ETSNet: A deep neural network for EEG-based temporal–spatial pattern recognition in psychiatric disorder and emotional distress classification
Introduction
The use of neurophysiological signals, such as the electroencephalogram (EEG), has been widely adopted for the study of brain dysfunction in psychiatric disorders. Machine learning techniques have also been utilized to develop EEG-based psychiatric diagnostic systems. With approximately 450 million people suffering from psychiatric disorders worldwide, including schizophrenia, bipolar disorder, and depression [1], reliable risk prediction and diagnosis using EEG remains a challenge due to the complexity and high level of noise in EEG signals and the significant differences between healthy and mentally ill individuals.
Previous research on computer-aided diagnosis (CAD) systems for schizophrenia has focused on distinguishing individuals with schizophrenia (SZ) from healthy controls (HC). However, such systems can lead to misdiagnosis in individuals who share genetic characteristics and phenotypes (i.e., endophenotypes [2]) with SZ but do not have the disorder. To improve the accuracy of the CAD system, genetic susceptibility to schizophrenia must be taken into consideration. Including features that differentiate biological relatives of SZ (REL) from their probands can reduce the risk of misdiagnosis. Although REL shares genetic traits with SZ, they do not exhibit the disease-specific traits that are present in SZ and absent in HC [3]. As such, it is ideal to develop a CAD system for diagnosing SZ by using features derived from both REL and SZ and HC [4], [5].
Machine learning techniques, such as support vector machine (SVM) and logistic regression, have been applied to EEG signals to diagnose schizophrenia [6], [7], [8], [9], [10]. However, these studies have only utilized straightforward EEG features at the scalp sensor level, which can limit the accuracy of the diagnosis due to the spatial resolution limitations of sensor-level signals. To improve the classification accuracy, brain source-level signals with high spatial and temporal resolution should be utilized, as they provide more rich and detailed data [11].
There has been a growing interest in the use of deep learning in EEG analysis for neuropsychiatric disorders. Deep learning has the potential to provide deeper insights into EEG analysis and contribute to more effective and timely treatment of these illnesses. Recent studies utilizing deep learning models have shown significant improvements in the accuracy of classifying individuals with SZ versus HC compared to conventional machine learning methods [12], [13], [14]. Given the advancements in deep learning, it is likely that this technology will significantly enhance the CAD system for SZ diagnosis through the application of deep-learning techniques.
The aim of this work is to develop a deep learning model for the categorization of individuals with SZ, REL, and HC. The paper’s contributions are summarized as follows:
- •
This study is the first to explore an end-to-end solution for EEG-based classification of SZ, REL, and HC to enhance the schizophrenia diagnostic classification system, as depicted in Fig. 1.
- •
The focus is on obtaining high spatial and temporal resolution characteristics from EEG data by computing whole-brain cortical source signals using a realistic head model built from individual magnetic resonance imaging (MRI) data.
- •
A novel EEG Temporal Spatial Network (ETSNet) is proposed for the cortical source signals of regions of interest (ROIs) by incorporating a Squeeze-and-Excitation block and fusing networks generated in the eyes-open (EO) and eyes-closed (EC) resting states. Two variants of ETSNet are developed: single (ETSNet), and fusion network (ETSNet).
- •
Extensive experiments are conducted to evaluate the performance of the proposed deep neural model in classifying SZ, REL, and HC, as well as differentiating EO vs. EC resting states. ETSNet achieved a classification accuracy of 99.57% for the 3 classes, SZ, REL, and HC. ETSNet achieved an accuracy of 93.15% in distinguishing EO vs. EC resting states.
- •
The performance of ETSNet is evaluated on two publicly available datasets, SEED [15] and IDD [16]. The results show improved classification accuracy on these datasets with ETSNet.
Section snippets
Related work
The primary objective of classifiers in every domain is pattern recognition. For instance, in computer vision, global and local feature recognition are used to categorize images [24] and use a variety of encryption based algorithms. In contrast to image domain, EEG data presents several challenges, including high inter- and intra-subject variance, a lack of standardization in data collection systems, and noise, which can negatively impact the classifier’s ability to generalize. The type of EEG
Method
We present our end-to-end solution for EEG-based classification of schizophrenia (SZ), related disorders (REL), and healthy controls (HC) with the aim of improving the diagnostic classification system for schizophrenia (refer to Fig. 1). To demonstrate the efficacy of our model, we utilized three datasets: psychiatric disorders, intellectual and developmental disorder and emotional datasets. The primary dataset, which is not publicly accessible, includes EEG data from participants diagnosed
IDD and SEED datasets
As previously noted, the experiments utilizing the IDD and SEED datasets serve to demonstrate the effectiveness of the proposed neural network model for EEG data classification, not just for psychiatric disorders. These additional datasets provide evidence that the model is suitable for classifying a variety of EEG data. The performance of ETSNet was evaluated against other state-of-the-art research in this field using publicly available datasets, the details of which are listed in Table 7.
The
Discussion
In this study, we introduced two novel 1D CNN models, ETSNet and ETSNet, for EEG classification. The results demonstrate the effectiveness of these models in classifying psychic disorders, as well as in other EEG domains such as intellectual development disorders and emotions. Our experiments on EEG signals showed remarkable performance, which can be attributed to several factors.
