limited by cross‐sectional designs, heterogeneous populations, and reliance on indirect markers without direct validation.

Specifically for insomnia, a study has documented enlarged perivascular spaces [27] and reduced ALPS indices [28], though often based on single indirect markers or visual scales. While recent contrast‐enhanced MRI has suggested delayed glymphatic clearance [29], this approach has low signal‐to‐noise ratios and limited temporal resolution [12, 29]. Studies comprehensively assessing multiple complementary markers in the same individuals or focused on middle‐aged and older women are lacking despite their high insomnia burden.

The absence of a reference standard for glymphatic assessment [12] and limited understanding of relationships between different indirect markers underscore the value of comprehensive multiparametric evaluation. By assessing multiple markers spanning structural morphology (ChP volume), diffusion properties (ALPS), and functional coordination (BOLD‐CSF coupling, NBM volume) within women aged 45–70, this study aimed to:

1. Characterize whether multiple indirect neuroimaging markers potentially related to neurofluid dynamics differ between middle‐aged and older women with chronic insomnia versus age‐matched healthy controls.

2. Examine relationships between these markers and clinical measures of sleep quality and cognitive function.

3. Evaluate inter‐relationships among markers to assess whether they provide convergent or complementary information; and

4. Assess the predictive utility of these markers for distinguishing insomnia patients from controls.

The hypothesis was that ID samples would exhibit convergent evidence of altered neurofluid dynamics, specifically characterized by ChP enlargement and reductions in the DTI‐ALPS, BOLD‐CSF coupling, and NBM volume.

2. Methods

2.1. Participants

From September 2023 to June 2024, female patients with ID, aged 45–70 years, were recruited from a tertiary hospital, alongside age‐matched healthy controls (HC) from the local community. Inclusion criteria for the ID group comprised an insomnia severity index (ISI) ≥ 8 (indicating clinically significant insomnia) and sufficient cognitive capacity to complete all assessments. Exclusion criteria were: (i) secondary insomnia etiology (malignancy, autoimmune disorders, etc.); (ii) medication‐induced insomnia; (iii) chronic pain conditions; (iv) severe sleep apnea syndrome; (v) restless legs syndrome; (vi) personal or familial parasomnia history; (vii) cognitive impairment or psychiatric comorbidities; (viii) substance use disorders; (ix) suicidal ideation; (x) hemodynamic instability; (xi) habitual hypnotic use; and (xii) contraindications to MRI.

2.2. Experimental Procedure

All eligible participants underwent a comprehensive assessment protocol comprising neuroimaging and behavioral evaluations. The MRI session lasted approximately 35 min, incorporating structural, functional, and diffusion imaging sequences. Subsequently, participants completed a battery of behavioral assessments (approximately 120 min), categorized into three domains:

1. Psychological evaluation: self‐rating anxiety scale (SAS) and self‐rating depression scale (SDS) were administered for preliminary psychological screening.

2. Sleep assessment: Subjective sleep quality was evaluated using the Pittsburgh Sleep Quality Index (PSQI) [30], insomnia severity via the ISI [31], and fatigue levels through the fatigue severity scale (FSS) [32].

3. Cognitive assessment: Global cognitive function was screened using the mini‐mental state examination (MMSE). Additionally, trained examiners administered the auditory verbal learning test (AVLT) and the Stroop color‐word test. SQ, with 20 years of experience in cognitive assessment, trained KW (8 years of experience) and HD (6 years of experience) to ensure standardized data collection.

   The AVLT yielded measures of immediate recall (AVLT‐I) and delayed recall (AVLT‐R). The Stroop paradigm consisted of three subtests: word reading (Str‐A) measuring verbal processing speed, color naming (Str‐B) assessing color processing speed, and color‐word interference (Str‐C) evaluating inhibitory control.

2.3. Image Acquisition

MR images were acquired using a 3.0 T MR scanner (MAGNETOM Prisma, Siemens, Erlangen, Germany) equipped with a 64‐channel head coil. Conventional T2‐weighted turbo spin‐echo and T2‐fluid‐attenuated inversion recovery turbo spin‐echo sequences were obtained to exclude organic brain lesions. For resting‐state functional MRI (rs‐fMRI) acquisition, an echo‐planar imaging (EPI) sequence was employed with the following parameters: field of view (FOV) = 244 mm × 244 mm, echo time (TE) = 30 ms, repetition time (TR) = 500 ms, slice thickness = 3.5 mm, voxel dimensions = 3.5 × 3.5 × 3.5 mm³, 35 slices, flip angle = 60°, with a total of 960 volumes acquired. The parameters for the 3D T1‐weighted magnetization‐prepared rapid acquisition gradient echo sequence were as follows: inversion time (TI) = 1100 ms, TR = 2530 ms, TE = 2.98 ms, FOV = 256 mm × 256 mm, slice thickness = 1 mm, voxel dimensions = 1 × 1 × 1 mm³, flip angle = 7°, comprising 192 sagittal slices. To achieve higher angular resolution, a MAGNETOM Prisma 3 T MRI scanner (Siemens, Erlangen, Germany) equipped with a 64‐channel head coil was employed to acquire brain diffusion‐weighted MRI (dMRI) datasets using spin‐echo echo‐planar imaging sequence using simultaneous multi‐slice (multiband factor 2) and in‐plane parallel imaging (GRAPPA factor 2). A multi‐shell q‐space sampling scheme was used, comprising 99 diffusion‐encoding directions distributed over 12 b‐values (0, 300, 350, 650, 950, 1000, 1350, 1650, 1700, 2000, 2700, 3000 s/mm²). Additional parameters were phase‐encoding direction = anterior–posterior (AP), acquisition, TR/TE = 4200/72 ms, FOV = 220 mm × 220 mm, slice thickness = 2 mm, voxel size = 2 × 2 × 2 mm³, NEX = 1, acquisition time = 7 min 31 s. A pair of b0 images with reverse phase encoding (PA) was also acquired to correct for EPI distortion.

