Selective peripheral nerve recordings from nerve cuff electrodes using convolutional neural networks.

Area of Science:

• Neuroscience
• Bioelectronics
• Machine Learning

Background:

• Peripheral nervous system recording is crucial for bioelectronics but faces chronic challenges in humans.
• Multi-contact nerve cuff electrodes offer improved selectivity for nerve signal recording.
• Convolutional Neural Networks (CNNs) can analyze complex spatiotemporal patterns in neural data.

Purpose of the Study:

• To develop a CNN-based method for associating peripheral nerve recordings with specific neural pathways.
• To utilize spatiotemporal patterns from multi-contact nerve cuff recordings for neural signal decoding.
• To assess the CNN's performance in classifying compound action potentials (CAPs) and predicting physiological measures.

Main Methods:

• Implanted 56-channel nerve cuff electrodes on the sciatic nerve of nine rats.
• Evoked afferent activity in different fascicles (tibial, peroneal, sural) using mechanical stimuli.
• Applied a CNN to analyze CAPs and a recurrent neural network to predict joint angles.

Main Results:

• Achieved high classification accuracy and F1-score ([Formula: see text] and 0.747 ± 0.114) with the optimized CNN.
• Demonstrated accurate tracking of ankle joint angles with a mean Pearson correlation coefficient of [Formula: see text].
• Validated the ability of CAP-based classification to decode meaningful physiological information.

Conclusions:

• The proposed CNN method enables accurate classification of peripheral nerve signals.
• This approach facilitates the tracking of physiological measures like joint angles from neural recordings.
• Results show promise for developing more effective and intuitive neuroprosthetic systems.