Exploring Machine Learning Techniques for Schizophrenia Diagnosis: A Comprehensive Review

Abstract

Background:

Schizophrenia is a prevalent neuropsychiatric disorder affecting a significant portion of the global population, with an increasing prevalence in the United States. This rise has led to more diagnoses, highlighting the need for efficient detection methods. Recently, machine learning has shown promise in detecting schizophrenia, enabling quicker and more accurate diagnoses.

Objectives:

This review aims to examine all physiological signals used in schizophrenia detection, inform researchers about available public datasets, and compare the advantages and disadvantages of these physiological signals.

Method:

A comprehensive search was conducted using databases such as Scopus, Web of Science, and the National Institutes of Health (NIH), focusing on studies published between 2020 and 2024. Keywords included "schizophrenia," "EEG," "MRI," and "machine learning." A comparative analysis synthesized the findings.

Findings:

The primary finding of this review is that Electroencephalogram (EEG) signals have been identified as the most effective choice for schizophrenia detection following a thorough comparative analysis. Additionally, it has been observed that detection accuracy is significantly influenced by the mental state or task during signal capture. Another notable outcome is that Support Vector Machines (SVM) yield superior results when used in conjunction with EEG signals. The limited size of available datasets has hindered the use of Deep Convolutional Neural Networks (CNN).

Significance:

This review offers a detailed comparison of various physiological signals used for schizophrenia detection, presenting a novel perspective. It serves as a valuable resource for novice researchers, guiding their decision-making process. The comprehensive presentation of publicly available datasets will benefit all researchers in the field. Identifying research gaps and providing future research directions are significant contributions of this paper.