P0544Development and Validation of the SYSU-Score for MRI-Based Transmural Healing Assessment in Crohn’s Disease: A Dual-Center Study

Abstract Background Standardizing magnetic resonance enterography (MRE)-defined transmural healing (TH) remains challenging in Crohn’s disease (CD) despite its prognostic superiority. We aimed to evaluate seven conventional MRE-defined TH criteria and develop a machine-learning-optimized model for improved TH assessment. Methods In this dual-center study, 460 active CD patients with 1,263 MRE scans were stratified into three cohorts. Cohort 1 (n = 341) enabled retrospectively dual-metric comparison (attainment rate/prognostic protection) of seven MRE-defined TH criteria. Leveraging their strengths, we developed five machine-learning models for TH assessment to identify the optimal one. External validation was performed in prospective Ustekinumab (n = 92) and Upadacitinib (n = 27) cohorts. Results Among seven conventional TH criteria, magnetic resonance index of activity (MaRIA), C-score, and simplified MaRIA (sMaRIA) demonstrated higher attainment rates (23.75%/28.74%/41.64%) and lower disease progression rates (14.81%/18.37%/25.35%). Random forest (RF) model consistently outperformed other machine-learning models across cohorts: Cohort 1 (AUC, 0.82 vs. 0.71-0.81), Ustekinumab (AUC, 0.83 vs. 0.73-0.81), and Upadacitinib (AUC, 0.77 vs. 0.60-0.77) cohorts. Dual-metric evaluation identified RF-model and C-score as clinically applicable tools. Notably, RF-model (namely SYSU-score) demonstrated superior protective effects versus C-score: in Ustekinumab cohort, SYSU-score-defined TH showed lower disease progression risk (HR = 0.07, P 0.001) than C-score (HR = 0.15, P 0.001); Upadacitinib cohort validated SYSU-score as the only system demonstrating significant protection (HR = 0.23, P 0.05). Conclusion We established a validated machine-learning-derived TH criterion (namely SYSU-score) integrating strengths of conventional MRE-defined systems. SYSU-score-defined TH status conferred significant protection against disease progression with robust prognostic discrimination, advancing standardized TH assessment for clinical implementation. Conflict of interest: Ms. Zheng, Qingzhu: No conflict of interest Zhang, Ruonan: No conflict of interest He, Weitao: No conflict of interest Huang, Lili: No conflict of interest Feng, Shi-ting: No conflict of interest Li, Xuehua: No conflict of interest