A lightweight learning-based decoding algorithm for intraneural vagus nerve activity classification in pigs.

Area of Science:

• Bioelectronic Medicine
• Neuroscience
• Biomedical Engineering

Background:

• Closed-loop neuromodulation of the autonomic nervous system (ANS) is promising for treating disorders.
• Computational demands of current decoding methods hinder real-time adaptation for clinical applications, particularly in cardiovascular and respiratory diseases.

Purpose of the Study:

• To develop a lightweight, learning-based decoder for real-time classification of cardiovascular and respiratory challenges from vagus nerve (VN) signals.
• To optimize decoder performance for computational efficiency and accuracy suitable for closed-loop neuromodulation.

Main Methods:

• Intraneural electrodes were used to acquire VN signals from anesthetized pigs.
• A decoder utilizing signal temporal windowing, handcrafted features, and a Random Forest (RF) model was developed.
• Temporal windows from 50 ms to 1 s were evaluated to mimic pseudo real-time conditions.

Main Results:

• High balanced accuracy (BA) was achieved across various temporal window durations.
• An optimal temporal window of 500 ms yielded >86% BA with a computational execution time of ~6.8 ms.
• The developed RF algorithm outperformed state-of-the-art methods in both BA and computational efficiency.

Conclusions:

• The lightweight decoder demonstrates feasibility for real-time decoding tasks using a single intraneural interface.
• This approach is a significant step towards implementing closed-loop neuromodulation protocols for selective VN modulation.
• The findings pave the way for advanced bioelectronic medicine therapies targeting ANS disorders.