Participants

This study analyzed diffusion tensor imaging data from the PPMI database (https://www.ppmi-info.org/access-data-specimens/download-data, RRID:SCR_006431) as of November 20, 202542,43. The PPMI study was approved by the institutional review boards of all participating centers, and written informed consent was obtained from all participants prior to study enrollment. The study was conducted in accordance with the Declaration of Helsinki. All participants underwent comprehensive clinical evaluations at baseline and throughout longitudinal follow-up.

Clinical assessments

Clinical characteristics were assessed using standardized instruments available in the PPMI dataset. Clinical assessments included the UPDRS, which assesses cognitive and neuropsychiatric symptoms (UPDRS I), activities of daily living (UPDRS II), motor function (UPDRS III), and treatment-related motor complications (UPDRS IV)44.

Disease severity in PD patients was assessed according to the H-Y staging scale45. Demographic details of participants stratified by H-Y staging are summarized in Supplementary Table 3.

RBD status in prodromal individuals was confirmed by polysomnography at enrollment, and participants were categorized as Pro-RBD or Pro-Non. In PD patients, RBD status was assessed using RBDSQ46. PD patients with RBDSQ scores exceeding one standard deviation above the HC mean were classified as PD patients with probable RBD (PD-pRBD), whereas those with scores within or below one standard deviation of the HC mean were classified as PD-Non41. Demographic details of participants stratified by RBD status are summarized in Supplementary Table 4.

Cognitive status was evaluated using MoCA20. PD patients were categorized as PD with possible dementia (PD-pD, MoCA ≤ 21), PD with possible mild cognitive impairment (PD-pMCI, MoCA = 22–25), or PD-Normal (MoCA ≥ 26)47. Demographic details of participants stratified by cognitive status are summarized in Supplementary Table 5.

Depressive symptoms were assessed using GDS-15. PD patients were categorized as PD-ND (GDS-15 < 5), PD-MD (GDS-15 = 5–8), or PD-MTSD (GDS-15 > 8). Demographic details of participants stratified by depression are summarized in Supplementary Table 6.

Olfactory function was assessed using UPSIT19. PD patients were categorized as having anosmia (PD-Anos, UPSIT ≤ 18), severe hyposmia (PD-SH, UPSIT 19-24), moderate hyposmia (PD-ModH, UPSIT 25-29), mild hyposmia (PD-MH, UPSIT 30-33), or normal olfaction (PD-Norm, UPSIT ≥ 34). Demographic details of participants stratified by olfactory function are summarized in Supplementary Table 7.

Anxiety severity was assessed using the State-Trait Anxiety Inventory (STAI), including STAI-State (STAI-S) and STAI-Trait (STAI-T)48. PD patients were categorized as having no state anxiety/no trait anxiety (PD-NSA/PD-NTA, STAI-S/STAI-T < 39), mild state anxiety/mild trait anxiety (PD-MSA/PD-MTA, STAI-S/STAI-T, 40-49), or moderate state anxiety/moderate trait anxiety (PD-MoSA/PD-MoTA, STAI-S/STAI-T, 50-59). Demographic details of participants stratified by anxiety symptoms are summarized in Supplementary Tables 8 and 9.

DTI acquisition

DTI scans for the 1,861 participants included in this study were acquired across multiple research centers using Siemens Prisma_fit, Philips Achieva dStream, and GE DISCOVERY_MR750 scanners. Detailed DTI acquisition parameters are available on the PPMI website (http://www.ppmi-info.org/study-design/research-documents-and-sops/). The following DTI acquisition protocols were used: (1) DTI_B0_PA, DTI_revB0_AP, DTI_B700_64dir_PA, DTI_B1000_64dir_PA, DTI_B2000_64dir_PA: slice thickness = 2 mm; number of slices = 72; field of view = 230 × 230 mm; repetition time = 3800 ms; echo time = 90–100 ms; flip angle = 90°; number of directions = 64 per shell; b-values = 0, 700, 1000, and 2000 s/mm2; (2) DTI_RL, DTI_LR: slice thickness = 2 mm; number of slices = 80; field of view = 256 × 256 mm; repetition time = 10,000 ms; echo time = 80 ms; flip angle = 90°; number of directions = 32; b-values = 0 and 1000 s/mm2. Demographic details of participants stratified by DTI protocols are summarized in Supplementary Table 10. No significant differences in ALPS indices were detected between two protocols.

ALPS index calculation

DTI images were preprocessed and ALPS indices were calculated using DSI Studio (Hou build, July 15, 2025; http://dsi-studio.labsolver.org). Raw DICOM files were converted to NIFTI format using MRIcron (https://www.nitrc.org/projects/mricron) and imported into DSI Studio, where they were subsequently converted to SRC format. Image correction included compensation for phase distortions and eddy currents using TOPUP/EDDY, followed by motion correction to align all time points to the first b0 image, thereby correcting for involuntary head movements during acquisition. Images were subsequently reconstructed in MNI space using Q-Space Differential Reconstruction with a diffusion sampling length ratio of 1.25 and aligned to the Human ICBM152_adult template. This process generated color-coded fractional anisotropy maps and diffusion tensor maps along the X, Y, and Z axes.

