What earlier Parkinson's imaging resources established and where QPN-NC fits
PPMI is described as the most widely used open PD data resource and has driven many biomarker discovery efforts, but it primarily focuses on early and prodromal stages [1]. Independent datasets are needed to improve sample size and representation in demographics and disease span, and to enable replication and new statistical models [1]. QPN-NC was designed as a cross-sectional cohort of 202 PD and 69 older adults with multimodal MRI and extensive clinical and neuropsychological evaluations, explicitly to investigate neural correlates of motor and non-motor symptoms and facilitate comparisons with other PD studies [1].
Earlier single-center work illustrates the kind of evidence QPN-NC is meant to contextualize. A study of 31 early-stage PD patients and 30 matched controls found bilateral caudate atrophy and widespread white matter microstructural alterations, with cognitive deficits in memory and processing speed [2]. Another multimodal MRI study of 73 PD, 32 PSP, and 36 controls reported that substantia nigra pars compacta volume and red nucleus susceptibility differentiated PSP from PD with AUCs of 0.883 and 0.892, respectively [6]. These findings establish candidate structural and microstructural markers, but their small samples and single-site designs make replication in independent cohorts essential.
How QPN-NC is built: clinical depth, five pipelines, and harmonized annotations
QPN-NC includes anatomical, diffusion, susceptibility-weighted, and functional MRI sequences, with clinical assessments including MDS-UPDRS parts I to IV for PD participants, MoCA, and a neuropsychological battery [1]. Data availability varies by assessment: MRI was available for all 271 participants, Hoehn and Yahr and MDS-UPDRS Part III for 142, MDS-UPDRS Parts I, II, and IV for 95, MoCA for 149 PD and 48 controls, and the neuropsychological battery for 174 PD and 61 controls [1]. This variability is important because it constrains which analyses can be run and how missingness must be handled.
The release processes imaging through five neuroimaging pipelines and provides harmonized data annotations to facilitate search and cohort matching across other PD datasets [1]. It also shares the tools and software used to enable data discovery, reproduce image processing, and help other studies adopt community standards [1]. The authors explicitly frame this as adopting FAIR principles and using Nipoppy and Neurobagel tools to simplify discovery and reusability [1]. For readers accustomed to assembling processing steps from disparate manuscripts and webpages, this is the central practical advance.
What QPN-NC can and cannot settle about cross-cohort reproducibility
The paper reports preliminary analyses showing expected widespread tissue loss in cortex and subcortical volumes in PD, but no significant cortical thickness differences after correction for multiple comparisons; subcortical volumetric analysis showed significant decrease in the left amygdala and bilateral putamen [1]. These results are consistent with prior reports of cortical and subcortical atrophy in PD, though the authors note that effect sizes remain small and stage-specific atrophy patterns need further investigation [1]. The negative cortical thickness finding is a useful calibration point: it shows that a well-powered multimodal cohort can still yield null results for widely discussed markers.
Cross-cohort generalization is the harder problem. A 2026 study on unsupervised domain adaptation for multi-cohort PD diagnosis used PPMI and a hospital cohort and found that model attention could shift away from the corpus callosum, a region repeatedly associated with early white matter degeneration in PD, when adaptation direction changed [4]. The authors argued that their curriculum-guided framework kept attention more consistent with known PD white matter patterns, but they also noted that heatmaps included non-relevant areas due to limited attribution resolution [4]. This is validation evidence that scanner and protocol heterogeneity can change what a model learns, which is exactly the problem harmonized annotations and shared pipelines are meant to reduce.
Competing evidence from UK Biobank shows how far afield biomarker claims can travel. In 68,508 participants with color fundus photography, optic disc pallor was higher in prevalent PD, with each standard deviation increase in global pallor associated with a 39% increase in the odds of being in the prevalent PD group [3]. However, the association between pallor and PD duration was strongly influenced by five individuals with duration over 15 years, and removing them rendered all associations non-significant [3]. This is a cautionary parallel: large cohorts can produce statistically significant but fragile associations, and QPN-NC's cross-sectional design means it cannot resolve temporal ordering either.
