2021/01/18 by Talha Siddique, Siddique, Talha, Md Shaad Mahmud +1
Neuroscience · Medicine · #EEG and Brain-Computer Interfaces #ECG Monitoring and Analysis #Functional Brain Connectivity Studies
paper · pdf · doi:10.48550/arxiv.2101.07128
Functional Near-Infrared Spectroscopy (fNIRS) is a non-invasive form of\nBrain-Computer Interface (BCI). It is used for the imaging of brain\nhemodynamics and has gained popularity due to the certain pros it poses over\nother similar technologies. The overall functionalities encompass the capture,\nprocessing and classification of brain signals. Since hemodynamic responses are\ncontaminated by physiological noises, several methods have been implemented in\nthe past literature to classify the responses in focus from the unwanted ones.\nHowever, the methods, thus far does not take into consideration the uncertainty\nin the data or model parameters. In this paper, we use a Bayesian Neural\nNetwork (BNN) to carry out a binary classification on an open-access dataset,\nconsisting of unilateral finger tapping (left- and right-hand finger tapping).\nA BNN uses Bayesian statistics to assign a probability distribution to the\nnetwork weights instead of a point estimate. In this way, it takes data and\nmodel uncertainty into consideration while carrying out the classification. We\nused Variational Inference (VI) to train our model. Our model produced an\noverall classification accuracy of 86.44% over 30 volunteers. We illustrated\nhow the evidence lower bound (ELBO) function of the model converges over\niterations. We further illustrated the uncertainty that is inherent during the\nsampling of the posterior distribution of the weights. We also generated a ROC\ncurve for our BNN classifier using test data from a single volunteer and our\nmodel has an AUC score of 0.855.\n