2.4. Markers Derivation

Four neuroimaging markers were selected as indirect proxies for neurofluid dynamics relevant to brain clearance. The spatial distributions of four markers are summarized in Figure 1B and Figure 2D, and they are conceptually abstracted in Figure 2A.

FIGURE 1.

(A) Processing pipeline and basis of The diffusion tensor imaging analysis along the perivascular space (DTI‐ALPS) using Fractional anisotropy (FA) to quantify diffusivity of projection fibers (blue), association fibers (green), and commissural fibers (red) representing different white matter tract orientations. (B) Anatomical localization of multimodal glymphatic markers. Spatial distribution of all glymphatic system measurements: ventricular system (light blue), DTI‐ALPS regions of interest (dark blue), ChP in lateral ventricles (green), and fourth ventricle slice location for CSF signal extraction (red). This comprehensive anatomical framework demonstrates the integrated approach for assessing different components of the Glymphatic system. (C) ChP volume quantification. Automated segmentation workflow using a 3D U‐NET deep learning model to extract ChP volume from T1‐weighted images registered to MNI standard space. (D) BOLD‐CSF coupling analysis. Resting‐state functional MRI approach measuring temporal synchronization between whole‐brain BOLD signals and cerebrospinal fluid signals in insomnia patients. CSF = cerebrospinal fluid; BOLD = blood oxygen level‐dependent; DTI‐ALPS = diffusion tensor image analysis along the perivascular space; ChP = choroid plexus; NBM = nucleus basalis of Meynert.

FIGURE 2.

(A) Conceptual illustrations of four glymphatic system dysfunction indicators in insomnia: (i) CSF‐Brain desynchronization with disrupted coordination between CSF (red) and brain activity(yellow); (ii) Altered CSF‐perivascular spaces; (iii) Choroid plexus enlargement(green); and (iv) Cholinergic atrophy affecting glymphatic regulatory pathways(pink). (B) Group comparisons of glymphatic indicators between ID and HC. BOLD‐CSF coupling, ALPS, ChP ratio, and NBM volume. (C) BOLD‐CSF coupling temporal dynamics in HC (blue) and ID (red). Left: Global coupling showing reduced correlation in insomnia. Right: Rate of change (−d/dt) of coupling. Vertical lines mark lag = 2.5 s; shaded areas represent standard error. (D) Location of NBM regions in the glymphatic model (top) and corresponding mask overlay on standard T1 MRI (bottom). Between groups comparisons were performed by ANCOVA, adjusting age, education, BMI, SAS, SDS, HBP, and TIV. ANCOVA = analysis of covariance; ID = insomnia disorder; HC = healthy controls; PSQI = Pittsburgh Sleep Quality Index; ISI = Insomnia Severity Index; FSS = Fatigue Severity Scale; BMI = body mass index; SAS = self‐rating anxiety scale; SDS = self‐rating depression scale; TIV = total intracranial volume; CSF = cerebrospinal fluid; BOLD = blood oxygen level‐dependent; DTI‐ALPS = diffusion tensor image analysis along the perivascular space; ChP = choroid plexus; NBM = nucleus basalis of Meynert.

2.4.1. DTI‐ALPS

The DTI‐ALPS was computed using the standard, openly available DTI‐ALPS pipeline (Figure 1A), which has been cross‐vendor test–retest validated in healthy adults [33], whose preprocessing was conducted using MRtrix3 [34] and FSL [35], included denoising, Gibbs‐ringing removal, susceptibility distortion correction (topup) and eddy‐current and motion correction.

A single diffusion tensor fitting was applied to the multi‐shell data using FSL's “dtifit” to extract the directional diffusivity components Dxx, Dyy, Dzz required for ALPS calculation. The resulting parametric maps were registered to the JHU‐ICBM‐FA‐1 mm template using linear (flirt) and nonlinear (fnirt) registration. Four regions of interest (ROI) were defined in template space using the JHU‐ICBM white matter atlas: bilateral superior corona radiata (coordinate left: 53, 47, 40; right: 25, 47, 40) representing projection fibers, and bilateral superior longitudinal fasciculus (coordinate left: 59, 47, 40; right: 19, 47, 40) representing association fibers. Directional diffusivity values were extracted from these ROIs to calculate the ALPS index as the bilateral mean ratio:

$$\text{ALPS}=\text{mean}\left(\text{Dxxproj}+\text{Dxxassoc}\right)/\text{mean}\left(\text{Dyyproj}+\text{Dzzassoc}\right),$$

where Dxxproj represents medial‐lateral diffusivity in projection fibers, Dxxassoc represents medial‐lateral diffusivity in association fibers, Dyyproj represents anterior–posterior diffusivity in projection fibers, and Dzzassoc represents superior–inferior diffusivity in association fibers.

To be consistent with prior a protocol [36] and because the single‐tensor model underlying DTI‐ALPS, and most prior DTI‐ALPS studies, assume single‐shell low‐b‐value data; we deliberately verified that our multi‐shell acquisition did not bias the index that the DTI‐ALPS was recomputed on a single‐shell subset (b ≤ 1000 s/mm²) through the identical pipeline, and the two estimates were highly consistent (Figure S2). We therefore refer to our measure as the DTI‐ALPS throughout. The estimation was fully automatic except for quality control. Registration quality was manually checked by three experienced radiologists (SQ has 25 years of experience; KW has 8 years of experience; QG has 6 years of experience) independently. Inter‐rater agreement on registration acceptability was 100%. 7 (6 HC, 1 ID) participants with evident misregistration were excluded from analysis.

