ADVANCED SCHIZOPHRENIA DIAGNOSIS USING EEG SIGNAL PROCESSING AND DEEP LEARNING TECHNIQUES

Abstract

Schizophrenia is a chronic neuropsychiatric disorder affecting perception, cognition, and behavior. Globally it impacts about 1% of the population. In India, nearly 4–5 million people are estimated to live with schizophrenia, with many cases undiagnosed due to stigma and limited specialists. Traditionally, diagnosis relies on clinical interviews and behavioral assessment. The objective is to develop an automated system that diagnoses schizophrenia using EEG signals transformed into Markov Transition Field images and classified through deep learning models for accurate, objective decision support. In the manual system, psychiatrists diagnose schizophrenia through patient interviews, behavioral observation, medical history review, and standardized questionnaires such as DSM-based clinical assessments, often supported by basic EEG visual inspection. Manual diagnosis is subjective, time-consuming, and highly dependent on clinician experience. EEG interpretation lacks quantitative analysis, leading to inter-observer variability, delayed diagnosis, misclassification, and difficulty in identifying subtle neurological patterns. The motivation of this research is to overcome subjectivity and inconsistency in manual diagnosis by introducing automated EEG feature extraction and learning-based classification. By capturing hidden temporal patterns using Markov Transition Fields and deep models, the system improves reliability, scalability, and early detection accuracy. The proposed system converts EEG signals into Markov Transition Field images that preserve temporal dynamics of brain activity. Deep learning models such as VGG16 and Vision Transformer (ViT) extract discriminative spatial–temporal features from these images. Multiple classifiers VGG16-MTF, VGG16-NC, VGG16-KNN-RNC, and ViT-MTF Perceptron are employed to evaluate and enhance diagnostic performance. Ensemble and transformer-based learning reduce noise sensitivity and improve generalization. This machine learning–driven framework enables accurate, objective, and scalable schizophrenia diagnosis, supporting clinicians with reliable EEG-based decision assistance