♾ Neurophysiologically Realistic Environment for Comparing Adaptive Deep Brain Stimulation Algorithms in Parkinson's Disease

Our team’s latest paper, presented at the top-tier KDD 2025 (A*) conference, focuses on Parkinson’s disease and developing stimulation algorithms to mitigate symptoms such as tremor, bradykinesia, and freezing of gait.
Parkinson’s symptoms can be effectively reduced with deep brain stimulation (DBS) — an electrode implanted in the basal ganglia delivering ~150 Hz pulses.
Adaptive DBS (aDBS) goes a step further by adjusting the stimulation pattern in real time for greater effectiveness 🌝

Insights:
📍 Current state-of-the-art aDBS algorithms are overly simple. Novel machine learning methods are rarely applied, and clinicians seldom use aDBS in practice due to the lack of testing tools.
📍 The dynamics of pathological neural activity can be approximated by a simple coupled-oscillator model — the Kuramoto model — which can be used to train machine-learning-based aDBS algorithms.
📍 Brain activity is highly non-linear, stochastic, and non-stationary — advanced ML methods must account for this complexity.
📍 Reinforcement learning (RL), which optimizes a reward function through interaction with the environment, is well-suited for mitigating PD symptoms. To support RL research, we developed our model as an easy-to-use Gymnasium environment, called DBS-Gym.

🧠 Thoughts:
Although Parkinson’s disease is not a normal part of aging, but the same tools could be applied to modulate brain dynamics in healthy aging. Brain stimulation is a broad and promising area of neuroscience that can potentially support cognitive function in the aging brain.

📜paper
🖥code
💬community