Abstract A050: Diagnostic accuracy of artificial intelligence powered by machine learning in differentiating WHO grade I and II meningiomas

Abstract

Abstract Primary objective To evaluate the diagnostic accuracy of AI and ML algorithms in differentiating WHO Grade l and Grade ll meningiomas based on preoperative imaging data. Secondary objective To compare the performance of different machine learning models in terms of sensitivity, specificity, and overall accuracy for tumor grading, and explore the feasibility of integrating the best-performing model into clinical workflows. Methods Patients with histopathologically confirmed WHO Grade I or II meningiomas who underwent preoperative MRI were retrospectively included. Radiomics features were extracted from T2-weighted, post-contrast T1-weighted (T1c), and apparent diffusion coefficient (ADC) sequences. The dataset was stratified into training (80%) and testing (20%) subsets. Logistic regression with L2 regularization, random forest, and gradient boosting classifiers were trained using cross-validation for hyperparameter optimization. Model performance was assessed on the independent test set using accuracy, precision, recall, F1-score, and confusion matrices. Results Gradient boosting achieved the best performance on T2 (accuracy 0.68, F1-score 0.68) and T1c sequences (accuracy 0.74, F1-score 0.69), while random forest performed best on ADC maps (accuracy 0.65, F1-score 0.63). Logistic regression consistently underperformed across modalities. Post-contrast T1-weighted imaging provided the strongest predictive features, whereas ADC-derived features were less robust. Conclusion Radiomics-based ML models can non-invasively differentiate WHO Grade I from Grade II meningiomas, with ensemble methods outperforming logistic regression. Gradient boosting and random forest showed the greatest potential, particularly on T1c and T2 sequences. These findings highlight the clinical utility of AI for preoperative grading and underscore the need for multicenter validation, standardized radiomics pipelines, and integration with molecular biomarkers to enable clinical translation. Citation Format: Fahad B. Albadr, Marwan A. Almalki, Faris A. Albassam, Mohammed A. Safhi, Walaa F. Almutawa, Renad A. Alduwayan, Salman T. Althunayan, Sarah A. Alajaji, Salma M. Alsaadoun, Waha A. Almathami, Metab A. Alkubeyyer. Diagnostic accuracy of artificial intelligence powered by machine learning in differentiating WHO grade I and II meningiomas [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Brain Cancer; 2026 Mar 23-25; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2026;86(6_Suppl):Abstract nr A050.

Ähnliche Arbeiten