Meeting Banner
Abstract #0856

Repeated Split Sample Validation to Assess Logistic Regression Model of DTI in Differentiating Glioblastomas from Brain Metastases

Sumei Wang1, Sang Joon Kim1, 2, Matthew R. Voluck1, Ronald L. Wolf1, Donald M. ORourke3, Harish Poptani1, Elias R. Melhem1, Sungheon Kim4

1Radiology, University of Pennsylvania, Philadelphia, PA, United States; 2Radiology, University of Ulsan Asan Medical Center, Seoul, Korea, Republic of; 3Neurosurgery, University of Pennsylvania, Philadelphia, PA, United States; 4Radiology, New York University School of Medicine, New York, NY, United States

One hundred and twenty nine glioblastomas and 74 brain metastases were included in this study. FA and MD were measured from the enhancing and immediate peritumoral region. FA values from the enhancing and immediate peritumoral regions for glioblastomas were significantly higher than those for brain metastases. Repeated split sample validation confirmed that FA and MD from the enhancing part is a robust model for the distinction between glioblastomas and brain metastases.

Keywords

acceleration adhesion amounts anisotropy array assess assumed automatically best better brain breast cancer cells challenging channel characteristic classification classifier coefficient coefficients coil colon combined components computed confirmed constant contrast curve curves dataset determine diagnosis differentiate differentiated differentiating differentiation diffusion diffusivity distinction edema employed enhanced enhancement enhancing enlarged esophageal examination exhibit extensive extracellular flair fractional full head house immediate include included indicated indicates lesions logistic lung maps mass materials matrix measured measures medical medicine melanoma metastases metastasis metrics migration mixed model multivariate needs neurosurgery operative orientation overproduced parallel part patients population post previous primary prior processing produce radiology randomly receiver regression remaining remains renal repeated reported representative republic ring robust routines sample sang sarcoma scanner school segmented semi sensitivity serve shot significantly sites slice software solid space specificity spin split splitting statistically studies subject subsequent substrate surgery table tissue together training trio tumor types underwent unpaired useful validated validation validations wolf