Wavelet denoising and ANN/SVM decoding of a self-paced forelimb movement based on multi-unit intra-cortical signals in rats

Brain-computer interfacing (BCI) has the potential to rehabilitate patients suffering from severe motor deficits. A key element in the success of BCI systems is the availability of sophisticated signal processing techniques for decoding intended movements. The first objective of the present work was to investigate the usefulness of ANN/SVM techniques to decode a self-paced forelimb movement based on multi-unit intracortical recordings in rats. The second objective was to evaluate if wavelet denoising of the raw intra-cortical signals had a positive effect on the decoding. Four rats were first trained to hit a response paddle. Multiple recording sessions where performed after implanting a 16ch microwire array in the M1. An artificial neural network (ANN) and a support vector machine (SVM) were designed for decoding the intra-cortical signals, i.e. to detect if a `Hit´ had occurred. The results showed that denoising of the intra-cortical recordings consistently yielded a lower misclassification error rate (statistically different). We also found that the ANN and the SVM resulted in similar classification results (no significant difference). The classification error rate showed variability between days and between rats (average between 23% - 31% (ANN) and 23% - 30% (SVM)). Our results indicate that the non-linear decoding methods both perform equally well, and as such, the quality of intra-cortical signals is of high importance in the design of a robust and reliable BCI system.