ESTRO 2024 - Abstract Book

S86

Invited Speaker

ESTRO 2024

Figure 1: The output of each AI-based application is subject to a QA workflow composed of case-specific and routine QA.

3468

Quality check of auto-segmentation during clinical use

Tomas Janssen

NKI, Radiotherapy, Amsterdam, Netherlands

Abstract:

In recent years, advance artificial intelligence (AI) based auto-segmentation has rapidly entered radiotherapy. While initially primarily a research tool, more and more commercial or in-house developed solutions are used in clinical practice. While auto-segmentation certainly has a significant potential benefit, its widespread clinical use also raises complex and novel questions like • How to evaluate the benefit of auto-segmentation in clinical practice? • What biases are introduced by auto-segmentation? Why would these biases be (not?) acceptable in clinical practice? • To what extent is pre-clinical model evaluation of auto-segmentation representative of the use of the model in practice? • what can we learn from time-trends in the use, time-gain and corrections of auto-segmentation in practice? • Should we worry about skill degradation?

In this talk I will discuss these questions, using both a conceptual reflection as practical examples from clinical experience.

3469