ESTRO 2021 Abstract Book

S581

ESTRO 2021

All the AUTO plans resulted to be within clinical tolerance, showing a mean target coverage variation of 1.4% with a maximum value of 3.7% and a minimum of 0.7% when compared with MAN plans. On the other hand, OARs sparing showed slightly bigger differences. The mean (max;min) variations were 7.7% (20.6%;0%) and 6.5% (15.5%;0) for rectum V50 and V60 respectively. Bladder V65 mean (max;min) variation was -1.8% (5.0%;- 20.2) and bowel V195 mean (max;min) variation was -0.8 cc (7.4cc; -7,2cc). Dosimetric indicators showed no relevant difference in plan comparison, with better results in favor of AUTO plans in terms of homogeneity and gradient index. VAS scoring (figure 2) has confirmed equal or lower plan quality for AUTO plans.

Conclusion The proposed workflow allowed a fast and accurate automatic generation of treatment plans that can be considered comparable to manual plans and, even in the worst-case scenario, with difference of minor clinical impact. Due to the reduced computational time required to generate a plan, further manual optimization can also be easily applied, taking advantages of the time saved by this innovative workflow that could significantly support the entire MRgRT team. PD-0751 The effect of organ-at-risk contour variations on automatically generated treatment plans for NSCLC F. Vaassen 1 , C. Hazelaar 1 , R. Canters 1 , S. Peeters 2 , S. Petit 3 , W. van Elmpt 1 1 Maastro Clinic, Physics Innovation, Maastricht, The Netherlands; 2 Maastro Clinic, Clinic, Maastricht, The Netherlands; 3 Erasmus MC, Physics, Rotterdam, The Netherlands Purpose or Objective Quality of automatic contouring is generally assessed by comparison with manual delineations on the geometrical level, but the effect of contour differences on the resulting treatment plan (i.e. dose distribution) remains unknown. It is hypothesized that many contouring inaccuracies have no influence on the dose distribution, indicating there is no need to correct them. This study investigated how contouring inaccuracies propagate into dosimetric errors of organ-at-risk (OAR) in the thorax region. Materials and Methods OARs of twenty lung cancer patients were manually (C MC ) and automatically contoured using deep learning (C DC ) and atlas-based (C AC ) contouring, after which user-adjustments (C ADC /C AAC ) were made to make them clinically acceptable. For each contour set, an automated treatment plan was generated