Mim maestro workstation

Two clinical target volumes (CTV) were defined for isolation of all four pulmonary veins [12 (link)]: a right CTV (CTVR) which included the right superior and inferior pulmonary vein antrums (RSPVA and RIPVA) and a left CTV (CTVL) which included the left superior and inferior pulmonary vein antrums (LSPVA and LIPVA). Four regions of interest (ROI) were defined as a 5 mm surface on axial view at the most antero-superior intersection of each pulmonary vein with the left atrium. These ROIs representing the RSPVA, RIPVA, LSPVA, and LIPVA were contoured on each of the eight cardiac phases by a same observer with particular care to ensure ROI was segmented at the same location in all phases. Maximal displacement of the centroid of each ROI was derived automatically in a MIM Maestro workstation (MIM Software Inc., Cleveland, OH) in the mediolateral (ML), anteroposterior (AP), and superoinferior (SI) directions.
Data were collected in an encrypted electronic database. Student T-test was used to compare maximal displacements. Analyses were completed using the SPSS statistics package (IBM Corp, 2013. IBM SPSS Statistics for Windows, Version 22.0. Armonk, NY). 

All MR images were reviewed in MIMMaestro workstation (MIM Software Inc, Cleveland, OH) by a 10-year experienced rectal MRI radiologist who delineated the gross tumor volume (GTV) following the guidelines defined in ICRU n.83 [28 (link)]. The segmentation process was performed manually and the radiomics analysis was focused on the entire volume. All DICOM files containing the MR images and the corresponding radiotherapy (RT) Structure files were imported in Moddicom, an open-source R (R Core Team, Vienna, Austria) statistical software package [29 ]. Images were pre-processed with the Laplacian of Gaussian (LoG) convolution kernel filter to decrease the high-frequency MRI signal noise and reduce the impact of large variations of signal. The size of the standard deviation (σ) in the LoG filter was scanned from 0.1 to 1.0 with a step-size of 0.05. To search for potential GTV features related to outcome prediction, five groups of features were extracted from the GTV on each pre-processed image, including statistical, morphological, grey-level co-occurrence matrix (GLCM), grey-level run length matrix (GLRLM), grey-level size zone matrix (GLSZM). In addition, three potential clinical features, i.e., clinical T-stage (cT), clinical N-stage (cN) and age, were also acquired for later analysis.

Several radiation oncologists with over 5 years of professional experience manually delineated the gross tumor volumes (GTVs). The contoured GTVs were cross-checked slice-by-slice by different experienced radiation oncologists. Since manual delineation is prone to interobserver variability, an independent radiation oncologist delineated the GTVs for 50 randomly selected patients with ESCC to evaluate and confirm the reproducibility of the radiomic features and to reduce the effect of the uncertainty in the manual tumor delineation on feature extraction. All delineation tasks were performed in on a MIM Maestro Workstation (version 6.8.2, MIM Software Inc., USA).