Accurate estimation of diffusion kurtosis (DK) imaging parameters of the prostate

Abstract

Purpose: To evaluate the performance of maximum likelihood (ML) estimation of diffusion kurtosis (DK) imaging parameters in the prostate and compare the estimated parameters to those measured using least squares (LS) estimation. Materials and methods: The institutional review board issued a waiver of informed consent for this Health Insurance Portability and Accountability Act (HIPAA)-compliant, retrospective study of forty-two patients (median [Md] age=61 years; range: 43-74 years) who underwent magnetic resonance imaging (MRI) between September and October 2016. Diffusion-weighted MRI (DW-MRI) at nine b-values (0-2000 s/mm2 ) were acquired using a 3-T whole-body MRI unit (Discovery MR750; GE Medical Systems, Waukesha, WI) equipped with an eight-channel phased array coil for signal reception. Diffusion coefficient (D) and kurtosis (K) were estimated from the normal appearing prostate peripheral zone and prostate cancer regions of interest (ROIs). The parameters were estimated by fitting the measured MR signal intensities as a function of b-value, using LS and ML algorithms. An estimate of the noise was obtained on the b=0 images in an artifact-free ROI in the rectum. Simulations were also carried out to assess the properties of the two estimators in a range of signal-to-noise ratios. Results: For benign ROIs, the mean D ± standard deviation, (1.88±0.52)×10-3 mm2 /sec, and mean K (0.79±0.20), measured using LS estimation, differed significantly from the mean D (1.96±0.48)×10-3 mm2 /sec and mean K (0.68±0.21), measured using ML estimation (P<0.001 for both). For malignant ROIs, the mean D (1.48±0.38)×10-3 mm2 /sec and mean K (0.94±0.20), measured using LS estimation, differed significantly from the mean D (1.54±0.36)×10-3 mm2 /sec and mean K (0.81±0.19), measured using ML estimation (P<0.001 for both). Simulations demonstrate that ML minimizes the bias estimate of DK parameters within the signal-to-noise ratio range of 5-15. Conclusion: By incorporating the noise level, the ML estimation increases the accuracy of DK parameter estimation. In vivo results with phased array coils showed significant differences in DK parameter estimates with ML as compared with the standard LS estimation.