| ABSTRACT |
Emotional deficits detection is one of the key areas of research in neuropsychiatric disorders assessment and cognitive disorders. Many of the cognitive/neuropsychiatric disorder patients are highly subjected to emotional impairments. The present clinical assessment tool is mainly based on patient¿s oral/verbal response to estimate the emotional impairment level. This method is method does not give ground truth results of actual impairment status since it¿s subjective. Hence, development of automated emotional impairment detection method is highly inevitable for better clinical assessment by psychologist/neurologist/psycho physiologist. This present work aims to develop an emotional deficits impairment tool for Parkinson¿s disease (PD) patients using Electroencephalogram (EEG) signals. EEG is used to record the ongoing activity of central nervous system (CNS), reflect the underlying true emotional state of a person. In particular, non-motor symptoms, including disruptions in emotion information processing, have been found in over 50% of newly diagnosed PD patients. Cumulating evidence indicates that individuals with PD have deficits in recognizing emotions from prosody, facial expressions, and event related potential (ERP), although not all findings have been consistent. This product is aim to work on detecting emotional impairments (happiness, sadness, fear, anger, disgust, and surprise) of PD patients compare with healthy controls from EEG signals using machine learning approach. A set of EEG data has been acquired from 20 PD and 20 normal control (NC) people from Hospital UKM, Malaysia. We have proposed a data acquisition protocol to induce the above mentioned emotions using audio-visual stimuli and EEG signals of the patients were acquired using 14 channel EEG data acquisition system at a sampling frequency of 256 Hz. The acquired signal are preprocessed (removal of noises and artifacts from EEG signals) and a set of statistical features (higher order spectral (HOS) information) were derived and mapped to corresponding emotions on PD and NC using different machine learning approaches. From the experimental result, we found that (a) bispectrum feature is superior to other three kinds of features; (b) high frequency bands (alpha, beta and gamma) play a more role in emotion activities than low frequency bands (delta and theta) in both the groups and; (c) the trajectory of emotion changes can be visualized by reducing subject-independent features with manifold learning.
|
