Integrating retinal and brain imaging biomarkers for diagnosis of parkinson’s disease: findings from a time-lagged cross-sectional persian cohort study

Abstract

Parkinson’s disease (PD) is a multisystem neurodegenerative disorder with both central and peripheral manifestations. This study aimed to comprehensively characterize structural and functional alterations in the retina and brain, as well as systemic blood biomarkers, in patients with clinically diagnosed PD compared with healthy controls, using a multimodal imaging approach. Blood-derived biomarkers and OCT data were collected from 29 PD patients and 25 healthy participants over an average interval of 2.7 years. MRI scans were performed using a Philips 1.5 Tesla scanner. Retinal and brain structural and perfusion metrics were analyzed using linear regression models adjusted for age, with statistical significance determined through false discovery rate (FDR) correction. The diagnostic performance was assessed using logistic regression, LASSO-penalized logistic regression model and ROC curve analysis. The study indicated that reduced ALT serum is the only significant hematological marker (p = 0.0253). Structural MRI revealed significant volume reductions in the left amygdala and caudate (p = 0.0241 and 0.0035 respectively); However, these findings necessitate careful interpretation due to limitations in spatial resolution. Perfusion MRI showed lower cerebral blood flow in gray matter and the whole brain (p = 0.02) in PD patients. The best imaging biomarker (PCASL total CBF) achieved an AUC of 0.79. A LASSO-penalized logistic regression model integrating left amygdala volume, left caudate volume, and SGPT achieved a significantly elevated AUC of 0.954, with 100% sensitivity and 86.7% specificity. The integration of central and peripheral biomarkers encompasses complementary aspects of Parkinson’s disease pathology, providing a more comprehensive diagnostic framework than any singular modality.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1038/s41598-026-66128-6.

Introduction

PD is a chronic, multisystem neurodegenerative disorder characterized not only by motor symptoms, including bradykinesia, resting tremor, rigidity, and postural instability, but also by a broad spectrum of non-motor manifestations, such as cognitive impairment, sleep disturbances, autonomic dysfunction, and visual abnormalities. As the second most common neurodegenerative condition after Alzheimer’s disease, PD is primarily defined by the progressive degeneration of dopaminergic neurons in the nigrostriatal pathway and consequent dopamine depletion in the striatum^1–6. Given global demographic ageing trends, the prevalence of PD is rising; it is projected that by 2050, over 12 million individuals worldwide will be affected⁷.

Early diagnosis of PD, prior to the full emergence of clinical symptoms, is critical for improving therapeutic outcomes, underscoring the urgent need for accessible, non-invasive biomarkers⁸.

PD is defined by the gradual buildup of alpha-synuclein aggregates in the central nervous system. The retina serves as a valuable lens for examining PD pathology due to its common neuroectodermal origin with the central nervous system and, importantly, its vulnerability to the same alpha-synuclein-driven neurodegenerative processes that define PD in the brain^9,10. The retina can be readily examined, facilitating non-invasive identification of pathological changes. Importantly, postmortem examinations reveal that phosphorylated alpha-synuclein aggregates accumulate in retinal ganglion cells and inner nuclear layers, following patterns that correlate with the severity of brain pathology. Ortuño-Lizarán et al. (2018) reported that phosphorylated α-synuclein in the retina serves as a biomarker of PD pathology severity⁶. Likewise, London et al. (2013) referred to the retina as ‘a window to the brain’ because it exhibits the same protein misfolding pathology, despite structural differences from the brain¹¹. This shared molecular mechanism, rather than anatomical resemblance, provides the biological groundwork for using retinal imaging as a biomarker of PD progression.

Evidence suggests that structural alterations in the retina may be detectable years before the onset of overt clinical symptoms of PD^8,12,13. Optical coherence tomography (OCT), a precise and non-invasive imaging modality with micrometer-level resolution, enables the quantitative assessment of these retinal changes^10,14–16.

Concurrently, structural and functional magnetic resonance imaging (MRI), including neuromelanin-sensitive MRI (NM-MRI) and arterial spin labelling (ASL), provides valuable insights into pathological and vascular alterations in the brain across different stages of the disease^17–21. NM-MRI can accurately visualize the degeneration of dopaminergic neurons in the brainstem^22–25, while ASL quantitatively measures cerebral blood flow (CBF) without requiring exogenous contrast agents, offering high safety and repeatability^26–31.

In parallel with advances in neuroimaging, blood-based biomarkers have attracted increasing attention for their accessibility, low cost, and potential to reflect systemic metabolic, inflammatory, and neurodegenerative processes in PD. Alterations in lipid profiles, hepatic enzymes, and immune cell counts have been inconsistently reported in PD, suggesting a complex interplay between peripheral physiology and central pathology^32,33.

Recent studies have explored the integration of various neuroimaging modalities, particularly in the context of assessing the efficacy of artificial intelligence^34–38, The systematic integration of multiple imaging modalities, particularly retinal imaging, structural brain MRI, neuromelanin-sensitive MRI, and perfusion imaging, in conjunction with comprehensive blood-based assessments in patients with PD, remains an unexplored area of investigation^39–45. In this study, we utilized a cross‑sectional study with temporally staggered acquisitions in which retinal imaging and blood biomarker analysis were conducted approximately 2.7 years before comprehensive brain imaging within the same cohort. This approach addresses a significant gap in the existing literature, which has largely overlooked the simultaneous assessment of retinal changes alongside advanced cerebral markers. We aimed to evaluate the complementary pathophysiological value of these modalities. Specifically, this investigation explores whether collective alterations across retinal, cerebral, and hematological systems reflect the multisystem nature of PD, ultimately contributing to the development of more comprehensive and temporally informed diagnostic frameworks.

