Enhancing Mathematical Optimization in Intensity-Modulated Radiation Therapy with Artificial Intelligence and Machine Learning

Abstract

A key technique used in cancer treatment is called Intensity-Modulated Radiation Therapy (IMRT), which requires accurate delivery of doses to target the tumors and reduce exposure to healthy tissues. Mathematically, optimal beam angles to maximize objectives are introduced as decision variables and the objective for an IMRT plan is used both linear or nonlinear functions of fluence intensities. Yet these techniques struggle with multi-objective optimization in complex settings and adapting to clinical data that is collected in real-time. In this paper, a novel AI-ML framework is introduced in which mathematical optimization integrated with the ethical dimensions of Artificial Intelligence (AI) and Machine Learning (ML)-enabled optimization leads to better treatment results for IMRT. The method uses reinforcement learning (RL) for adaptive optimization of the dose based on actual patient feedback, a deep-learning approach to predictive beam angle selection, and GA-based multi-objective optimization. Primary resulting metrics include tumour coverage, organ-at-risk (OAR) sparing, treatment planning time, and computational speed. The augmented framework integrates direct patient-specific information such as tumour geometry and biological markers to assist the treatment plan with a personalized approach, thereby improving precision and efficiency. We show experimentally that our method is considerably faster, with a slight improvement in dose distribution conformity and tumour coverage while causing less damage to normal tissues. Their AI-ML enhanced optimization framework proposes a revolutionary solution for IMRT, circumventing the disadvantageous feature of traditional methods and paving the way to more efficient personal cancer therapies. These results suggest that AI-based methods could help transform how radiation therapy is performed for cancer patients.