Characterization of individual retinal ganglion cell responses using K-means clustering method

The information to be transmitted along the nervous system is encoded with the rate of fire of the neurons expressing the number of action potentials in a temporal range. Findings from experimental studies in the development of visual prosthetic systems, as a neuroprosthetic device, are of critical importance. The determination of the various working intervals required for the development of electronic units is carried out initially by experiments using animal subjects following acute experiments on human subjects. Current implantable retinal implants generate large volume data which is required to be sorted to simultaneously provide strong neural control signals from each electrode. Spike sorting refers to the process that raw electrophysiological data is transferred to interpretable presentation of neural spikes. The first step in the analysis of neural activity for sorting recorded in vitro from retinal tissue is the detection and isolation of the action potentials to be used for further processing stages. It provides to clarify the understanding of the retinal response to electrical stimulation. In this study, spike activity is detected and isolated using the neural activity recorded from in vitro retina experiment. Next, data is pre-processed and sorted using k-means clustering method to distinguish real spikes. Moreover, Davies-Bouldin index is used to determine optimal separation of spike activity, which efficiently results 4 clusters. One of these clusters refers artifacts caused by electrical stimulation and others are related to real spikes with different properties. It is concluded that neural activity could be successfully sorted and more efficient approaches developed.