User-Centered Annotation Workflow for Multidisciplinary Medical Image Datasets

Margaréta Strýčková

Supervisor(s): Ing. Erika Váczlavová

Slovak Technical University

Abstract: The trend of digital pathology introduces significant challenges for the development of medical imaging datasets with structured and standardized medical data. Existing dataset creation methodologies often lack a unified pipeline to ensure that clinical annotations produce structured data. This results in inconsistent outputs and also imposes a significant time burden on medical experts. In this paper, we outline the current workflow for creating multidisciplinary datasets, placing a particular emphasis on the annotation phase. We explore how to adapt the annotation phase in a user-centered manner that respects the real clinical workflow of medical domain experts. This ensures that the output data are standardized and can be utilized not only by the informatics domain, for example, for AI training, but also by medical experts as reliable data storage and for educational and diagnostic purposes. This thereby creates motivation for the construction of such datasets, while simultaneously increasing adaptability and reducing the refusal rate for the digitalization of their processes. Unlike current approaches that physically separate visual annotation tools from textual reporting protocols, we propose a modified workflow for the creation of annotations using defined templates and standards that are directly integrated into the visual annotation interface through progressive templates. To evaluate our approach, we conducted a comparative user study (with 7 participants) using a domain transfer methodology. Our results demonstrate that our proposed workflow eliminates data omissions and context bleeding, significantly reducing the expert burden compared to conventional methods, while consistently generating structured ground-truth data for multimodal datasets with multidisciplinary utility.
Keywords: Human-Computer Interaction
Full text:
Year: 2026