An artificial intelligence-based model for cell killing prediction: development, validation and explainability...

Area of Science:

• Medical Physics
• Computational Biology
• Radiation Oncology

Background:

• Accurate prediction of radiation-induced cell killing is crucial for effective cancer treatment.
• Existing radiobiological models have limitations in predicting cell survival across diverse radiation types and cell lines.

Purpose of the Study:

• To develop and validate ANAKIN, an Artificial Intelligence-based model for predicting radiation-induced cell killing.
• To compare ANAKIN's predictive accuracy against established clinical models.

Main Methods:

• ANAKIN was trained and tested on 513 cell survival experiments from the PIDE database.
• The model was evaluated for predicting biological endpoints across various ion beams and cell lines.
• Explainable Artificial Intelligence (XAI) techniques were employed to interpret model predictions.

Main Results:

• ANAKIN demonstrated high accuracy in predicting cell survival fractions for diverse radiobiological conditions.
• The model outperformed the Microdosimetric Kinetic Model and the Local Effect Model (LEM version III).
• XAI analysis confirmed ANAKIN reproduces known biological patterns, including the overkilling effect.

Conclusions:

• ANAKIN offers a powerful and accurate AI-driven approach for predicting radiation-induced cell killing.
• The model's explainability enhances its utility and trustworthiness in radiobiology.
• ANAKIN shows potential for improving radiation therapy planning and understanding radiobiological responses.