- •
The use of 78 cortical ROI signals instead of scalp EEG signals, which provide more accurate representations of
Conclusion and future work
This article presents a novel approach for classifying individuals with schizophrenia (SZ), relatives of individuals with schizophrenia (REL), and healthy controls (HC) using an EEG-based deep learning model. We computed the major brain source signals of 78 cortical regions of interest (ROIs) from resting EEG recordings, which were then used as input for our deep neural network. Our experiments showed that the optimized model achieved an accuracy of 99.57% for group classification and 93.15%
CRediT authorship contribution statement
Syed Jawad H. Shah: Conceptualization, Methodology, Software, Writing - Original Draft, Writing - Review & Editing. Ahmed Albishri: Conceptualization, Methodology, Software, Writing - Original Draft, Writing - Review & Editing. Seung Suk Kang: Conceptualization, Methodology, Software, Data Curation, Writing - Original Draft, Writing - Review & Editing. Yugyung Lee: Conceptualization, Methodology, Software, Writing - Original Draft, Writing - Review & Editing. Scott R. Sponheim:
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (51)
- et al.
Is the P300 wave an endophenotype for schizophrenia? A meta-analysis and a family study
Neuroimage
(2005) - et al.
Automated detection of schizophrenia using nonlinear signal processing methods
Artif. Intell. Med.
(2019) - et al.
EEG dataset of individuals with intellectual and developmental disorder and healthy controls under rest and music stimuli
Data Brief
(2020) - et al.
Emotional EEG classification using connectivity features and convolutional neural networks
Neural Netw.
(2020) - et al.
A novel multi-modal machine learning based approach for automatic classification of EEG recordings in dementia
Neural Netw.
(2020) - et al.
EEG based multi-class seizure type classification using convolutional neural network and transfer learning
Neural Netw.
(2020) - et al.
3D image recognition using new set of fractional-order Legendre moments and deep neural networks
Signal Process., Image Commun.
(2021) - et al.
Spherical splines for scalp potential and current density mapping
Electroencephalogr. Clin. Neurophysiol.
(1989) - et al.
Dysfunctional neural processes underlying context processing deficits in Schizophrenia
Biol. Psychiatry Cognit. Neurosci. Neuroimaging
(2019) - et al.
Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain
Neuroimage
(2002)
Abnormal cortical neural synchrony during working memory in schizophrenia
Clin. Neurophysiol.
T-distributed stochastic neighbor embedding (t-SNE): A tool for eco-physiological transcriptomic analysis
Mar. Genom.
Emotion recognition based on EEG feature maps through deep learning network
Eng. Sci. Technol. Int. J.
EEG-based emotion recognition using tunable Q wavelet transform and rotation forest ensemble classifier
Biomed. Signal Process. Control
A review of telemental health in international and post-disaster settings
Int. Rev. Psychiatry
The endophenotype concept in psychiatry: etymology and strategic intentions
Am. J. Psychiatry
Genetic and disorder-specific aspects of resting state EEG abnormalities in schizophrenia
Schizophrenia Bullet.
The use of neurophysiological endophenotypes to understand the genetic basis of schizophrenia
Dialogues Clin. Neurosci.
A machine-learning framework for robust and reliable prediction of short-and long-term treatment response in initially antipsychotic-naïve schizophrenia patients based on multimodal neuropsychiatric data
Transl. Psychiatry
Machine learning identification of EEG features predicting working memory performance in schizophrenia and healthy adults
Neuropsychiatric Electrophysiol.
Identification of major psychiatric disorders from resting-state electroencephalography using a machine learning approach
Front. Psychiatry
Machine-learning classification using neuroimaging data in schizophrenia, autism, ultra-high risk and first-episode psychosis
Transl. Psychiatry
Addressing measurement issues in electroencephalography studies of meditations as alternative interventions of posttraumatic stress disorder
Psychol. Trauma Theory Res. Pract. Policy
Deep convolutional neural network model for automated diagnosis of schizophrenia using EEG signals
Appl. Sci.
Time-frequency analysis of scalp EEG with Hilbert-Huang transform and deep learning
IEEE J. Biomed. Health Inf.
Cited by (6)
Classification of Parkinson's disease EEG signals using 2D-MDAGTS model and multi-scale fuzzy entropy
2024, Biomedical Signal Processing and ControlFractal Spiking Neural Network Scheme for EEG-Based Emotion Recognition
2024, IEEE Journal of Translational Engineering in Health and MedicineA Hybrid Approach Of Differential Evolution And Multistage LSTM For Diagnosis Of Psychiatric Disorder Using EEG
2023, Proceedings of the 2023 IEEE International Conference on Computer Vision and Machine Intelligence, CVMI 2023Research on the Recognition of Motor Imaging EEG Signals Based on the PSD and the SVM
2023, 2023 IEEE 3rd International Conference on Data Science and Computer Application, ICDSCA 2023EEG Based Emotion Classification Using Two Layer Convolutional Neural Network
2023, 2023 IEEE International Conference on Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering, MetroXRAINE 2023 - Proceedings