Dynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysis

Parkinson’s disease (PD) is a multifactorial neurodegenerative disorder with complex progression. This study aims to analyze electroencephalography (EEG) connectivity patterns to better understand PD progression and stage of the disease using machine learning. Resting-state, eyes-closed EEG recordin...

Full description

Saved in:
Bibliographic Details
Main Authors: Caroline L Alves, Loriz Francisco Sallum, Francisco Aparecido Rodrigues, Thaise G L de O Toutain, Patrícia Maria de Carvalho Aguiar, Michael Moeckel
Format: Article
Language:English
Published: IOP Publishing 2025-01-01
Series:Journal of Physics: Complexity
Subjects:
Online Access:https://doi.org/10.1088/2632-072X/adf58a
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849394001127407616
author Caroline L Alves
Loriz Francisco Sallum
Francisco Aparecido Rodrigues
Thaise G L de O Toutain
Patrícia Maria de Carvalho Aguiar
Michael Moeckel
author_facet Caroline L Alves
Loriz Francisco Sallum
Francisco Aparecido Rodrigues
Thaise G L de O Toutain
Patrícia Maria de Carvalho Aguiar
Michael Moeckel
author_sort Caroline L Alves
collection DOAJ
description Parkinson’s disease (PD) is a multifactorial neurodegenerative disorder with complex progression. This study aims to analyze electroencephalography (EEG) connectivity patterns to better understand PD progression and stage of the disease using machine learning. Resting-state, eyes-closed EEG recordings were acquired from 31 individuals: 16 healthy controls (HCs) and 15 PD patients. The PD group was stratified by disease duration into early-stage (1–3 years, n  = 9) and advanced-stage (6–12 years, n  = 6). EEG was recorded using a 32-channel Biosemi Active-Two system (512 Hz), with signals segmented into non-overlapping 10 s windows. Functional connectivity matrices were constructed using multiple metrics, including coherence, Pearson, Spearman, canonical correlation, and Ledoit–Wolf shrinkage. Machine learning models were applied for both binary (PD vs HC) and multiclass (HC vs early vs advanced PD) classification. Interpretability was achieved using Shapley Additive Explanations (PD) methodology, and the most discriminative neural connections were statistically validated using the Wilcoxon test with Bonferroni correction. Our approach achieved high accuracy in classifying PD stages, with coherence emerging as the optimal metric for capturing synchronized neural activity. SHAP values revealed critical brain regions and connectivity patterns associated with disease progression. Statistical validation confirmed the significance of these connections across disease stages. Early-stage PD exhibited neural connectivity patterns similar to HCs, while advanced stages showed distinct connectivity changes. The findings highlight the utility of EEG connectivity and machine learning in staging PD, offering insights into PD pathogenesis and progression. SHAP-enhanced model interpretability ensures reliable identification of key neural connections, supporting personalized diagnostics and therapeutic strategies.
format Article
id doaj-art-231e771defcd413fb4cf11b2d96129ab
institution Kabale University
issn 2632-072X
language English
publishDate 2025-01-01
publisher IOP Publishing
record_format Article
series Journal of Physics: Complexity
spelling doaj-art-231e771defcd413fb4cf11b2d96129ab2025-08-20T03:40:13ZengIOP PublishingJournal of Physics: Complexity2632-072X2025-01-016303500610.1088/2632-072X/adf58aDynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysisCaroline L Alves0https://orcid.org/0000-0003-4708-1330Loriz Francisco Sallum1Francisco Aparecido Rodrigues2https://orcid.org/0000-0002-0145-5571Thaise G L de O Toutain3Patrícia Maria de Carvalho Aguiar4Michael Moeckel5Laboratory for Hybrid Modeling, Aschaffenburg University of Applied Sciences , Aschaffenburg, GermanyInstitute of Mathematical and Computer Sciences, University of São Paulo , São Paulo, BrazilInstitute of Mathematical and Computer Sciences, University of São Paulo , São Paulo, BrazilHealth Sciences