2.4.2. BOLD‐CSF Coupling

Adhering to the protocol detailed in previous studies [37, 38] (Figure 1D), both CSF and BOLD signals were derived from rs‐fMRI data preprocessed using DPABI software [39]. Author RW conducted the automated preprocessing pipeline (4 years of experience). The CSF extraction code was written by author CL (7 years). Two postdocs, XZ (8 years) and LH (9 years), reviewed the code and preprocessing pipeline. The initial 10 volumes were discarded for magnetization equilibration. Preprocessing included slice timing correction, rigid‐body motion correction (six‐parameter transformation), linear detrending, and bandpass filtering (0.01–0.1 Hz) to isolate low‐frequency fluctuations. For CSF signal extraction [14], individual T1‐weighted images were registered to MNI152 space using SPM12 for automated tissue segmentation. CSF probability maps were thresholded to create binary masks. To minimize partial volume effects from adjacent gray matter, CSF signals were extracted from the fourth ventricle, identified as the first non‐zero voxels in the CSF mask progressing from inferior to superior along the z‐axis. The visual inspection of each mask was performed by two radiologists, KW (8 years of experience) and HD (6 years of experience). In cases of uncertainty, SQ (25 years of experience) was consulted. Participants with unsuccessful mask generation were excluded.

For BOLD signal extraction [14], preprocessing included nuisance regression (white matter and six motion parameters) to reduce non‐neuronal physiological noise. Following regression and detrending, functional images were normalized to MNI space using the DARTEL algorithm for improved anatomical alignment. Images were smoothed using a 6 mm FWHM Gaussian kernel to increase signal‐to‐noise ratio. Global gray matter BOLD signals were extracted by averaging across all gray matter voxels.

BOLD‐CSF coupling was quantified as the cross‐correlation coefficient between mean global gray matter BOLD and fourth ventricle CSF time series. Negative values (typical in resting‐state data) indicate inverse phase relationships where BOLD increases correspond to CSF decreases, potentially reflecting coordinated vascular‐fluid dynamics. Values closer to zero indicate reduced temporal synchronization.

2.4.3. ChP Volume

T1‐weighted structural images underwent nonlinear registration to ICBM‐MNI 152 space using Advanced Normalization Tools (ANTs) [40], minimizing inter‐individual variability in lateral ventricular morphology and enhancing consistency of ChP spatial localization (Figure 1C). ChP segmentation employed an automated deep learning approach [41] using a 3D U‐Net convolutional neural network trained specifically for ChP identification.

The deep learning model was trained using a patch‐based approach. From MNI‐normalized T1 images, 64 × 64 × 64 voxel cubic patches centered on ChP regions identified in probabilistic atlases were extracted. This strategy focuses computational resources on relevant regions while providing sufficient spatial context. Data augmentation (random rotations, flips, intensity perturbations) expanded the training dataset and improved generalization.

Following automated segmentation, ChP masks were inverse transformed to each participant's native T1 space to ensure volumetric measurements reflect actual tissue volumes. Absolute ChP volumes (mm³) were calculated by summing all voxels within binary masks and multiplying by voxel dimensions. To account for individual differences in brain and head size, absolute volumes were normalized to total intracranial volume (TIV), calculated using SPM12's [42] unified segmentation algorithm. TIV was computed as the sum of gray matter, white matter, and CSF volumes. Relative ChP volume was calculated as: (absolute ChP volume/TIV) × 100. Both total bilateral and individual hemispheric ChP volumes were calculated to assess potential asymmetries. Author RW performed the automatic ChP volume extraction pipeline. The extraction output was visually inspected by experienced radiologist KW (with 8 years' experience) and HD (with 6 years' experience) under radiologist SQ (with 25 years' experience) supervision.

2.4.4. Nucleus Basalis of Meynert Volume

NBM volume was quantified using the SPM Anatomy Toolbox [42], which implements cytoarchitectonically defined probabilistic maps of magnocellular basal forebrain cell groups based on postmortem stereotaxic analysis [43]. These maps were created through histological examination with cellular boundaries identified using immunohistochemical markers. Volumetric data was extracted for the Ch4 cell group (NBM), which is the largest component and provides primary cholinergic innervation to neocortex, while Ch1‐3 project predominantly to hippocampal structures. For each participant's normalized T1 image in MNI space, the Ch4 probabilistic mask was applied and total volume of voxels exceeding a probability threshold (typically p > 0.25) was calculated. Consistent with a previous study [18], absolute volumetric measurements (mm³) are reported rather than normalized values. This reflects two considerations: the relatively small NBM volume makes normalized values susceptible to greater measurement error, and absolute volumes may better reflect actual cholinergic neuron populations, as cell numbers are not expected to scale linearly with overall brain size. The extraction pipeline of NBM volume was conducted by author RW. The visual inspection of the outputs was performed by two experienced radiologists (KW with 8 years' experience, HD with 6 years' experience) and guided by SQ (with 25 years' experience).

2.5. Statistical Analyses

Statistical analyses were performed using MATLAB software. For inter‐marker associations between BOLD‐CSF and the other markers (DTI‐ALPS, ChP volume, NBM volume), partial correlation analyses were conducted controlling age, education, SAS, SDS, high blood pressure (HBP), and body mass index (BMI) (p < 0.05). Based on evidence that basal forebrain cholinergic neurons regulate cerebrovascular dynamics and CSF flux [18], the relationship between NBM volume and BOLD‐CSF coupling was compared between insomnia patients and healthy controls. Multiple linear regression was employed to test for group × NBM volume interaction effects on coupling (p < 0.05). Following significant interactions, group‐stratified partial correlations controlling for age, education, blood pressure, and BMI were calculated. Additionally, Ch1‐3 cholinergic cell groups [43]. For visualization of the interaction effect, NBM volume was categorized into three equally sized tertiles (low, medium, and high). Additionally, Ch1‐3 cholinergic cell groups [44], which project primarily to hippocampal regions, were examined to assess whether coupling relationships are specific to cortically‐projecting (NBM/Ch4) versus hippocampally‐projecting (Ch1‐3) cholinergic systems.

