Deep learning\u2013based segmentation of brain tumours is increasingly integrated into radiotherapy planning pipelines, promising faster workflows and reduced inter-rater variability. At the same time, automatic contours remain fallible and require careful review by radiation oncologists, raising questions about how such tools actually shape clinical decision-making. In particular, there is growing interest in visualising model uncertainty, yet little empirical work on how this information affects clinicians\u2019 behaviour, trust, and cognitive load.<\/p>\n\n\n\n
In this talk, I will present the rationale and study design of an ongoing project on human\u2013AI interaction in tumour delineation. The project will combine qualitative interviews with controlled decision tasks, in which clinicians review AI-generated brain tumour segmentations with and without uncertainty information. I will outline how we plan to measure inter- and intra-rater variability, editing patterns, task duration, subjective cognitive load, and trust in the system. Finally, I will discuss anticipated challenges and how the results could inform guidelines for safe, efficient, and transparent deployment of AI-based segmentation in routine radiotherapy practice.<\/p>\n","protected":false},"excerpt":{"rendered":"
Agata Mosinska, PhD CogNes JU<\/p>\n","protected":false},"featured_media":27289,"template":"","other_events_category":[53],"class_list":["post-27211","other_events","type-other_events","status-publish","has-post-thumbnail","hentry","other_events_category-neuropizza"],"yoast_head":"\n