Automated detection of schizophrenia using nonlinear signal processing methods
Introduction
Malfunction of the brain by disease or disorder affects normal activity in humans [[1], [2], [3]]. Schizophrenia(sz) is a chronic disorder which affects the thinking ability as well as general behavior. The report of the World Health Organization (WHO) substantiates that sz is a severe mental disorder, and more than 21 million people worldwide are affected by it [4]. Yet, WHO has also stated that sz is treatable, and early or post diagnosis may be helpful to identify its severity and stage. Detection and treatment of sz is essential in patients, since it creates substantial inconvenience in regard to thinking, memory, perception, and other living activities. If left untreated, it is an unalterable process which damages the human behavioral abilities in its later stages [5]. Early as well as post discovery of sz may help during implementation of possible treatment methods to cure or limit the effects. Most mental disorders such as sz can be assessed by signaling [7] or imaging techniques [6]. Recently, a number of non-invasive techniques have been proposed and implemented by investigators to identify sz based on the Electroencephalogram(EEG) acquired using multi-channel sensor arrays [8,9]. The assessment and confirmation accuracy of sz from the EEG pattern depends mainly on the tool considered to examine these signals.
The imaging techniques, such as Magnetic Resonance Imaging (MRI) and Computed Tomography (CT), are costly and require additional recording and computational time as compared to signaling procedures such as EEG [[10], [11], [12], [13]]. An EEG signal acquired using an appropriate electrode can be useful to reveal essential detail regarding brain activity, and examination of these signals may help to detect the condition of the brain [14,15,25]. During the clinical diagnostic process, the EEG is obtained by placing the electrodes at predefined scalp sections. Recently, EEG patterns were extensively used to inspect for maladies such as dementia, Alzheimer's disease, sleep disorder, epilepsy, sz, Parkinson’s disease and other brain related disorders [16,17,21,23,52,53].
Recent studies provide insights to the classification of sz based on EEG patterns [18]. Table 1 highlights the summary of ADT systems employed for the detection of sz using EEG signals. Kim et al. [26] extracted EEG signals from 21 gold cup electrodes, positioned according to the 10–20 international standards. Horizontal and vertical eye movements of participants were studied. After pre-processing, five frequency bands were chosen for analysis. For each of the five bands, the spectral power of the EEG was computed using Fast Fourier Transformation, after which the Analysis of Variance(ANOVA) method was employed to study EEG power deviations. The Receiver Operating Curve (ROC) analysis technique was used to determine the diagnostic performance of a test, utilized in distinguishing between normal and sz patients. The highest classification accuracy of 62.2% was obtained for delta power. Dvey-Aharon et al. [20] discussed a Time-Frequency transformation based evaluation of the EEG signal to examine for sz. In this work, a Stockwell approach was implemented to convert the EEG signal into an image, and then feature extraction and classification procedures were incorporated to attain improved results. The top five unique electrodes were reported to have a prediction accuracy between 92.0% and 93.9%, with F2 portraying to be the best electrode. Johannesen et al. [27] acquired EEG recordings from participants using a 64 electrode system. Participants were required to press one of two response buttons, using either their right or left index finger, to indicate whether a particular letter was presented in the previous set. The signals were analysed using the Brain Vison Analyser software and segmented via four stages of processing: pre-stimulus baseline, encoding, retention and retrieval. At each of the four stages of processing, time-frequency data(squared wavelet coefficients, binned and averaged according to correct versus incorrect response accuracy) was retrieved for the five frequency bands examined. Statistical analyses were conducted on spectral power measured at the Frontal, Central and Occipital locations. Feature selection was done using the wrapper method [22]. The 1-norm Support Vector Machine (SVM) classifier was utilized to classify correct and incorrect trials in data with the SVM Model 1, achieving a classification accuracy of 84%. The SVM Model 2 was implemented to classify normal versus the sz condition in correct trial data, achieving a classification accuracy of 87%. Santos-Mayo et al. [28] analysed the EEG-ERP signals of participants who were involved in an auditory task. The Brain Vision equipment, in compliance with the 10–20 international standards, was used to record the brain signals. After acquisition, the signals were pre-processed using EGGLAB [24], after which 16 time-domain features and four frequency-domain features were extracted per electrode, for each participant. Features were selected via linear discriminant analysis using J5 and Mutual Information Feature Selection(MIFS) coupled with the Double Input Symmetrical