Materials and methods

Data collection

This study was conducted in accordance with ethical principles and approved by the Ethics Committee of Mashhad University of Medical Sciences (IR.MUMS.MEDICAL.REC.1403.318). Data were obtained from the Persian Organizational Cohort Study (POCS). OCT data were collected between 2017 and 2023 at Imam Reza Hospital, Mashhad University of Medical Sciences. The dataset included blood-based biomarkers and OCT-derived parameters. Retinal measures analyzed comprised thickness and volume of the foveal, parafoveal, and perifoveal regions, peripapillary retinal nerve fiber layer (pRNFL), ganglion cell complex (GCC), and structural indices of the optic nerve head from both eyes.

Inclusion and exclusion criteria

To ensure methodological consistency and sample homogeneity, strict inclusion and exclusion criteria were defined based on internationally recognized standards in neurodegenerative and neuro-ophthalmic imaging research. PD patients were required to have a confirmed diagnosis by a neurologist with at least 20 years of clinical experience, be under levodopa therapy, and not be using antipsychotic medications known to affect retinal structure. Additionally, participants must have been free of dementia and significant ocular pathologies, including advanced cataract, diagnosed or treated glaucoma, prior ocular surgery, age-related macular degeneration, or diabetic retinopathy. From a systemic health perspective, PD patients could not have uncontrolled diabetes (HbA1c > 7.5%), severe hypertension (systolic > 160 mmHg or diastolic > 100 mmHg), or a history of stroke or cerebrovascular disease.

Healthy controls were required to have no history of neurodegenerative disorders, significant ocular disease, ocular surgery, severe systemic illness, autoimmune conditions, long-term use of antipsychotic medications, or cerebrovascular events. Standard exclusion criteria for both groups included inability to undergo OCT or MRI scans with acceptable quality, significant changes in general health status that could confound study outcomes, or withdrawal of consent. After applying these criteria, 29 patients with PD and 25 healthy controls were included in the final analysis.

Spectral-domain optical coherence tomography

Retinal imaging was performed using the Optovue device (Fremont, CA, USA; software version V2018.1.1.63) following a standardized scanning protocol. Macular scans (centered on the fovea) and peripapillary scans (for optic nerve head assessment) were conducted by a trained operator and subsequently reviewed for image quality, ensuring the absence of segmentation errors or motion artifacts. Only high-quality scans were included in the final dataset.

Retinal regions were segmented according to the Early Treatment Diabetic Retinopathy Study (ETDRS) grid: the fovea (1-mm radius circle centered on the fovea), parafovea (1–3 mm ring divided into superior, inferior, nasal, and temporal quadrants), and perifovea (3–6 mm ring). Thickness and volume measurements for these regions were extracted separately for the right eye (OD) and left eye (OS). pRNFL thickness was measured using a circular scan around the optic disc, with values recorded for four quadrants and eight detailed sectors (ST, SN, NU, NL, IN, IT, TL, TU). The GCC within the macular region and volumetric parameters of the optic nerve head (total volume, rim volume, cup volume) were also analyzed.

Magnetic resonance imaging

All MRI examinations were performed approximately 2.7 years after OCT imaging, using a Philips Ingenia Ambition S 1.5 Tesla scanner (Philips Healthcare, Best, The Netherlands) at Nasle Omid Medical Imaging Center.

Structural T1-weighted spin echo (T1W-SE)

T1-weighted spin echo (2D spin echo) images were acquired with the following parameters: repetition time (TR) = 580.71 ms; echo time (TE) = 15 ms; flip angle = 69°; slice thickness = 5 mm; slice gap = 1 mm (resulting in 6 mm spacing between slices); number of signal averages = 2; sequence type = spin echo (SE); sequence variant = standard spin echo (SS); scan options = other; receiver coil = multi-channel head coil (MULTI COIL); field of view (FOV) = 240 × 189 mm²; in-plane resolution = 1.04 × 1.04 mm²; acquisition matrix = 232 × 232; reconstruction matrix = 512 × 512; pixel bandwidth = 109 Hz; phase encoding direction = anterior-posterior; imaging plane = oblique axial aligned with anterior-posterior commissure line.

Neuromelanin-sensitive imaging with magnetization transfer contrast (T1W-TSE + MTC)

Neuromelanin-sensitive imaging was performed using a 2D turbo spin echo with magnetization transfer contrast (T1W-TSE + MTC) with the following parameters: TR = 600.00 ms; TE = 12 ms; flip angle = 90°; slice thickness = 2.8 mm; slice gap = 0.3 mm (resulting in 3.1 mm spacing between slices); number of signal averages = 5; echo train length = 3; sequence type = spin echo (SE); sequence variant = magnetization transfer contrast (MTC); scan options = flow compensation (FC); receiver coil = multi-channel head coil (MULTI COIL); field of view (FOV) = 240 × 240 mm²; in-plane resolution = 1.08 × 1.08 mm²; acquisition matrix = 223 × 223; reconstruction matrix = 480 × 480; pixel bandwidth = 223 Hz; phase encoding direction = anterior-posterior.

Perfusion imaging using 3D pseudo-continuous arterial spin labeling with fat saturation (3D-PCASL + FS)

Cerebral perfusion was assessed using 3D pseudo‑continuous arterial spin labeling with fat saturation (3D-pCASL) with the following parameters: TR = 3886.45 ms; TE = 15.36 ms; flip angle = 90°; slice thickness = 8 mm; no gap between slices; number of signal averages = 1; echo train length = 225; sequence type = gradient echo (GR); sequence variant = spectral-spatial k-space encoding (SK); scan options = fat saturation (FS); receiver coil = multi-channel head coil (MULTI COIL); field of view (FOV) = 240 × 240 mm²; in-plane resolution = 3.0 × 3.0 mm²; acquisition matrix = 45 × 45; reconstruction matrix = 80 × 80; pixel bandwidth = 3490 Hz; phase encoding direction = right-left; label distance = 93.36 mm; post-labeling delay = 1800 ms; Labeling duration: 1650ms; background suppression = enabled; number of slices = 15; total acquisition time = 4 min 18 s.

