(AS1) Clinical Outcomes Using Pre-operative Prediction of Ki-67 Proliferative Index in Convexity Meningiomas Using a Radiomics-derived Machine Learning Algorithm
Neurosurgery Resident, PGY-7 Thomas Jefferson University Hospital Philadelphia, Pennsylvania, United States
Disclosure(s):
Omaditya (Goldey) Khanna, MD: No financial relationships to disclose
Introduction: The WHO grading system for meningiomas is inadequate in its ability to predict tumor aggressiveness and inherent risk of recurrence. In this study, we seek to identify convexity radiomic signatures using pre-operative multiparametric MRI using Ki-67 as a prognostic marker of clinical outcomes in convexity meningiomas, independent of WHO grade.
Methods: A retrospective analysis was conducted of all surgically-resected meningiomas between 2012-2018. Pre-operative MRI scans underwent high-throughput radiomic feature extraction and subsequent development of a support vector machine algorithm to stratify meningiomas based on Ki-67 < 5% and ≥5%, and clinical outcomes were assessed based on performance of this model.
Results: A total of 90 convexity meningiomas are included (WHO grade I N=73, grade II N=12, grade III N=5). The overall rate of recurrence was 15.6% (grade I: 12.5%, grade II: 16.7%, grade III: 80%). On multivariable analysis of several various clinical, histopathologic and radiographic findings, higher WHO grade (HR 3.99 (95% CI: 1.12-14.18)) and elevated Ki-67 >5% (HR 4.65 (95% CI: 1.11-19.41)) were identified as factors that portend an increased risk of recurrence after surgical resection. A total of 46 high-performing radiomic features (1 morphologic, 7 intensity-based, and 38 textural) were identified and used to train a machine learning model. A radiomic signature score was calculated based on machine learning algorithm and normalized via sigmoid function Prediction of Ki-67 by machine learning classifier revealed shorter progression free survival (PFS) for meningiomas with Ki-67 >5% compared to tumors with Ki-67 < 5% (log-rank: p< 0.0004), which corroborates divergent patient outcomes seen using histopathological Ki-67 (log-rank: p=0.0004).
Conclusion : In this study, we trained a machine learning model to stratify convexity meningiomas based on Ki-67, independent of WHO grade, and evaluate its predictive utility as a marker of outcomes. Radiomics-derived diagnostics could provide clinicians with enhanced tumor diagnostics that could deliver personalized, patient-specific treatment strategies.