An fNIRs study to classify stages of learning from visual stimuli using prefrontal hemodynamics

Abstract

A wide range of research exists on fMRI imaging and psychological assessment based memory and/or learning studies. However, absence of literature is observed in fNIRs based memory and learning research. This paper provides a novel study of prefrontal hemodynamic changes of subjects engaged in multiple trial paired-associate learning. The direct measure of prefrontal hemodynamic is collected by fNIRs machine. The raw signals are pre-processed (to filter out artifacts) to extract 144 features for each feature pool which are reduced to 36 using principal component analysis (PCA). From three pools of features, the most relevant feature pool is sorted out considering algorithms' classification performance. Learning stages are classified from ‘ZERO’ learning using three conventional classifiers (RBF-SVM, LSVM and LDA). Experimental analysis revels RBF-SVM algorithm has the highest performance in classification of learning trials which reaches over 93%. Analysis of hemodynamic features shows greater total hemoglobin load in orbitofrontal (OFC) and medial prefrontal cortex (mPFC) in initial learning trials which shits to dorsolateral (DLPFC) and ventrolateral prefrontal cortex (VLPFC) areas when learning is complete. We also observe the engagement of working memory in initial learning stages. This findings also can be useful to justify low learning ability among individuals with neurovascular deficits.