Where the evidence stops: cross-sectional design, single-center sampling, and clinical translation
The QPN-NC release is explicitly cross-sectional because only about 25% of participants have had follow-up visits; longitudinal data are promised in future releases [1]. The authors also note irregular visit schedules with variable intervals between MRI and clinical assessments, limited to a maximum of six months, and provide precise age at each assessment to help analysts filter by interval [1]. These design features mean QPN-NC cannot by itself validate progression biomarkers or establish causal relationships between imaging and clinical change.
Sample composition further bounds generalization. QPN-NC includes 202 PD and 69 controls recruited from the Quebec Parkinson Network registry, with controls primarily recruited from family members of clinic patients [1]. The PD group is 66% male and the control group 36% male, and participants were evaluated in their usual ON medication state [1]. These characteristics are not necessarily flaws, but they define the population to which findings apply and limit direct extension to unselected community samples or drug-naive patients.
The broader literature shows why multimodal biomarkers remain promising but unproven for clinical translation. A four-year longitudinal quantitative MRI study found that posteroventral substantia nigra iron increased with disease duration and was inversely correlated with nigral neuromelanin and striatal DaT changes in PD, with QSM outperforming R2* as a longitudinal marker [5]. A machine learning study using structural and functional MRI reported 94.7% accuracy and ROC-AUC of 0.98 for detecting beginning cognitive impairment in 38 PD patients, but the authors noted that a model using clinical and functional MRI alone reached similar accuracy [7]. These results support the potential of multimodal markers while underscoring that QPN-NC's contribution is infrastructural: it provides a reproducible substrate for testing whether such markers replicate across cohorts, not proof that they are clinically ready.
About These Sources
This research page is built on 7 peer-reviewed studies — published from 2025 to 2026, 7 from 2024 or later — selected as the most relevant from 10 studies that passed quality screening, drawn from 88 papers retrieved from a database of over 500 million.
Sources used in this answer
A preprocessed multimodal neuroimaging dataset from Quebec Parkinson Network for biomarker discovery
QPN-NC releases a cross-sectional cohort of 202 PD and 69 older adults with multimodal MRI, extensive clinical and neuropsychological evaluations, five processing pipelines, and harmonized annotations for cross-cohort matching [1].
Fronto-caudate and callosal microstructural alterations: unveiling multimodal MRI biomarkers in early Parkinson’s disease
In 31 early-stage PD patients and 30 matched controls, this precursor study found bilateral caudate atrophy, widespread white matter microstructural alterations, and selective memory and processing speed deficits [3].
Optic Disc Pallor in Parkinson's Disease: A UK Biobank Study
This competing UK Biobank study of 68,508 participants found higher optic disc pallor in prevalent PD, but the association with PD duration was driven by five individuals with duration over 15 years and became non-significant when they were removed [4].
A curriculum-guided unified framework for robust unsupervised domain adaptation on multi-cohort Parkinson's disease diagnosis
This validation study on unsupervised domain adaptation across PPMI and a hospital cohort found that model attention to the corpus callosum varied by adaptation direction, highlighting scanner and protocol heterogeneity as a barrier to robust multi-cohort PD diagnosis [5].
Early brain iron changes in Parkinson's disease and isolated rapid eye movement sleep behaviour disorder: a four-year longitudinal multimodal quantitative MRI study
This precursor four-year longitudinal quantitative MRI study found that posteroventral substantia nigra iron increased with disease duration and was inversely correlated with nigral neuromelanin and striatal DaT changes in PD, with QSM outperforming R2* [7].
Multimodal MRI biomarkers optimize differentiation between progressive supranuclear palsy and Parkinson's disease.
This precursor study of 73 PD, 32 PSP, and 36 controls found that substantia nigra pars compacta volume and red nucleus susceptibility differentiated PSP from PD with AUCs of 0.883 and 0.892, respectively [8].
A multimodal MRI framework employing machine learning for detecting beginning cognitive impairment in Parkinson’s disease
This precursor study of 38 PD patients reported 94.7% accuracy and ROC-AUC of 0.98 for detecting beginning cognitive impairment using structural and functional MRI features, though a model with clinical and functional MRI alone reached similar accuracy [10].