The Non-Local Bootstrap – Estimation of Uncertainty in Diffusion MRI
Diffusion MRI is a noninvasive imaging modality that allows for the estimation and visualization of white matter connectivity patterns in the human brain. However, due to the low signal-to-noise ratio (SNR) nature of diffusion data, deriving useful statistics from the data is adversely affected by different sources of measurement noise. This is aggravated by the fact that the sampling distribution of the statistic of interest is often complex and unknown. In situations as such, the bootstrap, due to its distribution-independent nature, is an appealing tool for the estimation of the variability of almost any statistic, without relying on complicated theoretical calculations, but purely on computer simulation. In this work, we present new bootstrap strategies for variability estimation of diffusion statistics in association with noise. In contrast to the residual bootstrap, which relies on a predetermined data model, or the repetition bootstrap, which requires repeated signal measurements, our approach, called the non-local bootstrap (NLB), is non-parametric and obviates the need for time-consuming multiple acquisitions. The key assumption of NLB is that local image structures recur in the image. We exploit this self-similarity via a multivariate non-parametric kernel regression framework for bootstrap estimation of uncertainty. Evaluation of NLB using a set of high-resolution diffusion-weighted images, with lower than usual SNR due to the small voxel size, indicates that NLB is markedly more robust to noise and results in more accurate inferences.
KeywordsBootstrap Sample Kernel Regression Diffusion Signal Diffusion Magnetic Resonance Imaging Rician Noise
Unable to display preview. Download preview PDF.
- 3.Coupé, P., Yger, P., Prima, S., Hellier, P., Kervrann, C., Barillot, C.: An optimized blockwise nonlocal means denoising filter for 3-D magnetic resonance images. IEEE Transaction on Medical Imaging 27, 425–441 (2008)Google Scholar
- 5.Davison, A., Hinkley, D.: Bootstrap Methods and their Application. Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press (1997)Google Scholar
- 6.Efron, B., Tibshirani, R.J.: An Introduction to the Bootstrap. Monographs on Statistics and Applied Probablilty. Chapman and Hall (1994)Google Scholar
- 7.Fan, J., Gijbels, I.: Local Polynomial Modelling and Its Applications. Monographs on Statistics and Applied Probablilty. Chapman and Hall (1996)Google Scholar
- 9.Härdle, W.: Applied Nonparametric Regression. Cambridge University Press (1992)Google Scholar
- 11.Härdle, W., Müller, M.: Multivariate and semiparametric kernel regression. In: Schimek, M.G. (ed.) Smoothing and Regression: Approaches, Computation, and Application. Wiley & Sons, Inc., Hoboken (2000)Google Scholar
- 14.Jeurissen, B., Leemans, A., Tournier, J.D., Sijbers, J.: Can residual bootstrap reliably estimate uncertainty in fiber orientation obtained by spherical deconvolution from diffusion-weighted MRI? In: Proceedings 14th Annual Meeting of the Organization of Human Brain Mapping (2008)Google Scholar
- 16.Johansen-Berg, H., Behrens, T.E. (eds.): Diffusion MRI — From Quantitative Measurement to In-Vivo Neuroanatomy. Elsevier (2009)Google Scholar
- 25.Silverman, B.: Density Estimation for Statistics and Data Analysis. Monographs on Statistics and Applied Probablilty. Chapman and Hall (1998)Google Scholar