Objective: To develop machine-learning neural biomarkers to predict cognition in Parkinson’s disease (PD).

Background: Variable levels of cognitive impairment are found in PD[1,2], but optimal neural biomarkers have not been established[3]. Machine-learning studies report that structural MRI (sMRI) can predict cognition in PD[4,5]. EEG source functional connectivity (EEG FC) also has some success[6-14]. Here, we benchmark the relative value of EEG FC against sMRI and their combination to assess whether EEG FC has the potential as an economical cognition biomarker for PD.

Method: We used MDS level II neuropsychological assessments[15] of five cognitive domains (attention and working memory, executive function, visuospatial, episodic memory, and language) in 136 PD patients and 51 healthy controls (HC), 74 females, mean age 71.5 years (SD=7.5), in the NZ Longitudinal Parkinson’s Progression Study[16]. We then examined global Z, the mean derived from the five domains[17]. EEG source-space FC from resting state (rs) (64 channels) was quantified using power-based Amplitude Envelope Correlation (AEC) and phase-based debiased weighted Phase Lag Index (dwPLI) for six frequency bands (delta 0.5-4, theta 4-8, alpha 8-13, beta 13-30, gamma1 30-49, gamma2 51-80 Hz). sMRI metrics from T1w imaging were cortical thickness, surface area, subcortical and FreeSurfer summations. Machine-learning algorithms (Elastic Net; Partial Least Square (PSL); stacking[18]) assessed prediction of global Z from EEG FC and sMRI.

Results: The best single predictive index was alpha AEC with r=.44 [95%CI .33–.54], R²=.17 [95%CI .05–.27]). This was followed by beta AEC, theta AEC, cortical thickness, and alpha dwPLI. However, when stacking different features we obtained slightly improved prediction with stacked AEC frequency bands r= .47 [CI .36–.57], R²=.19 [CI .07–.29]); this was not improved by including sMRI. When stacking dwPLI bands, prediction was r= .43 [CI .33–.53], R²=.15 [CI .01–.26]), improved slightly when including sMRI to r= .48 [CI .38–.58], R²=.21 [CI .10–.32]).

Conclusion: Power-based (AEC) rsEEG FC is a sufficient predictive biomarker for cognition in PD. Multimodal approaches added only small gains. Thus, EEG alone is a promising and relatively cheap biomarker of cognition in PD for potential large-scale applications or when the availability of MRI is an issue.

References: 1. Svenningsson P, Westman E, Ballard C, Aarsland D. Cognitive impairment in patients with Parkinson’s disease: diagnosis, biomarkers, and treatment. The Lancet Neurology. 2012;11(8):697-707.
2. Aarsland D, Batzu L, Halliday GM, et al. Parkinson disease-associated cognitive impairment. Nat Rev Dis Primers. 2021;7(1):1-21.
3. Droby A, Nosatzki S, Edry Y, et al. The interplay between structural and functional connectivity in early stage Parkinson’s disease patients. J Neurol Sci. 2022;442:120452.
4. Almgren, H., Camacho, M., Hanganu, A., Kibreab, M., Camicioli, R., Ismail, Z., … & Monchi, O. (2023). Machine learning-based prediction of longitudinal cognitive decline in early Parkinson’s disease using multimodal features. Scientific reports, 13(1), 13193.
5. Beheshti I, Ko JH. Predicting the occurrence of mild cognitive impairment in Parkinson’s disease using structural MRI data. Front Neurosci. 2024;18:1375395.
6. Betrouni N, Delval A, Chaton L, et al. Electroencephalography‐based machine learning for cognitive profiling in Parkinson’s disease: preliminary results. Mov Disord. 2019;34(2):210-217.
7. Chaturvedi M, Bogaarts JG, Kozak VV, et al. Phase lag index and spectral power as QEEG features for identification of patients with mild cognitive impairment in Parkinson’s disease. Clin Neurophysiol. 2019;130(10):1937-1944.
8. Sweeney A, Devereux B, Ong C, et al. The identification of mild cognitive impairment in Parkinson’s disease using EEG and machine learning. Alzheimers Dement. 2020;16:e040432.
9. Geraedts VJ, Koch M, Contarino MF, et al. Machine learning for automated EEG-based biomarkers of cognitive impairment during deep brain stimulation screening in patients with Parkinson’s disease. Clin Neurophysiol. 2021;132(5):1041-1048.
10. Jennings JL, Peraza LR, Baker M, et al. Investigating the power of eyes open resting state EEG for assisting in dementia diagnosis. Alzheimers Res Ther. 2022;14(1):109.
11. Gschwandtner U, Bogaarts G, Roth V, Fuhr P. Prediction of cognitive decline in Parkinson’s disease (PD) patients with electroencephalography (EEG) connectivity characterized by time-between-phase-crossing (TBPC). Sci Rep. 2023;13(1):5093.
12. Parajuli M, Amara AW, Shaban M. Screening of Mild Cognitive Impairment in Patients with Parkinson’s Disease Using a Variational Mode Decomposition Based Deep-Learning. In: 2023 11th International IEEE/EMBS Conference on Neural Engineering (NER). IEEE; 2023:1-6.
13. Mostile G, Quattropani S, Contrafatto F, et al. Testing Machine Learning Algorithms to Evaluate Fluctuating and Cognitive Profiles in Parkinson’s Disease by Motion Sensors and EEG Data. Comput Struct Biotechnol J. 2025.
14. Sasidharan D, Sowmya V, Gopalakrishnan EA. Significance of gender, brain region and EEG band complexity analysis for Parkinson’s disease classification using recurrence plots and machine learning algorithms. Phys Eng Sci Med. 2025:1-17.
15. Litvan I, Goldman JG, Tröster AI, Schmand BA, Weintraub D, Petersen RC, Mollenhauer B, Adler CH, Marder K, Williams‐Gray CH, Aarsland D. Diagnostic criteria for mild cognitive impairment in Parkinson’s disease: Movement Disorder Society Task Force guidelines. Movement disorders. 2012 Mar;27(3):349-56.
16. MacAskill MR, Pitcher TL, Melzer TR, Myall DJ, Horne KL, Shoorangiz R, Almuqbel MM, Livingston L, Grenfell S, Pascoe MJ, Marshall ET, Marsh S, Perry SE, Meissner WG, Theys C, Le Heron CJ, Keenan RJ, Dalrymple-Alford JC, Anderson TJ. The New Zealand Parkinson’s progression programme. J R Soc N Z. 2022 Aug 14;53(4):466-488.
17. Horne KL, MacAskill MR, Myall DJ, Livingston L, Grenfell S, Pascoe MJ, Young B, Shoorangiz R, Melzer TR, Pitcher TL, Anderson TJ, Dalrymple-Alford JC. Neuropsychiatric Symptoms Are Associated with Dementia in Parkinson’s Disease but Not Predictive of it. Mov Disord Clin Pract. 2021 Feb 18;8(3):390-399.
18. Wolpert DH. Stacked generalization. Neural networks. 1992 Jan 1;5(2):241-59.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/machine-learning-prediction-of-cognition-in-parkinsons-disease-benchmarking-eeg-source-functional-connectivity-and-structural-mri/