SUBTYPING SCHIZOPHRENIA VIA MACHINE LEARNING BY USING STRUCTURAL NEUROIMAGING

Schizophrenia is a heterogeneous disorder with diverse clinical presentations and neuroanatomical alterations. Despite recent advances, we still lack a working hypothesis for the pathophysiology of schizophrenia. One reason might be the heterogeneous neuroanatomy of the patients. Data-driven approaches leveraging structural neuroimaging and machine learning have emerged as transformative tools for unraveling this enigma. Recent studies employing clustering techniques have identified robust neuroanatomical subtypes independent of traditional symptom-based frameworks. These data-driven methods reveal distinct cortical and subcortical patterns, aligning with disease progression variations, cognitive function, and treatment outcomes. Novel trajectory-based models suggest that schizophrenia may originate from distinct neuroanatomical regions and follow divergent paths of progression, emphasizing the importance of understanding these patterns in the context of disease staging. These findings provide a foundation for improving diagnostic precision, understanding disease mechanisms, and tailoring interventions. However, further validation with longitudinal data and methodological standardization are necessary to translate these insights into clinical practice and develop personalized treatment approaches.