Unveiling Functional Biomarkers in Schizophrenia: Insights from Region of Interest Analysis Using Machine Learning

Abstract

Background: Schizophrenia is a complex and disabling mental disorder that represents one of the most important challenges for neuroimaging research. There were many attempts to understand these basic mechanisms behind the disorder, yet we know very little. By employing machine learning techniques with age-matched samples from the auditory oddball task using multi-site functional magnetic resonance imaging (fMRI) data, this study aims to address these challenges. Methods: The study employed a three-stage model to gain a better understanding of the neurobiology underlying schizophrenia and techniques that could be applied for diagnosis. At first, we constructed four-level hierarchical sets from each fMRI volume of 34 schizophrenia patients (SZ) and healthy controls (HC) individually in terms of hemisphere, gyrus, lobes, and Brodmann areas. Second, we employed statistical methods, namely, t-tests and Pearson's correlation, to assess the group differences in cortical activation. Finally, we assessed the predictive power of the brain regions for machine learning algorithms using K-nearest Neighbor (KNN), Naive Bayes, Decision Tree (DT), Random Forest (RF), Support Vector Machines (SVMs), and Extreme Learning Machine (ELM). Results: Our investigation depicts promising results, obtaining an accuracy of up to 84% when applying Pearson's correlation-selected features at lobes and Brodmann region level (81% for Gyrus), as well as Hemispheres involving different stages. Thus, the results of our study were consistent with previous studies that have revealed some functional abnormalities in several brain regions. We also discovered the involvement of other brain regions which were never sufficiently studied in previous literature, such as the posterior lobe (posterior cerebellum), Pyramis, and Brodmann Area 34. Conclusions: We present a unique and comprehensive approach to investigating the neurological basis of schizophrenia in this study. By bridging the gap between neuroimaging and computable analysis, we aim to improve diagnostic accuracy in patients with schizophrenia and identify potential prognostic markers for disease progression.

Keywords

• functional magnetic resonance imaging (fmri)
• schizophrenia
• feature selection
• region of interest (roi)
• machine learning