Clinical Commissioning and Implementation of An In-House Artificial Intelligence (AI) Tool for Automated Head and Neck Intensity Modulated Radiation Therapy (IMRT) Treatment Planning
X Li1*, Y Sheng1, QJ Wu1, Y Ge2, C Wang1, D Brizel1, Y Mowery1, J Lee1, W Wang1, H Stephens1, D Yang1, F Yin1, Q Wu1, (1) Duke University Medical Center, Durham, NC, (2) University of North Carolina at Charlotte, Charlotte, NC
Presentations
SU-E-BRB-4 (Sunday, 7/10/2022) 1:00 PM - 2:00 PM [Eastern Time (GMT-4)]
Ballroom B
Purpose: Many Artificial Intelligence (AI) algorithms have been developed for radiation treatment planning and have demonstrated promising plan quality in research environment. This study described the clinical commissioning process of an in-house AI algorithm for head-and-neck (HN) intensity-modulated radiation therapy (IMRT) planning in our clinic.
Methods: First, a graphical user interface (GUI) was developed to integrate the AI algorithm within the clinical treatment planning system (TPS). This GUI allowed the AI algorithm to be executed remotely on a designated server workstation. The server workstation was configured specifically for the AI algorithm with graphics processing unit (GPU) support, minimizing impact on clinical workstations. The optimal fluence maps, which were predicted by the AI algorithm, were imported into the TPS for dose calculation, followed by an optional automatic fine-tuning. Planners can visually examine the plan dose distribution and make further adjustments as clinically needed. Second, 36 recent cases were retrospectively collected for commissioning. The AI plans for the commissioning cases were compared to the clinical plans regarding critical dose-volume endpoints and 3D dose distribution. Third, the AI planning workflow and performance were evaluated and reviewed by experienced physicians and physicists.
Results: The average plan generation time including user interactions was 10-15min/case. The AI dose distribution was comparable to clinical plans, with slightly better dose conformity (V_100%/V_PTV, 1.10±0.06 versus 1.16±0.09 in clinical plans) and total MU (1609.6±270.7 versus 1979.7±562.0, normalized to 44 Gy prescriptions). AI plans’ BODY D_1cc was higher than clinical plans’ (109.8±1.2% versus 108.3±2.0%). Organ-at-risk (OAR) sparing was comparable. Based on physician and physicist feedback, the AI planning workflow was reasonable, and overall performance was satisfactory.
Conclusion: The in-house AI IMRT planning tool was commissioned for HN in our clinic. The commissioning process demonstrates outstanding performance and robustness of the AI tool and provides sufficient validation and documentation for clinical use.
Funding Support, Disclosures, and Conflict of Interest: Varian medical systems (master research agreement)
Keywords
Modeling, Radiation Therapy, Treatment Planning
Taxonomy
Contact Email