Periventricular diffusivity at the ventricular-parenchymal interface across healthy controls, episodic migraine, and chronic migraine: a cross-sectional multimodal MRI study

Abstract

Background

Chronic migraine (CM) carries high disability, yet interictal MRI phenotypes that distinguish CM from episodic migraine (EM), particularly in medication-overuse headache (MOH)-inclusive cohorts, remain incompletely characterized. We tested whether periventricular diffusivity (PVeD), an indirect MRI marker of tissue-water properties at the ventricular-parenchymal interface, is associated with CM and headache burden.

Methods

In this cross-sectional 3.0-T multimodal MRI study, 289 adults with complete structural MRI, diffusion MRI, and resting-state fMRI were analyzed: 67 healthy controls, 162 EM, and 60 CM. MOH was diagnosed and retained in the CM group. MRI was acquired during verified interictal evening sessions; participants with headache within 24 hours before or after MRI were excluded. The primary contrast was age- and sex-adjusted CM versus EM for PVeD. Secondary analyses examined MOH strata, headache-frequency strata, partial Spearman correlations, Firth logistic regression for CM status, and HIT-6 models.

Results

Age- and sex-adjusted PVeD was lower in CM than EM (adjusted difference, −0.011; 95% CI, −0.017 to −0.006; standardized difference, −0.71; p < 0.001) and healthy controls (−0.014; 95% CI, −0.021 to −0.007; standardized difference, −0.88; p < 0.001), with no EM-control difference. Both CM participants with and without MOH had lower PVeD than EM. In migraine participants, lower PVeD correlated with higher headache frequency, HIT-6, and MIDAS and with larger choroid plexus volume and smaller thalamic, nucleus accumbens, and putaminal volumes. In the expanded clinical-plus-MRI Firth model, each 1-SD higher PVeD was associated with lower CM odds (OR, 0.59; 95% CI, 0.38 to 0.89; p = 0.011). Higher PVeD was associated with lower HIT-6 after adjustment for headache frequency or MOH.

Conclusions

CM was associated with lower PVeD at the ventricular-parenchymal interface, and lower PVeD was linked to patient-centered burden in this MOH-inclusive cohort. PVeD is not yet a diagnostic or treatment-selection biomarker, but may help define an MRI phenotype for future chronification-risk and treatment-response studies after longitudinal and external validation.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s10194-026-02427-7.

Background

Headache disorders affected an estimated 2.9 billion people worldwide in 2023, and migraine accounted for most headache-attributed disability, including 40.9 million years lived with disability [1]. Chronic migraine (CM), defined by headache on at least 15 days per month for more than 3 months with migraine features or migraine-specific treatment response on at least 8 days per month, represents a high-burden clinical state within this spectrum [2–5]. CM is not simply a higher-frequency form of episodic migraine (EM); it is associated with greater disability, psychiatric and sleep comorbidity, medication overuse, and greater need for preventive treatment and care [3–8]. Because ICHD-3 recommends assigning both CM and medication-overuse headache (MOH) diagnoses when both criteria are met, imaging studies that exclude all MOH cases gain internal control but may lose clinical representativeness [5, 6]. Imaging phenotypes relevant to CM should therefore be tested in MOH-inclusive cohorts, with MOH-stratified analyses used to distinguish migraine chronification from medication-overuse-related confounding.

Mechanistic models of migraine chronification have traditionally emphasized trigeminovascular activation, central sensitization, descending pain modulation, and repeated attack-related changes in sensory and affective circuits [3, 4]. However, these circuit-based frameworks do not fully explain why some patients develop persistent high-frequency disease or how attack-related neurovascular, inflammatory, and CSF-facing processes may interact with brain tissue properties. Neurofluid biology has therefore emerged as a complementary framework. Experimental migraine models link cortical spreading depolarization and calcitonin gene-related peptide (CGRP) signaling to altered paravascular or meningeal lymphatic fluid handling [9, 10]. Human MRI studies have begun to probe this biology using diffusion tensor imaging analysis along the perivascular space (DTI-ALPS), dynamic contrast-enhanced meningeal lymphatic imaging, enlarged perivascular-space burden, choroid plexus (ChP) volume, and global BOLD-CSF coupling [11–20]. Because these metrics sample different compartments and are indirect MRI proxies, they should not be interpreted as direct measurements of glymphatic solute clearance or treated as interchangeable neurofluid readouts [12, 13, 21–23].

DTI-ALPS is the most widely used diffusion MRI proxy in migraine, but reported findings have been inconsistent across chronic, episodic, and high-frequency migraine samples [14–16]. This variability is not only a sampling issue; DTI-ALPS depends on deep white matter region-of-interest placement, fiber architecture, crossing fibers, and vascular orientation, and validation studies have questioned its use as a direct surrogate for glymphatic clearance [21–23]. The key unresolved question is therefore not whether migraine has a single glymphatic MRI signature, but whether CM is associated with a compartment-specific, anatomically localized diffusion phenotype that remains detectable after accounting for MOH, ventricular anatomy, image quality, mood and sleep symptoms, and other neurofluid-related MRI markers.

The ventricular-parenchymal interface provides an anatomically plausible target because it lies at a CSF-facing boundary distinct from the deep white matter regions sampled by DTI-ALPS. Periventricular diffusivity (PVeD), estimated with the EstPVeD workflow, was recently described in Alzheimer disease cohorts as a diffusion signal near the periventricular interface associated with amyloid burden and cognitive outcomes [24]. Its relevance to migraine chronification has not been tested. In this cross-sectional, interictal 3.0-T multimodal MRI study of healthy controls, EM, and MOH-inclusive CM, we tested the primary hypothesis that PVeD would be lower in CM than EM. Secondary objectives were to determine whether the PVeD-CM association was robust to MOH, lateral ventricle volume, PVeD quality assurance, mood and sleep symptoms, clinical duration, and other neurofluid-related MRI markers, and to determine whether PVeD was associated with patient-centered headache impact. All analyses were framed as cross-sectional associations rather than diagnostic validation, causal inference, or evidence of direct CSF-interstitial exchange.