Between‐group differences in neuroimaging markers were assessed using ANCOVA, with age, education level, SAS/SDS scores, BMI, HBP, and TIV as covariates. To investigate relationships between neuroimaging markers (BOLD‐CSF coupling, ALPS, ChP ratio, and NBM volume) and cognition, an interaction‐effect‐driven stratified analysis approach was employed, recognizing that insomnia might alter fundamental brain‐cognition relationships. When significant group‐by‐cognition interactions were detected (p < 0.05), separate within‐group partial correlation analyses were conducted; otherwise, data were pooled. Similarly, for examining relationships with sleep measures (ISI, PSQI, FSS), 12 interaction effect analyses were performed between neuroimaging markers and sleep scores. For non‐significant interactions, partial correlations controlling age, education, SAS, SDS, blood pressure, and BMI were conducted. For Visualization of the group‐by‐NBM interaction, NBM volume was stratified into tertiles (low, medium, high; 33rd and 67th percentiles). Analyses involving more than five comparisons were FDR‐corrected (p < 0.05).

Binary classification of ID and HC groups was performed using binomial logistic regression models with MATLAB's glmfit (binomial distribution, logit link function). Data preprocessing included standardizing heterogeneous data types, handling missing values, filtering effective samples, and Z‐score normalization to eliminate dimensional differences and improve model stability. Four core neuroimaging markers: BOLD‐CSF coupling, ALPS, ChP ratio, and NBM volume were utilized. An exhaustive feature combination strategy generated all possible feature subsets from univariable to multivariable (2 ⁿ  − 1 = 15 model combinations), systematically exploring both independent and synergistic predictive capabilities of different markers. Models were evaluated using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity metrics, with optimal features ultimately identified based on AUC values. The DeLong test was used to compare AUCs between the best combined model and the demographic baseline model, with ΔAUC reported to demonstrate the classification gain of the combined model. Given the nested structure among candidate models [44], stepwise Likelihood Ratio Tests (LRT) were applied to quantify the independent incremental contribution of each biomarker to model fit. All tests were two‐sided with α = 0.05.

3. Results

3.1. Demographic and Clinical Characteristics

The demographic and clinical characteristics of participants are presented in Table 1. The study included 46 HCs and 28 IDs, all females. No significant differences were found between groups in age (p = 0.139), education (p = 0.294), HBP (p = 0.241), LBP (p = 0.968), BMI (p = 0.052), or MMSE scores (p = 0.135). As expected, the ID group demonstrated significantly higher scores on all sleep and mood‐related measures, including SAS, SDS, ISI, PSQI, and FSS.

TABLE 1.

Demographic and clinical characteristics.

Characteristics	HC group (n = 46)	ID group (n = 28)	p	Remark
Gender, female (%)	46 (100.0)	28 (100.0)	1.000	—
Education, years	10.85 ± 3.35	10.85 ± 12	0.294	—
Age, years	56.3 ± 5.5	58.2 ± 4.9	0.139	—
Duration, months	—	52.7 ± 9.36	—	—
HBP, mmHg	128.1 ± 18.3	123.1 ± 15.9	0.241	—
LBP, mmHg	80.8 ± 12.3	80.9 ± 9.3	0.968	—
BMI, kg/m2	24.0 ± 3.1	22.5 ± 2.9	0.052	—
MMSE score	28.0 ± 2.5	28.8 ± 1.5	0.135	—
SAS score	26.0 ± 3.55	32.6 ± 5.98	0.000***	ID > HC
SDS score	27.4 ± 5.65	37.2 ± 8.19	0.000***	ID > HC
ISI score	4.87 ± 2.08	17.7 ± 2.5	0.000***	ID > HC
PSQI score	4.0 ± 1.9	14.1 ± 3.4	0.000***	ID > HC
FSS score	19.98 ± 7.18	44.7 ± 9.9	0.000***	ID > HC

Note: Data are presented as mean ± standard deviation for continuous variables and as frequency (percentage) for categorical variables. p values were calculated using independent samples t‐test for continuous variables and Fisher exact test for categorical variables. *p < 0.05, **p < 0.01, ***p < 0.001.

Abbreviations: BMI = body mass index; FSS = Fatigue Severity Scale; HBP = high blood pressure; HC = healthy controls; ID = insomnia disorder; ISI = Insomnia Severity Index; LBP = low blood pressure; MMSE = mini‐mental state examination; PSQI = Pittsburgh Sleep Quality Index; SAS = self‐rating anxiety scale; SDS = self‐rating depression scale.

3.2. Relationships Among Neuroimaging Markers

Correlation matrices across all markers are presented in Figure 3A and Table S5. Although ChP Ratio showed nominal correlations with NBM (partial r = −0.308, p = 0.013) and Ch1‐3 (partial r = −0.254, p = 0.043), neither survived FDR correction (FDR p = 0.120 and 0.194, respectively).

FIGURE 3.

(A) Heatmap showing cross‐group partial correlations between glymphatic system indicators. (B) Interaction effect between NBM volume and diagnostic group on BOLD‐CSF coupling. (C) Forest plot of partial correlations between glymphatic system indicators and sleep quality measures. (D) Scatter plots showing significant associations between glymphatic system indicators and sleep measures after FDR correction. All partial correlations were performed using Pearson's correlation, adjusted for age, education, BMI, SAS, SDS, HBP, TIV. ID = insomnia disorder; HC = healthy controls. †p < 0.05, ††p < 0.01, †††p < 0.001 (uncorrected); *p < 0.05, **p < 0.01, ***p < 0.001 (FDR‐corrected). CSF = cerebrospinal fluid; BOLD = blood oxygen level‐dependent; DTI‐ALPS = diffusion tensor image analysis along the perivascular space; ChP = choroid plexus; NBM = nucleus basalis of Meynert; PSQI = Pittsburgh Sleep Quality Index; ISI = Insomnia Severity Index; FSS = Fatigue Severity Scale.