Image processing

T1-weighted images were processed using the CAT12 toolbox within the SPM environment⁴⁶. Volumetric analyses were performed using the following atlases: Neuromorphometrics (for basal ganglia, white matter, and ventricles), Thalamic_nuclei (for thalamic subregions), and SUIT (for cerebellar structures). Cortical surface-based analyses were conducted using the Desikan-Killiany 40-region atlas (DK40)⁴⁷, as illustrated in Fig. 1.

Fig. 1.

Structural Processing Pipeline of T1W-SE MRI Images in CAT12. Structural analysis of MRI images is performed using the CAT12 toolbox, through which cortical surface and volumetric data are ultimately extracted utilizing dedicated neuroanatomical atlases.

NM-MRI images were analyzed using 3D Slicer software⁴⁸. Images were first converted to NIfTI format and imported into 3D Slicer. Regions of interest (ROIs) were manually delineated using the Level Tracing tool in the Segment Editor module to segment the substantia nigra (SN). Background ROIs (BG) (4-mm diameter) were placed bilaterally in the Cerebral Peduncle (CP) (Fig. 2)^49,50. Signal intensities from SN and CP segments were extracted using the Segment Statistics module and used to calculate signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR)^49,51:

Fig. 2.

Segmentation of the SN and Definition of the ROI in the CP. This figure was previously published in Nikparast et al., Photodiagnosis and Photodynamic Therapy, 2025; licensed under CC BY(44).

PCASL images were processed using the BASIL toolbox (part of FSL 6.0.7.18 running on Ubuntu) to generate CBF maps. Absolute CBF values were extracted for the whole brain, gray matter, and white matter, as illustrated in Fig. 3.

Fig. 3.

Perfusion Image Processing Pipeline. Initially, control and label images are subtracted from each other to generate a difference image. Subsequently, absolute cerebral perfusion is calculated and quantified using the BASIL toolbox.

Statistical analysis

All statistical analyses were performed using R software (version 4.5.0). Because all outcome variables were continuous, group comparisons were conducted using linear regression models, which are mathematically equivalent to analysis of covariance (ANCOVA) when covariates are included. For each outcome, models included group (PD vs. healthy control) and age as a covariate. To assess whether age-related associations differed between groups, a group × age interaction term was additionally included. P-values for the main effects of group and age, as well as for the interaction term, were reported.

To control for multiple comparisons, false discovery rate correction was applied separately within each domain containing multiple related outcomes (e.g., optical coherence tomography, structural MRI, and hematological/biochemical parameters). For neuromelanin-sensitive MRI and CBF indices, which comprised only a small number of predefined outcomes (two and three variables, respectively), no multiple-comparison correction was applied.

Diagnostic performance of imaging biomarkers was evaluated using logistic regression models adjusted for age. Receiver operating characteristic (ROC) curves were constructed, and the area under the curve (AUC) was calculated to quantify discrimination between PD patients and healthy controls. Optimal cutoff values were determined using Youden’s index, with corresponding sensitivity and specificity reported to enable standardized comparison across imaging modalities.

Results

Demographic characteristics

A total of 54 participants were included: 29 patients with PD (53.7%) and 25 healthy controls (46.3%). Of these, 31 (57.4%) were male and 23 (42.6%) were female. The Kolmogorov–Smirnov test confirmed normal age distribution. Independent t-tests indicated a significant difference in mean age between groups (p < 0.05). Therefore, age was included as a covariate in all subsequent analyses of OCT and MRI parameters. Linear regression models with a group × age interaction term were used to test potential moderation effects (Table 1).

Table 1.

Demographic and clinical characteristics of patients with Parkinson’s disease and healthy controls.

Variable	Healthy Controls	Parkinson’s Disease Patients	P-Value
Number of Participants and Percentage (%)	25 (46.3%)	29 (53.7%)
Sex (F/M)	11/14	12/17	0.846
Age (OCT)	50.00 ± 5.82	55.83 ± 11.25	0.024
Age (MRI)	52.68 ± 5.79	58.52 ± 11.13	0.022
UPDRS		47.45± 30.61

Note: Age differed significantly between groups, while sex distribution did not. A p-value < 0.05 was considered statistically significant.

Hematological and biochemical parameters

After adjustment for age and correction for multiple comparisons, serum glutamic-pyruvic transaminase (SGPT) was the only hematological or biochemical parameter that differed significantly between PD patients and healthy controls (q-value = 0.0253) (Table 2). No significant group × age interactions were observed, indicating that the association between ALT and disease status was consistent across the age range.

Table 2.

Hematological parameters in patients with Parkinson’s disease and healthy controls.

