The Impact of Artificial Intelligence Auto Contouring on Resident Education.
Authors
Affiliations (6)
Affiliations (6)
- University of California San Francisco, Department of Radiation Oncology. Electronic address: [email protected].
- Stanford University, Department of Radiation Oncology.
- University of California San Francisco, Department of Radiation Oncology.
- University of California Berkeley, Haas School of Business.
- University of California San Francisco, Department of Radiation Oncology; Bakar Computational Health Sciences Institute, University of California, San Francisco, San Francisco, California; UCSF-UC Berkeley Joint Program in Computational Health, San Francisco, California.
- University of California San Francisco, Department of Radiation Oncology. Electronic address: [email protected].
Abstract
The integration of artificial intelligence (AI) auto contouring (AAC) in radiation oncology has streamlined the delineation of organs at risk (OARs). Assessing OAR contours is a vital skill in radiation oncology. This study assesses the impact of AAC on residents' contouring education and skill acquisition and its consequences for educational programming METHODS: We conducted a cross-sectional survey of residents and resident-facing faculty at two tertiary centers that implemented AAC within the prior year. Respondents completed anonymous Likert (1-5) and free-text items; group differences were analyzed with two-sample t-tests (p≤0.05). Free-text comments underwent thematic analysis. Responses were received from 24/30 residents (80%) and 20/35 faculty (57%). Compared with faculty, residents more often reported that AAC improved anatomic understanding (Residents: 3.9 vs Faculty: 2.1, p<0.001) and overall education (4.2 vs 2.3, p<0.001). Both groups agreed AAC reduced time spent contouring (4.6 vs 4.4, p=0.39) and improved workflow from simulation to plan approval (4.5 vs 4.0, p=0.08). Perceived AAC contour quality was neutral (3.33 vs 2.85, p=0.11). AAC was viewed as improving familiarity with standardized OAR nomenclature (4.3 vs 3.3, p=0.001) and contributing positively to clinic (4.7 vs 3.7, p<0.001) and resident well-being (4.6 vs 3.6, p<0.001).Faculty comments highlighted inaccurate or incomplete contours and uncertainty about residents' systematic review or correction of AAC output, raising concerns about reduced practice with de novo delineation and CT anatomy. Residents acknowledged AAC's imperfections but emphasized time savings and the ability to redirect effort toward other educational activities. Residents and faculty diverge on AAC's educational value, particularly its effect on anatomic learning. However, both recognize benefits for workflow and well-being. Improving the integration and understanding of AAC-derived OARs during contouring will be crucial for improving resident training and ensuring high-quality care delivery in the AI era.