3.3. Alterations in Neuroimaging Markers in Female ID Patients

Table 2 and Figure 2B report neuroimaging measurements for both groups. DTI‐ALPS showed no significant differences between groups (p = 0.85, Cohen's d = 0.07). Subsequent comparisons of projection and association fiber variables also revealed no significant differences between groups (Table S3). Total ChP/TIV was significantly larger in ID compared to HC (1.72% ± 0.42% vs. 1.55% ± 0.39%), primarily due to right ChP enlargement (0.92% ± 0.24% vs. 0.82% ± 0.21%). For BOLD‐CSF coupling (Figure 2C), the average maximum negative correlation across all subjects occurred when CSF lagged BOLD by five time points (2.5 s). Analysis of correlation coefficients at this 2.5 s lag showed that BOLD‐CSF coupling was significantly less negative in ID (−0.18 ± 0.20 vs. −0.32 ± 0.17), indicating altered neurovascular‐CSF coordination. Additionally, NBM volume was significantly reduced in ID compared to HC (201.75 ± 17.66 mm³ vs. 217.47 ± 21.04 mm³). ANCOVA statistics for all covariates across models are reported in Table S6.

TABLE 2.

Glymphatic system measurements.

Measurements	HC group (n = 46)	ID group (n = 28)	p	Cohen's d	Remark
DTI‐ALPS	1.77 ± 0.21 (n = 40)	1.70 ± 0.19	0.85	0.07	—
Total ChP volume/TIV, %	1.55 ± 0.39 (n = 46)	1.72 ± 0.42 (n = 28)	0.04*	0.71	ID > HC
BOLD‐CSF coupling	−0.32 ± 0.17 (n = 46)	−0.18 ± 0.20 (n = 28)	< 0.001***	1.26	ID > HC
NBM volume, mm3	217.47 ± 21.04 (n = 46)	201.75 ± 17.66 (n = 28)	0.03*	−0.74	ID < HC

Note: Data are presented as mean ± standard deviation. ANCOVA controlling for age, education, BMI, SAS, SDS, HBP, and TIV. Cohen's d denotes the covariate‐adjusted effect size estimated from the ANCOVA model. *p < 0.05, **p < 0.01, ***p < 0.001.

Abbreviations: BOLD‐CSF = blood oxygen level‐dependent cerebrospinal fluid; ChP = choroid plexus; DTI‐ALPS = diffusion tensor image analysis along the perivascular space; HC = healthy controls; ID = insomnia disorder; NBM = nucleus basalis of Meynert; TIV = total intracranial volume.

3.4. Relationship Between Neuroimaging Markers and Cognitive Function

None of the correlations between the four neuroimaging markers (BOLD‐CSF coupling, ALPS, ChP ratio, NBM volume) and six cognitive tests survived FDR correction (Table S1).

However, based on uncorrected p‐values, three significant associations emerged (Figure S1): DTI‐ALPS negatively correlated with the reaction times on the Stroop A task (congruent condition) in all subjects (r = −0.32). ChP ratio showed a positive tendency with the reaction times on the Stroop B task (incongruent condition) in the ID group (r = 0.54) but not in HC (r = −0.03, p = 0.88). Additionally, NBM volume correlated positively with MMSE scores in HC (r = 0.37) but not in ID (r = −0.13, p = 0.57).

3.5. Relationship Between Neuroimaging Markers and Sleep Measures

Partial correlation analyses between neuroimaging markers and sleep measures showed consistent patterns as illustrated in Figure 3C and Table S2. BOLD‐CSF coupling and ChP ratio predominantly showed positive correlations with poor sleep quality, increased insomnia symptoms, and higher fatigue levels, while NBM volume demonstrated negative correlations. DTI‐ALPS showed no significant associations with any sleep measures (PSQI: p = 0.98; FSS: p = 0.93; ISI: p = 0.65). In detail (Figure 3D), BOLD‐CSF coupling exhibited significant (after FDR correction) positive correlations with PSQI (r = 0.34), FSS (r = 0.40), and ISI (r = 0.41). ChP ratio demonstrated significant (after FDR correction) positive correlations with FSS (r = 0.35) and ISI (r = 0.32). NBM volume showed marginally significant negative correlations with FSS and ISI (both r = −0.25, p = 0.05), though these did not survive FDR correction.

3.6. Relationship Between BOLD‐CSF Coupling and NBM

A significant group‐by‐NBM interaction effect was found (Figure 3B). However, this interaction effect did not reach significance for the Ch 1–3 subregion (p = 0.059), nor was it a correlation with BOLD‐CSF coupling observed (p = 0.532).

3.7. Performance of Insomnia Classification Models

Models' performance results are shown in Table S4 and Figure 4A,B. The highest classification performance was achieved by the four‐marker model combining BOLD‐CSF coupling, ALPS, NBM volume, and ChP ratio (AUC = 0.785, accuracy = 0.719, sensitivity = 0.615, specificity = 0.789). Compared to the demographic baseline model (AUC = 0.634), this combined model yielded a ΔAUC of +0.151 (DeLong test: Z = 2.174*).

FIGURE 4.

Glymphatic indicators performance (A) Receiver operating characteristic curves displaying the top four models for ID classification. (B) Feature versus Performance plot showing the relationship between the number of glymphatic biomarkers included in the model (x‐axis) and classification performance measured by area under the curve (AUC) (y‐axis). CSF = cerebrospinal fluid; BOLD = blood oxygen level‐dependent; DTI‐ALPS = diffusion tensor image analysis along the perivascular space; ChP = choroid plexus; NBM = nucleus basalis of Meynert.

