591. Combining structural neuroimaging and cognitive markers for the identification of clinical high-risk and first-episode schizophrenia

Abstract

Abstract Background Early identification of individuals at clinical high risk (CHR) for psychosis is crucial for timely intervention and better outcomes. Structural brain changes and cognitive impairments are independently linked to the risk of psychosis; however, the combined utility of these factors for early-stage identification remains to be thoroughly investigated. Aims & Objectives This study aimed to investigate cortical and subcortical structural differences in individuals with first-episode schizophrenia (FES), CHR, and healthy controls (HC) using structural MRI. Additionally, the study sought to integrate MRI features with cognitive performance scores and construct machine learning models to assess their potential in improving group classification accuracy for early psychosis detection. Method Data were collected from 43 CHR individuals, 36 FES patients, and 48 HC. Structural MRI data were preprocessed using the Data Processing & Analysis for Brain Imaging on Surface (DPABISurf) pipeline to extract cortical surface area, cortical thickness, cortical gray matter (CGM) volume, and subcortical volume. Group differences were analyzed using ANCOVA, controlling for age, sex, education, and intracranial volume (TIV). Support vector machine (SVM) models were constructed to classify participants based on MRI features combined with cognitive scores from the MATRICS Consensus Cognitive Battery (MCCB). Model performance was evaluated using AUC, sensitivity, specificity, and F1-score. Results Both the FES and CHR groups exhibited significant reductions in cortical surface area and CGM volume, particularly in the occipital, parietal, and prefrontal cortices. Structural abnormalities were most pronounced in the FES group. The surface area of the left anterior prefrontal cortex decreased progressively. In contrast, the surface area of the right primary visual cortex and CGM volumes of the left intraparietal sulcus and dorsoventral transition area increased progressively from FES to CHR to HC. In pairwise classification, the model combining MRI features with verbal learning scores achieved the highest performance for FES vs. HC (AUC = 0.952, specificity = 1.000, sensitivity = 0.571). For CHR vs. HC, the model combining MRI features with working memory scores performed best (AUC = 0.958, specificity = 0.875, sensitivity = 0.889). However, classification between FES and CHR showed limited discriminative ability (AUC range: 0.375–0.500). Discussion & Conclusions These findings highlight the potential of integrating structural neuroimaging and cognitive biomarkers for the early identification of psychosis. Significant structural changes in regions such as the prefrontal cortex and visual cortex, along with cognitive impairments in working memory and verbal learning, may serve as promising biomarkers for psychosis risk. However, machine learning results reveal limited capacity to differentiate between FES and CHR, suggesting the presence of shared neurobiological features in the early stages of psychosis. Future research should incorporate multimodal data and larger sample sizes to further validate these potential biomarkers.