Removing rician bias in diffusional kurtosis of the prostate using real‐data reconstruction

Abstract

Purpose To compare prostate diffusional kurtosis imaging (DKI) metrics generated using phase‐corrected real data with those generated using magnitude data with and without noise compensation (NC). Methods Diffusion‐weighted images were acquired at 3T in 16 prostate cancer patients, measuring 6 b‐values (0‐1500 s/mm2), each acquired with 6 signal averages along 3 diffusion directions, with noise‐only images acquired to allow NC. In addition to conventional magnitude averaging, phase‐corrected real data were averaged in an attempt to reduce rician noise‐bias, with a range of phase‐correction low‐pass filter (LPF) sizes (8‐128 pixels) tested. Each method was also tested using simulations. Pixelwise maps of apparent diffusion (D) and apparent kurtosis (K) were calculated for magnitude data with and without NC and phase‐corrected real data. Average values were compared in tumor, normal transition zone (NTZ), and normal peripheral zone (NPZ). Results Simulations indicated LPF size can strongly affect K metrics, where 64‐pixel LPFs produced accurate metrics. Relative to metrics estimated from magnitude data without NC, median NC K were lower (P < 0.0001) by 6/11/8% in tumor/NPZ/NTZ, 64‐LPF real‐data K were lower (P < 0.0001) by 4/10/7%, respectively. Conclusion Compared with magnitude data with NC, phase‐corrected real data can produce similar K, although the choice of phase‐correction LPF should be chosen carefully.

Quantitative diffusion MRI of the abdomen and pelvis

“…This effect may confound the measurement of quantitative diffusion parameters, for example, by introducing bias (underestimation) in ADC 27,44–46 (Figure 7). In more advanced nonmonoexponential signal models, noise floor effects may lead to an artifactual estimation of restricted diffusion (e.g., overestimation of the kurtosis parameter K ) 47 . A variety of image reconstruction, filtering, and postprocessing techniques have been proposed to ameliorate noise floor effects in quantitative diffusion MRI 27,47–49 …”

“…In more advanced nonmonoexponential signal models, noise floor effects may lead to an artifactual estimation of restricted diffusion (e.g., overestimation of the kurtosis parameter K). 47 A variety of image reconstruction, filtering, and postprocessing techniques have been proposed to ameliorate noise floor effects in quantitative diffusion MRI. 27,[47][48][49] Other artifacts and limitations include fat suppression errors, gradient nonlinearities, partial volume effects, and general imaging artifacts.…”

“…47 A variety of image reconstruction, filtering, and postprocessing techniques have been proposed to ameliorate noise floor effects in quantitative diffusion MRI. 27,[47][48][49] Other artifacts and limitations include fat suppression errors, gradient nonlinearities, partial volume effects, and general imaging artifacts.…”