Association of radiomic skeletal muscle features on prostate T1-weighted MRI with major adverse cardiac events in prostate cancer patients.

Abstract

34 Background: Prostate cancer (PCa) is the most common malignancy among men, yet cardiovascular disease (CVD) remains a leading cause of morbidity and mortality. Recent studies show that baseline skeletal-muscle strength predicts major adverse cardiac events (MACE). As prostate MRI (pMRI) is routine in PCa care, we evaluated whether radiomic features (RFs) of pelvic skeletal muscle predict MACE. Methods: We retrospectively analyzed 225 men with PCa (2010–2025). Baseline 3 T T1-weighted pMRI scans were used to segment the obturator internus, externus, and pectineus muscles. Thirty scans were manually annotated to train an nnUNetv2 model; remaining scans were auto-segmented and verified under radiologist supervision. Haralick/GLCM, gradient-derived, and CoLlAGe (co-occurrence of local anisotropic gradient orientations) features were extracted per IBSI standards. Correlated RFs were removed, significance tested (Wilcoxon p < 0.05), and top five selected by mRMR; univariate Cox models identified RFs with highest concordance index. Results: Of 225 patients, 52 (23.1 %) developed MACE during a median follow-up of 1,641 days. After FDR correction, CoLlAGe Entropy [Gradient Orientations (ColEGO) S4N3; q = 0.012] and ColEGO (S3N3; q = 0.028) remained significant. Five-fold cross-validation AUC = 0.64 ± 0.11 (L1-logistic) and 0.61 ± 0.03 (XGBoost). Conclusions: Pelvic skeletal-muscle RFs from routine pMRI associate with MACE in PCa. These RFs may enable noninvasive cardio-oncology risk stratification, supporting evidence that muscle strength predicts cardiac events. Prospective validation is warranted. RF (IBSI-style descriptive name) Context Test / HR (95 % CI) p/q value Direction Clinical Interpretation ColEGO S4 N3 Binary U = 3179, AUC = 0.65 p = 0.001, q = 0.01 Decreased in MACE Lower entropy means more homogeneous fiber pattern and healthier architecture. ColEGO S3 N3 Binary U = 3373, AUC = 0.63 p = 0.006, q = 0.03 Decreased in MACE Lower entropy reflects preserved muscle integrity and lower CVD risk. GLCM Correlation [Haralick (GLCMH) 3 S7 N3] Binary U = 5431, AUC = 0.60 p = 0.024 Increased in MACE Higher correlation suggests fibrotic/lipid texture linked to adverse remodeling. Gradient Magnitude Variance (GMV) S4 N3 Binary U = 3568, AUC = 0.6 p = 0.024 Decreased in MACE Lower variance suggests smoother edges and uniform density in healthy muscle. GLCMH 3 S8 N4 Binary + Survival U = 3605, AUC = 0.6 / HR = 0.68 (0.50–0.93) p = 0.030 / 0.016 Decreased in MACE (lower hazard) Greater uniformity implies preserved myofiber structure and reduced event risk. GLCMH 3 S8 N3 Survival HR = 0.47 (0.22–0.99) p = 0.046 Decreased in MACE (lower hazard) Predictable texture patterns associate with lower cardiovascular risk. GMV S12 N3 Survival HR = 1.12 (1.01–1.25) p = 0.039 Increased in MACE (higher hazard) Larger-scale variance indicates heterogeneous architecture consistent with fibrosis.