Use of wavelet transform coefficients for spike detection for a Robust Intracortical Brain Machine Interface

A common problem in Brain-Machine Interface (BMI) is the variations in neural signals over time, leading to significant decrease in decoding performance if the decoder is not re-trained. However, frequent re-training is not practical in real use case. In our work, we found that a temporally more robust system may be achieved through the use of wavelet transform in feature extraction. We used wavelet transform coefficients as means to detect spikes in neural recordings, in contrast to conventional amplitude threshold methods. Using offline data as the preliminary testbed, we showed that decoding based on firing rates determined from four levels of wavelet transform decomposition resulted in a decoder with 6–12% improvement in accuracy sustained over four weeks after training. This strategy suggests that wavelet transform coefficients for spike detection may be more temporally robust as features for decoding, and offers a good starting point for further improvements to tackle nonstationarities in BMI.

Recommended citation: G. C. Lee, C. Libedinsky, C. Guan, and R. So, “Use of wavelet transform coefficients for spike detection for a robust intracortical brain machine interface,” in 2017 8th International IEEE/EMBS Conference on Neural Engineering (NER). IEEE, 2017, pp. 540–543.