Objective: To evaluate the diagnostic value of quantitative electroencephalogram (qEEG) and structure Magnetic resonance (sMR) in identifying mild cognitive impairment in Parkinson’s disease (PD-MCI) via machine learning model.

Background: Many single-marker based machine learning models are used to identify mild cognitive impairment (MCI) from patients with Parkinson’s disease (PD), however, we still don’t know the diagnostic value of multimodal markers.

Method: Patients with PD-MCI and nondemented PD (PD-ND) were recruited, propensity score was used to match the two groups to control the confounding factor. Overall, qEEG features of the power spectrum, sample entropy, and sMR features of cerebral cortex were evaluated, machine learning models of support vector machine (SVM), decision tree (DT), and logical regression (LR) were established to compare the diagnostic value. Feature selection model with qEEG and sMR features was established to sort the vital characteristics

Results: A total of 23 pairs of PD-MCI and PD-ND were recruited after propensity score matching. QEEG and sMR analysis revealed a statistically significant increase in the power spectrum of PD-MCI in 7 leads and waves and changes in 8 cortical regions. In all the models of SVM, DT, and LR, the qEEG features have better performance in prediction accuracy and AUC compared to the sMR features. The sequence showed that the θ wave in P3 of the qEEG feature had a marked impact on the model for classification.

Conclusion: The diagnostic value of quantitative electroencephalogram (EEG) features combining the power spectrum, and sample entropy was higher than that of the cortical structural imaging features in identifying PD-MCI using machine learning. “Composite marker” may be valuable for the individualized diagnosis of PD-MCI.

« Back to MDS Virtual Congress 2020

MDS Abstracts - https://www.mdsabstracts.org/abstract/multimodal-eeg-mri-based-machine-learning-model-in-identifying-parkinsons-disease-with-mild-cognitive-impairment/