A Morphological Peak-Detector for Single-Unit Neural Recording Acquisition Systems

Abstract

The increasing demand for measurement systems in neuroscience with the ability to acquire signals at neuron-level resolution has led to the development of techniques based on innovative organic biosensors. These single-unit extracellular neural recording systems provide useful information on neural behavior in terms of the action potential (AP) firing mechanism. In this work, we propose a processing technique specifically designed to accurately and reliably extract single-cell APs generated by free and attached membranes from signals acquired by extracellular recording devices. The simultaneous presence of overlapped APs, due to the nonideal coupling between cell and surface device, involves a distortion of the signal acquired by the device. Such distortion makes APs detection challenging. The proposed approach consists of a morphological peak detector characterized by high selectivity and based on three stages: a denoising phase for reducing wideband noise performed by the empirical mode decomposition (EMD) technique, and a threshold-based classifier for the identification of all the possible peaks that could correspond to an AP and a morphological classifier based on the support vector machine technique that improves the performances of the whole peak detection algorithm increasing its selectivity. The proposed method was designed on simulated data obtained by taking cell-to-cell variability into account in order to adapt the proposed method to the heterogeneity of the biological reality rather than the average behavior of the cells. This point is crucial to design an approach suitable for applications <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in vivo</i> . Finally, the morphological classifier was tested on simulated and semisimulated data obtained from experimental acquisitions measured from the axons of a giant squid in response to current stimulation, achieving F1-score > 89% for both scenarios.