SNR Estimation in Fast Dynamic Imaging Using Bootstrapped Statistics

Introduction: Evaluation of local signal-to-noise ratio (SNR) is essential for many image-based analyses. Novel pulse sequence technology for dynamic imaging, such as HYPR [1], requires an alternative to conventional SNR estimation methods, since the SNR can vary in unexpected ways over very small distances; the typical assumption of near constant noise level throughout the object of interest [2] no longer applies. Other methods [3] need some model of this noise behavior for accurate results. Here we propose a new technique that accurately estimates pixel-wise SNR using a single raw data set and collected noise data. Our method is based on bootstrap statistics. Estimates of the SNR statistics are derived by synthesizing multiple image repetitions by randomly reordering the collected noise data and combining it with the original raw data. This work validates the bootstrap statistics method by comparing its SNR estimation with conventional techniques and demonstrates its extension in HYPR angiography.