Spherical ROIs (radius = 3 mm) were placed bilaterally at the level of the lateral ventricles (Fig. 2a). Projection fibers were represented by the superior corona radiata, while association fibers were represented by the superior longitudinal fasciculus. ROI center coordinates were as follows: left projection fibers (53, 48, 49), left association fibers (59, 48, 49), right projection fibers (26, 48, 49), and right association fibers (20, 48, 49). Minor manual adjustments were made if necessary to optimize placement. The ALPS indices for the left and right hemispheres were calculated using the following formula8:

$${\rm{ALPS}}\,{\rm{index}}=\frac{{D}_{{xx},{proj}}+{D}_{{xx},{assoc}}}{{D}_{{yy},{proj}}+{D}_{{zz},{assoc}}}$$

where \({D}_{{xx},{proj}}\) is the X-axis diffusivity of projection fibers, \({D}_{{xx},{assoc}}\) is the X-axis diffusivity of association fibers, \({D}_{{yy},{proj}}\) is the Y-axis diffusivity of projection fibers, and \({D}_{{zz},{assoc}}\) is the Z-axis diffusivity of association fibers (Fig. 2b). The mean of the left and right ALPS indices (mean ALPS index) was used as the final measure. The ALPS index is an indirect MRI-based marker associated with perivascular fluid dynamics and has been proposed to reflect brain clearance-related processes in humans, with higher and lower values indicating relatively greater or reduced function.

CPV segmentation and normalization

CPV were obtained from 3D T1-weighted MRI using the ChP-Seg pipeline (https://github.com/princeleeee/ChP-Seg) via Docker. This pipeline employs a two-stage 3D U-Net to segment the lateral ventricles followed by the ChP, with preprocessing steps including reorientation, resampling, and skull stripping. A human-in-the-loop active-learning strategy49 was applied, with manual corrections of challenging cases to improve segmentation accuracy. Segmented ChP masks were restored to the original image space, and absolute volumes were calculated from voxel counts (Fig. 2C)49,50. Normalized CPV (NCPV) was calculated as the ratio of CPV to total intracranial volume (ratio×103)22,51. Detailed information regarding the acquisition parameters of T1-weighted, 3D volumetric sequence is available on the PPMI website (http://www.ppmi-info.org/study-design/research-documents-and-sops/). Key parameters were as follows: plane = sagittal; slice thickness = 1.0 mm; number of slices = 192; voxel size = 1.0 × 1.0 mm in plane resolution; phase encode direction = anterior-posterior; matrix = 256 × 256; field of view = 256 mm; and scan time ≈ 7 min.

Cortical thickness analysis

Cortical thickness data were obtained from 428 PD patients with appropriate 3D T1-weighted MRI data in the PPMI database (270 males, 158 females; mean age = 65.30 ± 9.29 years; age range = 36–86). Cortical thickness was calculated using FreeSurfer (version 7.2.3) with the automated recon-all processing pipeline applied to T1-weighted MRI scans52. This pipeline performs standardized preprocessing, cortical surface reconstruction, and tissue segmentation, with Nipoppy feature extraction tools (https://nipoppy.readthedocs.io/en/latest/user_guide/extraction.html) tracking all runtime parameters to ensure analytical consistency and reproducibility. Cortical regions were parcellated according to the Desikan-Killiany atlas, and thickness measurements for each cortical subregion were obtained for subsequent analyses. To ensure data quality, all cortical reconstructions were visually inspected. Scans with major segmentation errors, evident misclassification, or inclusion of non-brain tissue were excluded.

Statistical analysis

All statistical analyses were performed using IBM SPSS Statistics v26.0. The normality of continuous variables was assessed through visual inspection of histograms and Q-Q plots, along with skewness and kurtosis measures. Normally distributed variables were compared using one-way ANOVA followed by Bonferroni-corrected post hoc pairwise comparisons. Non-normally distributed variables were assessed using the Kruskal-Wallis test, and categorical variables were compared using the chi-square test. Partial correlation analyses were conducted to evaluate associations between the ALPS index or NCPV and clinical scales, and cortical thickness, with age, sex, and imaging sequences included as covariates. False discovery rate correction was applied to account for multiple comparisons. Analysis of covariance (ANCOVA) was used to perform cross-sectional group comparisons of ALPS indices or NCPV and clinical scale scores while controlling for age, sex, and imaging sequences. ROC curves were employed to assess the diagnostic performance of the ALPS index, NCPV, and their combination in distinguishing functional groups and assessing their potential utility as early biomarkers. Structural equation modeling was used to evaluate potential mediation effects. Mediation analyses were conducted using Model 4 of the SPSS PROCESS macro with bootstrap resampling (5,000 iterations) to assess indirect effects and their 95% CI. Mediation effects were considered significant if the 95% CI did not include zero.