Computer-Extracted Texture Features to Distinguish Cerebral Radionecrosis from Recurrent Brain Tumors on Multiparametric MRI: A Feasibility Study

• DOI: 10.3174/ajnr.a4931
• Publication Date: 2016-09-15T23:13:17Z
• Authors: P. Tiwari; P. Prasanna; L. Wolansky; M. Pinho; M. Cohen; A.P. Nayate; A. Gupta; G. Singh; K.J. Hatanpaa; A. Sloan; L. Rogers; A. Madabhushi
• Keywords: Fluid-attenuated inversion recovery

<h3>BACKGROUND AND PURPOSE:</h3> Despite availability of advanced imaging, distinguishing radiation necrosis from recurrent brain tumors noninvasively is a big challenge in neuro-oncology. Our aim was to determine the feasibility radiomic (computer-extracted texture) features differentiating on routine MR imaging (gadolinium T1WI, T2WI, FLAIR). <h3>MATERIALS METHODS:</h3> A retrospective study tumor performed 9 months (or later) post-radiochemotherapy 2 institutions. Fifty-eight patient studies were analyzed, consisting training (<i>n</i> = 43) cohort one institution and an independent test 15) another, with surgical histologic findings confirmed by experienced neuropathologist at respective Brain lesions manually annotated expert neuroradiologist. set extracted for every lesion each sequence: gadolinium FLAIR. Feature selection used identify top 5 most discriminating sequence cohort. These then evaluated support vector machine classifier. The classification performance compared against diagnostic reads neuroradiologists who had access same sequences FLAIR) as <h3>RESULTS:</h3> On cohort, area under receiver operating characteristic curve highest FLAIR 0.79; 95% CI, 0.77–0.81 primary 22); 0.79, 0.75–0.83 metastatic subgroups 21). Of 15 holdout classifier identified 12 correctly, while neuroradiologist 1 diagnosed 7 8 respectively. <h3>CONCLUSIONS:</h3> preliminary results suggest that may provide complementary information improve distinction recurrence both tumors.