A preliminary study of a photon dose calculation algorithm using a convolutional neural network.

Area of Science:

• Medical Physics
• Computational Biology
• Radiotherapy

Background:

• Improving dose calculation accuracy and efficiency is crucial for radiation therapy.
• Traditional algorithms like collapsed cone convolution/superposition (CCCS) have limitations in speed and accuracy.

Purpose of the Study:

• To develop and evaluate a novel dose calculation method using convolutional neural networks (CNNs).
• To enhance the efficiency of existing dose calculation algorithms, such as CCCS.
• To improve the accuracy of three-dimensional dose distribution calculations.

Main Methods:

• Utilized three-dimensional total energy release per unit mass (TERMA) and electron density (ED) distributions as inputs for a CNN-based dose calculation.
• Trained the CNN model using 7500 samples from CT images for monoenergetic photon models (0.5-6 MeV).
• Validated the CNN-Dose algorithm against the CCCS algorithm using a 6-MV spectrum and 1512 test samples.

Main Results:

• The CNN-Dose algorithm achieved high accuracy, with mean values of 99.56% (3%/2 mm) and 98.57% (2%/2 mm) across various anatomies.
• The method demonstrated comparable accuracy to the CCCS algorithm while offering potential efficiency improvements.
• Results met clinical requirements, indicating usability under linear accelerator (linac) conditions.

Conclusions:

• The proposed CNN-based method accurately calculates 3D dose distributions using TERMA and ED.
• This approach offers a significant potential for improving efficiency in radiation therapy treatment planning.
• The method is adaptable for other traditional dose algorithms and has applications in adaptive radiotherapy and in vivo verification.