A comparison of Monte Carlo dose calculation denoising techniques

Area of Science:

• Medical Physics
• Computational Imaging
• Radiotherapy Physics

Background:

• Monte Carlo (MC) methods are crucial for accurate radiotherapy dose computation.
• Statistical noise in MC dose distributions necessitates denoising for clinical application.
• Existing denoising methods vary in effectiveness and computational efficiency.

Purpose of the Study:

• To compare the performance of novel and established MC denoising techniques.
• To evaluate denoising effectiveness using metrics like mean-square-error (MSE) and Van Dyk criteria.
• To assess the impact of denoising on MC dose computation time and accuracy.

Main Methods:

• Implementation and comparison of 3D wavelet thresholding, content adaptive mean-median-hybrid (CAMH) filtering, locally adaptive Savitzky-Golay (LASG) filtering, anisotropic diffusion (AD), and iterative reduction of noise (IRON).
• Testing on diverse phantom and CT-based MC dose distributions with varying noise levels.
• Quantitative evaluation using MSE, maximum difference, and Van Dyk pass/fail criteria against reference low-noise distributions.

Main Results:

• LASG demonstrated the best MSE reduction for three of four test cases, with up to a factor of 16 improvement on phantom data.
• Wavelet thresholding showed superior performance for head and neck IMRT cases, reducing maximum error more effectively than LASG.
• All evaluated methods generally accelerated MC results toward statistically accurate dose distributions.

Conclusions:

• Denoising algorithms significantly improve MC radiotherapy dose computation accuracy and efficiency.
• LASG and wavelet thresholding are effective MC denoising methods, with optimal choice depending on the specific dose distribution characteristics.
• Further development and application of these denoising techniques can enhance clinical radiotherapy planning.