TY - GEN
T1 - The non-local bootstrap - Estimation of uncertainty in diffusion MRI
AU - Yap, Pew Thian
AU - An, Hongyu
AU - Chen, Yasheng
AU - Shen, Dinggang
PY - 2013
Y1 - 2013
N2 - 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.
AB - 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.
UR - http://www.scopus.com/inward/record.url?scp=84901252332&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-38868-2_33
DO - 10.1007/978-3-642-38868-2_33
M3 - Conference contribution
C2 - 24683985
AN - SCOPUS:84901252332
SN - 9783642388675
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 390
EP - 401
BT - Information Processing in Medical Imaging - 23rd International Conference, IPMI 2013, Proceedings
T2 - 23rd International Conference on Information Processing in Medical Imaging, IPMI 2013
Y2 - 28 June 2013 through 3 July 2013
ER -