AbstractID: 7786 Title: 3-D anisotropic diffusion and wavelet filtering of Monte
Carlo dose distribution
The Monte Carlo transport is the most accurate method for radiotherapy dose
calculations; however, within limited time, Monte Carlo calculation introduces
statistical noise into the result. In our work, two- and three-dimensional
post-processing denoising methods were studied to reduce the statistical fluctuation of
the Monte Carlo dose distributions. Wavelet denoising method, anisotropic diffusion
method and pixel-wise adaptive Wiener method were investigated and compared.
The MCNP4C Monte Carlo code with a special lattice geometry patch was used to
generate dose distributions with different level of statistical noise in homogeneous
and inhomogeneous phantom. All the denoising results were compared with the
results obtained with a large number of simulated histories (typically 10 9). The mean
square error method was used to evaluate the difference between the distributions.
Extensive studies were performed to investigate importance and sensitivity of the
results to the input parameters of the investigated filters.
It is shown in our work that both anisotropic diffusion and wavelet denoising method
can reduce the statistical noise significantly while well preserving important gradients
of the dose distribution. The adaptive Wiener method was found to be inferior to the
other two methods. The anisotropic diffusion method worked better than the wavelet
method in most of the investigated cases and is considered preferable method for
noise reduction in Monte Carlo calculated dose distributions.