Investigation of automatic treatment planning (ATP) optimization using an Adaptative Neuro-Fuzzy AI system

Abstract

Automatic treatment planning (ATP) plays a critical role in modern radiation therapy, where optimizing treatment plans to achieve high-quality dosimetric outcomes remains a challenging task. This dissertation investigates the development and implementation of an Adaptive Neuro-Fuzzy Inference System (ANFIS) for ATP optimization, integrating the strengths of fuzzy logic and artificial neural networks to enhance inverse planning in Intensity-Modulated Radiation Therapy (IMRT).The study explores the limitations of conventional Fuzzy Inference Systems (FIS) in treatment planning and demonstrates how ANFIS improves upon them by learning and adapting decision rules through a data-driven approach. The methodology focuses on three primary aims: (1) developing an ANFIS-based AI framework for guiding inverse planning in IMRT, (2) generating and optimizing ANFIS rules to enhance dose distribution and treatment plan quality, and (3) evaluating ANFIS performance using clinical prostate and head \& neck cancer cases.The proposed system is trained using a dataset of optimized treatment plans, leveraging a plan scorecard to assess and refine dosimetric objectives. Comparative evaluations between ANFIS, conventional FIS, and human planners demonstrate that ANFIS achieves superior accuracy in dose-volume histogram (DVH) metrics, improved organ-at-risk (OAR) sparing, and enhanced target dose conformity. The results highlight ANFIS's potential in reducing reliance on manual tuning, increasing planning efficiency, and providing interpretable decision-making insights for clinical application.This dissertation contributes to the field of medical physics and AI-driven treatment planning by introducing a novel, adaptable, and interpretable framework for ATP optimization. The findings suggest that ANFIS-guided planning can significantly improve clinical workflows, paving the way for more automated and efficient radiotherapy treatment planning systems.