Methods

Study design, setting, ethics approval, and consent

This single-center, cross-sectional observational study was approved by the Medical Ethics Committee of the Second Affiliated Hospital of Wenzhou Medical University (approval No. 2024-K-137–01) and was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent. Trial registration was not applicable because no intervention was assigned. The primary PVeD contrast and major sensitivity-analysis tiers were specified before outcome modeling. Reporting followed STROBE guidance [25]. Detailed acquisition, processing, and statistical parameters are provided in the Supplementary Methods.

Participants and diagnostic Classification

Adults aged 18–65 years were recruited between April 2024 and December 2025. Participants with migraine were consecutively recruited from the Headache Clinic of the Second Affiliated Hospital of Wenzhou Medical University. Healthy controls (HCs) were recruited from the local community and hospital-affiliated volunteer pools with group-level frequency matching by age and sex. Right-handedness was required at screening. HCs were required to have no history of migraine or recurrent headache, no clinically relevant medical or psychiatric history, no clinically diagnosed sleep disorders, and no medication use within 3 months before MRI, including analgesics.

Migraine diagnoses were assigned by board-certified headache neurologists using a structured interview based on the International Classification of Headache Disorders, 3rd edition (ICHD-3) [5]. EM included participants with migraine with or without aura and fewer than 15 headache days per month during the preceding 3 months. CM was defined as headache on at least 15 days per month for more than 3 months, with migraine features or migraine-specific acute-treatment response on at least 8 days per month, and not better accounted for by another diagnosis. MOH was diagnosed according to ICHD-3 and retained as a clinically relevant comorbidity within the CM group. Preventive migraine medication was permitted only when the medication agent, regimen, and dose had been stable for at least 3 months before MRI; initiation, discontinuation, or dose/regimen changes during this interval were not allowed. Current preventive medication use, acute medication frequency in days per month, aura status, disease duration, and days since the last headache before MRI were recorded. Participants were not excluded solely because of MOH or stable preventive medication use; these variables were addressed using stratified, covariate-adjusted, and sensitivity analyses.

Unless otherwise specified, exclusion criteria were applied to both migraine and healthy-control participants. Exclusion criteria were other primary or secondary headache disorders, with the prespecified exception of MOH in the CM group; clinically relevant comorbid pain disorders, including chronic musculoskeletal pain, fibromyalgia, neuropathic pain, or other chronic pain disorders; major neurologic, systemic, or psychiatric disorders other than migraine-associated anxiety or depressive symptoms in migraine participants; substance or alcohol abuse; MRI contraindications; significant structural brain abnormalities; white matter hyperintensities greater than Fazekas grade 1; and motion-related or modality-specific image-quality failure. Participants with any headache, including low-intensity or background headache, within 24 hours before or after MRI were excluded to reduce acute-state effects. Image-quality exclusions were determined before statistical outcome modeling.

Study size

Study size was determined by consecutive recruitment during the prespecified enrollment period. No formal a priori power calculation was performed because no CM-specific PVeD effect-size estimate was available. Interpretive sensitivity power calculations were performed to contextualize detectable effect sizes for the primary CM-EM PVeD contrast and migraine-cohort correlation analyses; these calculations did not alter the prespecified, estimate-based modeling framework and were not used to support confirmatory inference for subgroup, threshold, or exploratory analyses.

Clinical Assessment and MRI state control

HIT-6, MIDAS, PHQ-9, GAD-7, and PSQI were administered on the scanning day. Headache frequency was the average number of headache days per month during the preceding 3 months. Among participants with migraine, headache frequency was assessed using headache diaries in 47/222 participants (21.2%) and by structured clinician interview in the remaining 175/222 participants (78.8%). MIDAS used the same interval. Acute medication exposure was summarized as medication-use days per month, preventive exposure as use within 3 months before MRI, and severe headache impact as HIT-6 score ≥ 60. All MRI scans were acquired between 19:00 and 23:00 to reduce diurnal variability in neurofluid-related physiology. Participants were instructed to maintain a regular sleep-wake schedule during the preceding week and avoid alcohol and caffeine for 24 hours. Headache-free status for 24 hours before and after MRI was verified by diary review and follow-up contact; days since last headache were recorded for sensitivity analyses.

MRI acquisition and quality control

MRI was performed on the same 3.0-T scanner (Ingenia Elition X, Philips Healthcare, Best, the Netherlands) using the same 32-channel head coil. The protocol included 3D T1-weighted imaging (TR/TE = 8.1/3.7 ms; 1-mm isotropic voxels), T2-FLAIR (TR/TE = 8000/120 ms; 5-mm slices), multi-shell diffusion MRI (TR/TE = 3109/89 ms; 2-mm isotropic voxels; 56 directions across b = 1000 and 2000 s/mm2 plus b = 0; reverse phase-encoded b = 0 for TOPUP), and resting-state fMRI (TR/TE = 2000/30 ms; 3.2-mm isotropic voxels; 200 volumes). Full parameters are provided in the Supplementary Methods. During resting-state fMRI, participants were instructed to keep their eyes closed, remain awake, minimize head movement, and avoid structured thought. Wakefulness was assessed by technologist observation during scanning and post-scan interview. Visual and quantitative QA assessed artifacts, anatomic abnormalities, registration, segmentation, susceptibility correction, diffusion signal quality, and motion. Exclusion criteria included unusable data in any modality, major segmentation or registration failure, and resting-state fMRI mean FD_Power > 0.3 mm. Group labels were not used to determine inclusion. Uncertain cases were resolved by consensus review with a senior investigator or neuroradiologist.

