Speaker
Description
Dr. Jintao Ren is a radiotherapy AI postdoc researcher at Aarhus University Hospital and Aarhus University, Denmark. He is working on deep learning for medical image segmentation, tumor delineation, and adaptive radiotherapy. His recent work includes CT/PET lesion segmentation and MRI-guided head-and-neck gross tumor volume segmentation. His research focuses on translating AI contouring methods into clinically useful, safe, and trustworthy radiotherapy workflows
Summary
Auto-contouring has become one of the most visible applications of artificial intelligence in radiotherapy. What began as an academic segmentation problem is now entering routine clinical workflows through commercial software, treatment planning system integration, and adaptive radiotherapy platforms. This talk reviews the current clinical status of AI-based auto-contouring, with a focus on where the technology is genuinely useful, where caution remains essential, and how departments can implement it safely.
The central message is that auto-contouring is no longer simply a research promise. For many organs at risk and selected guideline-based lymph node structures, modern deep learning systems can generate clinically useful draft contours, reduce repetitive workload, improve structure naming consistency, and accelerate the start of treatment planning. However, the most important question has shifted from “Can AI draw contours?” to “When can we safely trust, edit, and approve them?”
A key distinction is made between organ-at-risk segmentation and target delineation. Organs at risk are primarily anatomical structures, whereas gross tumor volumes and clinical target volumes often require multimodality imaging, pathology, surgical information, disease biology, institutional guidelines, and physician judgment. For this reason, autonomous target contouring remains much less mature than OAR contouring, particularly in complex sites such as head and neck cancer, lung cancer, postoperative cases, and adaptive radiotherapy.
The talk will also discuss common failure modes, including small structures, poor soft-tissue contrast, imaging artifacts, unusual anatomy, domain shift, CBCT image quality, and over-reliance on plausible but incorrect contours. Finally, it proposes a practical framework for safe clinical adoption: local commissioning, structure-specific validation, human-in-the-loop review, contour QA, model version control, and clear responsibility for final approval.
Auto-contouring should therefore be understood not as a replacement for clinicians, but as a clinical co-pilot: powerful, useful, and increasingly necessary, but only safe when embedded in a validated and governed radiotherapy workflow.