2025 Proffered Presentations
S010: MACHINE LEARNING WITH RADIOMICS TO PREDICT WHO PATHOLOGIC GRADE OF MENINGIOMAS FROM PREOPERATIVE MRI: A MULTICENTER STUDY
Syed M Adil, MD; Pranav I Warman, BSE; Kaizar Rangwala; Andreas Seas, BS; Tanner J Zachem, BSE; Jihad Abdelgadir, MD, MSc; Evan Calabrese, MD, PhD; Patrick J Codd, MD; Anoop Patel, MD; Ali Zomorodi, MD; Duke University
Introduction: Meningiomas are the most common primary brain tumor. Their management, including surgical aggression, adjuvant therapies, and surveillance strategy, depends on their World Health Organization (WHO) pathologic grade. Currently, there are no methods to reliably predict this preoperatively. Radiomics are a set of features quantifying radiographic phenotypes and have shown promise in predicting clinical outcomes for other pathologies. Here, we apply machine learning to radiomics features derived from preoperative MRIs to predict meningiomas’ WHO pathologic grade.
Methods: Preoperative MRIs from 700 patients with meningiomas from six hospitals were obtained. Each MRI was segmented into three separate masks: enhancing tumor, non-enhancing/necrotic tumor, and surrounding FLAIR abnormality. The segmentation masks were processed to create radiomics features describing the image intensity, shape, and texture. These features, without other clinical variables, were fed into a custom machine learning pipeline to predict binarized pathologic grade of the tumor (Grade 1 versus Grade 2/3). Training, testing, and validation data were separated using nested cross-validation, and model performance was determined by area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity.
Results: The total cohort comprised 525 (75%) meningiomas that were grade 1, 158 (23%) that were grade 2, and 17 (2%) that were grade 3. The final machine learning model achieved an AUROC of 0.70 (95% confidence interval [CI]: 0.65 to 0.76), sensitivity of 68.5% (95% CI: 51.7% to 85.3%) and specificity of 69.3% (95% CI: 58.7% to 80.0%).
Conclusion: Machine learning applied to meningioma radiomics may allow prediction of WHO grade preoperatively to help guide surgical strategy, surveillance decisions, and patient counseling. Future iterations may improve the model’s fidelity by incorporating more radiomic features and other clinical variables.