Image processing

T1-weighted images were processed with FreeSurfer recon-all pipeline for subcortical volumes and lateral ventricle volume [26]. ChP volume was segmented with a pretrained T1-weighted 3D U-Net pipeline [27]. White matter perivascular-space volume was estimated from T1-weighted images using a Frangi filter pipeline in Quantitative Imaging Toolkit with ANTs-derived white matter masks [28]. Volumetric measures were residualized for estimated total intracranial volume [29]. Diffusion MRI underwent MP-PCA denoising and Gibbs-ringing correction in MRtrix3, TOPUP susceptibility correction, EDDY correction, and QSDR reconstruction to ICBM152 in DSI Studio [30]. Tensor measures were fitted from b ≤ 1000 s/mm² data. DTI-ALPS used fixed 5-mm-diameter spherical MNI-space ROIs at the lateral ventricle body, verified to avoid CSF and gray matter contamination; the bilateral average was prespecified. PVeD was quantified with the published EstPVeD workflow in QSDR-reconstructed MNI space using MATLAB 2023b, SPM12, and CAT12 [24]. The individualized candidate periventricular area was intersected with the participant-specific white matter mask; subject-level PVeD was the median transverse tensor ratio in the final bilateral periventricular white matter mask. The QA index of PVeD was used only in sensitivity analyses. Resting-state fMRI was processed in FSL, and gBOLD-CSF coupling was defined as the cross-correlation between global gray matter BOLD and inferior edge-slice CSF signals at the group-level negative-peak lag [31–33]. PVeD, DTI-ALPS, and gBOLD-CSF were interpreted as indirect MRI proxies. The multimodal MRI-derived markers and proxy locations are illustrated in Figs. 1B and 1C.

Fig. 1.

Participant flow and multimodal MRI-derived neurofluid-related markers. Panel a shows participant enrollment, peri-MRI headache exclusions, MRI review and quality-control exclusions, and the final analytic sample of 289 participants (67 HCs, 162 EM, and 60 CM, including 22 with MOH and 38 without MOH). Panel B shows representative T1-weighted MRI, resting-state fMRI, and diffusion MRI images and the corresponding derived markers: ChP volume, PVS volume, gBOLD-CSF coupling, DTI-ALPS, and PVeD. Panel C is a conceptual schematic of CSF-facing compartments, perivascular spaces, interstitial-neurofluid interfaces, perivenous compartments, and MRI proxy locations. The schematic is conceptual and does not imply that these MRI markers directly measure CSF secretion, glymphatic exchange, or solute clearance

Statistical analysis

Analyses were performed in R version 4.5.3, two-sided tests, and alpha = 0.05 unless specified. Continuous variables are reported as mean ± SD or median [IQR], and categorical variables as n (%). Baseline comparisons used analysis of variance or Kruskal-Wallis tests and chi-square or Fisher exact tests; baseline p values were descriptive. Analyses used complete-case data; no imputation was performed. Headache-frequency-stratified analyses categorized migraine participants into low-frequency EM (LFEM; <5 headache days/month), frequent EM (FEM; 5–14 headache days/month), and CM [34, 35]. FEM was used as a broader operational frequent-episodic-migraine stratum and was not intended to denote the narrower HFEM construct [34–36]. As a threshold-sensitivity analysis, EM was additionally classified as lower-frequency EM (<8 headache days/month) or HFEM (8–14 headache days/month), with CM defined as ≥ 15 headache days/month [36]. MOH-stratified analyses categorized migraine participants into EM, CM without MOH, and CM with MOH. Methodological and anatomic sensitivity analyses additionally adjusted for the QA index, lateral ventricle volume, or both.

The primary analysis compared age- and sex-adjusted PVeD between CM and EM using general linear models with HC3 robust standard errors; healthy-control contrasts contextualized the group pattern. Estimated marginal means provided adjusted means and pairwise contrasts. Pairwise p values were Bonferroni-adjusted within prespecified contrast families, and standardized mean differences used residual SD. Secondary imaging families used false-discovery-rate screening followed by Bonferroni-corrected pairwise contrasts.

As an age-balance sensitivity analysis, we performed complete 1:2 age/sex propensity-score matching for the primary CM-EM PVeD contrast. Propensity scores were estimated from age and sex, and nearest-neighbor matching was performed without replacement on the logit propensity score with a caliper of 0.2 SD. To address potential residual attack-state effects, we added direct linear PVeD sensitivity models for the primary CM-EM contrast with additional adjustment for days since the last headache before MRI. These models were also repeated after excluding participants scanned within ≤ 2 or ≤3 days of the last headache. As an aura-related sensitivity analysis, we repeated the primary CM-EM PVeD contrast with additional adjustment for aura status and after excluding participants with aura.

Partial Spearman correlations in migraine participants used residualized ranks adjusted for age and sex, with sensitivity adjustment for headache-frequency strata or MOH and Benjamini-Hochberg correction. CM status was modeled with Firth penalized logistic regression, with continuous imaging variables z-standardized [37]. Hierarchical models adjusted for age and sex; PHQ-9, GAD-7, PSQI, and disease duration; neurofluid MRI markers; and the expanded clinical-plus-MRI set. HIT-6 was modeled as a continuous outcome using multivariable linear regression; assumptions were checked by residual inspection, and HC3 sensitivity analyses yielded unchanged PVeD inference. HC-normative low-PVeD and incremental AUC analyses were exploratory and non-diagnostic.

Results

Clinical characteristics

The final sample included 289 participants: 67 healthy controls, 162 with EM, and 60 with CM. Age and sex did not differ significantly across groups (age p = 0.062; sex p = 0.082). Compared with EM, CM had greater headache frequency (18.50 [15.00, 26.00] vs 3.00 [2.00, 4.00] days/month), higher HIT-6 (66.00 [63.00, 68.00] vs 51.00 [49.00, 55.00]), higher MIDAS (48.50 [45.00, 71.00] vs 9.00 [6.00, 12.00]), longer disease duration, greater preventive-medication exposure, greater acute-medication frequency, and shorter time since the last headache before MRI. MOH was present in 22 CM participants (36.7%) and no EM participants. Participant flow is shown in Fig. 1A, and demographic, clinical, and core neuroimaging characteristics are summarized in Table 1.

Table 1.

