ESTRO meets Asia 2024 - Abstract Book

S41

Invited Speaker

ESTRO meets Asia 2024

Division of Cancer Sciences, University of Manchester, Manchester, United Kingdom

Abstract

Treatments for tumours located in the thorax may be affected by breathing, cardiac and gastro-intestinal motion. This impact is multiple: motion will not only affect the delivery of radiation (both inter and intra-fraction) but also image quality, registration and contouring. In addition, some approaches may modify the internal anatomy (e.g. deep inspiration breath hold) This presentation will focus on current and upcoming developments to assess, handle and mitigate breathing motion in lung cancer, as applicable to both normo-fractionated and hypofractionated treatments.

541

AI for target volume and organ at risk delineation

Jin Sung Kim

Dept. of Radiation Oncology, Yonsei University, Seoul, Korea, Republic of

Abstract

Advancements in artificial intelligence (AI) are revolutionizing radiation oncology, particularly in the delineation of organs at risk (OAR) and target volumes. Accurate delineation is crucial for effective treatment planning, minimizing radiation exposure to healthy tissues while maximizing the dose to the tumor. This lecture aims to provide an overview of the latest AI-driven contouring technologies and their applications in modern radiation oncology. Traditional manual contouring is time-consuming and subject to inter-observer variability. AI technologies, particularly deep learning algorithms, offer solutions by automating the contouring process with high precision and consistency. Convolutional neural networks (CNNs) and other machine learning models have been trained on vast datasets of medical images, enabling them to identify and delineate OAR and target volumes with remarkable accuracy. These models are continuously learning and improving, making them indispensable tools in clinical settings. Recent advancements have seen the integration of large language models (LLMs) with imaging data to enhance contouring capabilities. LLMs, known for their proficiency in understanding and generating human-like text, are being adapted to interpret and analyze medical imaging data. By leveraging the strengths of both visual and textual data, LLM-based AI systems can provide more comprehensive and accurate delineation results. One of the key areas of ongoing research is the development of hybrid models that combine CNNs for image analysis with LLMs for contextual understanding. These models can process a patient's imaging data alongside their medical history, clinical notes, and other relevant information to produce highly tailored and precise contouring. This approach not only improves the accuracy of delineation but also enhances the overall workflow efficiency in radiation oncology departments. In this lecture, we will explore case studies and clinical trials showcasing the efficacy of these AI technologies. We will discuss the practical implementation of AI contouring systems, including the challenges and solutions in integrating these technologies into existing clinical workflows. Additionally, we will delve into the future directions of AI in radiation oncology, focusing on the potential of LLM-based models to further revolutionize target volume delineation.

543