Data-driven ion-independent relative biological effectiveness modeling using the beam quality Q

Area of Science:

• Radiation Oncology
• Medical Physics
• Radiobiology

Background:

• Relative biological effectiveness (RBE) modeling is crucial for ion therapy treatment planning.
• The conventional linear energy transfer (LET) model has limitations in predicting RBE across different ions.
• The beam quality Q concept offers a potential alternative for ion-independent RBE modeling.

Purpose of the Study:

• To investigate the validity of the Q concept for modeling RBE across a broad range of linear energy transfer (LET) values, including the overkilling region.
• To develop and evaluate data-driven neural network (NN) models for predicting RBE using Q and other relevant parameters.
• To compare the performance of Q-based models against established models like the local effect model (LEM IV).

Main Methods:

• Utilized the particle irradiation data ensemble (PIDE) for *in vitro* experimental data.
• Developed low-complexity neural network (NN) models to predict RBE using LET, Q, and the linear-quadratic photon parameter (αx/βx).
• Compared model prediction power and ion dependence, and benchmarked the optimal NN model against LEM IV.

Main Results:

• NN models using αx/βx and Q as input outperformed models using LET for predicting RBE at reference photon doses (2-4 Gy) and near 10% cell survival.
• The Q-based NN model demonstrated no significant ion dependence (p > 0.5).
• The prediction power of the Q model was comparable to that of the established LEM IV.

Conclusions:

• The Q concept is validated for RBE modeling in a clinically relevant LET range, including the overkilling region.
• A data-driven Q model provides accurate RBE predictions comparable to mechanistic models, irrespective of the particle type.
• The Q concept can potentially reduce RBE uncertainties in treatment planning by enabling knowledge transfer between different ions.