Stepwise LRT showed that adding NBM volume to the demographic baseline significantly improved model fit (χ ² = 9.003**), and the further addition of BOLD‐CSF coupling yielded an additional significant improvement (χ ² = 4.763*), resulting in a two‐marker model with AUC = 0.769 and accuracy = 0.672. The subsequent addition of ALPS did not reach significance (χ ² = 0.945, p = 0.331), nor did the further inclusion of ChP ratio (χ ² = 0.081, p = 0.775).

4. Discussion

This study examined multiple indirect neuroimaging markers potentially related to neurofluid dynamics in middle‐aged and older women with chronic insomnia, using a multiparametric approach assessing complementary structural, diffusion, and functional markers within the same individuals. Three of four markers showed significant group differences: BOLD‐CSF coupling was reduced in insomnia patients, suggesting altered temporal coordination between neurovascular and cerebrospinal fluid oscillations; ChP volume was enlarged, potentially reflecting inflammatory changes, vascular remodeling, or compensatory CSF production upregulation; and NBM volume was reduced, indicating cholinergic system alterations. The DTI‐ALPS, calculated from multi‐shell diffusion data, showed no group differences. These markers correlated with sleep quality measures and showed differential inter‐marker relationships across groups, with a combined classification model achieving moderate discriminative performance.

4.1. BOLD‐CSF Coupling: Desynchronization, Compensation, or Vascular Dysfunction?

The physiological meaning of significantly decreased temporal correlation between BOLD and CSF signals in insomnia patients is uncertain and multiple interpretations are plausible. One potential interpretation is impaired clearance coordination. Reduced coupling might reflect disrupted coordination between neural activity, hemodynamic responses, and CSF flow that normally facilitates waste clearance during sleep [14]. If slow‐wave sleep coordinates these processes [14], chronic sleep disruption might decouple them. However, an alternative interpretation involves compensatory mechanisms. Altered coupling could represent an adaptive response attempting to maintain clearance despite sleep disruption. Acute sleep deprivation studies show increased coupling [14, 22], potentially reflecting compensatory enhancement. Chronic insomnia might involve impaired compensation manifesting as reduced coupling. A third possibility is vascular dysfunction. Changes in cerebrovascular reactivity, vascular compliance, or blood flow regulation—common in insomnia and associated with chronic stress—might alter hemodynamic‐CSF relationships independently of clearance function [24].

Notably, coupling reductions have been observed in severe obstructive sleep apnea patients [24] but not narcolepsy [45], suggesting specificity to sleep fragmentation rather than sleep–wake regulation disorders broadly. However, whether reduced coupling causes, results from, or simply correlates with sleep disruption cannot be determined from cross‐sectional data. The positive correlations between altered coupling and sleep symptom severity (PSQI, ISI, FSS) indicate the relationship between neurovascular‐CSF coordination and sleep quality, but causality remains to be established.

4.2. ChP Enlargement: Primary Dysfunction or Inflammatory Response?

ChP enlargement in insomnia patients aligns with findings in other sleep disorders [23] and recent associations with poor sleep quality [20]. However, ChP volume changes likely reflect complex pathophysiology distinct from clearance function per se. ChP enlargement has been reported across diverse neuropsychiatric conditions [19], suggesting non‐specific inflammatory or vascular responses rather than clearance‐specific pathology. The ChP volume changes potentially relate to: neuroinflammation (chronic sleep disruption increases inflammatory markers; ChP may enlarge in response to inflammatory signaling), or blood‐CSF barrier dysfunction structural changes in barrier integrity [19].

The positive correlations between ChP volume and sleep disturbance severity (ISI, FSS) but not PSQI may indicate associations with specific sleep disruption features rather than general sleep quality. The finding of ChP enlargement in female insomnia patients compared to HC aligns with recent work by Kikuta et al., who also observed increased ChP volume in older adults with poor sleep quality [20]. However, Becker et al. found that while ChP volume correlated with age and cognition in healthy adults, it showed no association with PSQI sleep quality [46]. This may be due to two factors: their cohort lacked clinical sleep disturbances, and they used only PSQI for assessment, whereas in the current study, ChP volume correlated significantly with ISI and FSS measures but not with PSQI. The ChP harbors the brain's most prominent circadian clock, and its desynchronization with sleep cycles may impair metabolite clearance [47]. This mechanism would explain the consistent ChP enlargement observed across various sleep disorders including narcolepsy [23]. However, reverse causation remains plausible—ChP pathology might contribute to sleep difficulties through inflammatory mediators or altered CSF composition.

4.3. DTI‐ALPS: Null Findings

Our diffusion data were acquired with a multi‐shell protocol, the DTI‐ALPS index was computed with the standard, cross‐vendor‐validated pipeline [33]. Although DTI‐ALPS is conventionally derived from single‐shell diffusion data, fitting a single tensor to data spanning a wide range of b‐values could in principle alter the directional diffusivity estimates relative to a conventional single‐shell acquisition. However, because DTI‐ALPS represents a ratio of directional diffusivities, systematic effects of the higher b‐value components on the single‐tensor fit are expected to largely cancel between numerator and denominator. We therefore directly examined whether our multi‐shell acquisition biased the index. The sensitivity analysis using conventional single‐shell subset (b ≤ 1000 s/mm²) was strongly agreed with our primary estimate and gave the same null finding (Pearson r = 0.945, p < 0.001; Figure S2), and the single‐shell subset yielded the same null between‐group finding (ANCOVA F = 0.113, p = 0.74), indicating the robustness of DTI‐ALPS estimation against various shell‐design of acquisition schemes. As DTI‐ALPS lack a reference standard and its specificity for perivascular fluid movement remains unestablished [12, 16], the null DTI‐ALPS result does not contradict the significant differences observed in other markers (BOLD‐CSF coupling, ChP volume NBM volume), but may instead reflect the distinct physiological processes that these markers represent.

