BIOMEDICAL SIGNAL PROCESSING AND CONTROL, vol.79, no.2, pp.1-10, 2023 (SCI-Expanded)
Parkinson’s disease (PD) is an incurable nervous system disease that
affects millions of people all around the world. The loss of smell is
one of the first symptoms that come into prominence in the early
diagnosis of PD. The main motivation of this study is to provide a more
accurate diagnosis in the early period of the disease using chemosensory
electroencephalography (EEG) signals, which are difficult to study and
also less studied. For this purpose, we proposed a hybrid feature
extraction method called EEMD_VAR that combines Ensemble Empirical Mode
Decomposition (EEMD) and Vector Autoregressive Model (VAR). In contrast
to conventional feature extraction methods, the proposed method is to
prevent arbitrary selection of features and to determine the number of
features. The pre-processed EEG signals were decomposed using EEMD and
the obtained intrinsic mode functions (IMFs) used as independent
variables in VAR. The coefficients of the VAR model were employed as
features in frequently used supervised classification algorithms. The
performance metrics of the EEMD_VAR were compared to the performance
metrics of the autoregressive (AR) model and Hjorth parameters. The
maximum classification accuracy of the proposed method was 100% using
artificial neural networks (ANN) in C2 electrode, while the AR method
and Hjorth parameters only obtained a maximum of 72%. The other metrics
also corroborate the proposed method's ability to perform well in the
classification. In addition, the higher results from right side
electrodes may lead to the conclusion that the right side of the brain
is more sensitive to odor stimuli.