Researchers have widely used extracellular recordings as a technique of paramount importance due to its wide usage in cognitive studies, health technologies, and prosthetics and orthotics research. To extract the required information from this technique, a critical and crucial step, called spike sorting, must be performed on the recorded signal. By this method, it is possible to analyze a single neuron (single-unit activity) and investigate its specifications, such as the firing rates and the number of action potentials (spikes) of an individual neuron. Here we introduce a novel idea of a user-friendly interactive, offline, and unsupervised algorithm called ION-Decoding. This platform extracts and aligns the spikes using a high-resolution alignment method, and the clusters can be atomically identified and manually edited. The entire procedure is performed using the minimum number of adjustable parameters, and cluster merging was performed in a smart, intuitive way. The ION-Decoding algorithm was evaluated by a benchmark dataset, including 95 simulations of two to twenty neurons from 10 minutes simulated extracellular recordings. There was not any significant relationship between the number of missed clusters with the quality of the signal (i.e., the signal-to-noise ratio (SNR)) by controlling the number of neurons in each signal (p_value=0.103). Moreover, the number of extra clusters was not significantly dependent on the parameter SNR (p_value=0.400). The accuracy of the classification method was significantly associated with the decomposability index (DI) (p_value<0.001). A number of 77% of the neurons with the DI higher than 20 had the classification accuracy higher than 80%. The ION-Decoder significantly outperformed Wave Clus in terms of the number of hits (p_value=0.017). However, The Wave Clus algorithm significantly outperformed the INO Decoder when the false-positive error (FP) was considered (p_value=0.001). The ION-Decoding is thus a promising single-channel spike sorting algorithm. However, our future focuses on the improvement of the cluster representative identification and FP error reduction.
bioRxiv Subject Collection: Neuroscience