Towards zero-latency neurofeedback

Neurofeedback (NFB) is a real-time paradigm, where subjects monitor their own brain activity presented to them via one of the sensory modalities: visual, auditory or tactile. NFB has been proposed as an approach to treat neurological conditions and augment brain functions. In many applications, especially in the automatic learning scenario it is important to decrease NFB latency, so that appropriate brain mechanisms can be efficiently engaged. To this end, we propose a novel algorithm that significantly reduces feedback signal presentation in the electroencephalographic (EEG) NFB paradigm. The algorithm is based on the least squares optimization of the finite impulse response (FIR) filter weights and analytic signal reconstruction. In this approach, the trade-off between NFB latency and the accuracy of EEG envelope estimation can be achieved depending on the application needs. Moreover, the algorithm allows to implement predictive NFB by setting latency to negative values while maintaining acceptable envelope estimation accuracy. As such, our algorithm offers significant improvements in cases where subjects need to detect neural events as soon as possible and even in advance.