Abstract:Schizophrenia (SZ) is a complex neuropsychiatric disorder affecting approximately 1% of the global population. The early diagnosis of SZ, with electroencephalograph (EEG) signals using deep learning (DL), can help in timely interventions which may mitigate the risk of progression to clinical psychosis. This study introduces five novel machine learning (ML)/DL-based frameworks for identifying SZ using EEG signals. The first framework involves extracting complexity features using discrete wavelet transform (DWT) of the EEG signal. In the second framework, to capture the interrelatedness among the EEG channels the complexity features are computed using multivariate empirical mode decomposition (MEMD). In both of these frameworks, the complexity features extracted are transformed into their 2D representation which uses convolutional neural network (CNN) based model for classification. Various CNN models, including conventional CNN and pretrained models were used for this purpose. In the third framework, to obtain the benefit of multiple view of the EEG signal, the complexity features extracted from DWT and MEMD features in vector representation were fused using concatenation. The combined feature was integrated with a feedforward neural network (FFNN). To obtain the optimized multiview feature set principal component analysis (PCA) was used on the concatenated feature set in the fourth framework. Finally, in the fifth framework, to further optimize the fusion of DWT and MEMD feature set canonical correlation analysis (CCA) based approach was proposed. This study is one of the first to apply a 2D representation of entropy features extracted from DWT and MEMD transformed signals for diagnosis of SZ. Furthermore, this is the first study to propose an optimized multiview feature derived from the fusion of 1D-DWT and 1D-MEMD transformed complexity features using PCA and CCA for identifying SZ from healthy control (HC). The classification performance of various CNN models in the proposed DWT-based framework ranges from 57.14 to 86.59%. Similarly, for the MEMD-based framework, CNN performance falls within the range of 54.19–85.19%. It is worth noting that for both of these frameworks standard CNN demonstrates superior performance compared to the pretrained models. For DWT and MEMD based complexity features, the highest accuracy achieved by simple CNN was 86.59% and 85.19% respectively. In the third framework, incorporating the multiview feature set with the FFNN model yields an impressive classification accuracy of 88.18%. Finally, the optimized multiview feature set obtained using PCA and CCA in the fourth and fifth framework further enhances the classification accuracy to 90.15% and 90.64% respectively. Thus, the study have proposed a computationally efficient model that outperforms computationally complex counterparts ensuring optimal resource utilization for SZ identification.

Classification of EEG-Based Brain Connectivity Networks in Schizophrenia Using a Multi-Domain Connectome Convolutional Neural Network

Sparsity‐guided multiple functional connectivity patterns for classification of schizophrenia via convolutional network

EEG-based schizophrenia detection using fusion of effective connectivity maps and convolutional neural networks with transfer learning

Classification of Schizophrenia by Combination of Brain Effective and Functional Connectivity

Assisting Schizophrenia Diagnosis Using Clinical Electroencephalography and Interpretable Graph Neural Networks: a Real-World and Cross-Site Study

Automatic identification of schizophrenia based on EEG signals using dynamic functional connectivity analysis and 3D convolutional neural network

Classification of schizophrenia patients using a graph convolutional network: A combined functional MRI and connectomics analysis

3D convolutional neural network for schizophrenia detection using as EEG-based functional brain network

Detection of schizophrenia using hybrid of deep learning and brain effective connectivity image from electroencephalogram signal

Multi feature fusion network for schizophrenia classification and abnormal brain network recognition

Multiview EEG signal analysis for diagnosis of schizophrenia: an optimized deep learning approach

Multi-modal deep learning from imaging genomic data for schizophrenia classification

Recognition of the Multi-class Schizophrenia Based on the Resting-State EEG Network Topology

Multiple Functional Connectivity Networks Fusion for Schizophrenia Diagnosis

Multimodal Classification of Schizophrenia Patients with MEG and fMRI Data Using Static and Dynamic Connectivity Measures

Discriminative analysis of schizophrenia patients using an integrated model combining 3D CNN with 2D CNN: A multimodal MR image and connectomics analysis

Multi-Site Diagnostic Classification of Schizophrenia Using Discriminant Deep Learning with Functional Connectivity MRI.

Investigating the Interpretability of Schizophrenia EEG Mechanism through a 3DCNN-based Hidden Layer Features Aggregation Framework

Automatic Diagnosis of Schizophrenia in EEG Signals Using CNN-LSTM Models

A Modified Convolutional Neural Network for Resting-State EEG-Based Schizophrenia Classification with Weighted Electrodes.

Integrating Multi-scale Feature Representation and Ensemble Learning for Schizophrenia Diagnosis.