Metaheuristic-Driven Deep Learning Framework for Optimizing Radiotherapeutic Planning

Introduction: Humans are the utmost creatures living on this planet; besides being able to express their thoughts and feelings, they possess cognitive and behavioural intricacies. Further, they have immunity which allows them to fight disease-causing germs. Still, it seldom happens that the immunity of humans fails to fight against pathogens, as in the case of cancer, thus requiring interference from foreign influence. Objectives: Radiotherapeutic Planning is an important aspect that aids cancer treatment by optimizing the choice of dosage and beam orientation. This research aims to propose a novel methodology to deal with this problem by employing Artificial Intelligence, thereby improving the efficiency of Radiotherapeutic Planning. Methods: A Residual Network (ResNet) is used to identify the mapping between the anatomy of the patient and the therapeutic variables. The ResNet, trained with a relatively small dataset of nearly 15 subjects, showed better performance compared to state-of-the-art approaches using large datasets. Additionally, the neuronal parameters of the ResNet were concurrently updated using Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO). The proposed methodology and its variants were compared to Columnar Generation-based Optimization and a standard Convolutional Neural Network (CNN). Results: The results suggested that the Column Generation (CG) method produced outcomes are similar to those generated by the ResNet and its variants optimized via PSO and GWO. Conclusion: From the experiments presented, it is evident that the performance analysis shows that Deep Learning-based therapeutic paradigms are comparable to Column Generation Optimization in Radiotherapeutic Planning.