Abstract Book

S1026

ESTRO 37

Purpose or Objective To automate optimization process in RayStation treatment planning system (TPS) for head and neck treatments by means of specific Phyton script. Material and Methods Genetic algorithm (GA) was implemented for head and neck VMAT treatments. The chromosomes of the algorithm were max equivalent uniform dose functions (maxEUD) for: larynx, oral cavity, both parotid glands and inner ears. Ten initial couples were randomly generated and a maximum of five reproductions were permitted. The fitness function included coverage of PTVs and mean dose of parotid glands, larynx, oral cavity and inner ears with different weights. The script was tested for five patients and two different dose prescriptions (both simultaneous integrated boost): four patients with three dose levels (70 Gy, 63 Gy and 54 Gy) in 35 fractions; one with two (66 Gy, 59.4 Gy) in 33 fractions. A comparison with corresponding plans created with Monaco TPS (M, Elekta) was carried out. M plans had the same initial template with active multicriterial function for OARs. All the plans were evaluated with a score of PlanIQ (Sun Nuclear) in terms of target coverage, conformity index of PTVs and constraints of OARs. Results In figure are reported the totals scores (TS) obtained. Script plans scores were always higher to M ones (mean TS: 78.7±27.7 for GA and 73.3±28.1 for M). Mean score of PTV1.1 coverage (max = 15) was 12.0±6.7 for GA and 11.5±6.5 for M. The values of other PTVs were (max = 10): for PTV1.2, 8.1±1.9 for GA and 4.1±3.2 for M; for PTV1.3, 9.4±1.0 for GA and 9.5±0.5 for M. Maximum doses of spinal cord, brainstem and optic pathways were always below the requested value both for GA and M. For parallel OARs, GA spared parotid glands more (average score of summed parotid glands: 5.6±2.0 for GA and 4.4±2.5 for M). For other OARs, the results obtained with GA and M were very similar to each other.

Figure 1. Distribution of plan quality indices for a prostate (top) and lung (bottom) case. Blue points correspond to treatment plans generated using the proposed formulation; the blue surface is the convex hull. Red and green points correspond to plans generated using penalty-based functions; red points violate clinical criteria. The magenta circle marks the corner of all best values. Top: The PTV conformity index (CI) is here defined as the average dose of a 1 cm ring around the PTV. Conclusion For the two studied patient cases, the plan quality indices obtained using the proposed formulation are deemed equivalent to those obtained using penalty-based functions. When disregarding plans that violate clinical criteria, the quality indices of plans obtained using the proposed formulation are superior. This outcome confirms our previous preliminary results. We conclude that the proposed formulation has an automating potential when applied to DMLC planning. EP-1894 Genetic algorithm based script for automation of head and neck VMAT treatment planning E. Gallio 1 , A. Alparone 2 , C. Fiandra 3 , C. Vecchi 2 , G. Balestra 4 , R. Ragona 3 , U. Ricardi 3 1 A.S.O.U. S. Giovanni Battista di Torino - Molinette, S.C. Fisica Sanitaria, Torino, Italy 2 Tecnologie Avanzate s.r.l, Electronics, Torino, Italy 3 University of Torino, Radiation Oncology, Torino, Italy 4 Politecnico di Torino, Electronics and Telecomunication, Torino, Italy

Conclusion Automatic planning gives results clinically comparable with manual ones reducing considerably the workload and increasing the overall quality of plans. These preliminary results must be confirmed by further ongoing investigation with a larger Head and Neck plans dataset and planned with already available commercial versions of automated treatment planning solution by different vendors, too. EP-1895 Automated Trajectory Planning for Breast Treatment Using TORUS C. Locke 1 , K. Horst 1 , K. Balazy 1 , K. Bush 1 1 Stanford Cancer Center, Radiation Oncology, Stanford, USA