Depression Detection Based on Geometrical Features Extracted from SODP Shape of EEG Signals and Binary PSO

Abstract

Late detection of depression is having detrimental consequences including suicide thus there is a serious need for an accurate computer-aided system for early diagnosis of depression. In this research, we suggested a novel strategy for the diagnosis of depression based on several geometric features derived from the Electroencephalography (EEG) signal shape of the second-order differential plot (SODP). First, various geometrical features of normal and depression EEG signals were derived from SODP including standard descriptors, a summation of the angles between consecutive vectors, a summation of distances to coordinate, a summation of the triangle area using three successive points, a summation of the shortest distance from each point relative to the 45-degree line, a summation of the centroids to centroid distance of successive triangles, central tendency measure and summation of successive vector lengths. Second, Binary Particle Swarm Optimization was utilized for the selection of suitable features. At last, the features were fed to support vector machine and k-nearest neighbor (KNN) classifiers for the identification of normal and depressed signals. The performance of the proposed framework was evaluated by the recorded bipolar EEG signals from 22 normal and 22 depressed subjects. The results provide an average classification accuracy of 98.79% with the KNN classifier using city-block distance in a ten-fold cross-validation strategy. The proposed system is accurate and can be used for the early diagnosis of depression. We showed that the proposed geometrical features are better than extracted features in the time, frequency, time-frequency domains as it helps in visual inspection and provide up to 17.56% improvement in classification accuracy in contrast to those features.

EEG-based spatio-temporal relation signatures for the diagnosis of depression and schizophrenia

“…This enhances the generalizability but increases the risk that features of the clinical EEG recording environment might impact results. Recently, progress was made to accurately classify healthy subjects and patients with depression and schizophrenia by automated geometrical feature extraction of EEG signals [64][65][66][67][68][69][70][71][72][73][74] . The accuracy of these methods to separate these groups is non-inferior to our presented method, although their algorithm uses multiple feature extractions and thus the physiological interpretation is difficult.…”

Recognizing seizure using Poincaré plot of EEG signals and graphical features in DWT domain

“…Summation of triangle area using consecutive points (STA) For every three consecutive points of the shape of the Poincaré plot in 2-D space, there is only one triangle. To measure the covered area by shape in 2-D space, STA is defi ned, which can be formulated as follows (28,30): where m is the number of Poincaré plot arrays (see Equation 3); [X (i) Y (i)], [X (i+1) Y (i+1)] and [X (i+2) Y (i+2)] indicate how to coordinate three consecutive points of Poincaré plot shape in a 2-D space. Figure 5 shows the STA as a geometrical feature.…”

“…The drawback of the RPS technique lies in the time-consuming nature of the procedure, particularly because the MI and FNN calculations are highly burdensome. Although SODP, unlike RPS, can illustrate the EEG signals without any parameter calculation, SODP illustrates the variability of the signal, not the signal itself (23,30). In other words, SODP cannot illustrate the complex nature of the signal (12).…”

“…These geometrical features also could decode the chaotic behavior of muscle diseases by plotting electromyogram (EMG) signals in a 2-D complex plane (36). In other works (28,30), the geometrical features of RPS and SODP of EEG signals have been used to detect depressed subjects. Also, the geometrical properties have shown an acceptable performance also in classifying RPS of EEG signals in normal individuals and those suffering from schizophrenia (8).…”

Few-Electrode EEG from the Wearable Devices Using Domain Adaptation for Depression Detection

“…When effective feature selection is performed on the EEG, the support vector machine (SVM), linear discriminant analysis (LDA), naive Bayes (NB), k-nearest neighbors (kNN), and decision tree (D3) can be used to make predictions with better results [ 27 ]. Akbari et al suggested a novel strategy for the diagnosis of depression based on several geometric features derived from the EEG signal shape of the second-order differential plot (SODP) [ 28 ]. With the development of deep learning, many researchers are using deep network models for the detection of depression.…”