Institute , Federal University of Bahia, Bahia, BrazilHospital Israelita Albert Einstein , São Paulo, Brazil; Department of Neurology and Neurosurgery, Federal University of São Paulo , SãoPaulo, BrazilLaboratory for Hybrid Modeling, Aschaffenburg University of Applied Sciences , Aschaffenburg, GermanyParkinson’s disease (PD) is a multifactorial neurodegenerative disorder with complex progression. This study aims to analyze electroencephalography (EEG) connectivity patterns to better understand PD progression and stage of the disease using machine learning. Resting-state, eyes-closed EEG recordings were acquired from 31 individuals: 16 healthy controls (HCs) and 15 PD patients. The PD group was stratified by disease duration into early-stage (1–3 years, n  = 9) and advanced-stage (6–12 years, n  = 6). EEG was recorded using a 32-channel Biosemi Active-Two system (512 Hz), with signals segmented into non-overlapping 10 s windows. Functional connectivity matrices were constructed using multiple metrics, including coherence, Pearson, Spearman, canonical correlation, and Ledoit–Wolf shrinkage. Machine learning models were applied for both binary (PD vs HC) and multiclass (HC vs early vs advanced PD) classification. Interpretability was achieved using Shapley Additive Explanations (PD) methodology, and the most discriminative neural connections were statistically validated using the Wilcoxon test with Bonferroni correction. Our approach achieved high accuracy in classifying PD stages, with coherence emerging as the optimal metric for capturing synchronized neural activity. SHAP values revealed critical brain regions and connectivity patterns associated with disease progression. Statistical validation confirmed the significance of these connections across disease stages. Early-stage PD exhibited neural connectivity patterns similar to HCs, while advanced stages showed distinct connectivity changes. The findings highlight the utility of EEG connectivity and machine learning in staging PD, offering insights into PD pathogenesis and progression. SHAP-enhanced model interpretability ensures reliable identification of key neural connections, supporting personalized diagnostics and therapeutic strategies.https://doi.org/10.1088/2632-072X/adf58abrain networkParkinson diseaseage stagemachine learning
spellingShingle Caroline L Alves
Loriz Francisco Sallum
Francisco Aparecido Rodrigues
Thaise G L de O Toutain
Patrícia Maria de Carvalho Aguiar
Michael Moeckel
Dynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysis
Journal of Physics: Complexity
brain network
Parkinson disease
age stage
machine learning
title Dynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysis
title_full Dynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysis
title_fullStr Dynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysis
title_full_unstemmed Dynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysis
title_short Dynamical patterns of EEG connectivity unveil Parkinson’s disease progression: insights from machine learning analysis
title_sort dynamical patterns of eeg connectivity unveil parkinson s disease progression insights from machine learning analysis
topic brain network
Parkinson disease
age stage
machine learning
url https://doi.org/10.1088/2632-072X/adf58a
work_keys_str_mv AT carolinelalves dynamicalpatternsofeegconnectivityunveilparkinsonsdiseaseprogressioninsightsfrommachinelearninganalysis
AT lorizfranciscosallum dynamicalpatternsofeegconnectivityunveilparkinsonsdiseaseprogressioninsightsfrommachinelearninganalysis
AT franciscoaparecidorodrigues dynamicalpatternsofeegconnectivityunveilparkinsonsdiseaseprogressioninsightsfrommachinelearninganalysis
AT thaisegldeotoutain dynamicalpatternsofeegconnectivityunveilparkinsonsdiseaseprogressioninsightsfrommachinelearninganalysis
AT patriciamariadecarvalhoaguiar dynamicalpatternsofeegconnectivityunveilparkinsonsdiseaseprogressioninsightsfrommachinelearninganalysis
AT michaelmoeckel dynamicalpatternsofeegconnectivityunveilparkinsonsdiseaseprogressioninsightsfrommachinelearninganalysis