Leveraging Advanced AI Technologies for Radiotherapy Dose Calculation: A Narrative Review

Abstract

Context: Accurate radiotherapy dose calculation is critical for optimizing treatment efficacy and minimizing toxicity. Traditional algorithms, while clinically validated, often struggle with complex anatomical variations and heterogeneous tissue compositions. Recent advances in artificial intelligence (AI) offer promising alternatives for enhancing dose prediction accuracy and workflow efficiency. Objectives: This review aims to critically appraise the current landscape of AI-based radiotherapy dose calculation methods, comparing their performance, interpretability, and clinical applicability across various algorithmic families. Methods: A comprehensive literature search was conducted using PubMed, Scopus, and IEEE Xplore databases, focusing on studies published between 2015 and 2025. Included articles were categorized into six AI domains: Machine learning (ML), deep learning (DL), reinforcement learning (RL), Bayesian models, fuzzy logic systems, and evolutionary algorithms. Comparative analysis was performed based on dosimetric accuracy, computational efficiency, explainability, and integration with treatment planning systems (TPS). Results: The DL models, particularly convolutional neural networks (CNNs) and transformer-based architectures, demonstrated superior performance in dose prediction for head and neck, prostate, and lung cancers. The RL approaches showed potential in adaptive planning scenarios, while Bayesian and fuzzy logic models offered enhanced interpretability. Evolutionary algorithms were effective in multi-objective optimization but required extensive computational resources. Despite promising results, most studies lacked external validation and standardized benchmarking. Conclusions: The AI-driven dose calculation methods represent a transformative shift in radiotherapy planning. However, challenges remain in clinical translation, including algorithm transparency, regulatory approval, and integration with existing workflows. Future research should prioritize multi-institutional validation, hybrid model development, and human-AI collaboration frameworks to ensure safe and effective deployment.