An Online Learning Framework for Neural Decoding in Embedded Neuromodulation Systems.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Advancements in brain-computer interfaces (BCIs) enable real-time neural signal decoding for adaptive closed-loop neuromodulation.
• Current neuromodulation systems often require high-power computational platforms, hindering portability and real-time applications.

Purpose of the Study:

• To introduce RONDO (Recursive Online Neural DecOding), a resource-efficient framework for real-time neural decoding.
• To enable adaptive closed-loop neuromodulation in resource-limited settings.

Main Methods:

• Developed RONDO, a framework utilizing dynamic updating schemes in online learning with recurrent neural networks (RNNs).
• Supported various RNN architectures (simple RNN, LSTM, GRU) for flexible adaptation.
• Evaluated RONDO's performance against offline learning methods.

Main Results:

• RONDO's adaptive model updating improved neural decoding accuracy by 35% to 45% compared to offline learning.
• The framework operated within real-time constraints of neuroimaging devices, eliminating the need for cloud or high-performance computing.
• Dynamic updating ensured high accuracy with minimal updates, enhancing energy efficiency and robustness.

Conclusions:

• RONDO offers a scalable, adaptive, and energy-efficient solution for real-time closed-loop neuromodulation, independent of cloud computing.
• Its flexibility positions it as a valuable tool for clinical and research applications, advancing personalized neurostimulation and adaptive BCIs.