From LQ to AI-BED-Fx: A Unified Multi-Fraction Radiobiological and Machine-Learning Framework for Gamma Knife...

Area of Science:

• Radiosurgery
• Medical Physics
• Artificial Intelligence

Background:

• Current Gamma Knife radiosurgery (GKS) decision-making relies on physical dose metrics, neglecting crucial radiobiological factors like fractionation and DNA repair.
• Existing radiobiological models are inadequate for multi-fraction GKS (3- and 5-fractions).
• Biologically Effective Dose (BED) shows promise for predicting radiosurgical response, but a unified framework is missing.

Purpose of the Study:

• To develop AI-BED-Fx, the first multi-fraction extension of the Jones-Hopewell radiobiological model for GKS.
• To compute fraction-resolved BED for 1-, 3-, and 5-fraction GKS.
• To assess the predictive value of BED versus physical dose for treatment outcomes across different pathologies.

Main Methods:

• Developed AI-BED-Fx incorporating α/β ratio, repair kinetics, and lesion-specific parameters.
• Generated synthetic cohorts for arteriovenous malformation (AVM), meningioma (MEN), vestibular schwannoma (VS), and brain metastasis (BM).
• Trained machine-learning models to compare physical dose and BED predictions; included Bayesian estimation and a neural-network surrogate for BED prediction.

Main Results:

• AI-BED-Fx generated pathology-specific BED distributions and dose-response relationships.
• BED was the dominant predictor for AVM and meningioma, while physical dose and volume were key for VS and BM.
• The AI model accurately recovered biological parameters and BED calculations, showing high fidelity.

Conclusions:

• AI-BED-Fx offers a unified, biologically explicit framework for modeling GKS, applicable to single and multi-fraction treatments.
• The predictive utility of BED is pathology-dependent, requiring consideration of repair kinetics and biology.
• This framework integrates radiobiology and machine learning, paving the way for biologically adaptive, AI-guided radiosurgery.