Investigating Feasibility of Blind Optimization of a Nonlinear Neural Spike Enhancement Filter

This study investigates feasibility of automated parameter selection for a noise-enhanced nonlinear filter (NF) designed for enhancement of intracortical neural spike recordings. Using standard particle swarm optimization (PSO) based on objective functions independent of true spike locations, NF and solver parameters impacting signal-to-noise ratio (SNR) enhancement are identified. The optimized NF, blinded to true spike locations, achieves over 0.86 dB SNR enhancements, slightly below the maximum attainable when optimizing based on SNR using true spike locations. Results on a publicly accessible synthetic extracellular dataset demonstrate that peak-to-root-mean-square (RMS) amplitude-based optimization, with manually selected threshold levels, yields a maximum average spike detection sensitivity of 99.81% and positive predictivity of 99.89%. These findings represent the first instance of blindly optimizing parameters for a noise-enhanced nonlinear neural spike-enhancement filter.