Robust intra-trial t-SNE Brain State Detection for Multiband fMRI Data

T-distributed stochastic neighbor embedding (tSNE) based postprocessing techniques have been used for visualization of temporal information from functional Magnetic Resonance imaging (fMRI) data. With advancements in multiband pulse sequence, finer temporal sampling can be achieved during fMRI data acquisition. The synergy between fast sampled fMRI data and t-SNE can be used to detect intra-trial brain state changes which might correspond to mind wandering or task switching. In this article we propose an intra-trial brains state detection algorithm suitable for multiband fMRI data which is less susceptible to noise and random fluctuations. The proposed approach was tested on a gambling decision-making task to identify the instance of brain state change during the task. Functional connectivity analysis revealed that the switching from decision making phase to reward anticipation phase caused a brain state change which was reliably and repeatedly detected by the proposed algorithm. The proposed technique can be used to validate task switching or quantify the degree of involvement during task trials.