Neural Spike Compression Using Feature Extraction and a Fuzzy C-Means Codebook

The live implantation of micro-electrode arrays allows the observation of populations of neural activity. There is a strong need for unsupervised adaptable neural spike extraction and spike-sorting methods, which would significantly reduce data throughput and allow for implantation and wireless transmission. We present a method for neural spike feature-extraction that involves the generation and use of a fuzzy c-means codebook. A dual-threshold spike detector isolates action potentials, and arithmetic circuits are used to extract features. Simulations show that only seven features are needed for effective template matching. Feature extraction improves neural signal compression by 87% compared to spike detection alone, possibly enabling wireless capability and hence implantation.