2321 - Feasibility of Automated Radiotherapy Treatment Planning to Increase Patient Throughput in Limited Resource Settings

A. Kavuma¹, S. Kibudde¹, Y. Hao², T. Zhao², and B. Sun³; ¹Uganda Cancer Institute, Kampala, Uganda, ²Washington University School of Medicine in St. Louis, Department of Radiation Oncology, St. Louis, MO, ³Department of Radiation Oncology, Dan L. Duncan Comprehensive Cancer Center, Baylor College of Medicine, Houston, TX

Purpose/Objective(s): Advanced radiotherapy (RT) techniques relies on high-quality treatment planning process (TPP), ensuring adequate RT dose to target volume (TV) and maximal sparing of organs-at-risk (OAR). With the RT advancement in the past few decades, the TPP has become more sophisticated and labor intensive, creating barriers for rapid deployment in low and middle-income countries (LMICs). We aimed to establish the feasibility of implementing AI based auto-planning to facilitate clinical efficiency and increase patient throughput in an LMIC. Materials/

Methods: 25 planning CT structure datasets including five for each site of: head and neck, esophagus, prostate, cervical and rectum cancers were randomly sampled. Auto-segmentation using web-based AI platform was performed on all CTs for contouring. Radiation oncologists edited the OAR contours and defined the TV basing on RTOG guidelines. VMAT Manual-plans (MP) were executed by experienced RT planners. The Auto-plans (AP), which involves matching the patient’s structure with corresponding model structures before optimization were generated using RapidPlan® supported AI knowledge-based models for each site. The plans were evaluated by comparison of different dosimetric parameters, estimation of average planning time (ATP), and quantification of plan-quality utilizing Plan-Score card.
Results: Table 1 below summarizes mean-value-dosimetric parameters of 50 plans. The average planning time (minutes) were 28.8±2.5 and 43.2±13.5, while the plan-scores were 74.3±10.2% and74.2±9.8% for AP and MP respectively. Variations in OARs doses were observed for AP and MP cases, however the differences were not significant. For prostate cases, as much as both the AP and MP methods achieve the rectum: V50Gy = 50% constraint, the AP predicted lower doses than the corresponding MP (14.5±4.1 vs 22.4±4.9: p=0.019).
Conclusion: AI-based AP-generated technique offers adequate dose coverage, homogeneity and conformity to the TV, for single and simultaneous-integrated-boost plans, while sparing the OAR. Also, APs are more consistent (less dependent on planner experience), saves planning time, allow dose reduction to the pelvic OAR and holds tremendous potential for improving radiotherapy care in LMICs. Table 1

Site	Planning method	Dmax (%) p=0.07	D98% p=0.09	D2% p=0.17	HI p=0.36	CI p=0.10	MU/dose ratios p=0.26	APT (minutes) p<0.0001	Plan score (%) p=0.50
Head& Neck	AP	111.6	95.3	104.5	0.09	0.98	281	29.0	69.4
	MP	111.2	96.6	107.6	0.11	0.98	288	45.5	67.4
Esophagus	AP	111.0	97.7	106.6	0.09	0.97	282	25.0	83.0
	MP	108.4	99.0	105.5	0.05	0.99	297	41.5	84.2
Prostate	AP	108.8	96.7	103.7	0.07	0.98	379	31.5	86.4
	MP	107.3	99.0	106.1	0.06	0.99	342	44.0	81.9
Cervical	AP	109.7	98.4	106.9	0.08	0.99	359	29.5	75.1
	MP	108.6	98.6	106.4	0.08	0.99	339	42.0	79.2
Rectum	AP	111.3	97.0	104.1	0.07	0.99	434	29.0	57.7
	MP	109.9	97.9	106.5	0.08	0.99	361	43.0	58.4
(HI= Homogeneity Index, CI=Conformity Index, p=statistical significance, MU=monitor units, D2% & D98% are maximum dose covering 2% & 98% of the TV)