ESTRO 37 Abstract book

S1176

ESTRO 37

Conclusion Textural features are less affected by different b-values, especially the GLCM-based features. Besides, b=0, 20, 50, 100, 200s/mm 2 is the highest reproducible DWI image group. The use of these stable radiomic features to analyze the reproducible DWI images of patients with hepatic cirrhosis can acquire more robust and accurate results. EP-2130 Automatic EPID image quality evaluation using a Standard Imaging QC-3 phantom. A. Prado 1 , M. Leonor 1 , A. Ferrando 2 1 Hospital Universitario 12 de Octubre, Radiofísica y Protección Radiológica, Madrid, Spain 2 Hospital Universitario 12 de Octubre, Servicio de Oncología Radioterápica. Sección de Radiofísica., Madrid, Spain Purpose or Objective To obtain a set of parameters in order to evaluate image quality of QC-3 phantom EPID images and to set tolerances for those parameters for QA purposes. Material and Methods Measurements were carried out on a 6 MV Varian Unique (Varian Medical Systems. Palo Alto, CA.) with an amorphous silicon Portal Vision unit. The phantom used was a Standard Imaging QC-3 phantom with spatial frequencies of 0.1, 0.201, 0.25, 0.435 and 0.766 lp/mm for the bar pattern sectors. It also has six sectors with different thicknesses so as to achieve different contrasts. All images were taken using 6 MU. In order to evaluate parameter consistency 20 images were obtained in consecutive days. The phantom was aligned with the lasers and was located on the EPID surface, which was set at SSD=160 cm. To analyze the images an Image J macro was developed. This macro automatically creates all needed ROIs, obtains the relevant data and performs the calculations. First, standard deviations of black and white sectors were obtained in order to calculate the noise as σ 2 n =0.5·(σ 2 b + σ 2 w ). For each bar pattern sector its standard deviation was obtained (σ i ). After that, a noise subtraction was performed just subtracting σ 2 n to each of the σ i 2 and applying the square root. These results were normalized to the σ obtained in a ROI with half black and half white regions. The corresponding 0 frequency value was force to be 1. A Rodbard function was fitted to data with R 2 >0.98, which is reasonable for obtaining a CTF estimation. The area under the curve was calculated utilizing the Simpson rule. Five parameters were evaluated, namely the area under the curve up to the highest frequency in the pattern and the frequencies for which the CTF values were 0.9, 0.5, 0.4 and 0.3, respectively. Results Results obtained are depicted in figure 1. Time evolution of all parameters studied during the considered period are shown. Standard deviations of all 20 images for each parameter were obtained (table 1). As a consequence, it was considered as reasonable to set a 3σ tolerance for every parameter (table 1) in order to compare future results when performing EPID QA at our institution.

Table 1: Average values obtained with their standard deviations (STD) and tolerances. Conclusion Image quality of EPID images was evaluated using the Standard Imaging QC-3 phantom. Several parameters were selected for time evaluation in order to set tolerances for future QA comparison. It is expected that we would be able to anticipate strong variations and lack of calibration of EPID devices as of deviations from the set tolerances in subsequent measurements. EP-2131 Deep learning based rectum tumor auto segmentation on MRI T2 image J. Lu 1 , J. Wang 1 , W. Hu 1 , Z. Zhang 1 , L. Shen 1 , Y. Sun 1 , G. Qing 1 1 Fudan University Cancer Hospital, Radiation Oncology, Shanghai, China Purpose or Objective Manually contouring gross tumor volume (GTV) is a crucial and time-consuming process in rectum cancer radiotherapy. This study aims to develop a deep learning based rectum tumor auto segmentation algorithm on MRI T2 images. Material and Methods Ninety-five patients with locally advanced (cT3-4 and/or cN1-2) rectal cancer treated with neoadjuvant chemoradiotherapy followed by surgery were enrolled in this study. The MRI T2 images were acquired before treatment on a 3.0 T MR scanner. A very deep convolutional networks (VGG16) based full convolution network (FCN) was established as training model (Figure 1). Considering the 3D structure of the MRI image, 5 image slices were simultaneously input into the network.

A two-phase training process was implemented to increase training efficiency. The existence of GTV on an image slice was identified in phase 1, and segmented in phase 2. Data were randomly separated into training (90%) and validation (10%) dataset for 10 folder cross- validation. The training dataset were input for FCN model training and the model performances were assessed on validation dataset. The segmentations were performed by an experienced radiation oncologist (reader 1). Additionally, 20 patients were double contoured by an independent radiologist (reader 2) for performance evaluation. Four indices were calculated to evaluate the similarity of automated and manual segmentation, including Hausdorff distance (HD), average surface

Figure 1: Time evolution of evaluated parameters.