Medu-48. Mri Textural Features Can Differentiate Pediatric Posterior Fossa Tumors
Year: 2017
Type: conference-abstract
Abstract: BACKGROUND: Medulloblastoma (MB), Ependymomas (EP), & Gliomas are the most common posterior fossa (PF) tumors in children. Reliable, non-invasive biomarker that can distinguish these tumor types would significantly aid clinical management. We recently developed a new computerized descriptor, CoLlAGe, that distinguishes subtly-different pathologies in adult brain tumors. CoLlAGe measures the orientation of local intensity gradients as an estimate of lesion heterogeneity. In this work, we hypothesize that CoLlAGe can identify subtle micro-architectural differences between MB, EP and Glioma on routine MRI scans. METHODS: A retrospective cohort of 59 studies (MB = 22, EP = 12, Glioma = 25) was acquired. After registration, scans were age normalized, skull-stripped, and corrected for intensity inhomogeneities. Every lesion was then expert delineated to obtain a region of interest (ROI). 52 CoLlAGe features were extracted from every ROI for every MRI sequence (T1w, T2w, FLAIR). Wilcoxon rank sum test was used to identify top 10 CoLlAGe features within a training set of 30 patients (MB = 11, EP = 6, G = 13), for differentiating MB from (EP+Glioma), & Glioma from (MB+EP). Using top features, random forest classifiers were built for every sequence, and validated on an independent set of 29 studies (MB = 11, EP = 6, G = 12) by using area under receiver operating characteristic curve (AUROC). RESULTS: On the independent set, we found that sum variance and entropy of CoLlAGe (p< 0.002) on T2w protocol could reliably distinguish MB from (EP+Gliomas) (AUROC = 0.88), and entropy and sum average of CoLlAGe on T2w (p < 0.0003) could differentiate Gliomas from (MB+EP) (AUROC = 0.72). CONCLUSION: Our results suggest that entropy feature of CoLlAGe, a measure of tumor heterogeneity, enabled discrimination of MB from (EP+Glioma). Prospective validation of these features in a larger cohort may improve personalized treatment management for PF tumors.
Source: Neuro-Oncology
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W2620413290