A Scalable Framework for Closed-Loop Neuromodulation with Deep Learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Conventional closed-loop neuromodulation often relies on biased biomarker detection and stimulation application.
• There is a need for flexible, data-driven, and automated systems for optimizing neuromodulation interventions.
• Existing methods may lack scalability and require personalized performance data.

Approach:

• Developed a novel deep learning framework for automated closed-loop neuromodulation.
• The framework is data-driven, scalable, and agnostic to stimulation technology (tACS, tDCS, tFUS, TMS).
• Utilized identified periods of responsiveness to trigger interventions, without needing personalized ground-truth data.

Key Points:

• Demonstrated the framework using the open-sourced GX dataset with concurrent physiological (ECG, EOG) and neuronal (EEG) measures.
• The system achieved 88.26% correct intervention applications, with 11.25% of trials showing potential for mistimed stimulation.
• The deep learning approach (Convolutional Neural Networks - CNNs) leverages identified periods of responsiveness.

Conclusions:

• The proposed framework enables flexible, automated, and multimodal closed-loop neuromodulation.
• It demonstrates adaptability and feasibility for both clinical and nonclinical applications.
• This unifying approach supports diverse datasets and emerging stimulation technologies.