VQ-DoseNet: A vector quantized model for stochastic radiotherapy dose prediction.

Area of Science:

• Medical Physics
• Artificial Intelligence in Medicine
• Radiotherapy Research

Background:

• Radiotherapy treatment planning is complex and time-consuming.
• Current deep learning dose prediction models lack variability, limiting clinical application.
• Stochasticity is essential for capturing dose distribution variations in radiotherapy.

Purpose of the Study:

• To introduce stochasticity into deep learning-based radiotherapy dose prediction.
• To develop a novel vector quantized model (VQ-DoseNet) for probabilistic dose prediction.
• To improve the efficiency and clinical acceptance of radiotherapy treatment planning.

Main Methods:

• Proposed a novel vector quantized model (VQ-DoseNet) for stochastic dose prediction.
• Incorporated input feature perturbation to generate multiple plausible dose distributions.
• Evaluated model performance against state-of-the-art methods on in-house and OpenKBP datasets.

Main Results:

• VQ-DoseNet achieved a mean absolute error (MAE) of 0.106 Gy on an in-house dataset.
• Reported dose and DVH scores of 3.608 ± 1.267 Gy and 1.329 ± 1.934 Gy on the OpenKBP dataset.
• Generated stochastic dose distributions consistent with ground truth and within clinical constraints.

Conclusions:

• VQ-DoseNet successfully introduces stochasticity into dose prediction, maintaining high accuracy.
• The model generates clinically relevant, variable dose distributions, enhancing treatment planning.
• This approach offers a promising method for improving radiotherapy efficiency and patient outcomes.