A Node-Based Workflow Management System for End-to-End and Human-in-the-Loop Medical Image Processing

Traditional medical image processing pipelines often suffer from fragmented tools, tedious manual interactions, and a lack of reproducibility, hindering clinical translation. This paper introduces a comprehensive, node-based visual workflow management module integrated into the KoGa platform (our image preparation system), enabling reproducible, automated, and human-in-the-loop medical image curation, anonymization, segmentation, and report generation. The proposed system abstracts complex backend computational tasks into atomic, reusable processing nodes (e.g., DICOM/NIfTI converters, PyDeface anonymizers, SAM2/MedSAM2 segmentation engines, and MedGemma-1.5 visual-language report generators). Clinicians and researchers can visually design complex pipelines via an interactive drag-and-drop user interface. A subjective usability evaluation involving 12 clinical participants across five representative workflows yielded high ratings for learnability ( $4.8\pm 0.4$ ), perceived efficiency ( $4.6\pm 0.5$ ), and overall satisfaction ( $4.8\pm 0.4$ ). These findings characterize perceived usability rather than objective reductions in annotation time or workload. By balancing zero-code automation with structured expert checkpoints, the workflow management system fosters clinical interpretability and accelerates the creation of high-quality AI-driven data pipelines.