Abstract Book

S222

ESTRO 37

Material and Methods 15 stereotactic radiosurgery (SRS) cases were employed to evaluate contouring between CT and T1W+C images in cranial region while the MR data in HN region of 10 healthy volunteers were collected to compare that between T1W and T2W images. Anatomical contouring by 4 and 3 observers in cranial and HN region respectively as well as image registration (Figure 1) were then performed. Using those data sets, we quantified: (1) inter-observer variability within each imaging technique by analysing the organ volumes; (2) inter-observer variability between two imaging techniques by comparing the mean and standard deviation (SD) of organ volumes as well as overlapping ratio (R overlap ) and; (3) spatial variation in OAR delineation by dice similarity coefficient (DSC).

Results 11 and 23 OARs were contoured in cranial and HN region respectively. Inter-observer variabilities were shown in 11 (100%), 8 (73%), 21 (91%) and 16 (70%) OARs on CT, T1W+C, T1W and T2W images respectively. When comparing between different techniques (details shown in tables), T1W+C showed comparable or smaller inter- observer variability than CT in 72% and 63% of cranial structures in volumetric and locational aspects respectively. For HN region, 78% and 70% of OARs exhibited no significant delineation difference in respective terms of volume and location when comparing between T1W and T2W images. DSC values were higher than 0.8 for organs with volume exceeding 8cm 3 in cranial region (CT & T1W+C), and those with volume exceeding 5cm 3 in HN region (T1W & T2W).

Conclusion MR modalities showed a better inter-observer agreement than CT in OAR contouring, especially on T1W+C and T2 images. The inter-observer variabilities between T1W and T2W images were comparable, suggesting the complementary use of both images in OAR delineation during MR-only planning. The moderate to high occurrence of inter-observer variations in all MR modalities revealed the urge of increasing contouring consistency among observers especially for very small OARs. Peer review and training possibly aid observers to gain consensus. Currently, MR-only planning consisting very small organs should consider the use of planning organ-at-risk volume (PRV). OC-0424 Multi-atlas auto-segmentation for head and neck OARs: accuracy and time efficiency D. Gugyerás 1 , A. Farkas 1 , M. Peto 2 , J. Hadjiev 1 , A. Gulyban 3 , F. Lakosi 1 1 University of Kaposvar- Health Center, Radiotherapy, Kaposvar, Hungary 2 Kaposvar University, Faculty of Pedagogy, Kaposvar, Hungary 3 University Hospital of Liège, Radiation Oncology, Liege, Belgium Purpose or Objective to test time efficiency and accuracy of multi-atlas based auto-segmentation (m-ABAS) (Mirada Medical, Oxford, UK) in Head and Neck (H&N) organs at risk (OAR) contouring Material and Methods 57 patients with hypopharynx/larynx cancer were selected and divided into two groups: ideal population (T1-T2/N0-1, n=30 pts) and clinical scenario (≥T3/ ≥N2a, n=27 pts). All manual delineation was performed by an expert RTT using Eclipse (version 13.0, Varian Medical, Palto Alto, US), including semi-automated options (“ Man” ). The baseline multi-atlas for both groups consisted of the first 10 reference delineated cases using Mirada RTx (version 1.6.3, Mirada Medical, Oxford, UK). For the next 10 patients (11-20th) m-ABAS contours were generated (“ Mir ”) followed by manual correction