Blood Factor	Mean ± SD
	Control	PD	p (Group)	p (Age)	p (Group × Age)	q-value (Group, FDR)
WBC (1000/µL)	6.07 ± 1.35	6.18 ± 1.33	0.6999	0.5857	0.8968	0.8664
RBC (Mil/µL)	4.95 ± 0.49	4.76 ± 0.43	0.2591	0.4643	0.5949	0.3955
HGB (g/dL)	14.65 ± 1.37	14.77 ± 1.35	0.5350	0.0901	0.5024	0.7052
HCT (%)	42.36 ± 3.33	41.95 ± 3.58	0.8978	0.3515	0.9495	0.9643
MCV (fl.)	86.05 ± 6.87	88.22 ± 3.51	0.2101	0.8801	0.5096	0.3584
MCH (pg)	30 ± 2.88	31.43 ± 1.69	0.0378	0.3697	0.4651	0.0991
MCHC (g/dL)	34.73 ± 1.43	35.62 ± 0.96	0.0092	0.1344	0.3755	0.0874
PLT (1000/µL)	252.4 ± 85.48	208.29 ± 57.93	0.0255	0.2471	0.0444	0.0924
LY (%)	40.03 ± 7.82	32.87 ± 12.87	0.0371	0.9435	0.8454	0.0991
MO (%)	4.26 ± 1.46	3.3 ± 1.75	0.1490	0.1907	0.2465	0.3087
GR (%)	55.07 ± 7.5	62.88 ± 13.76	0.0217	0.6694	0.5846	0.0924
RDWCV (%)	11.24 ± 0.68	11.28 ± 0.59	0.9415	0.6590	0.8283	0.9751
RDWSD (fl.)	38.22 ± 2.82	39.85 ± 3.18	0.1673	0.8045	0.5442	0.3234
PCT (%)	0.2 ± 0.06	0.17 ± 0.04	0.0121	0.3609	0.1496	0.0874
MPV (fl.)	8.18 ± 0.64	8.13 ± 0.62	0.9872	0.3421	0.0362	0.9872
PDW (fl.)	17.14 ± 1.15	17.07 ± 0.62	0.4688	0.0623	0.0960	0.6475
GLUC (mg/dL)	102.84 ± 55.26	104.81 ± 20.36	0.8902	0.9759	0.6882	0.9643
TG (mg/dL)	152.28 ± 81.37	125.9 ± 82.15	0.3727	0.5901	0.7279	0.5405
CHOL (mg/dL)	194.48 ± 46.18	166.62 ± 36.27	0.0237	0.4981	0.1949	0.0924
HDLC (mg/dL)	50.24 ± 15.67	40.52 ± 9.65	0.0157	0.5147	0.7050	0.0909
BUN (mg/dL)	30.32 ± 7.9	35.1 ± 10.27	0.2309	0.0148	0.1453	0.3720
Cerat (mg/dL)	1.03 ± 0.16	1.07 ± 0.22	0.7986	0.2426	0.7809	0.9263
SGOT (U/L)	22.52 ± 5.61	18.95 ± 4.27	0.0410	0.1518	0.0168	0.0991
SGPT (ALT)(U/L)	27.64 ± 13.71	13.33 ± 10.17	0.0009	0.2906	0.2405	0.0253
ALP (IU/L)	201.32 ± 62.07	269.1 ± 69.79	0.0036	0.3298	0.9942	0.0515
GGT (U/L)	38 ± 34.52	41.43 ± 32.13	0.7170	0.8836	0.7333	0.8664

Note: p-values are derived from age-adjusted linear regression models including group and group × age interaction terms. FDR-adjusted p-values (q-values) for the group effect were calculated using the Benjamini–Hochberg method.

Optical coherence tomography parameters

In age-adjusted analyses, PD patients exhibited higher peripapillary retinal nerve fiber layer (pRNFL) thickness in the temporal (uncorrected p = 0.0236) and temporal-lower (uncorrected p = 0.0324) sectors of the left eye compared with controls. However, these differences did not remain significant after FDR correction for multiple comparisons. No significant group × age interactions were detected for any OCT parameter, providing no evidence for differential age-related effects between groups (Table S1). An exploratory correlation analysis between motor asymmetry (derived from lateralized UPDRS Part III scores) and interocular retinal thickness asymmetry showed no significant association (r = − 0.058, p = 0.4).

Structural MRI parameters

Following adjustment for age and correction for multiple comparisons, significant group differences were observed for left amygdala volume (q-value = 0.0241) and left caudate volume (q-value = 0.0035) only (Table S2). No significant group × age interactions were detected for either structure, suggesting that the observed group differences were stable across the examined age range and were not driven by differential age-related trajectories.

Neuromelanin-sensitive MRI

No significant group differences were identified in signal-to-noise ratio (SNR) or contrast-to-noise ratio (CNR) after adjustment for age, indicating no detectable alterations in neuromelanin-sensitive MRI measures in this cohort (Table 3).

Table 3.

Neuromelanin-sensitive MRI parameters (SNR and CNR) in patients with Parkinson’s disease and healthy controls.

neuromelanin factor	Control Mean ± SD	PD Mean ± SD	p (Group)	p (Age)	p (Group× Age)
CNR	3.6 ± 0.77	3.5 ± 0.49	0.4517	0.5734	0.7245
SNR	17.93 ± 2.94	18.78 ± 2.62	0.6557	0.1951	0.7706

Note: Results are based on linear regression models including group (PD vs. control), age (continuous), and a group × age interaction term.

Perfusion MRI

Patients with PD demonstrated significantly lower CBF in gray matter (p = 0.022) and whole brain (p = 0.028) compared with healthy controls after adjustment for age (Table 4). No significant group × age interactions were observed for perfusion measures.

Table 4.

Cerebral blood flow (CBF) Indices in patients with Parkinson’s disease and healthy controls.

Mean ± SD
CBF	Control	PD	p (Group)	p (Age)	p (Group×Age)
Total_Mean_CBF	54.41 ± 9.99	43.92 ± 15.2	0.0275	0.0777	0.8140
GM_Mean_CBF	71.19 ± 14.08	55.45 ± 21.47	0.0216	0.0330	0.9299
WM_Mean_CBF	44.71 ± 7.02	38.27 ± 12.55	0.0641	0.4371	0.7833

Note: Results are based on linear regression models including group (PD vs. control), age (continuous), and a group × age interaction term.

Diagnostic performance of imaging biomarkers

Age-adjusted logistic regression models were used to evaluate the discriminatory performance of imaging biomarkers (Table 5). In addition, to assess the value of multimodal integration, a LASSO-penalized logistic regression model was developed using candidate biomarkers from OCT, structural MRI, and hematological domains.

Table 5.

ROC analysis of imaging biomarkers for discrimination between Parkinson’s disease and healthy controls.

