ESTRO 36 Abstract Book
S553 ESTRO 36 _______________________________________________________________________________________________
Results MR_ST was significantly more accurate than other group according to DSC of 0.646 ± 0.094 compared to 0.564 ± 0.102, 0.298 ± 0.109, 0.39 ± 0.254 respectively for MR_MV, CA and BM in optic chiasm. DCS scores in pituitary gland were following, 0.624 ± 0.055 in MR_ST, 0.582 ± 0.052 in MR_MV, 0.514 ± 0.140 in CA and 0.28 ± 0.24 in BM respetively. Brainstem was showed similar DSC score as 0.89 ± 0.021, 0.892 ± 0.017, 0.842 ± 0.038 and 0.73 ± 0.156 respectively for MR_ST, MR_MV, CA and BM. Conclusion Most of auto delineated contours was smoothened in advance. Among 4 groups, DSC of MR based was the highest. Even though auto contouring is conducted by different users, it shows certain shape and included similar region when we use same subject’s data. ABS software takes more effort and time to use in the first place. However, MR based ABS would have better auto contouring accuracy compared with MBS and CT based ABS in brain cancer. In addition STAPLE has provided better results for smaller volumes based on my study. PO-1003 A analysis of safety of whole brain radiotherapy with Hippocampus avoidance in brain metastasis Y. Han 1 , J. Chen 1 , G. Cai 1 , X. Cheng 1 , Y. Kirova 2 , W. Chai 3 1 Shanghai Jiao Tong university-ruijin hospital, radiaton oncology, Shanghai, China 2 Institute Curie- Paris- France, Department of Radiation Oncology, Paris, France 3 Shanghai Jiao Tong university-ruijin hospital, Department of Radiology, Shanghai, China Purpose or Objective Purpose : Whole brain radiotherapy (WBRT) remains reference treatment in patients with brain metastasis (BM), especially with multiple lesions. Hippocampus avoidance in WBRT (HA-WBRT) offers the feasibility of less impaired cognitive function than conventional WBRT and better intracranial control than SBRT. Oncological safety is critical in defining the proper role of HA-WBRT. The study aims to investigate the frequency of intracranial substructure involvement based on large series of radiological data and to optimize the margin definition in treatment planning. Material and Methods Methods : Consecutive patients with diagnosis of BM from enhanced MRI between 03/2011 and 07/2016 diagnosed and treated in RuiJin Hospital were analyzed. Lesions of each patient were confirmed by a senior radiologist and the closest distances from tumor to the hippocampal area were measured and analyzed by radiation oncologist. Results Results : A total of 226 patients (pts) (115 males and 111 females) with 1080 metastatic measurable lesions have been studied. The distribution of the primary tumours was as following: 72.6% lung cancers (LC) (n=164), 19.9% breast cancer (BC) patients (n=45) and 7.5% from other malignancies (n=17). Seventy-one pts were diagnosed with BM before or simultaneously with their primary malignancy. In the case of others 155 pts, the latency of BM appearance was as following: 14 months in LC pts (n=100), 59 months in BC pts (n=42). Totally, 758 (70.2%) lesions were situated beyond the tentorium. The median diameter of the lesions was 10 mm (1.2mm-162mm). The situation of the lesions was as following: 322 (29.8%) in the cerebellum, 268 (24.8%) in the frontal lobe, 168 (15.6%) in the temporal lobe, 128(11.9%) in the parietal lobe, 131 (12.1%)in the occipital lobe, 45 (4.2%)in the thalamus and 18 (1.6%) in the brainstem. After measuring the closest between the lesions and the Hippocampus in every case, the pts with lesions close to this zone (n=45) were classifed into 3 catogories: 7 (3.1 %) at 5 mm or less, 13 (5.7%) within 10 mm or less and 19 (8.4%) at 20 mm or less (Fig1). 45 patients received WBRT only and 18 of 45
patients who had complete radiological follow-up after WBRT in the same hospital were founded progress of BM. The median follow-up was 11 months. Only one new lesion was observed in area of Hippocampus (less than 5 mm).
Conclusion Conclusion : With complete radiological diagnosis, HA- WBRT can be delivered with oncological safety. Proper margin definition of HA in delineation is to be confirmed with individual technique. PO-1004 Machine learning methods for automated OAR segmentation P. Tegzes 1 , A. Rádics 1 , E. Csernai 1 , L. Ruskó 2 1 General Electric, Healthcare, Budapest, Hungary 2 General Electric, Healthcare, Szeged, Hungary Purpose or Objective Manual contouring of organs at risk can take significant time. The aim of this project is to use machine learning to develop fully automated algorithms to delineate various organs in the head and neck region on CT images. Material and Methods Machine learning models were built based on 48 CT sequences of the head and neck region with 5 manually contoured organs from the Public Domain Database for Computational Anatomy. Data were randomly separated to 32 train, 8 cross-validation and 8 test sequences. Three different machine learning models were combined to achieve automated segmentation. The first step uses a support vector machine classifier to separate patient anatomy from all other objects (a). The second step applies slice-based deep learning classification to detect the bounding box around the organ of interest (b). The final step achieves voxel-level classification based on a fully connected neural net on the voxel intensities of suitably selected neighboring voxels (c). Very similar model architectures were trained for all the different organs.
Results The body contour detection has been previously trained on another dataset containing full-body images and achieved an average accuracy of 96.6%. The mean error of the bounding box edges was 3mm, the corresponding dice scores ranged from 72% to 94% depending on the organ of interest. The first results of the voxel level segmentation gave average dice values of 38% to 77% depending on the investigated organ, and several opportunities for further fine-tuning have been identified.
Made with FlippingBook