Interestingly, the absence of group differences in DTI‐ALPS contrasts with a previous DTI‐ALPS study reporting reductions in insomnia [28]. This contradiction cannot exclude cohort‐specific characteristics. Specifically, the cognitive function of above ID cohort is significantly declined whereas no significant difference is observed in our sample (MMSE score: HC = 28.0 ± 2.5; ID 28.8 ± 1.5, p = 0.135). Meanwhile, our finding align with another cohort of sleep disorder (narcolepsy) that has similar cognitive condition to our sample [45]. Whether DTI‐ALPS reflects a cognitive‐related glymphatic proxy rather than a sleep‐specific marker remains to be further investigated.

4.4. Reduced NBM Volume: Cholinergic‐Vascular Links or Independent Pathology?

Reduced NBM volume in insomnia patients and its interaction with BOLD‐CSF coupling represents potentially novel findings, but interpretation requires caution regarding mechanism and causality. The NBM provides widespread cholinergic innervation to cortex and regulates both sleep–wake states and cerebrovascular tone [48]. A recent study suggested basal forebrain cholinergic neurons modulate vascular pulsatility and CSF flow in rodents [18], providing a rationale for examining NBM in neurofluid context. However, NBM volume is not a glymphatic marker but rather an indicator of cholinergic neuron integrity with potential indirect relevance to vascular regulation. The differential NBM‐coupling relationship between groups (negative in controls, positive in insomnia) suggests disrupted cholinergic‐vascular coordination. However, multiple alternative explanations exist progressive neurodegeneration (insomnia might accelerate cholinergic degeneration through chronic stress or sleep deprivation), shared vulnerability (underlying factors such as genetics or vascular risk might affect both cholinergic system and sleep regulation), or measurement confounds (NBM is a small structure; volumetric measurements are sensitive to segmentation accuracy and atrophy in adjacent regions).

The specificity of NBM (Ch4) but not Ch1‐3 interactions aligns with differential projection patterns [43] but requires replication. This study reveals significant cholinergic atrophy in the NBM in ID compared to HC. This NBM structural decrease appears inconsistent with Ma et al., who reported no volumetric differences in any basal forebrain subregions in insomnia cohorts, despite documenting functional connectivity alterations in NBM regions [21]. This discrepancy likely stems from demographic differences—the current study examined middle‐aged/older females versus their younger adult sample. These contrasting findings suggest functional connectivity changes may precede structural atrophy in the NBM.

The NBM, a critical component of the ascending reticular activating system, influences sleep–wakefulness regulation [48]. Prior insomnia studies have identified functional NBM alterations with increased basal forebrain glucose metabolism [21]. The current study showed that cholinergic dysfunction in ID extends beyond functional alterations to include structural atrophy. A recent study by Lee et al. in Parkinson's disease patients demonstrated that anterior NBM structural reduction correlated with overall PSQI scores [49], which aligns with the current findings that NBM volume reduction positively correlates with both ISI and FSS scores, further illustrating the role of the NBM region in sleep–wake regulation. The effect of cholinergic NBM atrophy extends beyond sleep; NBM has been widely reported in neurodegenerative diseases and is highly correlated with cognitive performance in Alzheimer's disease patients [18, 50].

Importantly, cross‐sectional data cannot establish whether: (a) insomnia causes NBM atrophy affecting neurofluid regulation, (b) NBM degeneration causes sleep disturbances, or (c) shared factors affect both independently. The current results showed a significant group‐by‐NBM interaction effect on BOLD‐CSF coupling, demonstrating differential relationships between NBM volume and coupling strength in insomnia disorder versus healthy controls. Notably, this interaction was specific to NBM, as Ch1‐3 regions showed neither significant group interactions nor correlations with BOLD‐CSF coupling. These findings partially align with Chuang et al.'s observations that NBM uniquely correlates with BOLD‐CSF coupling while Ch1‐3 regions do not [18]. The differential impact reflects distinct projection patterns: NBM projects to cerebral cortex, while Ch1‐3 targets hippocampus and olfactory ball [43]. This selective association supports Chuang et al.'s cholinergic‐vascular unit concept, wherein NBM serves a function in cholinergic regulation of vascular tone and potentially CSF dynamics. However, NBM abnormalities demonstrate heterogeneity across different disorders [49, 50], necessitating careful examination of insomnia‐specific mechanisms. A previous animal model have established the causal role of cholinergic neurons in regulating arterial pulsation and CSF flow [18], suggesting that altered NBM‐BOLD‐CSF coupling in insomnia may reflect disrupted cholinergic regulation. Given this novel perspective, future studies should employ multiparametric approaches including NBM BOLD time‐series and fiber tractography to gather additional evidence.

4.5. Direct Sleep Associations Versus Complex Cognitive Relationships

This study identified trend‐level associations between neuroimaging markers and cognitive performance, including a negative correlation trend between DTI‐ALPS and reaction times on the Stroop task. This pattern aligns with previous findings that insomnia patients exhibit reduced processing speed in insomnia and with studies reporting task performance associations in OSA patients [25]. However, none of these cognitive‐glymphatic relationships survived FDR correction in the current analysis.

In contrast, neuroimaging markers showed highly consistent associations with all three subjective sleep measures, with BOLD‐CSF coupling and ChP ratio exhibiting positive correlations, and NBM volume showing negative correlations with PSQI, ISI, and FSS scores. These associations were not disease‐specific but represented universal relationships present in both ID and HC groups. Notably, BOLD‐CSF coupling positively correlated with PSQI, ISI, and FSS, indicating that lower neurovascular‐CSF synchrony corresponds to poorer sleep quality, higher insomnia severity, and greater fatigue. These findings align with previous studies on insomnia [26] and Alzheimer's disease [17]. Additionally, greater ChP enlargement was associated with higher insomnia severity and increased fatigue levels. Future studies should utilize more targeted measures to better characterize ChP function in sleep regulation.

