Peripheral Nerve Activation Evokes Machine-Learnable Signals in the Dorsal Column Nuclei

Area of Science:

• Neuroscience
• Somatosensory system research
• Computational neuroscience

Background:

• The brainstem dorsal column nuclei (DCN) are crucial for processing tactile and proprioceptive information.
• Understanding how ascending somatosensory information is encoded within the DCN is limited.
• Investigating signal features for predicting nerve origin is vital for mapping somatosensory pathways.

Purpose of the Study:

• To assess the predictive power of high-frequency (HF) and low-frequency (LF) DCN signal features (SFs) for identifying the stimulated peripheral nerve.
• To evaluate the robustness of DCN SFs and map their information content across the brainstem surface.
• To explore functional asymmetry within the DCN.

Main Methods:

• Recorded DCN surface potentials in urethane-anesthetized Wistar rats during sural and peroneal nerve stimulation.
• Extracted five salient SFs from a seven-electrode array.
• Applied a machine learning approach, specifically a supervised back-propagation artificial neural network (ANN), to analyze signal features and electrode positions.

Main Results:

• An ANN achieved up to 96.8 ± 0.8% accuracy in predicting the nerve of origin based on DCN SFs.
• High prediction accuracy was maintained even after reducing input features, highlighting the efficacy of specific HF and LF features.
• Feature-learnability demonstrated that DCN signals can differentiate bilateral nerve stimulation from a single midline electrode, indicating functional asymmetry.

Conclusions:

• Novel method developed for mapping information content across the DCN surface.
• DCN signal features are robust and effective for classifying peripheral nerve activation.
• The DCN exhibits functional asymmetry, challenging established notions of sub-cortical somatotopic symmetry.