Demographic, clinical, and core neuroimaging characteristics across healthy controls, episodic migraine, and chronic migraine

Characteristic	HC (n = 67)	EM (n = 162)	CM (n = 60)	p value
Demographics
Age, years	33.34 ± 9.30	33.81 ± 8.80	36.83 ± 10.56	0.062
Female sex, n (%)	48 (71.6%)	135 (83.3%)	51 (85.0%)	0.082
Clinical features
PHQ-9	2.00 [0.00, 5.00]	5.00 [3.00, 8.00]	7.00 [5.00, 11.50]	<0.001
GAD-7	1.00 [0.00, 4.00]	5.00 [2.00, 8.00]	5.00 [2.00, 10.00]	<0.001
PSQI	5.00 [3.00, 7.00]	6.00 [4.00, 9.00]	9.00 [6.00, 12.00]	<0.001
Aura, n (%)	-	33 (20.4%)	7 (11.7%)	0.134
Headache frequency, d/m	-	3.00 [2.00, 4.00]	18.50 [15.00, 26.00]	<0.001
Disease duration, years	-	7.00 [3.00, 10.00]	10.00 [5.00, 20.00]	0.003
Medication-overuse headache, n (%)	-	0 (0.0%)	22 (36.7%)	<0.001
Preventive medication use, n (%)	-	10 (6.2%)	15 (25.0%)	<0.001
Acute medication frequency, d/m	-	1.50 [0.00, 3.00]	9.00 [7.00, 17.00]	<0.001
Days since last headache before MRI, days	-	4.00 [3.00, 6.00]	2.00 [2.00, 3.00]	<0.001
HIT-6	-	51.00 [49.00, 55.00]	66.00 [63.00, 68.00]	<0.001
MIDAS	-	9.00 [6.00, 12.00]	48.50 [45.00, 71.00]	<0.001
Core MRI and QC markers
PVeD	0.5191 ± 0.0166	0.5165 ± 0.0164	0.5042 ± 0.0156	<0.001
PVeD QA index	0.62 ± 0.01	0.62 ± 0.01	0.62 ± 0.01	0.555
DTI-ALPS	1.75 ± 0.16	1.69 ± 0.17	1.60 ± 0.18	<0.001
gBOLD-CSF coupling	−0.32 ± 0.21	−0.28 ± 0.23	−0.10 ± 0.21	<0.001
mean FD_Power	0.16 ± 0.06	0.16 ± 0.05	0.17 ± 0.05	0.074
Volumetric MRI markers, displayed as raw mm3
Lateral ventricle volume	13424.22 ± 5910.94	11829.67 ± 3960.82	13276.47 ± 4507.52	0.011
PVS volume	3409.86 ± 1531.63	4317.97 ± 1558.55	3809.70 ± 1808.95	<0.001
ChP volume	1640.19 ± 518.78	1688.84 ± 456.10	1882.92 ± 443.55	<0.001
Thalamus volume	15580.53 ± 1448.92	14868.74 ± 1260.30	14387.08 ± 1585.29	<0.001
Nucleus accumbens volume	1131.35 ± 155.51	1086.54 ± 158.67	998.20 ± 146.21	0.001
Putamen volume	10353.05 ± 969.75	9990.87 ± 943.39	9500.83 ± 1143.49	0.002

Note. Values are mean ± SD, median [IQR], or n (%), as appropriate. p values are descriptive and were calculated using analysis of variance or Kruskal-Wallis tests for continuous variables and χ² or Fisher exact tests for categorical variables, as appropriate. Variables available only among participants with migraine were compared between EM and CM; healthy controls were not included for migraine-specific variables. Volumetric MRI markers are shown as raw mm³ in this descriptive table; adjusted analyses used eTIV-residualized volumes. Adjusted group contrasts are reported in the Results and Supplementary Tables. Higher gBOLD-CSF values indicate weaker negative coupling because values are closer to zero. Unless otherwise specified, PVeD and DTI-ALPS denote bilateral-average values. ChP = choroid plexus; CM = chronic migraine; DTI-ALPS = diffusion tensor imaging analysis along the perivascular space; EM = episodic migraine; gBOLD-CSF = global BOLD-CSF coupling; HC = healthy controls; HIT-6 = Headache Impact Test-6; MIDAS = Migraine Disability Assessment; MOH = medication-overuse headache; PHQ-9 = Patient Health Questionnaire-9; PSQI = Pittsburgh Sleep Quality Index; PVeD = periventricular diffusivity; PVS = perivascular spaces

Primary and stratified PVeD analyses

Age- and sex-adjusted PVeD differed across groups (global p < 0.001). Adjusted mean PVeD was 0.518 (95% CI, 0.514 to 0.522) in controls, 0.516 (95% CI, 0.513 to 0.519) in EM, and 0.504 (95% CI, 0.500 to 0.509) in CM. PVeD was lower in CM than EM (adjusted difference, −0.011; 95% CI, −0.017 to −0.006; standardized difference, −0.71; Bonferroni-adjusted p < 0.001) and controls (−0.014; 95% CI, −0.021 to −0.007; standardized difference, −0.88; p < 0.001). EM and controls did not differ (−0.003; 95% CI, −0.008 to 0.003; standardized difference, −0.17; p = 0.719). Using the observed CM-EM standardized difference and sample sizes, sensitivity power analysis yielded an approximate achieved power of 0.997, with an 80% minimum detectable standardized difference of 0.425 SD. The CM-EM contrast remained significant after adjustment for QA index, lateral ventricle volume, or both (all p < 0.001). Primary adjusted means and pairwise contrasts are shown in Fig. 2A–B and Supplementary Table S1.

Fig. 2.

Primary PVeD differences and stratified robustness analyses. Panel a shows age- and sex-adjusted PVeD means across HCs, EM, and CM. Panel B shows primary pairwise contrasts. Panel C shows MOH-stratified PVeD results, with CM participants with and without MOH compared with EM. Panel D shows headache-frequency strata contrasts. Panel E shows sensitivity of the CM-EM contrast after additional adjustment for the PVeD QA index, lateral ventricle volume, or both. Filled markers indicate Bonferroni-significant contrasts

In headache-frequency strata, CM had lower PVeD than LFEM (difference, −0.012; 95% CI, −0.018 to −0.006; p < 0.001) and FEM (−0.009; 95% CI, −0.018 to −0.001; p = 0.022), whereas frequent EM did not differ from LFEM (p = 0.965). In this frequency-stratified analysis, the migraine cohort included 124 participants with LFEM, 38 with FEM, and 60 with CM. In the additional HFEM-threshold sensitivity analysis, 153 EM participants were classified as lower-frequency EM and 9 as HFEM. Because the HFEM subgroup was small, contrasts involving HFEM were imprecise; CM nevertheless remained significantly lower than lower-frequency EM in PVeD (−0.0119; 95% CI, −0.0167 to −0.0072; Bonferroni p < 0.001; Supplementary Table S2).

