Dose-volume histogram prediction using density estimation.

Area of Science:

• Medical Physics
• Radiotherapy
• Machine Learning

Background:

• Accurate dose-volume histograms (DVHs) are crucial for consistent radiotherapy planning.
• Predicting DVHs for new patients can improve treatment quality.
• Current methods may have limitations in flexibility and performance with limited data.

Purpose of the Study:

• To develop a novel machine learning method for predicting dose-volume histograms (DVHs) for unseen patients.
• To leverage a probabilistic framework for DVH prediction in radiotherapy.
• To enhance the consistency and quality of radiotherapy treatment planning.

Main Methods:

• A machine learning approach using previous treatment plans to predict DVHs.
• Framing DVHs in a probabilistic setting by estimating joint probability distributions.
• Predicting DVHs by estimating feature distributions and marginalizing conditional probabilities.

Main Results:

• Proof-of-concept predictions for brainstems in acoustic schwannoma patients and lungs in lung cancer patients.
• Demonstrated similar prediction accuracy compared to previous methods with the same input data.
• The probabilistic method shows potential for improved ease of use and flexibility.

Conclusions:

• The proposed probabilistic machine learning method offers a new approach to DVH prediction.
• This method may overcome deficiencies of previous prediction techniques.
• The approach shows promise for improved performance, especially with limited training data.