Objective: This study aims to evaluate the wider frequency range (1-200 Hz) LFPs as the novel neural biomarkers for adaptive DBS. In parkinsonism rats, we evaluated the difference between the LFPs of beta-range versus broader spectrum as the biomarkers in parkinsonism rats under off- or on-levodopa conditions. Different machine learning algorithms were applied and accuracy of classifications of LFPs were evaluated and compared.

Background: Despite substantial motor improvement, conventional deep brain stimulation (DBS) limit its efficacy with the fixed stimulation parameters in the face of dynamic neural fluctuations. Adaptive DBS offers a promising strategy for refining DBS. However, current adaptive system utilizing only beta range (13-30 Hz) of local field potentials (LFPs) as the sole neural biomarker in PD.

Method: Twelve adult male rats with unilateral 6-OHDA lesions were divided into two groups: one receiving normal saline (NS) and the other levodopa (LD). Rats were monitored in an open field arena, where simultaneous video tracking and LFPs recordings were conducted. LFPs signals were acquired in two frequency ranges: the beta band (13–30 Hz), and the broad spectrum (1–200 Hz). Three machine learning algorithms—Support Vector Machine (SVM), Random Forests (RF), and XGBoost (XGB)—were applied to classify treatment conditions, with classification accuracy compared.

Results: Behaviorally, levodopa-treated rats demonstrated a significant increase in total distance traveled in the open field test and exhibited higher abnormal involuntary movement (AIMs) scores. Analysis of LFPs showed the area under the receiver operating characteristics curve (ROC-AUC) of the three machine learning algorithms to classify the status of parkinsonian rats receiving NS or LD was: (1) classifying by the beta band LFPs— SVM 0.836, RF 0.877, XGB 0.863; (2) classifying by the broad spectrum LFPs— SVM 0.863, RF 0.904, XGB 0.843. Machine learning classifier, particularly Random Forests, achieved higher ROC-AUC using the broad spectrum data than the beta band data, indicating that the broad spectrum may be a good LFPs biomarker for levodopa-induced changes and can well be used by the machine-learning algorithms.

Conclusion: These findings suggest that levodopa significantly modulates neural activity and behavior, with broad spectrum or beta band LFPs features serving as robust predictors of treatment response for adaptive DBS in PD.

Accuracy of different machine learning algorithms

References: 1. Amoozegar, S., Pooyan, M. & Roghani, M. Identification of effective features of LFP signal for making closed-loop deep brain stimulation in parkinsonian rats. Méd. Biol. Eng. Comput. 60, 135–149 (2022).
2. Oliveira, A. M. et al. Machine learning for adaptive deep brain stimulation in Parkinson’s disease: closing the loop. J. Neurol. 270, 5313–5326 (2023).

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/broad-spectrum-local-field-potential-activities-as-a-novel-biomarker-for-adaptive-deep-brain-stimulation/