Meeting Banner
Abstract #2980

Comparison of Three Non-Gaussian Diffusion Models for Differentiation of Pediatric Brain Tumors

Yi Sui1, 2, Guanzhong Liu1, He Wang3, Frederick C. Damen1, 4, Yuhua Li5, Xiaohong Joe Zhou1, 6

1Center for MR Research, University of Illinois Hospital & Health Sciences System, Chicago, IL, United States; 2Department of Bioengineering, University of Illinois at Chicago, Chicago, IL, United States; 3Applied Science Lab, GE Healthcare, Shanghai, China; 4Department of Radiology, University of Illinois Hospital & Health Sciences System, Chicago, IL, United States; 5Department of Radiology, Xinhua Hospital, Shanghai, China; 6Departments of Radiology, Neurosurgery and Bioengineering, University of Illinois Hospital & Health Sciences System, Chicago, IL, United States

A systematic comparison of non-Gaussian diffusion models is conducted in the context of differentiating pediatric brain tumors. Three non-Gaussian diffusion models - FROC, kurtosis and bi-exponential models - were selected for evaluation of their performance for differentiating low-grade from high-grade pediatric brain tumors. Our results suggested that FROC model had the best performance, although the other two models also produced excellent results. In conclusion, non-Gaussian diffusion models with high b-values (up to 4000 s/mm2) can provide valuable and reliable information for characterizing pediatric brain tumors.

Abstract PDF Presentation Video

Keywords

absolute according accuracy advantages among apparent applied assessment assumes audience basis best better binary bioengineering bold brain calculus cancer carried characteristic characterization characterizing children china clinical coefficient combination combining conducted considerable context contrast curves customized dataset determine developed deviation diagnosis differential differentiating differentiation diffusion distribution easily entire environmental error evaluate evaluated evaluation exhibited experimental exponential extracted falls find fitting flair fractional gave gold grade guided health healthy heterogeneity heterogeneous highlight hospital important increasingly individual input irrespective knowledge kurtosis labeled logistic maier maker markers measure median merit micro model models moderate mono neurosurgery offer operating others parametric particularly past patient pediatric percentage performance process produced proven providing quantify radiologists radiology receiver regression related reliable reported researchers respectively scanner science sciences selected serve shanghai short significance simple slices slightly software solid space statistical stretch studies suggest system systematic target tissue tissues treatment tumor tumors typically useful valuable written years