JU-PDDv1: A harmonized dataset for Parkinson's disease diagnosis from heterogeneous EEG recordings
Brain Network Disorders · 2026-09
DOI: 10.1016/j.bnd.2026.07.003 ↗Kurz gefasst
Fachartikel in der Zeitschrift Brain Network Disorders, 2026-09. Verfasst von Sankhadip Bera und 1 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 55 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Background
Parkinson's disease (PD) is a progressive neurodegenerative disease characterized by both motor and nonmotor symptoms for which early diagnosis of PD remains challenging. Electroencephalography (EEG) has emerged as a promising modality for PD diagnosis owing to its high temporal resolution, non-invasive nature, and cost-effectiveness. However, progress in EEG-based PD research has been hindered by the heterogeneity of publicly available datasets. Existing datasets vary widely in terms of their sampling rates, electrode montages, referencing schemes, amplitude units, and recording protocols, making cross-dataset evaluation and reproducible benchmarking extremely challenging.
Methods
We introduced Jadavpur University Parkinson's Disease Diagnosis version 1 (JU-PDDv1), a harmonized multi-source EEG dataset integrating six publicly available PD EEG datasets comprising 360 participants. By implementing a unified pre-processing pipeline, we established the necessity of standardizing EEG parameters to mitigate dataset-specific biases that impede cross-study comparisons. A subject-independent split was also provided to ensure fair evaluation. To benchmark our proposed JU-PDDv1 dataset, we tested standard deep learning architectures and a classical machine learning model.
Results
These baseline models achieved 57%–74% accuracy under subject-independent evaluation for binary PD classification. Intradataset and Leave-One-Dataset-Out (LODO) cross-dataset evaluations were conducted to evaluate the efficacy of the proposed harmonization framework. The harmonized JU-PDDv1 pipeline substantially improves the cross-dataset generalization, increasing the average LODO accuracy from 43.25% to 75.28% following harmonization. The results establish a reliable baseline performance and confirm the learnability of the harmonized dataset.
Conclusions
Overall, the JU-PDDv1 dataset offers a scalable resource that aims to reduce the pre-processing burden, supports cross-dataset generalization, and provides a strong platform for EEG-based PD research. The JU-PDDv1 dataset is publicly available from Kaggle.
Zitieren
APA
Bera, S., & Singh, P. K. (2026). JU-PDDv1: A harmonized dataset for Parkinson's disease diagnosis from heterogeneous EEG recordings. Brain Network Disorders. https://doi.org/10.1016/j.bnd.2026.07.003
Vancouver
Bera S, Singh PK. JU-PDDv1: A harmonized dataset for Parkinson's disease diagnosis from heterogeneous EEG recordings. Brain Network Disorders. 2026. doi:10.1016/j.bnd.2026.07.003
BibTeX
@article{Bera2026,
author = {Bera, Sankhadip and Singh, Pawan Kumar},
title = {JU-PDDv1: A harmonized dataset for Parkinson's disease diagnosis from heterogeneous EEG recordings},
journal = {Brain Network Disorders},
year = {2026},
doi = {10.1016/j.bnd.2026.07.003}
}
EUVIMED