Meeting Banner
Abstract #0840

Multivariate Pattern Analysis of in vivo MR Imaging Parameters for Detecting Transformations to a Higher Grade in Patients with Recurrent Low Grade Gliomas

Alexandra Constantin1, Llewellyn Jalbert1, Adam Elkhaled2, Rupa Parvataneni2, Annette Molinaro2, Joanna Phillips2, Soonmee Cha2, Susan M. Chang2, Sarah J. Nelson2

1University of California, Berkeley, Berkeley, CA, United States; 2University of California, San Francisco

A multivariate diagnostic model was built to estimate the probability that a recurrent low grade glioma had progressed to a higher grade based on in vivo MR imaging parameters. The model was able to discriminate between recurrent low grade gliomas that upgraded versus those that remained grade 2 with 93% cross-validation accuracy and 84% bootstrapping accuracy, based on the 75th percentile normalized choline height, the 25th percentile recovery to baseline of the perfusion susceptibility curve, the 75th percentile ratio of choline to n-acetylaspartate height, and the maximum choline height inside the T2 lesion.

Keywords

accuracy additive agent allow anatomic anatomical anatomy approaches attenuated automatic becoming bias bootstrapping build built cerebral chemotherapy choline class classification clinical comprised confirmation contrast contribution corrected cross cutoffs define designed desired detecting determine diagnosis diagnostic diffusion distinguish earlier edited enhancing ensemble errors examples extracted feature features final flair fluid fractional good grade guided height histological identify identifying important in vivo included incorrectly indicated individual initial injection input instance intensity invasive lactate leads learning leaves lesion lesions lipid localization logistic malignant manually maps median metabolic model models multivariate nelson normalized objective options originally output overall parametric patients pattern peak percentile performance physiological planning practice predict probability problem progressed progression protocol quality radiation recovery recurred recurrence recurrent regression remained repeatedly resection risks sample samples sampling scheduled selected selecting selection sets subtype supervised suppose surgery surgical susceptibility suspected table therapeutics tissue training transformation transformations tumor tumors unseen upon utility validation variables variance vector volume volumes wrapper