	AUC	Optimal Cutoff	Sensitivity	Specificity
OCT
Perifovea_Thick	0.671	0.49	60.9%	80.0%
Perifovea_Vol	0.670	0.49	60.9%	80.0%
Fovea_Thick	0.663	0.60	47.8%	92.0%
Fovea_Vol	0.663	0.60	47.8%	92.0%
Nervehead_Volume	0.663	0.51	58.3%	80.0%
para_Fovea_Thick	0.661	0.48	65.2%	76.0%
ParaFovea_vol__para_Fovea_vol	0.659	0.48	65.2%	76.0%
Average_pRNFL	0.652	0.53	58.3%	80.0%
Average_GCC	0.646	0.49	63.6%	73.9%
Structural MRI
GM_Absolute_Volume__cm3	0.754	0.48	79.3%	63.6%
CSF_Absolute_Volume__cm3	0.718	0.49	79.3%	63.6%
Brain_Thickness__mm	0.704	0.42	93.1%	45.5%
TIV__cm3	0.682	0.59	62.1%	77.3%
WM_Absolute_Volume__cm3	0.676	0.71	41.4%	100.0%
Neuromelanin MRI
CNR	0.710	0.57	55.6%	88.0%
SNR	0.695	0.51	74.1%	72.0%
Perfusion MRI (ASL)
Total_Mean_CBF	0.793	0.57	75.9%	88.0%
GM_Mean_CBF	0.789	0.59	75.9%	88.0%
WM_Mean_CBF	0.771	0.59	62.1%	92.0%

Note: Logistic regression was adjusted for age; optimal cutoff points were determined using Youden’s index.

Using the more parsimonious lambda.1se criterion, three variables were retained in the final integrated model: left amygdala volume, left caudate volume, and SGPT. This multimodal model demonstrated excellent diagnostic performance, with an AUC of 0.954, 100% sensitivity, and 86.7% specificity, outperforming all individual biomarkers (Table 6).

Table 6.

Multimodal LASSO model results.

Retained Biomarker	Penalized Coefficient
Left_Amygdala	−1.372
Left_Caudate	−2.899
SGPT	−0.002

Cross-validated AUC = 0.954 (95% CI: 0.887–1); Sensitivity = 100%; Specificity = 86.7%.

OCT: AUC values ranged from 0.646 to 0.671. Foveal thickness measures demonstrated high specificity (92%) but low sensitivity (47.8%), indicating limited screening utility but potential value for confirming disease presence.

Structural MRI: Gray matter volume showed the highest discrimination among sMRI parameters (AUC = 0.754; sensitivity = 79.3%; specificity = 63.6%), followed by cerebrospinal fluid volume (AUC = 0.718; sensitivity = 79.3%; specificity = 63.6%). White matter volume demonstrated perfect specificity (100%) but low sensitivity (41.4%), suggesting utility primarily as a confirmatory biomarker.

Neuromelanin-sensitive MRI: CNR achieved the highest overall performance among single imaging biomarkers (AUC = 0.710; sensitivity = 55.6%; specificity = 88%), followed by SNR (AUC = 0.695; sensitivity = 74.1%; specificity = 72%).

Perfusion MRI (PCASL): Whole-brain and gray matter cerebral blood flow (CBF) demonstrated the highest diagnostic accuracy among perfusion measures, with AUC values of approximately 0.79, sensitivity of 75.9%, and specificity of 88%. White matter CBF also showed high discriminatory performance (AUC = 0.771), characterized by high specificity (92%) and comparatively lower sensitivity, indicating stronger confirmatory than screening utility.

Discussion

PD is a classic example of a multisystem disorder that affects more than just the central nervous system. Although it has traditionally been recognized as a movement disorder due to dopamine loss in the nigrostriatal pathway, modern research shows that PD affects multiple organ systems. The primary pathological feature of PD, the aggregation of alpha-synuclein, follows a specific progression as outlined in the Braak staging, starting in the peripheral nervous system (specifically the olfactory bulb and enteric nervous system) and eventually moving to the brainstem and, finally, the neocortex. This progression accounts for the early emergence of non-motor symptoms that often appear years or even decades before the onset of motor symptoms, including gastrointestinal issues, olfactory loss, autonomic dysfunction, sleep disturbances, and peripheral nervous system effects^52,53. Our observations of retinal changes alongside brain alterations illustrate only one aspect of this broader systemic pathology.

The observed alterations in blood parameters, OCT, and structural, neuromelanin-sensitive, and perfusion MRI underscore the importance of multimodal imaging in the diagnosis of this disease.

Blood-based biomarkers

Blood-based biomarkers, due to their ease of access, low cost, and non-invasive nature, are attractive candidates for screening and longitudinal monitoring of patients⁵⁴. Following FDR correction, ALT emerged as the only blood parameter that remained significantly lower in our PD group. A plausible explanation for this isolated finding comes from Ikenaka and colleagues⁵⁵, who recently reported that ALT reduction in PD largely reflects a levodopa-driven depletion of vitamin B6. Long‑term levodopa therapy accelerates the methionine cycle through COMT-mediated metabolism, consuming the active form of vitamin B6, pyridoxal‑5′‑phosphate (PLP), as a necessary cofactor. Because ALT activity is especially sensitive to PLP availability, more so than AST, the drop in transaminase levels tends to be more pronounced for ALT, often producing a lower ALT/AST ratio. In their study, patients on intensified levodopa regimens showed further declines in ALT, and the COMT inhibitor Opicapone partially restored ALT levels while also reducing hyperhomocysteinemia, reinforcing the link to metabolic rather than hepatocellular origin. All our PD participants were receiving levodopa, which makes this mechanism particularly relevant. We realize, of course, that reduced physical activity, common in PD, can also influence ALT, and our cross‑sectional data cannot completely separate the two. Still, the convergence of our data with the treatment‑related metabolic model proposed by Ikenaka et al. suggests that the lower ALT we observed likely represents a levodopa‑induced metabolic signature, not a nonspecific or lifestyle‑driven change. Direct measurement of vitamin B6 levels and activity monitoring in future longitudinal designs would be needed to firmly disentangle these contributions.

Structural retinal changes

In recent years, the retina has emerged as a potential non-invasive window into CNS pathology in established PD.

Our observation of increased peripapillary retinal nerve fiber layer thickness in the temporal and temporal-lower sectors of the left eye in PD patients, compared with controls, is a noteworthy finding that warrants careful consideration in light of the existing literature. This result appears counterintuitive, given that numerous studies have reported retinal thinning in patients with PD. However, it’s important to note that these differences did not remain significant after applying false discovery rate correction for multiple comparisons^56–59.

