Shiou-Ping Lee, Presenter: Nothing to Disclose

Chung-Ming Cheng, Abstract Co-Author: Nothing to Disclose

  Fiber direction estimated from the high angular resolution diffusion imaging (HARDI) data is subject to uncertainty due to the sampling noise. To evaluate the effect of uncertainty on the fiber orientation for different reconstruction schemes, in this study, two types of wild bootstrap resampling methods were used for the constrained spherical deconvolution (CSD) and the Q-Ball imaging (QBI) reconstruction algorithms.

  A simulated two-tensor profile was used with the following parameters: fractional anisotropy (FA)=0.8, crossing angle=60º, b-value=3000 s/mm2, and SNR=25. A single HARDI dataset used 60 diffusion gradients data.   500 samples were generated. For each sample, a wild bootstrap procedure repeating 50 times was used to calculate the deviation angle between the gold standard and the estimated fiber orientation. The CSD-based wild bootstrap and QBI-based wild bootstrap used the maximal 4 SH order as a noisy data to reconstruct the FODs/ODFs in super-CSD, CSD and QBI.   For each FOD/ODF, the opaque surface corresponded to the mean FOD/ODF over 50 tests, the transparent surface corresponded to the mean + 2 standard deviations, and the blue line corresponded to the local maximal fiber amplitude. The non-parametric Kolmogorov-Smirnov Z test (p<0.05) was used to test if the mean deviation angles of two groups were significantly different from each other.  

  For CSD-based wild bootstrap analysis on Fig.1(left), the mean (±std) of the deviation angles of three methods were 4.015±1.146, 4.059±0.870 and 4.080±0.986, respectively. There was no significant difference between the deviation angles of super-CSD and QBI (P=0.059) as well as between those of CSD and QBI (P=0.082).   For QBI-based wild bootstrap analysis on Fig.1(right), the mean (±std) of the deviation angles of three methods were 5.049±1.121, 6.550±1.831 and 6.405±1.608, respectively. There was no significant difference between the deviation angles of CSD and QBI (P=0.162). 

  For both of the CSD-based and the wild bootstraps, there was no significant difference between the deviation angles of CSD and QBI. Additionally, of the CSD-based wild bootstrap, the super-CSD had the lowest mean deviation angle, which may potentially provide the more accurate fiber orientation under noise-induced variability.   

Of CSD-based wild bootstrap, super-CSD provided the more accurate fiber orientation for tractography researches.

Lee, S, Cheng, C, Two Types of Wild Bootstrap to Quantify Uncertainty Based on CSD and QBI.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11000580.html