Optimization of Radiotherapy Dosimetry using Machine Learning and Monte Carlo Simulations for Personalized Brain Cancer Treatment

In recent years, machine learning systems have achieved impressive performances comparable toand even superior to human level in various tasks. However, these advances have relied heavily onvast amounts of data, such as large volumes of labeled data or continuous evaluations in controlledenvironments. In contrast, human learning is predominantly unsupervised, guided by observationand interaction with the environment. Emulating this type of learning in machines constitutes acrucial challenge to achieve general artificial intelligence. This context has led to the incorporationof advanced technologies such as artificial intelligence (AI) and machine learning (ML) in the fieldof radiotherapy, where these tools can significantly improve diagnostic accuracy and the efficiencyof clinical processes, marking a milestone in cancer treatment.Radiotherapy fundamentally depends on precise dosimetric planning, a process that seeks tomaximize the dose delivered and prescribed to the tumor while minimizing the exposure of healthytissues. Automated treatment planning (ATP) has emerged as an innovative solution to improve thequality and consistency of radiotherapy plans, while reducing the workload of clinical staff. Toolssuch as RapidPlan have demonstrated that automation based on prior knowledge allows for thegeneration of personalized treatment plans by analyzing complex data patterns. Machine learningmodels, such as convolutional neural networks (CNN), stand out for their ability to optimize dosedistributions, achieving effective tumor irradiation with less affectation to organs at risk (OARs).On this basis, the present project seeks to contribute to the refinement of the standard radiotherapyplanning process by applying advanced methodologies.The project combines machine learning with Monte Carlo simulations to optimize radiotherapytreatments, offering a comprehensive and novel approach. Specifically, convolutional neural networkswill analyze clinical imaging data, like a computed tomography (CT), positron emission tomography(PET) scans and magnetic resonance imaging (MRI), to identify tumor regions and organs at risk.Monte Carlo simulations will complement this analysis by generating probabilistic radiation dosedistributions, allowing treatment parameters to be adjusted to maximize irradiation to the tumorand minimize adverse effects on healthy tissues. This hybrid system not only improves the efficiencyand precision of radiotherapy, but also offers a scalable and adaptable approach, with potentialapplications in diverse and challenging clinical settings.