However, given that all patients in our study were uniformly treated with Levodopa, this observation may be explained by the neuroprotective effects of dopamine replacement therapy on retinal structures.

The association between levodopa treatment and retinal morphology has been explored in previous research, yielding mixed outcomes. Sen et al. found that while PD patients typically show thinning of the retinal nerve fiber layer compared to healthy controls, there was no notable difference in RNFL thickness between those treated with Levodopa and those who were not, even though the treated group had more advanced disease⁶⁰.

Importantly, Gulmez Sevim et al. reported that levodopa users had thicker inferotemporal pRNFL and reduced macular volume compared to nonusers within the PD patient population⁶¹.

This result directly aligns with our finding of increased thickness in the temporal and temporal-lower sectors, providing substantial support for a levodopa-specific influence on these retinal regions.

The mechanism that may contribute to this potential protective effect involves dopamine’s role in retinal physiology. The human retina contains dopaminergic amacrine cells that communicate with ganglion cells, and dopamine is a principal neurotransmitter in retinal networks^62,63.

Postmortem investigations have indicated reduced dopamine levels in the retinas of PD patients^64–66, suggesting that dopaminergic dysfunction may extend to retinal neurons beyond the substantia nigra. Levodopa therapy might alleviate this deficit through various mechanisms.

Yavas et al. further add to this understanding, reporting that the rim area, rim volume, and pRNFL were significantly greater in the levodopa-treated cohort. In contrast, the dopamine agonist-treated group had the thinnest measurements. This comparative result suggests that Levodopa may uniquely maintain retinal structure, unlike other dopaminergic treatments.

Levodopa could exert neuroprotective effects on retinal neurons by activating all dopamine receptor subtypes, thereby reinforcing dopaminergic pathways in the retina⁶⁷. Specific subtypes of dopamine receptors, particularly D1 and D2, are distributed throughout retinal neurons.

While Newman-Tancredi and colleagues illustrated that dopamine agonists display inconsistent efficacy across various receptor subtypes, with no single agonist achieving full efficacy (100%) at all D2 receptor subtypes, Levodopa, a direct precursor of dopamine, shows complete (100%) efficacy at all dopamine receptor subtypes^68,69.

Our findings of increased pRNFL thickness in the temporal sectors of levodopa-treated PD patients suggest that dopaminergic therapy exerts neuroprotective effects on retinal structures.

The observed asymmetry in retinal involvement correlates with the well-established asymmetric characteristics of PD pathology. Typically, PD presents with motor symptoms that initiate unilaterally, often impacting one side of the body more significantly than the other. This asymmetry also extends to non-motor manifestations, including retinal changes. Recent studies have identified that asymmetric volume alterations in the retina may serve as a promising new surrogate marker for neurodegenerative changes associated with PD^70,71. To determine whether this retinal asymmetry corresponds to clinical motor laterality, we computed a motor asymmetry index from the available lateralized UPDRS Part III scores; no significant correlation was found with interocular pRNFL asymmetry (r = − 0.058, p = 0.4). While this null result may reflect the limitations of cross‑sectional motor subscores rather than disprove a true clinico‑retinal relationship, it cautions against a simplistic link between retinal and motor asymmetries.

This asymmetry may indicate the uneven degeneration of dopaminergic pathways that characterizes the progression of PD. It is important to note that dopamine plays a crucial role not only in the substantia nigra but also functions as a key neurotransmitter in the retina, particularly in amacrine and plexiform cells. Therefore, the asymmetric loss of dopaminergic neurons in the brain may correspond to the asymmetric changes observed in the retina^72,73.

Multiple studies have indicated that the temporal retina displays both structural and functional vulnerabilities in neurodegenerative diseases, especially in the context of PD^74–76.

The preferential involvement of the temporal pRNFL sector in our study can be attributed to the selective vulnerability of the papillomacular bundle (PMB) fibers located in this quadrant. These fibers primarily consist of small-caliber parvocellular axons, which have high metabolic demands and rapid firing rates. As a result, they are particularly prone to mitochondrial dysfunction and oxidative stress, which are significant pathogenic mechanisms in PD.Thus, the observed thickening in the temporal sector likely signifies a compensatory response in this metabolically vulnerable region^77–80.

Structural brain changes

PD is associated with widespread structural alterations extending beyond the substantia nigra. In this study, 20 structural parameters showed statistically significant group differences independent of age.

The reduction in gray matter volume and left entorhinal thickness aligns with previous findings⁸¹. Evidence suggests the entorhinal cortex is among the earliest regions affected in neurodegenerative diseases such as PD and Alzheimer’s, functioning as the primary gateway for information flow into the hippocampus^82–84. Prior studies have associated volume loss in this region with memory deficits and conversion to dementia in PD⁸⁵.

Thinning in frontal regions, including Right Lateral Orbitofrontal Thickness, Right Medial Orbitofrontal Thickness, Right Parsopercularis Thickness, Right Parstriangularis Thickness, and Right Superiorfrontal Thickness, is consistent with multiple prior reports^86–89.

Similarly, reductions in right anterior and posterior cingulate thickness align with the existing literature^90,91.

Volume reductions were also observed in subcortical structures, including the accumbens, left amygdala, bilateral caudate, and left hippocampus. Increased volume of the left inferior lateral ventricle may reflect surrounding parenchymal atrophy.

Pathophysiologically, these reductions may stem from alpha-synuclein accumulation in non-dopaminergic regions, an anatomical progression pattern described by Braak et al. in their staging of PD pathology⁵².

Notably, a significant group × age interaction was observed for left entorhinal thickness, right superior frontal thickness, and right caudate volume, suggesting that neuroanatomical differences intensify with age, possibly reflecting accelerated neurodegeneration in these regions.

In summary, structural MRI may serve as a valuable diagnostic and prognostic biomarker; however, it requires validation in larger cohorts and integration with clinical and functional data.