In MOH-stratified analyses, PVeD was lower in CM without MOH than EM (−0.011; 95% CI, −0.018 to −0.004; p < 0.001) and lower in CM with MOH than EM (−0.013; 95% CI, −0.022 to −0.004; p < 0.001); CM strata did not differ (p = 1.000). Analyses are shown in Figure 2 and Supplementary Figure S1.

The CM-EM PVeD difference was also preserved after complete 1:2 age/sex propensity matching (50 CM and 100 EM; adjusted difference, −0.0121; 95% CI, −0.0177 to −0.0066; p < 0.001; Supplementary Table S3). The CM-EM PVeD difference remained essentially unchanged after additional adjustment for pre-scan interval (−0.0112; 95% CI, −0.0163 to −0.0061; p < 0.001) and persisted in >2-day and >3-day pre-scan interval exclusion analyses (Supplementary Table S4). The CM-EM PVeD difference was also preserved after additional adjustment for aura status (−0.0116; 95% CI, −0.0163 to −0.0069; p < 0.001) and after excluding participants with aura (−0.0119; 95% CI, −0.0172 to −0.0067; p < 0.001; Supplementary Table S5).

Secondary MRI marker patterns

Secondary MRI markers showed a nonuniform pattern (Fig. 3 and Supplementary Table S6). DTI-ALPS was lower in CM than EM (adjusted difference, −0.090; 95% CI, −0.150 to −0.031; Bonferroni p < 0.001), CM than controls (−0.158; 95% CI, −0.228 to −0.087; p < 0.001), and EM than controls (−0.067; 95% CI, −0.125 to −0.010; p = 0.015). gBOLD-CSF coupling values were higher, indicating weaker negative coupling, in CM than EM (0.162; 95% CI, 0.083 to 0.242; p < 0.001) and controls (0.192; 95% CI, 0.098 to 0.286; p < 0.001). ChP volume was larger in CM than EM (251.225 mm3; 95% CI, 96.981 to 405.470; p < 0.001) and controls (357.856 mm3; 95% CI, 175.439 to 540.273; p < 0.001). PVS volume differed primarily because EM exceeded controls; CM-EM and CM-control contrasts were not significant. Nucleus accumbens volume was lower in CM than EM and controls after Bonferroni correction. Additional selected subcortical contrasts are shown in Figure 3Band Supplementary Table S6.

Fig. 3.

Group differences in neurofluid-related MRI markers and selected subcortical volumes. Standardized adjusted differences were estimated from age- and sex-adjusted models; volumetric markers used eTIV-residualized values. Panel a shows standardized adjusted differences for PVeD, DTI-ALPS, gBOLD-CSF coupling, ChP volume, and PVS volume in CM vs EM, CM vs HC, and EM vs HC contrasts. Panel B shows corresponding standardized adjusted differences for selected subcortical volumes. Filled circles indicate Bonferroni-significant pairwise contrasts; open circles indicate nonsignificant contrasts

Clinical and imaging correlates of PVeD and DTI-ALPS

For the migraine-cohort correlation analyses (n = 222), the approximate 80% minimum detectable correlation was |r| ≈ 0.187. Within participants with migraine, lower PVeD was associated with greater headache frequency (partial Spearman r = −0.31; 95% CI, −0.43 to −0.19; FDR q < 0.001), HIT-6 (r = −0.35; 95% CI, −0.46 to −0.22; q < 0.001), and MIDAS (r = −0.31; 95% CI, −0.43 to −0.19; q < 0.001) after age and sex adjustment. PVeD was not significantly associated with PHQ-9, GAD-7, or PSQI after FDR correction. Lower PVeD was also associated with larger ChP volume (r = −0.45; 95% CI, −0.55 to −0.33; q < 0.001) and with smaller thalamic (r = 0.26; 95% CI, 0.13 to 0.38; q < 0.001), nucleus accumbens (r = 0.31; 95% CI, 0.19 to 0.43; q < 0.001), and putaminal volumes (r = 0.24; 95% CI, 0.11 to 0.36; q < 0.001).

DTI-ALPS was correlated with PVeD (r = 0.33; 95% CI, 0.20 to 0.44; q < 0.001) and with clinical burden measures, including headache frequency, HIT-6, and MIDAS, but showed a narrower neurofluid and structural correlation profile than PVeD. After additional adjustment for headache-frequency strata, PVeD associations with clinical-burden measures attenuated and were no longer FDR-significant; after MOH adjustment, these clinical-burden associations remained significant but were smaller in magnitude. Across both adjustment frameworks, PVeD associations with ChP volume, PVS burden, and selected subcortical volumes remained. Correlation patterns are shown in Figure 4 and Supplementary Table S7.

Fig. 4.

PVeD and DTI-ALPS clinical-imaging correlation patterns. Panel a shows partial Spearman correlations for PVeD after age and sex adjustment, after additional headache-frequency strata adjustment, and after additional MOH adjustment. Panel B shows corresponding correlations for DTI-ALPS. Panel C summarizes associations that remained stable across adjustment frameworks. Colors indicate partial Spearman r values; markers indicate FDR-significant or near-significant cells as shown in the figure. Filled markers indicate FDR q < 0.05; open markers indicate near-significant cells as defined in the figure

Firth logistic regression and robustness analyses

In Firth penalized logistic regression restricted to participants with migraine, lower PVeD was associated with CM status across hierarchical models (Table 2). Each 1-SD higher PVeD was associated with lower odds of CM in the age- and sex-adjusted model (OR, 0.49; 95% CI, 0.35 to 0.68; p < 0.001), the clinical-adjusted model (OR, 0.48; 95% CI, 0.34 to 0.67; p < 0.001), the neurofluid-MRI model (OR, 0.61; 95% CI, 0.41 to 0.90; p = 0.012), and the expanded clinical-plus-MRI model (OR, 0.59; 95% CI, 0.38 to 0.89; p = 0.011). In the expanded model, GAD-7, PSQI, ChP volume, and gBOLD-CSF coupling were also associated with CM status; PHQ-9, DTI-ALPS, PVS volume, age, sex, and disease duration were not statistically significant.

