Comparison of Univariate and Multivariate EEG Signals Analysis for the Prediction of Freezing of Gait in Parkinson’s Disease Patients — The Association Specialists
  1. Schedule
  2. Posters - Tuesday
  3. Comparison of Univariate and Multivariate EEG Signals Analysis for the Prediction of Freezing of Gait in Parkinson’s Disease Patients

Comparison of Univariate and Multivariate EEG Signals Analysis for the Prediction of Freezing of Gait in Parkinson’s Disease Patients (441)

Tuan Nghia Nguyen 1 , Aluysius Maria Ardi Handojoseno 1 , Hung T Nguyen 1 , James M Shine 2 , Simon JG Lewis 2 , Yvonne Tran 1
  1. University of Technology, Sydney, Ultimo, NSW, Australia
  2. Parkinson’s Disease Research Clinic, Brain and Mind Research Institute, University of Sydney, Sydney, NSW, Australia

Freezing of Gait (FOG), a gait disturbance in Parkinson’s disease (PD) patients, is defined as a ‘‘brief, episodic absence or marked reduction of forward progression of the feet despite the intention to walk.’’ Recent neuroimaging studies suggested that the pathophysiology underlying FOG may be detectable from surface scalp electroencephalography (EEG). In this study, we investigate the performance of univariate and multivariate EEG measurements in predicting FOG episodes in PD patients. EEG signals from four cortical regions were recorded in sixteen PD patients with significant FOG (age between 56 and 84 years) during timed up-and-go tasks (TUG). Classification was determined  using Multilayer Perceptron Neural Network to discriminate EEG signals of transition to FOG from normal walking using univariate and multivariate Fourier transform based features.  Univariate measurement based on power spectral analysis captured significant differences between EEG signals of normal walking and transition to FOG with the overall accuracy classification rate of 80.9 % and 77.6 %  for 11 within-group patients and 5 out-group patients (whose data were not included in training or validation data), respectively. Univariate measurements of centroid frequency revealed the important role of beta sub-band in FOG which decreased significantly in all locations of electrodes during transition. The interdependence measurement using cross power spectral analysis slightly increased this classification accuracy to 81.2 % and 81.8 % for within-group patients and out-group patients, respectively. The result of this study shows that Freezing of Gait is related to changes across local networks, which can be detected by EEG more accurately using multivariate analysis compared to a univariate analysis. The shift in the beta power observed with the transition to FOG may be related to motor impairment whereas significant increases in the power spectral in frontal and central areas (especially in alpha and theta sub-bands) may reflect non-motor aspects of freezing.