The denoising of Monte Carlo dose distributions using convolution superposition calculations

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Imaging

Background:

• Monte Carlo (MC) dose calculations offer high accuracy but are computationally demanding.
• Convolution Superposition (CS) methods are faster but rely on approximations, leading to smoother, less detailed dose distributions.
• Bridging the accuracy of MC with the speed of CS is crucial for efficient radiotherapy planning.

Purpose of the Study:

• To investigate MC denoising techniques that leverage CS results to accelerate MC dose calculations.
• To develop and evaluate methods for combining CS data with MC simulations for improved dose distribution accuracy and reduced noise.
• To assess the computational efficiency and noise reduction capabilities of these hybrid approaches.

Main Methods:

• Developed two primary approaches for MC denoising guided by CS data.
• Approach 1: Iterative denoising of the residual difference between MC and CS images, initialized with CS data, using multi-scale methods (wavelets, contourlets).
• Approach 2: Frequency splitting using quadrature filtering to combine low-frequency MC components (scatter, high-dose regions) with high-frequency CS components (details), employing 3D Butterworth filters.

Main Results:

• Both MC denoising approaches were demonstrated on clinical lung and head and neck cancer cases.
• MC dose distributions were calculated using an open-source MC code with varying noise levels.
• The frequency-splitting technique effectively integrated CS-guided denoising, showing significant promise.

Conclusions:

• MC denoising techniques guided by CS information can accelerate dose calculations while reducing noise.
• The frequency-splitting method is particularly promising for enhancing computational efficiency and noise reduction in MC dose planning.
• These hybrid methods offer a viable path towards faster, more accurate dose distributions in radiation therapy.