A Bidirectional Neural Interface With Direct On-Device Neuromorphic Decoding for Closed-Loop Optogenetics

Area of Science:

• Neuroscience
• Biomedical Engineering
• Embedded Systems

Background:

• Closed-loop (CL) neuromodulation requires bidirectional interfaces for simultaneous neural monitoring and responsive stimulation.
• Implementing computationally intensive neural decoders in compact, wireless systems for freely moving animals is a significant challenge.
• Existing platforms often rely on tethered setups and external processors, limiting experimental freedom.

Purpose of the Study:

• To design and optimize a neural decoder integrated into a bidirectional wireless system for CL optogenetic experiments.
• To develop a resource-efficient platform capable of real-time neural decoding and stimulation control.
• To enable untethered, real-time neuroscience research in freely moving subjects.

Main Methods:

• Integrated a 32-channel electrophysiological recording system with neuromorphic feature extraction and dimensionality reduction (PCA) on a Spartan-6 FPGA.
• Implemented a nonlinear support vector machine (NL-SVM) decoder using spike-count features and leaky integrators, with k-means clustering for model size reduction.
• Validated decoder performance using non-human primate and rat motor cortex datasets, comparing accuracy against CNNs and Wiener filters.
• Demonstrated the system *in vivo* with wireless closed-loop optogenetic stimulation in rats.

Main Results:

• The integrated NL-SVM decoder achieved high accuracy (R²=0.85), comparable to CNNs (R²=0.87) and superior to Wiener filters (R²=0.81).
• The system demonstrated sub-millisecond inference with minimal memory and power requirements due to efficient feature extraction and dimensionality reduction.
• *In vivo* experiments showed successful wireless closed-loop optogenetic stimulation in rats with a variance accounted for (VAF) of 0.9148.
• The platform is fully self-contained, versatile, and resource-efficient for real-time applications.

Conclusions:

• This work introduces a novel, resource-efficient, and fully self-contained wireless platform for real-time closed-loop neuromodulation.
• The integrated neural decoder overcomes limitations of tethered systems, enabling greater experimental flexibility for neuroscience research.
• The developed system facilitates advanced untethered studies in freely moving animals, paving the way for more complex brain-machine interfaces.