Abstract
BACKGROUND: The preoperative identification of (isocitrate dehydrogenase) IDH-mutant low-grade gliomas (LGGs) is critical for personalized treatment planning. We aimed to develop a streamlined machine-learning model using key clinical features for rapid and accurate preoperative prediction. METHODS: A retrospective cohort of 418 adult glioma patients was partitioned into training (70%) and internal validation (30%) sets. (Support Vector Machine) SVM was selected as the optimal model after comparing 9 machine learning models. Six clinically significant features, ranked by predictive importance, were incorporated into the final SVM model. The model's generalizability was further validated using an independent external cohort (n = 206). RESULTS: The SVM model demonstrated high discriminative performance, achieving an (area under the receiver operating characteristic curve) AUC-ROC of 0.860 (internal validation) and 0.869 (external validation). SHAP (SHapley Additive exPlanations) analysis confirmed age as the most influential predictor, followed by edema and enhanced features, aligning with known biological associations in IDH-mutant LGGs. CONCLUSIONS: This SVM-based model provides a clinically practical tool for the preoperative identification of IDH-mutant LGGs, combining diagnostic reliability, interpretability, and minimal feature requirements. Its robust external validation underscores its potential utility in diverse clinical settings.