Empowering precision medicine: AI-driven schizophrenia diagnosis via EEG signals: A comprehensive review from 2002-2023

Mahboobeh Jafari, Delaram Sadeghi, Afshin Shoeibi, Hamid Alinejad-Rokny, Amin Beheshti, David López-García, Zhaolin Chen, U. Rajendra Acharya, Juan Manuel Górriz

TL;DR

This paper provides a comprehensive review of Artificial Intelligence (AI) techniques, specifically Machine Learning (ML) and Deep Learning (DL), applied to the diagnosis of Schizophrenia (SZ) using Electroencephalography (EEG) signals. It highlights the growing potential of AI in this field while also identifying critical challenges and future research directions necessary for its clinical translation.

What Problem Does This Paper Address?

Schizophrenia (SZ) is a complex mental disorder with significant cognitive, emotional, and behavioral impacts. Accurate and early diagnosis is crucial for effective treatment, but traditional diagnostic methods can be time-consuming and subjective. Electroencephalography (EEG) offers a promising, non-invasive way to capture brain activity, but analyzing EEG signals for SZ diagnosis is challenging due to artifacts, long recordings, and complex data. This paper addresses the need for a systematic overview of how AI, particularly ML and DL, is being used to overcome these challenges and improve SZ diagnosis from EEG data, while also pinpointing the hurdles that still need to be cleared for widespread clinical adoption.

Methods

The paper reviews a vast body of literature on AI-driven SZ diagnosis using EEG signals, categorizing the approaches into Machine Learning (ML) and Deep Learning (DL).

Machine Learning (ML) methods discussed include:

  • Feature Extraction: Time-domain (e.g., amplitude, variance), frequency-domain (e.g., PSD, coherence), time-frequency (e.g., DWT, CWT), and nonlinear features (e.g., entropy, fractal dimensions) are explored.
  • Dimension Reduction: Techniques like PCA and LDA are mentioned for simplifying data.
  • Classification Algorithms: Common classifiers like SVM, KNN, RF, and MLP are highlighted.

Deep Learning (DL) methods reviewed include:

  • Convolutional Neural Networks (CNNs): Specifically 1D-CNNs for time-series data and 2D-CNNs (requiring signal-to-image conversion).
  • Pre-trained Models: Leveraging models trained on large datasets like ImageNet.
  • Recurrent Neural Networks (RNNs): Including LSTM and GRU for modeling temporal dependencies.
  • Autoencoders (AEs): Used for feature extraction and dimensionality reduction.
  • Other DL Architectures: Attention mechanisms, graph models, multi-task learning, and federated learning are presented as future directions.

The paper also details preprocessing techniques (filtering, normalization, artifact removal) and discusses the PRISMA guidelines used for literature selection.

Datasets

The paper identifies and discusses two primary publicly available EEG datasets used in SZ diagnosis research:

  • RepOD dataset: Recorded by Olejarczyk et al. (2017), comprising EEG signals from 14 SZ patients and 14 healthy controls.
  • Kaggle dataset: Comprising EEG signals from 23 individuals with SZ and 22 healthy control subjects.

The authors note that while these datasets have been instrumental, the limited number of subjects and the lack of standardized recording protocols pose significant challenges. They also mention the use of "clinical datasets" which are often not publicly accessible, hindering comparative research.

Key Insights and Ideas

  • EEG as a Promising Modality: EEG's non-invasiveness, real-time monitoring, high temporal resolution, and cost-effectiveness make it a highly attractive modality for SZ diagnosis, despite its limitations in spatial resolution and susceptibility to artifacts.
  • AI's Transformative Potential: AI, especially ML and DL, has demonstrated significant promise in automating SZ diagnosis from EEG signals, offering improved accuracy and potentially earlier detection compared to traditional methods.
  • The ML vs. DL Trade-off: ML methods offer interpretability and efficiency with limited data, while DL excels at automated feature extraction and handling complex patterns but requires larger datasets and more computational power.
  • Data Scarcity is a Major Bottleneck: The limited availability of large, standardized, and multimodal EEG datasets is a critical barrier to advancing DL models for SZ diagnosis.
  • Explainability is Key for Clinical Trust: DL models are often "black boxes," and developing explainable AI (XAI) techniques is crucial for building clinician trust and facilitating the adoption of AI in clinical settings.
  • Multimodality Holds Untapped Potential: Integrating EEG with other neuroimaging modalities (like fMRI) could provide a more comprehensive understanding of SZ, but access to such multimodal datasets is currently limited.

