Prognostic Radiomic Markers of Posterior Fossa Ependymoma. (16th November 2020)

• Record Type: Journal Article
• Authors: Yecies, Derek W; Tam, Lydia; Han, Michelle; Jabarkheel, Rashad; Mankad, Kshitij; Lober, Robert; Cheshier, Samuel H; Vitanza, Nick; Hargrave, Darren; Jacques, Thomas; Aquilina, Kristian; Grant, Gerald A; Taylor, Michael D; Ramaswamy, Vijay; Yeom, Kristen
• Is Part Of: Neurosurgery. Volume 67(2010)Supplement 1
• Publication Date: 2020-11-16
• Subjects: Nervous system -- Surgery -- Periodicals; 617.48005
• DOI: 10.1093/neuros/nyaa447_575
• Languages: English
• ISSNs: 0148-396X

Abstract

Abstract: INTRODUCTION: Posterior fossa ependymoma (PFE) is a common pediatric brain tumor that is almost invariably assessed with MRI prior to surgical treatment. Advanced radiomic analysis has shown promise in stratifying risk and outcome in other pediatric brain tumors. METHODS: Tumor volumes were manually contoured on T1-post contrast and T2-weighted MR images for high-dimensional feature extraction using PyRadiomics. We extracted 900 features on each image series comprising first-order statistics, size, shape, and texture metrics calculated on the original, log-sigma, and wavelet transformed images. Progression free survival (PFS) served as our outcome of interest. 10-fold cross-validation of a LASSO Cox regression was used to predict PFS (glmnet, R Software). We analyzed performance of our model using clinical variable (age at diagnosis and sex) only, radiomics only, and radiomics plus clinical variable. Concordance (C) metric was used to assess performance of the Cox model. RESULTS: 93 children from five centers (median age 3.3 years; 59 males) met the inclusion criteria. The mean PFS was 50 months with 50% (47/93) of patients progressing. Six radiomic features were selected, all for T1 imaging, including 1 first-order kurtosis (log-sigma) and 5 texture features (3 wavelet, and 2 original). This model demonstrated significantly higher performance than a clinical model alone (C: 0.69 vs 0.58, P < .001). Adding clinical features to the radiomic features did not improve prediction ( P = .67). For the subset of patients with molecular subtyping (n = 48), adding this feature to the clinical plus radiomics models significantly improved performance over the clinical features alone (C = 0.79 vs. 0.66, P = .02). CONCLUSION: Our pilot study shows a potential role for MRI-based radiomics and machine learning for PFE risk stratification and as radiographic biomarkers.