Closed-loop optical stimulation and recording system with GPU-based real-time spike sorting

Closed-loop brain computer interfaces are rapidly progressing due to their applications in fundamental neuroscience and prosthetics. For optogenetic experiments, the integration of optical stimulation and electrophysiological recordings is emerging as an imperative engineering research topic. Optical stimulation does not only bring the advantage of cell-type selectivity, but also provides an alternative solution to the electrical stimulation-induced artifacts, a challenge in closedloop architectures. A closed-loop system must identify the neuronal signals in real-time such that a strategy is selected immediately (within a few milliseconds) for delivering stimulation patterns. Real-time spike sorting poses important challenges especially when a large number of recording channels are involved. Here we present a prototype allowing simultaneous optical stimulation and electro-physiological recordings in a closed-loop manner. The prototype was implemented with online spike detection and classification capabilities for selective cell stimulation. Real-time spike sorting was achieved by computations with a high speed, low cost graphic processing unit (GPU). We have successfully demonstrated the closed-loop operation, i.e. optical stimulation in vivo based on spike detection from 8 tetrodes (32 channels). The performance of GPU computation in spike sorting for different channel numbers and signal lengths was also investigated.