Benchmarking proton RBE models.

Area of Science:

• Medical Physics
• Radiation Oncology
• Radiobiology

Background:

• Accurate prediction of proton relative biological effectiveness (RBE) is crucial for optimizing proton therapy.
• Current phenomenological RBE models exhibit significant discrepancies due to varying assumptions and fitting datasets.

Purpose of the Study:

• To benchmark thirteen RBE models using a common dataset and fitting techniques.
• To evaluate model generalizability and robustness through cross-validation.
• To assess the impact of data weighting and biological endpoint choice on model performance.

Main Methods:

• Fitting thirteen RBE models to a comprehensive proton RBE dataset.
• Employing a cross-validation approach to assess model generalizability.
• Investigating the effects of weighted fits and single vs. multiple biological endpoints.

Main Results:

• Common dataset fitting reduced prediction differences, but significant disagreements remained.
• Weighted cross-validation fits performed poorly, indicating underestimated experimental uncertainties and over-fitting.
• The simplest linear LET-dependent model excelled for single survival levels, while complex tissue-dependent models were better for multiple levels.

Conclusions:

• Poor error estimation in dose-response parameters significantly impacts RBE model fitting.
• Heterogeneous datasets and inherent model complexities challenge accurate RBE prediction.
• Robust statistical validation is essential for developing reliable RBE models in proton therapy.