Date of Award
Spring 2025
Document Type
Thesis - SCU Access Only
Publisher
Santa Clara : Santa Clara University, 2025
Department
Bioengineering
First Advisor
Hamed Akbari
Abstract
Schizophrenia is a chronic psychiatric disorder characterized by hallucinations, delusions, cognitive deficits, and disorganized behavior. Despite its prevalence, diagnosis remains subjective and treatment response is highly individualized. This thesis presents a machine learning (ML) framework leveraging electroencephalography (EEG) data to develop an objective, scalable, and non-invasive diagnostic and predictive tool for schizophrenia. The model utilizes clinical EEG recordings from patients and healthy controls during a social decisionmaking task. A multi-stage preprocessing pipeline—including temporal data augmentation, wavelet-based denoising, and independent component analysis—was applied to enhance signal clarity and structure the data for deep learning. EEG segments were then tokenized and input into a Transformer-based neural network capable of both binary classification (schizophrenia vs. control) and pseudo-severity estimation. Trained using stratified group 5-fold cross-validation and optimized with early stopping, the model achieved an average accuracy of 90%, an AUC of 0.93, and an F1 score of 0.90. Continuous output scores from the model provide a gradient of symptom severity, offering insights into subclinical neural patterns. This dual-output approach enables both diagnostic support and real-time symptom tracking, representing a novel contribution to AI-assisted precision psychiatry. The results demonstrate that EEG-derived neural signatures, like altered gamma oscillations, impaired phase synchrony, and reduced eventrelated potentials, can be effectively leveraged by deep learning models to enhance clinical decision-making. This work establishes a foundation for scalable EEG-based diagnostics and creates the framework for further development of integrated neuroinformatics tools for psychiatric care.
Recommended Citation
Flinn, Ally, "EEG-based Machine Learning Framework For Schizophrenia Diagnosis And Predictive Modeling" (2025). Bioengineering Senior Theses. 137.
https://scholarcommons.scu.edu/bioe_senior/137
