Making the Robust Tensor Estimation Approach: "RESTORE" more Robust
暂无分享,去创建一个
Methods & Materials: The original RESTORE algorithm used a threshold criterion for outlier identification. The criterion is based on the estimation of signal standard deviation assuming signal variability is affected by the thermal noise in magnetic resonance images (MRIs). The signal standard deviation caused by the white noise is a constant; however, signal variability in non-DWIs are generally much greater than in DWIs in several regions of human brain. The higher signal variability in T2 weighted images may be caused by the high frequency spin inflow effects or due to physiological fluctuations. This fact can be easily observed by comparing the signal intensity in a series of DWIs and non-DWIs, or by comparing the statistics of residuals obtained from the nonlinear least squares fitting. Therefore, the same outlier identification criterion for DWIs may not be suitable for non-DWIs, and a wider tolerance of signal variability should be used for non-DWIs, i.e., the signal standard deviation for non-DWIs will be adjusted based on the following formula: DWIs DWIs non SD ratio signal kappa noise SD × × = − _ _