Abstract
Hippocampal deviations are a hallmark of schizophrenia, yet their regional specificity remain unclear. Neuroimaging studies have reported smaller volumes for each hippocampal subfields in schizophrenia compared to healthy controls but affected regions differ between studies. These conflicting findings highlight substantial heterogeneity within psychosis, which may be elucidated through more detailed sub-regional analyses. In this study, we aimed to determine whether patients with schizophrenia exhibit distinct volumetric alterations in specific hippocampal subfields compared to healthy controls. We analysed T1-weighted MRI data from the MCICShare project, employing the ComBat algorithm to harmonize data across multiple MRI platforms. Hippocampal subfields were segmented and quantified using the "Bayesian Segmentation with Histological Atlas". All computational analyses were performed on Google Colab Pro+ with Nvidia A100 GPUs. Multiple ANCOVAs were then conducted, with diagnosis as the independent variable and each hippocampal subfield volume as the dependent variable, controlling for sex, age, and estimated intracranial volume. To mitigate type I error inflation, a 5% false discovery rate (FDR) threshold was applied. After excluding segmentation errors, we included 108 patients with schizophrenia and 94 healthy controls in the final analysis. Among the examined subfields, only the right CA2 showed a significant volumetric difference after FDR adjustment (F = 8.562, P(FDR) =0.048, η(2)p=0.042). Our findings underscore the value of high-granularity segmentation approaches and highlight the potential importance of CA2 alterations in schizophrenia's pathophysiology, thereby guiding future research directions and clinical applications.