It is essential to highlight that the 5 mm slice thickness of the structural MRI scan renders volumetric assessments of small, compact subcortical structures vulnerable to partial volume effects (PVE). Consequently, although these structures appeared as significant predictors in our comprehensive diagnostic model, the interpretations associated with these particular subcortical outcomes should be approached with considerable caution and confirmed in future research employing high-resolution isotropic 3D imaging.

Neuromelanin-sensitive MRI

In this study, after adjusting for age, no statistically significant differences in SNR or CNR were observed between groups, despite the PD group having lower mean values.

This finding appears inconsistent with studies reporting significant reductions in the neuromelanin signal in PD^92,93.

No significant differences in NM-MRI measures (SNR or CNR) were observed between PD patients and healthy volunteers after adjusting for age, despite reduced NM signal in PD patients. This lack of significance is likely due to age-related NM dynamics within our cohorts.

In healthy aging, the NM-MRI signal increases to a peak around ages 50 to 60, attributable to NM buildup, before declining due to age-related neuron loss. Our healthy controls, with a mean age of 52, are beginning to enter this decline phase, where neurodegeneration starts to reduce the signal even in the absence of PD^94–96.

Individuals with PD, whose average age is 58.52, are somewhat older and have passed this peak, experiencing accelerated loss of neurons that leads to a quicker decline in signal. However, the continuous age-related deterioration in the control group may overlap with the changes related to PD, making it challenging to distinguish group differences in this research. In patients with PD, additional genetic and environmental factors, along with pathogenic NM levels, may contribute to the more rapid degeneration of susceptible neurons once the threshold is surpassed^97–99.

Several other factors may account for the absence of significant NM-MRI differences between the groups. Our sample size of 54 participants was relatively small. As noted in the meta-analysis by Cho et al., the diagnostic accuracy of NM-MRI tends to diminish in smaller studies¹⁰⁰. Furthermore, variations in imaging parameters and segmentation methods can affect the results. Coupled with age-related NM dynamics, these limitations likely reduced the statistical power to identify subtle changes^22,49,100. To validate these findings, larger longitudinal studies with standardized protocols are necessary.

Cerebral perfusion imaging

CBF is usually greater in gray matter compared to white matter because of fundamental differences in the structure of blood vessels and the composition of the tissues. Gray matter has a higher capillary density than white matter, with more extensive vascular branching to meet the significant metabolic needs of neuronal cell bodies and synaptic activity.

In contrast, the flow of blood in white matter is naturally limited by the dense organization of myelinated axons^101–104.

In our study, we observed a significant reduction in CBF in the gray matter of PD patients, independent of age, whereas white matter CBF showed no statistically significant reduction. This finding was consistent with some previous studies^30,105–109.

Although decreased gray matter perfusion corresponds with the known neurodegenerative changes affecting dopaminergic and various neuronal groups in PD, the lack of substantial white matter CBF reduction may indicate intricate compensatory mechanisms rather than maintained white matter integrity^110–112.

Recent findings indicate that white matter abnormalities in PD are characterized by ongoing demyelination, axonal degeneration, and disruption of the blood-brain barrier^113–117.

As white matter pathways deteriorate, the breakdown of structural barriers could lead to increased vascular permeability and altered perfusion patterns^118,119.

This phenomenon has been observed in both aging and neurodegenerative disorders, in which MRI-detected white matter hyperintensities frequently correlate with abnormal perfusion measurements^120–122. Additionally, the use of chronic dopaminergic medications (which is common in our PD group) has been demonstrated to affect cerebral hemodynamics, possibly altering regional perfusion patterns in gray and white matter areas differently^123–125.

Consequently, the lack of decreased white matter CBF in our group should not be considered indicative of preserved white matter integrity; instead, it may reflect a complex physiological response to fundamental structural changes, which pharmacological factors may further influence. To enhance our understanding of the dynamic interplay between white matter microstructural integrity and changes in perfusion during the progression of PD, longitudinal studies that integrate advanced diffusion MRI metrics with ASL perfusion imaging are necessary.

Diagnostic performance of biomarkers

In the statistical analyses of this study, some imaging indices, including NM-MRI parameters, although not statistically significant differences between the PD group and the healthy control group, had significant diagnostic ability in the ROC curve analysis(Figs. 4 and 5). This seemingly contradictory finding is quite explainable from a statistical perspective. Significance tests (such as ANCOVA) only examine the difference in group means and are affected by factors such as sample size, within-group variance, and data distribution. In contrast, the AUC criterion in ROC analysis, rather than focusing solely on the mean, evaluates the entire data distribution and the ability of an index to correctly distinguish diseased from healthy individuals. Therefore, even in situations where the difference in means is not statistically significant due to sampling limitations (such as small sample size), the index may still provide valuable diagnostic information at the individual level. This highlights the importance of simultaneously examining inferential measures (p-value) and diagnostic performance measures (such as AUC) in biomarker studies.

Fig. 4.

ROC curves assessing the diagnostic performance of retinal OCT parameters in differentiating Parkinson’s disease patients from healthy controls. (A) ROC curves for macular thickness; (B) ROC curves for macular volume; (C) ROC curves for global retinal layer measurements.

Fig. 5.

ROC curves evaluating the diagnostic performance of structural, neuromelanin-sensitive, and perfusion MRI biomarkers in differentiating Parkinson’s disease patients from healthy controls. (A) Structural MRI-derived volumetric and cortical thickness measures; (B) Neuromelanin-sensitive MRI indices; (C) CBF metrics from PCASL perfusion MRI.

In this study, the diagnostic performance of four imaging techniques was evaluated using AUC and age-adjusted logistic regression. Results indicate that functional biomarkers, particularly PCASL, outperform structural and retinal biomarkers. OCT parameters, despite their non-invasive advantages, demonstrated only moderate diagnostic performance. Structural MRI and NM-MRI provided valuable information but exhibited imbalanced sensitivity/specificity profiles. PCASL emerged as the only modality demonstrating optimal performance across statistical, clinical, and practical domains, potentially serving as the foundation for a paradigm shift from structural to functional diagnostics in PD.

