Objective: To determine whether preoperative electroencephalography (EEG) combined with deep learning can predict cognitive decline 1 year after subthalamic deep brain stimulation (STN-DBS) in patients with Parkinson’s disease (PD).

Background: Cognitive decline, particularly in verbal fluency and executive function, can occur after STN-DBS. Older age and lower preoperative cognitive performance have been suggested as risk factors. However, accurately predicting postoperative decline at the individual level remains challenging. Recently, preoperative EEG-based machine learning models have shown potential for predicting cognitive outcomes after STN-DBS [1], whereas the utility of deep learning models has not yet been established.

Method: We retrospectively collected preoperative and 1-year postoperative neuropsychological evaluations, and preoperative eyes-closed resting-state EEG data from patients with PD who underwent bilateral STN-DBS. Patients with postoperative intracranial hemorrhage, unstable psychiatric symptoms at the time of evaluation, or insufficient data were excluded from the study. Neuropsychological assessments included global cognition, attention and working memory, executive function, memory, and visuospatial function assessments. Postoperative cognitive decline was determined based on the MDS Level I criteria [2]. EEGNet, a convolutional neural network for EEG analysis, was trained using the epoched EEG data and postoperative cognitive outcomes [3]. Binary classification was performed for each EEG epoch, and the mean predicted class label (0 or 1) across epochs was calculated as the patient-level prediction score. Model performance was evaluated using nested leave-one-out cross-validation with hyperparameters optimized through an inner 6-fold cross-validation.

Results: Twenty-five patients, of whom 10 experienced cognitive decline, were included in the analysis. Based on the patient-level prediction scores, the model achieved an area under the receiver operating characteristic curve of 0.83.

Conclusion: Our study suggests that a preoperative EEG-based deep learning model can predict cognitive outcomes after STN-DBS in patients with PD. These findings may contribute to the identification of neurophysiological biomarkers for predicting postoperative cognitive decline.

References: 1 Geraedts, V. J., Koch, M., Kuiper, R., Kefalas, M., Bäck, T. H. W., van Hilten, J. J., Wang, H., Middelkoop, H. A. M., van der Gaag, N. A., Contarino, M. F., & Tannemaat, M. R. (2021). Preoperative electroencephalography-based machine learning predicts cognitive deterioration after subthalamic deep brain stimulation. Movement Disorders, 36(10), 2324–2334. https://doi.org/10.1002/mds.28661
2 Litvan, I., Goldman, J. G., Tröster, A. I., Schmand, B. A., Weintraub, D., Petersen, R. C., Mollenhauer, B., Adler, C. H., Marder, K., Williams-Gray, C. H., Aarsland, D., Kulisevsky, J., Rodriguez-Oroz, M. C., Burn, D. J., Barker, R. A., & Emre, M. (2012). Diagnostic criteria for mild cognitive impairment in Parkinson’s disease: Movement Disorder Society Task Force guidelines. Movement Disorders, 27(3), 349–356. https://doi.org/10.1002/mds.24893
3 Lawhern, V. J., Solon, A. J., Waytowich, N. R., Gordon, S. M., Hung, C. P., & Lance, B. J. (2018). EEGNet: A compact convolutional neural network for EEG-based brain-computer interfaces. Journal of Neural Engineering, 15(5), 056013. https://doi.org/10.1088/1741-2552/aace8c

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/prediction-of-cognitive-decline-after-stn-dbs-in-parkinsons-disease-using-preoperative-electroencephalography-and-a-deep-learning-model/