Background Microvascular obstruction (MVO) is strongly associated with adverse outcomes after ST-segment elevation myocardial infarction (STEMI). However, manual quantification of MVO is time-consuming and fails to capture the heterogeneity of microvascular injury. Purpose To evaluate an artificial intelligence (AI)-based model including automated MVO segmentation and radiomic feature extraction to decode microvascular damage heterogeneity, and to assess its ability to predict major adverse cardiovascular events (MACEs). Materials and Methods This multicenter retrospective study (June 2013-December 2023) included patients with STEMI and MVO who underwent cardiac MRI. A previously developed AI model was applied for automated MVO analysis, followed by least absolute shrinkage and selection operator (LASSO) regression for dimensionality reduction to construct a radiomic score (radscore). The primary outcome was MACEs, including cardiovascular death, myocardial reinfarction, malignant arrhythmia, and hospitalization for heart failure. Restricted cubic spline analysis was performed to examine the potentially nonlinear relationship between the radscore and MACE risk. Results Among the 843 patients with STEMI (median age, 60 years [IQR, 51-67 years]; 760 male patients; training set, n = 387; validation set, n = 166; external test set, n = 290), 190 experienced MACEs. The AI model segmented the MRI scans, from which 1595 radiomic features were extracted, and LASSO regression was used to obtain six features for constructing the radscore. Patients with MACEs had higher radscores than those without MACEs (mean, -0.98 ± 0.50 [SD] vs -1.42 ± 0.50; P < .001). Compared with conventional MVO volume quantification, the radscore demonstrated greater prognostic value. The radscore emerged as an independent predictor of MACEs (hazard ratio, 4.20 [95% CI: 3.19, 5.53]; P < .001) and contributed to optimizing risk stratification. Integrating the radscore with conventional variables enhanced prognostic performance, with the C index increasing from 0.77 (95% CI: 0.73, 0.81) for conventional variables alone to 0.80 (95% CI: 0.77, 0.83) for conventional variables plus radscore (P < .001). Conclusion AI-automated MVO radiomic analysis effectively predicted MACE risk and outperformed conventional quantitative assessment. © RSNA, 2026 Supplemental material is available for this article.