The results of our study indicate that multimodal imaging, integrating OCT and PCASL, can enhance the diagnostic characterization of PD by providing complementary structural and functional information. In addition to assessing individual modalities, we investigated whether the integration of central and peripheral indicators could enhance discrimination further. A LASSO-penalized logistic regression model, which incorporated putative biomarkers from blood, OCT, and structural MRI, maintained just three predictors: left amygdala volume, left caudate volume, and serum ALT. The resultant model had much superior discriminative performance compared to any individual biomarker, including the most effective unimodal MRI metric. The choice of two subcortical areas known to be impacted in early Parkinson’s disease, in conjunction with a levodopa-related metabolic marker, is physiologically coherent and implies that these modalities reflect partially distinct aspects of Parkinson’s disease pathology. This discovery offers empirical validation for the integrative approach proposed in this study: instead of substituting one modality for another, the amalgamation of structural, perfusion, and systemic metabolic data produces a more comprehensive and diagnostically potent understanding of the disease. Key limitations of this study include¹: reliance on conventional imaging and analytical methods, while advanced technologies such as OCT angiography (OCT-A), diffusion tensor imaging, and deep learning may offer improved sensitivity². It should be noted that the PD group was, on average, approximately six years older than the control group, Although all group comparisons were adjusted for age, this baseline imbalance could lead to residual confounding, particularly for biomarkers that are known to change with normal aging; such as brain volumes, cerebral perfusion, and retinal nerve fiber layer thickness. Therefore, while age-adjusted models help mitigate this issue, the possibility that some of the observed differences partly reflect age-related processes cannot be entirely excluded. Future studies with more closely age-matched samples would help to further disentangle disease-specific effects from those of aging³. A further limitation of the present study is that the control group consisted exclusively of healthy individuals. While this design permits characterization of disease‑related alterations in PD relative to normal aging, it does not provide information on the specificity of the observed imaging and blood biomarkers for PD versus other conditions that may present with overlapping motor features. In particular, the diagnostic performance metrics reported here (AUC, sensitivity, specificity) reflect discrimination between PD and health, not between PD and clinically relevant differential diagnoses such as atypical parkinsonism, vascular or drug‑induced parkinsonism, or essential tremor. The Persian Organizational Cohort Study from which our data were drawn did not include such comparison groups at the time of analysis. Future studies incorporating these clinical populations are essential to evaluate whether the multimodal biomarkers identified here can aid in differential diagnosis in real‑world neurological practice⁴. the cohort dataset did not include disease duration, side of initial motor onset, or quantitative dopaminergic medication doses. This precluded subgroup analyses that could have strengthened the interpretation of retinal and brain imaging findings⁵. It should also be noted that information on the use of cholesterol‑lowering medications (e.g., statins) was not systematically recorded in the cohort dataset, and therefore their potential influence on blood lipid levels could not be assessed)⁶. A significant methodological constraint of this work is the 5 mm slice thickness employed in the structural MRI acquisition. This parameter was intentionally selected as an essential clinical compromise to reduce overall scan time and alleviate significant motion artifacts, a vital factor for patients with PD who find it challenging to remain still during extended, multimodal imaging protocols (including T1-weighted, Neuromelanin-sensitive MRI, and PCASL). We recognize that whereas post-processing pipelines include interpolation, resampling cannot mitigate the physical partial volume effects (PVE) associated with thicker slice acquisition. Thus, volumetric assessments of small, compact subcortical structures, namely the left amygdala and caudate, identified as significant predictors in our integrated model, are vulnerable to partial volume effect-related error. Consequently, judgments related to these particular subcortical outcomes necessitate considerable caution and must be corroborated in forthcoming research employing high-resolution isotropic 3D scans optimized for subcortical morphometry.

Conclusion

Among several modalities, PCASL perfusion imaging demonstrated the highest diagnosis accuracy, emphasizing decreased gray-matter cerebral blood flow as a critical metric. Nonetheless, combination of essential central and peripheral markers, including left amygdala volume, left caudate volume, and serum ALT, significantly enhanced diagnostic discrimination (interpreted with caution due to MRI resolution limits)., nearing an almost flawless distinction between patients and controls. The choice of two subcortical areas known to be impacted early in Parkinson’s disease, alongside with a levodopa-related metabolic marker, is biologically coherent and suggests that structural brain modifications and systemic metabolic changes reflect complimentary facets of Parkinson’s disease pathogenesis. These findings illustrate that a genuinely integrated multimodal strategy, rather than dependence on any singular imaging or blood biomarker, can significantly improve diagnostic differentiation. This supports the broader adoption of combined central and peripheral biomarker panels for a more comprehensive characterization of Parkinson’s disease.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

F. N: Conceptualization, data curation, formal analysis, visualization, writing original draft. N. ASH: Methodology, statistical analysis, validation, writing review & editing. Z. G: Data collection, OCT processing, literature review. A. SH: Clinical diagnosis of PD patients, patient recruitment, clinical interpretation. N. SH: Ophthalmological assessment, OCT protocol design, quality control. H. Z: Supervision, project administration, funding acquisition, MRI protocol design, writing review & editing, corresponding author.

Funding

This work was supported by a research grant from Mashhad University of Medical Sciences, Iran.

Data availability

The datasets generated and analyzed during the current study are not publicly available due to restrictions imposed by the Ethics Committee of Mashhad University of Medical Sciences regarding participant confidentiality and data sharing policies. However, the datasets are available from the corresponding author (Dr. Hoda Zare, ZareH@mums.ac.ir) upon reasonable request and subject to approval by the Ethics Committee of Mashhad University of Medical Sciences.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

This study was conducted in accordance with the ethical standards of the Declaration of Helsinki and was approved by the Ethics Committee of Mashhad University of Medical Sciences (Approval Code: IR.MUMS.MEDICAL.REC.1403.318). Written informed consent was obtained from all participants prior to their inclusion in the Persian Organizational Cohort Study (POCS).