Table 2.

Hierarchical and expanded Firth penalized logistic regression models for chronic migraine status

Model or characteristic	OR (95% CI)	p value
Panel A. PVeD association across hierarchical Firth models
Model 1: age + sex	0.49 (0.35 to 0.68)	<0.001
Model 2: Model 1 + PHQ-9 + GAD-7 + PSQI + disease duration	0.48 (0.34 to 0.67)	<0.001
Model 3: Model 1 + DTI-ALPS + ChP + PVS + gBOLD-CSF	0.61 (0.41 to 0.90)	0.012
Model 4: Expanded clinical + MRI model	0.59 (0.38 to 0.89)	0.011
Panel B. Expanded prespecified model
PVeD, per 1 SD	0.59 (0.38 to 0.89)	0.011
Age, per year	0.99 (0.95 to 1.03)	0.618
Male sex vs female sex	0.96 (0.32 to 2.71)	0.937
PHQ-9, per point	0.94 (0.85 to 1.04)	0.252
GAD-7, per point	1.10 (1.01 to 1.22)	0.030
PSQI, per point	1.10 (1.00 to 1.20)	0.045
Disease duration, per year	1.04 (1.00 to 1.09)	0.055
DTI-ALPS, per 1 SD	0.72 (0.47 to 1.07)	0.101
ChP volume, per 1 SD	1.65 (1.11 to 2.52)	0.014
PVS volume, per 1 SD	0.98 (0.68 to 1.42)	0.909
gBOLD-CSF coupling, per 1 SD	2.16 (1.49 to 3.24)	<0.001

Firth penalized logistic regression compared CM with EM among participants with migraine. CM was coded as the outcome category and EM as the reference category. Model 1 adjusted for age and sex. Model 2 additionally adjusted for GAD-7, PHQ-9, PSQI, and disease duration. Model 3 adjusted for age, sex, DTI-ALPS, ChP volume, PVS volume, and gBOLD-CSF coupling. Model 4 was the expanded prespecified model including age, sex, GAD-7, PHQ-9, PSQI, disease duration, DTI-ALPS, ChP volume, PVS volume, gBOLD-CSF coupling, and PVeD. PVeD, DTI-ALPS, ChP volume, PVS volume, and gBOLD-CSF coupling were z-standardized within the EM+CM migraine analytic cohort; age, disease duration, and questionnaire scores were entered in their original units. ChP and PVS volumes were recomputed from raw volumes using eTIV residual correction before z-standardization. Female sex was the reference category. Higher gBOLD-CSF values indicate weaker negative coupling. ChP = choroid plexus; CM = chronic migraine; DTI-ALPS = diffusion tensor imaging analysis along the perivascular space; EM = episodic migraine; GAD-7 = Generalized Anxiety Disorder-7; gBOLD-CSF = global BOLD-CSF coupling; PHQ-9 = Patient Health Questionnaire-9; PSQI = Pittsburgh Sleep Quality Index; PVeD = periventricular diffusivity; PVS = perivascular spaces; eTIV = estimated total intracranial volume

Continuous PVeD sensitivity models are shown in Supplementary Table S8 and Supplementary Figure S4. They supported the main association after adding pre-scan interval (OR, 0.50; 95% CI, 0.32 to 0.76; p < 0.001), PVeD QA index (OR, 0.46; 95% CI, 0.31 to 0.66; p < 0.001), lateral ventricle volume (OR, 0.48; 95% CI, 0.33 to 0.67; p < 0.001), all three variables (OR, 0.48; 95% CI, 0.30 to 0.75; p < 0.001), and no-MOH restriction (OR, 0.50; 95% CI, 0.33 to 0.74; p < 0.001). Acute-medication-frequency models were treated as exploratory overadjustment and attenuated the PVeD estimate (OR, 0.97; 95% CI, 0.47 to 2.07; p = 0.932). Sample-derived tertile, spline, HC-normative threshold, and incremental AUC analyses are reported in the Supplement.

Patient-centered HIT-6 outcome models

In multivariable linear models among migraine participants (Table 3), each 1-SD higher PVeD was associated with lower HIT-6 after clinical adjustment (beta, −2.227; 95% CI, −2.994 to −1.461; p < 0.001). The association remained after adjustment for headache-frequency strata (beta, −0.673; 95% CI, −1.076 to −0.270; p = 0.001), MOH status (beta, −1.796; 95% CI, −2.509 to −1.082; p < 0.001), continuous headache frequency (beta, −0.764; 95% CI, −1.202 to −0.327; p < 0.001), days since last headache before MRI (beta, −1.972; 95% CI, −2.703 to −1.240; p < 0.001). In the joint model including DTI-ALPS, both PVeD (beta, −1.765; 95% CI, −2.558 to −0.971; p < 0.001) and DTI-ALPS (beta, −1.420; 95% CI, −2.235 to −0.604; p < 0.001) were associated with HIT-6.

Table 3.

