Stacking ensemble model combining MRI habitat radiomics and clinical factors predicts lymph node metastasis in early cervical cancer.

A Meta-learner integrating intratumoral and peritumoral MRI radiomics with clinicopathological variables predicted lymph node metastasis with AUCs of 0.915 in training, 0.895 in internal validation, and 0.875 in external testing, outperforming all base learners.

• The Meta-learner used a stacking ensemble of five base classifiers (including SVM) and an XGBoost second-level model, and DeLong tests confirmed it significantly outperformed each individual base learner (all P < 0.05).
• SVM-based ITH_integrated_score was the strongest single base learner, with AUCs of 0.882, 0.865, and 0.845 across the three cohorts.
• The study was retrospective with a prospective validation cohort (n=125); the 623-patient dataset included training (n=311), internal validation (n=187), and external validation cohorts.

Automated summary

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