PO-1004 Simulation of tissue dependent magnetic field susceptibility effects in MRI guided radiotherapy

• DOI: 10.1016/s0167-8140(19)31424-0
• Publication Date: 2019-05-20T06:09:04Z
• Authors: C. Kroll; M. Opel; C. Paganelli; F. Kamp; S. Neppl; G. Baroni; O. Dietrich; C. Belka; K. Parodi; M. Riboldi
• Keywords: 03 medical and health sciences; 0302 clinical medicine

Purpose or ObjectiveSegmentation of gross tumor volumes (GTVs) on nasopharyngeal carcinoma (NPC) MR images is an important basis for NPC radiotherapy planning.Manual segmentation GTVs a time-consuming and experience-dependent process in radiotherapy.This study aimed to develop simple deep learning based auto-segmentation algorithm segment T1weighted images. Material MethodsThis involved the analysis 510 from two datasets: (a) T1-weighted contrast-enhanced head neck (H&N) 305 patients (b) H&N 205 without obviously abnormal regions head.An FCN VGG16 was developed perform automatic GTVs.Data were randomly separated into training (90%) validation (10%) datasets.Additionally, 15 manually contoured by oncologists performance evaluation.Performance automated evaluated similarity manual Hausdorff distance (HD), average surface (ASD), Dice index (DSC), Jaccard (JSC). ResultsThe HD, ASD, DSC, JSC (mean±std) 16.18±7.93mm，2.42±1.38mm，0.71±0.12，0.57±0.13 dataset; these indices 14.21±4.73mm,1.51±0.98mm，0.83±0.08,0.72±0.12 between human radiation oncologists, respectively.The t-test indicated there no statistically significant difference concerning HD(p=0.67),ASD(p=0.46),DSC(p=0.17),JSC(p=0.16). ConclusionThe results suggested that GTV close segmentation's MRIs.However, performed better than segmentation.Thus, must be modified before being put use.