Objective: Our aim is to build an AI-based algorithm engine to deliver brain sensing-based data insights from deep brain stimulation (DBS) therapy systems that are intended to reduce caregiver burden and facilitate curation of patient-specific treatment plans. This work highlights prototype classifiers developed to identify patterns seen in neural sensing data across large cohorts of patients.

Background: Parkinson’s Disease (PD) is a disorder characterized by a multitude of symptoms; the heterogeneous patient population makes treatment difficult and largely driven by trial and error. With the introduction of chronic neural sensing data, there is a need and opportunity to derive data driven insights that can inform therapy decisions.

Method: We conducted retrospective analyses to capture variation in symptom presentation and prototyped algorithms to classify trends of clinically relevant physiology from chronic local field potentials (LFP) and therapy use from system logs. The outputs of these algorithms are objective metrics to facilitate insight into therapy efficacy and patient-specific neurophysiology.

Results: The first classifier focuses on the well-documented relationship between symptom severity, sleep quality, and long-term beta oscillations; results show how we might classify 24-hour circadian rhythm “strength” and “abnormal sleep.” A second classifier characterizes modulations in LFP, which can lead to suboptimal therapy and increase follow ups; results identify periods of “anomalous” LFP behavior and distinguish transient and persistent changes. A third classifier evaluates DBS therapy usage, identifying changes in programming that may affect the LFP response and interpretability thereof; results illustrate therapy usage trends and provide tools to correlate response metrics to stimulation parameters. These classifiers have the potential to provide valuable insights to improve identification of patient-specific DBS programming.

Conclusion: As DBS therapy moves towards a data-rich and data-driven therapy model, new demands arise for clinically actionable insights. Here we have demonstrated proof-of-concept algorithms that leverage high-quality chronic brain sensing technology to highlight key physiological and therapeutic characteristics that may be useful in clinical decision making. Future embodiments aim to provide comprehensive, AI-guided therapy automation.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/using-chronic-dbs-brain-sensing-data-to-develop-an-ai-based-engine-for-clinical-insights/