To find optimal b-value distributions for monoexponential, stretched exponential, kurtosis, and biexponential models of prostate cancer (PCa) diffusion-weighted imaging (DWI) using simulations and repeated DWI examinations.

Simulations aiming to minimize estimation accuracy error were performed. Ten PCa patients underwent in total four repeated 3-tesla DWI examinations using 12 equally spaced b values (0-2000 s/mm2 ). Normalized mean signal intensities of regions-of-interest placed in normal tissue and PCa were fitted. In total, 210 different b-value combinations consisting of six b values, 0 and 100 s/mm2 included in every b-value distribution, were evaluated in terms of accuracy and repeatability.

The simulations and in vivo DWI data suggest the optimal b-value distribution for the monoexponential model consists of four to five equally distributed b values in the range of 0 to 1200 s/mm2 . The parameters of the stretched exponential and kurtosis models are best estimated using five to seven b values in the ranges of 300 to 700 and close to 2000 s/mm2 , in addition to low b value. B-value distribution consisting of eight to 10 b values in the ranges of 0 to 100, 800 to 1200, and 1800 to 2000 s/mm2 is the preferred method for estimation of the biexponential model parameters of PCa DWI.

CONCLUSION:

The optimized b-value distributions demonstrated improved estimation accuracy and repeatability of DWI signal decay-derived parameters. Magn Reson Med 73:1954-1969, 2015. (c) 2014 Wiley Periodicals, Inc.