ESTRO 2024 - Abstract Book

S4486

Physics - Machine learning models and clinical applications

ESTRO 2024

Purpose/Objective:

In vivo dosimetry (IVD) aims to catch treatment errors, assist in treatment adaptation, and monitor the actual dose delivered to the patient [1]. For external beam radiotherapy (EBRT), the electronic portal imaging device (EPID) lends itself well for IVD and presents the advantage of passively measuring deliverd dose without needing further setup. In practice, IVD using EPID is rarely performed, as assessing the complex and multitude of EPID-generated information is too time-consuming to act upon treatment errors in time as required for dose-guided radiotherapy (DGRT). Recent artificial intelligence (AI) developments allow for rapid processing of these large quantities of complex EPID generated data to extract relevant information for further decision-making [2]. To promote IVD for clinical practice, we designed an AI-based system to evaluate treatment accuracy and to detect treatment errors by automatically analyzing EPID images, while presenting human-interpretable information to users.

Material/Methods:

To develop the prototype of this system, we employ a co-creation methodology based on design thinking principles, diverging from the conventional norms of clinical software development. Design thinking is a purposeful and iterative approach to innovation that places the utmost importance on addressing the specific needs and concerns of the end user. Unlike traditional development methods that typically follow a linear path towards creating a final product before presenting it to the end users, such as medical physicists (MPs) in our context, design thinking encourages a more collaborative and user-centered process from the very beginning [3]. A five-phase design sprint was set up with eight experts from both a clinical and commercial organization, including MPs, researchers, and user experience designers (Figure 1). As a first step, we defined a shared goal guiding the design choices. Our goal aimed at aiding MPs to confidently detect and verify treatment errors, including root cause analysis, and support their decision making before the next fraction to achieve safer treatment for every patient. For the empathize and test phase of our project, an additional six user-experts from both clinical and industry backgrounds were consulted. All five phases were completed in five full days.

Results: