DFC-SNN: A New Approach for the Recognition of Brain States by Fusing Brain Dynamics and Spiking Neural Network

Rich dynamics are the intrinsic features in brain activity, which could be characterized as sequences of multiple spatio-temporal activity events. However, how to efficiently apply brain dynamics for the recognition of brain states is still unclear and need more investigations. The spiking neural network (SNN) is a promising model with better performance in the pattern recognition of event streams. Thus, this paper proposes an algorithm framework for brain states recognition by fusing brain dynamics and SNN, where the brain dynamics are estimated as the dynamic functional connectivity (DFC) matrices. Through applying the DFC-SNN algorithm to the dataset of resting state electroencephalograph signals from healthy subjects and obsessive compulsive disorder patients, we observed that this algorithm was competent to perform the recognition of pathological brain states. It showed that the convergence of SNN model was rapid within less than 20 epochs, and the accuracy was 87.5% under optimal threshold of DFC matrices. In summary, this is the first attempting for the recognition of brain states via the aspect of brain dynamics. The algorithm framework would be beneficial for the applications of SNN model in the field of neuroscience.