Patient-centered outcome models for HIT-6 among participants with migraine

Outcome	Model	Predictor	Beta estimate (95% CI)	p value
Section A. PVeD-only patient-centered models
HIT-6	Clin-adj	PVeD, per 1-SD higher	−2.227 (−2.994 to −1.461)	<0.001
HIT-6	Clin-adj + frequency strata	PVeD, per 1-SD higher	−0.673 (−1.076 to −0.270)	0.001
HIT-6	Clin-adj + MOH	PVeD, per 1-SD higher	−1.796 (−2.509 to −1.082)	<0.001
HIT-6	Clin-adj + headache frequency	PVeD, per 1-SD higher	−0.764 (−1.202 to −0.327)	<0.001
HIT-6	Clin-adj + days since last headache before MRI	PVeD, per 1-SD higher	−1.972 (−2.703 to −1.240)	<0.001
Section B. Selected HIT-6 joint model including DTI-ALPS
HIT-6	Clin-adj + DTI-ALPS + PVeD	PVeD, per 1-SD higher	−1.765 (−2.558 to −0.971)	<0.001
HIT-6	Clin-adj + DTI-ALPS + PVeD	DTI-ALPS, per 1-SD higher	−1.420 (−2.235 to −0.604)	<0.001

Note. Values are β estimates from multivariable linear models among participants with migraine. HIT-6 was modeled as a continuous score. β estimates represent the difference in HIT-6 score per 1-SD higher MRI marker. Clin-adj indicates adjustment for age, sex, PHQ-9, GAD-7, PSQI, and disease duration; additional models added the covariate indicated in the model label. The joint model included PVeD and DTI-ALPS simultaneously. These models were used as patient-centered burden analyses and were not interpreted as clinical prediction models. Negative β estimates indicate lower HIT-6 scores per 1-SD higher MRI marker. CM = chronic migraine; DTI-ALPS = diffusion tensor imaging analysis along the perivascular space; HIT-6 = Headache Impact Test-6; MOH = medication-overuse headache; PHQ-9 = Patient Health Questionnaire-9; PSQI = Pittsburgh Sleep Quality Index; PVeD = periventricular diffusivity

Exploratory PVeD threshold and dose-response analyses

Exploratory HC-normative low-PVeD threshold analyses were consistent with enrichment of low PVeD among participants with greater migraine burden: the lower 5th-percentile threshold derived from healthy controls (PVeD < 0.4912) identified low-PVeD status in 12/162 EM participants (7.4%) and 12/60 CM participants (20.0%) and was associated with both CM status and severe HIT-6 burden in Firth logistic models. Complementary sample-derived tertile and restricted cubic spline analyses showed greater CM and severe HIT-6 burden at lower PVeD values; apparent changes in AUC and Brier score after adding continuous PVeD to clinical models are reported in the Supplementary Tables S9–S11 and Supplementary Figures S5–S6.

Discussion

This interictal, MOH-inclusive multimodal MRI study supports the primary hypothesis that PVeD is lower in CM than EM. The principal findings were that the CM-EM PVeD difference was present in CM participants with and without MOH, lower PVeD was associated with HIT-6 burden, and PVeD showed a correlation profile partly distinct from DTI-ALPS and other CSF-facing or subcortical MRI markers. These findings identify a localized ventricular-interface diffusion phenotype associated with CM burden.

Retaining MOH cases was an important feature rather than a nuisance design choice. MOH is common in CM, affects acute medication withdrawal and preventive treatment planning, and is explicitly co-diagnosed with CM in ICHD-3 when criteria for both disorders are met [5, 6]. Excluding all MOH cases can improve etiologic homogeneity but may produce a less representative CM sample. In this cohort, both CM strata had lower PVeD than EM, whereas CM with MOH did not differ from CM without MOH. This pattern argues against the primary PVeD finding being driven only by medication overuse. Because acute-medication days overlap with headache frequency, MOH classification, and treatment-seeking severity, acute-medication-frequency adjustment was interpreted as exploratory overadjustment rather than as a primary confounding-control model. This attenuation suggests overlap with the broader CM severity-medication-use phenotype, whereas the primary CM-EM association remained robust across MOH-stratified and sensitivity analyses.

The PVeD finding extends neurofluid-related MRI work by shifting the anatomic focus to the ventricular-parenchymal interface. PVeD was recently described in Alzheimer disease cohorts as a diffusion signal near periventricular white matter related to amyloid burden and cognitive outcomes [24]. The present study tests this ventricular-interface metric in migraine chronification, where the relevant biology is likely to differ from neurodegeneration. Lower PVeD should therefore be interpreted narrowly as an altered diffusion ratio near a CSF-facing periventricular boundary, not as a direct readout of CSF-interstitial exchange [12, 13, 24]. The CM-EM difference persisted after lateral ventricle adjustment, supporting the value of reporting PVeD with explicit QA and anatomic sensitivity analyses rather than treating it as a standalone biological biomarker.

The contrast with DTI-ALPS is methodologically informative. DTI-ALPS showed lower values across migraine groups, including an EM-HC difference, whereas PVeD showed a more CM-selective pattern with no primary EM-HC difference. Prior migraine DTI-ALPS studies have reported impaired, increased, or frequency-dependent patterns, likely reflecting differences in clinical phenotype, MOH handling, attack-state control, acquisition, ROI placement, and analytic implementation [14–16]. Current validation literature also cautions that DTI-ALPS is affected by deep white matter fiber architecture, crossing fibers, and vascular orientation and should not be treated as a direct glymphatic clearance measure [21–23]. In the present correlation atlas, PVeD retained adjustment-stable associations with choroid plexus volume, PVS volume, and selected subcortical volumes, whereas DTI-ALPS retained adjustment-stable associations chiefly with HIT-6 and PVeD and did not show FDR-significant structural ROI associations. These patterns support treating PVeD and DTI-ALPS as complementary compartment-specific diffusion proxies rather than interchangeable readouts of a single neurofluid process.

The ChP, PVS, and gBOLD-CSF findings support a multi-compartment interpretation but should not be over-read. ChP volume was higher in CM and remained associated with CM status in the expanded model, consistent with recent migraine studies reporting choroid plexus enlargement or group differences across migraine phenotypes [19, 20]. PVS burden showed a less CM-specific group pattern; therefore, PVeD-PVS correlations were interpreted only as exploratory marker-interrelationship findings rather than evidence that visible PVS burden and PVeD index the same neurofluid pathway [17, 18]. gBOLD-CSF coupling was associated with CM status, with higher values indicating weaker negative coupling, but this metric is sensitive to arousal, sleep physiology, vascular dynamics, respiratory physiology, and scanning state [31–33]. Together, these secondary markers support a multi-compartment MRI phenotype, but they remain indirect proxies and should not be interpreted as identifying a single causal pathway [12, 13, 21–23].

