2334 - Adaptive Lung Radiotherapy Planning with Expedited Constrained Hierarchical Optimization (ECHO) Empowered Knowledge-Based Models on Ethos(E³)

L. Kuo¹, E. D. Yorke¹, M. Savacool¹, A. Kuo², L. I. Cervino¹, M. Zarepisheh¹, A. Rimner³, P. Zhang¹, J. O. Deasy¹, J. M. Moran¹, and S. B. Lim¹; ¹Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, ²Herricks High School, New Hyde Park, NY, ³Department of Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, NY

Purpose/Objective(s): The study aims to investigate the benefits of adapting the optimization capabilities of the Expedited Constrained Hierarchical Optimization (ECHO) from a C-series LINAC platform to a ring gantry delivery system by incorporating a knowledge-based planning (KBP) model pretrained on ECHO-generated Pareto-optimal plans for online adaptive conventionally fractionated lung radiotherapy (oART) planning. Materials/

Methods: Seventy-six treatment plans for conventionally fractionated lung treatment (2 Gy x 30) were generated using our in-house ECHO system. 66 were randomly selected to train a commercial KBP model within our treatment planning system. The remaining 10 cases served as model validation. For each validation case, 4-5 adaptive plans with different beam were generated for the ring gantry delivery system with and without using the trained KBP model. Treatment plan quality was assessed using clinical dose metrics for target coverage and organ at risk (OAR) sparing. We calculated ratios of these metrics with and without the KBP model to evaluate its impact.
Results: Table 1 summarizes the comparison results. PTV coverage and OAR dose metrics from all plans met all department guidelines. Compared to plans without the KBP model, the mean ratios of the different metrics of targets ranged from 0.974 to 1.003 with statistically insignificant deviations (p>0.01). Esophagus mean dose, and heart mean dose and V_30Gy were significantly (p<0.0001, p=0.0004, p=0.0003, respectively) lower with KBP. Lungs (not GTV) showed minimal mean dose and V_20Gy/V_5Gy deviations. Spinal cord D_max was slightly higher (1.068±0.018), but statistically insignificant (p=0.018) and was within department’s guideline.
Conclusion: We successfully adapted ECHOs optimization capabilities for delivery in a ring gantry system by by incorporating a pre-trained KBP model based on ECHO-generated Pareto-optimal plans for conventionally fractionated lung . ECHO generates high quality training plans efficiently, allowing models to be updated when new clinical constraints arise, producing optimal planning goals for daily anatomy changes during initial and online oART. Abstract 2334 - Table 1: Comparison of Ethos Plan Quality Metrics: ECHO-KBP vs. No KBP

Targets	Metrics	Ratio	P value
CTV	D95	0.999±0.127	0.127
PTV	D95	0.995±0.070	0.070
	Dmax	1.003±0.044	0.044
	Dmin	0.974±0.041	0.041
	V100%	0.982±0.090	0.090
OARs
Esophagus	Dmean	0.915±0.000	<0.0001
Lungs not GTV	Dmean	0.990±0.029	0.028
	V20Gy	0.997±0.603	0.603
	V5Gy	0.998±0.660	0.660
Heart	Dmean	0.932±0.000	0.0004
	V30Gy	0.880±0.000	0.0003
Cord	Dmax	1.068±0.018	0.018