Relevance (DISR). The Multilayer Perceptron(MLP) and SVM classifiers were employed for classification. High classification rates of 93.42% and 92.23% were achieved with the J5 MLP and J5 SVM classifiers, respectively. Ibáñez-Molina et al. [8] also implemented sz examination based on the EEG. In this work, EEG recordings were extracted from participants while they were at rest and engaged in a naming task. The Neuroscan SynAmps 32-channel amplifier was employed for data acquirement. EEG signals at the resting phase were acquired prior to the task, while those from the task were extracted after each trial. In the resting phase, the segments were analysed using a moving window method, after which LZC was computed per window. After normalisation, the final LZC value was computed by calculating the average of the values obtained from the moving window method. A total of 80 EEG segments of 2 × 103ms were evaluated, at task, and then averaged to obtain the final Multiscale LZC value. Higher complexity values were reported in right frontal regions of patients who were at rest. V. Jahmunah et al. [51], developed an eleven-layered deep learning model for the classification of sz. Two CNN models were developed separately for subject base testing and non-subject base testing. In subject base testing, validation of the system was carried out in three phases: training, testing and validation of data. During training, k-fold validation was used, whereby the entire data was split into fourteen equal parts. Twelve parts were used for training, one was used for validation and another for testing. In non-subject base testing, 10-fold validation was conducted during training and the system was evaluated through the training and testing phases. Accuracies of 98.07% and 81.26% were yielded for non-subject base testing and subject base testing, respectively.
Section snippets
Data used
Fifteen minutes of EEG signals acquired from 14 patients with paranoid sz, encompassing seven males and seven females, with a mean age of 27.9 ± 3.3 and 28.3 ± 4.1 years, respectively, were collected from the Institute of Psychiatry and Neurology in Warsaw, Poland [19]. Fourteen healthy subjects within similar age and gender ratio were recruited from the same institute. In this study, a multi-channel (19-channel) EEG was adopted for the assessment. The electrodes used were Fp1, Fp2, F7, F3, Fz,
Pre-processing
Thirty second segments without artefacts were used for analysis. A 2nd order Butterworth filter was employed to preprocess the extracted EEG signals. The signals were segmented into nonoverlapping segments of 25 s, such that each segment consisted of 6250 × 19 sample points. This segmentation gave rise to 1142 EEG patterns, which were then grouped to form a new database of normal and sz class EEGs with a fixed length. Following segmentation, 12 features were extracted from the signals. Fig. 2
Results
Table 3 illustrates the 14 significant features identified with the t-test. The features were ranked based on the p-values. The Hjorth complexity, with the lowest p-value, is ranked first, portraying to be the most significant feature. Entropy, with the highest p-value, is ranked fourteenth, portraying to be the worst feature, for the classification of EEG signals. Analysing the p values (p < 0.05) from Table 3, it is clear that the 14 features are highly discriminatory. Hence the features
Conclusion
The proposed ADT involves the extraction of nonlinear features from signals, t-test based feature selection, and performance validation of the different classifiers reconnoitered. The SVM(RBF) classifier yielded the highest accuracy of 92.91% as compared to the other classifiers employed in this work. It achieved the best accuracy with 12 features, and portrays as the best classifier. This confirms that the proposed technique is expedient in the classification of normal versus sz cases.
Declaration of Competing Interest
The authors declare that they have no conflicts of interest.
References (53)
- et al.
Entropy based segmentation of tumor from brain MR images–a study with teaching learning based optimization
Pattern Recogn. Lett.
(2017) - et al.
Deep convolutional neural network for the automated detection and diagnosis of seizure using EEG signals
Comput Biol Med
(2018) - et al.
Automated approach for detection of ischemic stroke using Delaunay Triangulation in brain MRI images
Comput Biol Med
(2018) - et al.
Application of deep transfer learning for automated brain abnormality classification using MR images
Cogn Syst Res
(2019) - et al.
Application of multiresolution analysis for automated detection of brain abnormality using MR images: a comparative study
Future Gener Comput Syst
(2019) Characterization of focal EEG signals: a review
Future Gener Comput Syst
(2019)- et al.
Use of features from RR-time series and EEG signals for automated classification of sleep stages in deep neural network framework
Biocybern Biomed Eng
(2018) - et al.
Deep convolutional neural network for the automated detection and diagnosis of seizure using EEG signals
Comput. Biol. Med.
(2018) - et al.
EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis
J Neurosci Methods
(2004) - et al.
Diagnostic utility of quantitative EEG in un-medicated schizophrenia
Neurosci Lett
(2015)