Notably, neurovascular‐CSF synchronicity and the ChP share robust circadian properties, both serving as indicators potentially coupled with sleep–wake states [14, 50]. The ChP likewise exhibits circadian oscillator functions, participating in diurnal rhythm regulation through temporally controlled cerebrospinal fluid production patterns that maintain circadian homeostasis [47]. A prior study has established correlations between ChP volume and slow‐wave sleep percentage [46], suggesting that deep sleep disruption may potentially induce compensatory and inflammatory enlargement of this structure [19].

The robust sleep‐related associations and weaker cognitive correlations observed in the current study may be explained by the more direct relationship between sleep and neurofluid regulatory mechanisms, which primarily involve sleep–wake state transitions, whereas cognition depends on interactions across multiple neural networks. Furthermore, sleep exerts more immediate effects on neurofluid dynamics [22], necessitating longitudinal assessment to detect meaningful associations.

4.6. Multiparametric Neuroimaging Markers in Insomnia Classification

Although the four‐marker model combining BOLD‐CSF coupling, ALPS, NBM volume, and ChP ratio numerically achieved the highest classification performance, stepwise test revealed that the discriminative gain over the demographic baseline was primarily driven by NBM volume and BOLD‐CSF coupling, which each made statistically significant independent contributions to model fit. This pattern highlights the central role of cholinergic neurovascular integrity in insomnia‐related glymphatic dysfunction, consistent with recent evidence that basal forebrain cholinergic neurons regulate glymphatic flux. NBM atrophy and disrupted BOLD‐CSF coupling may therefore reflect functionally linked manifestations of the same underlying cholinergic neurovascular impairment [18], jointly capturing the largest proportion of classification variance. This integrated methodology is particularly relevant for ID, which involves both brain structural abnormalities and functional dysregulation [8, 24]. The subsequent addition of ALPS and ChP ratio did not yield significant incremental improvements, possibly reflecting limited sample size (ID = 28) may have been insufficient to capture meaningful differences in ΔAUC.

Although the four‐marker model combining BOLD‐CSF coupling, ALPS, NBM volume, and ChP ratio numerically achieved the highest classification performance, stepwise test revealed that the discriminative gain over the demographic baseline was primarily driven by NBM volume and BOLD‐CSF coupling, which each made statistically significant independent contributions to model fit. This pattern highlights the central role of cholinergic neurovascular integrity in insomnia‐related glymphatic dysfunction, consistent with recent evidence that basal forebrain cholinergic neurons regulate glymphatic flux. NBM atrophy and disrupted BOLD‐CSF coupling may therefore reflect functionally linked manifestations of the same underlying cholinergic neurovascular impairment, jointly capturing the largest proportion of classification variance. This integrated methodology is particularly relevant for ID, which involves both brain structural abnormalities and functional dysregulation [8, 24]. The subsequent addition of ALPS and ChP ratio did not yield significant incremental improvements, possibly reflecting limited statistical power given the current sample size, and future studies with larger cohorts are warranted to clarify the independent contributions of these markers.

4.7. Limitations

This study has several limitations. First, the sample focused exclusively on middle‐aged and elderly females, reflecting the highest insomnia prevalence in this demographic and providing biological homogeneity during the perimenopausal/postmenopausal period. While this design strengthened internal validity, sex and age diverse cohort with adequate power for stratified analyses are needed to establish generalizability. Second, the markers used here are indirect proxies. The null findings of DTI‐ALPS was associated with a negligible covariate‐adjusted effect size, suggesting the absence of a meaningful group difference rather than insufficient sensitivity, although a small effect cannot be entirely excluded; this underscores the need for larger samples and for validation against more direct glymphatic measures, and complementary mechanistic studies. Third, the cross‐sectional design precludes causal inference; Longitudinal and interventional studies are needed to establish temporal relationships and to test whether successful insomnia treatment normalizes these markers.

5. Conclusions

This multiparametric approach showed convergent evidence linking chronic insomnia to altered neurofluid dynamics markers in middle‐aged and older women, including reduced neurovascular‐CSF coordination, ChP enlargement, and cholinergic system volume reduction.

Author Contributions

R.W. and K.W. contributed equally to this work and share first authorship. Conceptualization: R.W., K.W., S.Q., and A.M.; Data curation: K.W., D.H., and S.Q; Formal analysis: R.W., K.W., Q.G., C.L., and J.Z.; Funding acquisition: S.Q; Methodology: R.W. and K.W.; Investigation: Q.G., C.L., X.Z., and J.Z.; Software: R.W., K.W., L.H., and J.E.; Supervision: K.W., S.Q., and A.M.; Visualization: R.W.; Writing – original draft: R.W. and K.W.; Writing – review and editing: J.E., J.A., S.Q., and A.M.

Funding

This work was supported by National Natural Science Foundation of China (82330058, T2341014); University of Macau Development Foundation Technology and Innovation Start‐up (UMDF‐TISF/2025/010/ICI).

Disclosure

The authors have nothing to report.

Ethics Statement

This research was carried out following the principles set forth in the Declaration of Helsinki and was approved by the Institutional Review Board at the First Affiliated Hospital of Guangzhou University of Chinese Medicine (approved number: K‐2025‐021).

Consent

Written informed consent was acquired from all participants after a thorough explanation of the experimental procedures. Participants were clearly informed of their right to withdraw from the study at any point without any repercussions.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Partial correlations between glymphatic measures and cognitive variables.

Table S2: Partial correlations between glymphatic measures and sleep variables.

Table S3: Group differences in projection fibers and association fibers.

Table S4: Performance comparison logistic regression models.

Table S5: Partial correlations among neuroimaging markers.

Table S6: ANCOVA F and p values for covariates on four neuroimaging markers.

Figure S1: Forest plot of partial correlations between glymphatic indicators and cognitive measures.

Figure S2: Correlation between multi‐shell DTI‐ALPS index and single‐shell DTI‐ALPS.

Data Availability Statement

Data will be made available to others on reasonable requests to the corresponding author.