Neural network-based classification of ENG recordings in response to naturally evoked stimulation

The paper evaluates two different neural network strategies for the classification of electroneurographic (ENG) recordings obtained in response to mechanical stimulations of a rat paw available from an open-access dataset. The first strategy is based on spiking neural networks (SNNs), a new class of artificial neural networks inspired by the information processing solutions of biological neurons. SNNs are considered to be promising for the classification of complex space-time models. The second strategy relies on convolutional neural networks (CNNs), a well-established tool for processing structured arrays of data such as images or space-time data. The existing effects of power line noise and other distortions generated by the non-idealities of the acquisition system (e.g., timing jitter) are removed by a suitable pre-processing. The accuracy and F1-score achieved by the SNN and the CNN are reported for a multi-class problem aiming to separate different types of somatosensory stimuli (i.e., nociception, plantar flexion, dorsiflexion, and touch, which are the mechanical stimulations of the dataset) and their different intensities.