Diffusion measurements and diffusion tensor imaging with noisy magnitude data.

PURPOSE

To compare an unbiased method for estimation of the diffusion coefficient to the quick, but biased, log-linear (LL) method in the presence of noisy magnitude data.

MATERIALS AND METHODS

The magnitude operation changes the signal distribution in magnetic resonance (MR) images from Gaussian to Rician. If not properly taken into account, this will introduce a bias in the estimated diffusion coefficient. We compare two methods by means of Monte Carlo simulations. The first one applies least-squares fitting of the measured signal to the median (MD) value of the probability density function. The second method is uncorrected LL estimation. We also perform a high-resolution diffusion tensor experiment.

RESULTS

The uncorrected LL estimator is heavily biased at low signal-to-noise ratios. The bias has a significant effect on image quality. The MD estimator is accurate and produces images with excellent contrast.

CONCLUSION

In the presence of noisy magnitude data, unbiased estimation is essential in diffusion measurements and diffusion tensor imaging.