Sparse fNIRS Feature Estimation via Unsupervised Learning for Mental Workload Classification
Recent studies have demonstrated that functional near-infrared spectroscopy (fNIRS) is a potential non-invasive system for human mental workload (MWL) evaluation in both off-line and on-line manners. While most of the studies have been based on supervised classification of different MWL levels, which requires much effort to collect labeled training data, investigation on unlabeled data seems to be more promising. In this paper, we developed unsupervised learning and classification techniques of fNIRS parameters to support human workload classification. In the experimental setup, five subjects engaged in ten-loop memorizing tasks that were devised into two MWL levels while fNIRS signals were being monitored over their frontal lobes. Independent component analysis (ICA) was applied on a set of unlabeled random fNIRS data to extract the basis and sparse functions. Then two-dimensional convolutional matrices, which were constructed as sets of convolutional coefficients of fNIRS signal with learned basis functions, were implemented as the inputs for MWL classification using convolutional neural network classifier. Study of generalized linear model demonstrated that basis functions extracted using ICA is more effective when illustrating the activation regions over measuring cortex than using the modeled hemodynamic response functions. Besides, ICA basis function demonstrates the sparseness so that it is superior to basis functions learned by the conventional method of principle component analysis (PCA) in mental classification and shows its potential for further study of fNIRS signals based on their hidden basis functions.
KeywordsFunctional near-infrared spectroscopy (fNIRS) Mental work-load (MWL) Unsupervised learning Spare features
- 2.Penny, W., Fladin, G., Trujillo-Barreto, N.: Chapter 25-Spatio-temporal models for fMRI. In: Statistical Parametric Mapping. Academic Press, London (2007)Google Scholar
- 3.John, R.H., Wilson, W.P.: EEG and Evoked Potentials in Psychiatry and Behavior Neurology. Butterworth-Heinemann (1999)Google Scholar
- 8.Cohen, L.: Convolution, filtering, linear systems, the Wiener-Khinchin theorem: generalizations. In: Proceedings of SPIE 1770, Advanced Signal Processing Algorithms, Architectures, and Implementations III, 378–393 (1992)Google Scholar
- 17.Sainath, T.N., Kingsbury, B., Saon, G., Soltau, H., Mohamed, A., Dahl, G., et al.: Deep Convolutional Neural Networks for Large-scale Speech Tasks. Neural Networks, 64, 39–48 (2015)Google Scholar