The subcortical associations provide biologic context rather than mechanistic proof. Thalamic involvement is consistent with sensory relay, thalamocortical processing, allodynia, and photophobia models of migraine, and prior MRI studies have reported thalamic structural or functional abnormalities in migraine [38, 39]. Nucleus accumbens findings are compatible with prior migraine work linking nucleus accumbens iron deposition with disease progression and chronicity [40]. The putaminal association is also compatible with evidence that pain-responsive putaminal territories contribute to sensory aspects of pain and are connected with sensory-motor cortical regions [41]. However, these correlations do not establish neurodegeneration, central sensitization, or a causal sequence and may reflect shared vulnerability, cumulative attack burden, treatment exposure, mood or sleep comorbidity, or residual confounding.

The main contribution of this study is phenotyping rather than clinical deployment. Three aspects strengthen that contribution: use of a real-world MOH-inclusive CM cohort with MOH-stratified robustness checks; verified interictal MRI acquired in a restricted evening window with multimodal imaging and image-quality sensitivity analyses; and parallel modeling of PVeD with other neurofluid-related MRI markers and HIT-6 outcomes. In current clinical practice, CM is diagnosed clinically using ICHD-3 criteria, and CGRP-targeting therapies may be used as first-line preventive options in appropriate patients [5, 7]. A cautious translational path would be to test whether PVeD predicts EM-to-CM transition, response after MOH management, preventive-treatment response, or recovery trajectories in longitudinal, externally validated cohorts.

Several limitations frame the interpretation. First, the cross-sectional design precludes inference about temporality, causality, mediation, or progression.

Second, the study was single-center, conducted in a clinical setting in China, restricted to right-handed adults, and lacked an external validation cohort. Generalizability to other populations, scanners, acquisition protocols, and broader CM phenotypes remain uncertain.

Third, the CM sample and several subgroup analyses were modest in size, particularly the MOH subgroup and the narrower HFEM-threshold sensitivity analysis, limiting the robustness of multivariable modeling, threshold analyses, and subgroup comparisons. In addition, headache frequency was not uniformly diary-based: 47/222 migraine participants (21.2%) used headache diaries, whereas 175/222 (78.8%) were assessed by structured interview, introducing potential recall bias. Although Firth penalization and prespecified model structures reduce small-sample bias, subgroup-specific estimates should be interpreted cautiously and require replication [37]. Although aura-adjusted and aura-exclusion sensitivity analyses supported the primary PVeD finding, aura-specific subgroup inference was limited by the small number of CM participants with aura.

Fourth, medication and attack-state confounding cannot be fully removed. CM participants had greater acute medication frequency, more preventive medication exposure, and shorter intervals since the last headache before MRI. Although participants with any headache within 24 hours before or after MRI were excluded, and sensitivity analyses additionally adjusted for pre-scan interval and repeated models after excluding participants scanned within ≤ 2 or ≤3 days of the last headache, residual preictal, postictal, or peri-ictal physiological effects on diffusion metrics cannot be fully excluded. Sensitivity analyses adjusted for medication variables and pre-scan interval and repeated exclusion models, but residual confounding remains plausible, particularly because treatment patterns may reflect both disease severity and access to care.

Fifth, all MRI markers were indirect and technically constrained. PVeD may be influenced by local ventricular morphology, template-space reconstruction, registration accuracy, partial volume, and periventricular white matter anatomy [24]. DTI-ALPS is sensitive to ROI placement and fiber architecture [21–23]. T1-based Frangi-derived PVS volume is not equivalent to dedicated T2/FLAIR PVS rating [28]. gBOLD-CSF coupling may have been affected by unmonitored sleep, respiratory, vascular, or CO2-related physiology [31–33].

Conclusions

In this interictal, MOH-inclusive migraine cohort, lower PVeD at the ventricular-parenchymal interface was associated with CM status and HIT-6 burden. These findings define a localized periventricular diffusion phenotype linked to chronic migraine burden. Longitudinal, multicenter validation is needed to determine whether PVeD is a trait, state, chronification-risk, or treatment-responsive imaging phenotype.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

ChP

Choroid plexus

CM

Chronic migraine

CSF

Cerebrospinal fluid

DTI-ALPS

Diffusion tensor imaging analysis along the perivascular space

EM

Episodic migraine

FD

Framewise displacement

FEM

Frequent episodic migraine

FDR

False discovery rate

GAD-7

Generalized Anxiety Disorder-7

gBOLD-CSF

Global BOLD-cerebrospinal fluid coupling

HC

Healthy control

HFEM

High-frequency episodic migraine

HIT-6

Headache Impact Test-6

ICHD-3

International Classification of Headache Disorders, 3rd edition

LFEM

Low-frequency episodic migraine

MIDAS

Migraine Disability Assessment

MOH

Medication-overuse headache

PHQ-9

Patient Health Questionnaire-9

PSQI

Pittsburgh Sleep Quality Index

PVeD

Periventricular diffusivity

PVS

Perivascular spaces

QA

Quality assurance

Author contributions

J.S. and Y.L. contributed to conceptualization, supervision, funding acquisition, and project administration. FW.F. and CY.D. contributed to methodology, data curation, formal analysis, and drafting of the manuscript. XQ.L., Z.T., XJ.Z., and XZ.L. contributed to investigation, validation, software implementation, and data interpretation. XH.L., CY.L., MQ.Y., T.X., YX.H., and K.L. contributed to participant recruitment, imaging data acquisition, quality control, and visualization. YG.C., KY.C., and BL.H. contributed to technical implementation, interpretation of imaging analyses, and critical revision of the manuscript for important intellectual content. All authors reviewed and approved the final manuscript, contributed to critical revision for important intellectual content, and agree to be accountable for all aspects of the work.

Funding

This study was supported by the Health Commission of Zhejiang Province (2024KY1275) and the Wenzhou Basic Scientific Research Program (YC20250597). The funders had no role in study design, data collection, data analysis, interpretation, manuscript preparation, or the decision to submit the manuscript.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This single-center, cross-sectional observational study was approved by the Medical Ethics Committee of the Second Affiliated Hospital of Wenzhou Medical University (approval No. 2024-K-137–01) and was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent.

Consent for publication

Not applicable; the manuscript does not contain identifiable individual person’s data.

Competing interests

The authors declare no competing interests.