Meeting Banner
Abstract #1920

Track-Density Imaging & Noise: When Super-Resolution Quality Does Not Yield Accuracy

SUMMA25Thijs Dhollander1, 2, Louise Emsell1, 3, Wim Van Hecke1, 3, Frederik Maes1, 2, Stefan Sunaert1, 3, Paul Suetens1, 2

1Medical Imaging Research Center (MIRC), K.U.Leuven, Leuven, Belgium; 2Department of Electrical Engineering (ESAT), K.U.Leuven, Leuven, Belgium; 3Department of Radiology, University Hospitals of the K.U.Leuven, Leuven, Belgium

Track-density imaging (TDI) is a recently proposed technique to achieve super-resolution from diffusion weighted imaging datasets, by performing massive fiber tracking and counting the tracks in each voxel of a high resolution grid. In this work, the effect of noise in the data on the final TDI is investigated. The findings indicate that caution is needed: many discovered patterns of structures might mostly be caused by noise rather than true anatomy, although they look plausible and do not resemble the traditional impression of noise. Tools such as bootstrapping, combined with certain maps can highlight regions where caution is due.

Keywords

accuracy achieved advised amongst amplitude anatomical anatomy arise automated available basically beyond blue bootstrap brain brightly calculating caused caution channels coefficient combined consistent constrained contrast corresponds count counting dataset datasets deconvolution defined density detailed diffusion directionally discovered distance distinct drops easily electrical equals examples extracted fact fiber free full general generated give gradient great greater grid healthy hidden highest impression indicates indicators initiate intensities investigation isotropic kept larger leaving length leverage lies local long looks made many mapped mapping maps mask might mostly movie noise numerical original others outcome overlaid package particular parts pathology pattern patterns phantom pipeline pitfall plausible potential press probabilistic properties quality quantification radius random rather reachability real realizations realized recovered requires residual resolution robust scale scanned scanner software specified spherical starts strategies structural structures super terminated terms thalamus though threshold thresholding track tracking tracks traditional true typical uniformly user validation variation varying volumes white whole wise yield