Key Results

The paper is a comprehensive review, so it doesn't present novel experimental results. Instead, it synthesizes findings from numerous studies. Key takeaways from the reviewed literature include:

  • High Accuracy Achieved: Many studies utilizing ML and DL techniques on EEG signals have reported high diagnostic accuracies, often exceeding 90% in some cases (as seen in Tables 3 and 4).
  • DL Outperforms ML in Some Aspects: DL models, particularly CNNs, have shown advantages in automated feature extraction and handling complex patterns, leading to improved performance in many studies.
  • ML Still Relevant: ML methods remain valuable due to their interpretability and efficiency, especially when dealing with smaller datasets.
  • Challenges Persist: Despite promising results, significant challenges remain, including dataset limitations, lack of standardized protocols, interpretability issues with DL, and hardware resource constraints.

Limitations and Future Work

Limitations: * Dataset Availability: The most significant limitation is the scarcity of large, diverse, and publicly accessible EEG datasets for SZ diagnosis. * Standardization: A lack of standardized EEG recording protocols across studies makes direct comparison difficult. * Interpretability of DL: Deep learning models are often considered "black boxes," hindering clinical trust and understanding. * Hardware Resources: The computational power and memory required for training complex DL models can be a barrier. * Multimodality Data Access: Datasets combining EEG with other neuroimaging modalities are not readily available.

Future Work: * Dataset Development: Creating larger, more diverse, and standardized EEG datasets, potentially including data from different SZ severity levels and interventional treatments. * Multimodal Data Integration: Developing AI techniques that can effectively integrate data from multiple neuroimaging modalities. * Novel ML/DL Architectures: Exploring advanced DL models like attention mechanisms, graph neural networks, multi-task learning, and federated learning. * Explainable AI (XAI): Developing and applying XAI techniques to enhance the interpretability of DL models for SZ diagnosis. * Hardware Optimization: Investigating efficient hardware solutions like FPGAs and ASICs, and exploring model compression techniques. * Rehabilitation Systems: Developing AI-based rehabilitation systems (BCI, neurofeedback) for SZ patients. * Uncertainty Quantification (UQ): Applying UQ methods to assess the reliability and confidence of AI models in SZ diagnosis.

Summary updated

Abstract

Schizophrenia (SZ) is a prevalent mental disorder characterized by cognitive, emotional, and behavioral changes. Symptoms of SZ include hallucinations, illusions, delusions, lack of motivation, and difficulties in concentration. While the exact causes of SZ remain unproven, factors such as brain injuries, stress, and psychotropic drugs have been implicated in its development. SZ can be classified into different types, including paranoid, disorganized, catatonic, undifferentiated, and residual. Diagnosing SZ involves employing various tools, including clinical interviews, physical examinations, psychological evaluations, the Diagnostic and Statistical Manual of Mental Disorders (DSM), and neuroimaging techniques. Electroencephalography (EEG) recording is a significant functional neuroimaging modality that provides valuable insights into brain function during SZ. However, EEG signal analysis poses challenges for neurologists and scientists due to the presence of artifacts, long-term recordings, and the utilization of multiple channels. To address these challenges, researchers have introduced artificial intelligence (AI) techniques, encompassing conventional machine learning (ML) and deep learning (DL) methods, to aid in SZ diagnosis. This study reviews papers focused on SZ diagnosis utilizing EEG signals and AI methods. The introduction section provides a comprehensive explanation of SZ diagnosis methods and intervention techniques. Subsequently, review papers in this field are discussed, followed by an introduction to the AI methods employed for SZ diagnosis and a summary of relevant papers presented in tabular form. Additionally, this study reports on the most significant challenges encountered in SZ diagnosis, as identified through a review of papers in this field. Future directions to overcome these challenges are also addressed. The discussion section examines the specific details of each paper, culminating in the presentation of conclusions and findings.
BibTeX
@article{jafari2024empowering,
  author = {Mahboobeh Jafari and Delaram Sadeghi and Afshin Shoeibi and Hamid Alinejad-Rokny and Amin Beheshti and David López-García and Zhaolin Chen and U. Rajendra Acharya and Juan Manuel Górriz},
  title = {{Empowering precision medicine: AI-driven schizophrenia diagnosis via EEG signals: A comprehensive review from 2002-2023}},
  journal = {Applied Intelligence},
  year = {2024},
  doi = {10.1007/s10489-023-05155-6},
}
Figure