Hippocampal subfield radiomics boosts aggression risk prediction in schizophrenia

A combined model integrating hippocampal subfield radiomics, whole-brain GMV, and clinical data achieved AUC 0.875 for predicting in-hospital aggression in schizophrenia, outperforming standalone imaging models (AUCs 0.681–0.729). Retrospective single-center study.

• Retrospective single-center cohort of 247 hospitalized schizophrenia patients, randomly split into training (70%) and test sets.
• Combined model achieved highest AUC 0.875 and lowest Brier score 0.119, with hippocampal radiomics score as the top contributing feature (SHAP analysis).
• Single-center, retrospective design; external validation needed before clinical use.

RadPigeon summaries are original and for information only. They are not clinical advice.