#OSIRIX LITE SEGMENTATION UPDATE#
#OSIRIX LITE SEGMENTATION MANUAL#
#OSIRIX LITE SEGMENTATION SOFTWARE#

#OSIRIX LITE SEGMENTATION SOFTWARE#

Nevertheless, most deep learning pipelines for semantic image segmentation generate color-coded segmentation maps stored as image files, while most free software programs for medical image analysis cannot use these files to generate ROI statistics of multimodal images stored as DICOM files. One growing area of development is the popularization of deep learning methods through the creation of user-friendly tools with a graphical interface. In recent years, numerous open-source software tools have been reported in the field of medical image processing. Similar results were obtained for the measurement of FDG uptake in the lungs ( Fig.4C-D).

#OSIRIX LITE SEGMENTATION MANUAL#

Nevertheless, the P-value of the correlation between manual measurement and that using segmentation maps continued to decrease when more ROIs were used ( Fig.4B). Increase in measurement accuracy (determined by the correlation coefficient) through increasing numbers of manually selected ROIs plateaued after more than 8 ROIs were used. For subcutaneous adipose tissue FDG uptake, the correlation coefficient and the -log of the P-value increased sharply once values from more than 4 ROIs were combined ( Fig.4A). In general, regardless the number of ROIs used, manually measured FDG uptake in the lungs and subcutaneous adipose tissue was well correlated with that calculated with segmentation maps using the MMMISA program ( Fig.4). Next, we tested the concordance of organ-specific FDG uptake measured using multiple manually selected ROIs versus FDG uptake determined using deep learning-generated segmentation maps. Increase in number of manually selected ROIs increases accuracy of organ-specific FDG uptake approximations compared to true organ-specific FDG uptake measured using deep learning-generated segmentation maps

The generalizability of such an approach for automated segmentation has previously been shown to be excellent. The final model was used to perform automatic semantic segmentation of the 4188 axial CT images from the 10 patients in less than 10 minutes. CT image-segmentation map pairs were then loaded into the DeepImageTranslator software to train a deep convolutional neural network as previously described in Ye et al. Thirty-six training samples were considered more than sufficient since we have previously shown that models can be trained to achieve high accuracy with as little as 17 images. Thirty-six axial slices were randomly chosen from the 4188 axial CT images from the 10 patients for manual semantic segmentation with the GIMP (GNU Image Manipulation Program) software of the background, lungs, bones, brain, subcutaneous and visceral adipose tissue, and other soft tissues by labelling these regions in black (RGB=), yellow (RGB=), white (RGB=), cyan (RGB=), red (RGB=), green (RGB=), and blue (RGB=). Furthermore, we also compare measurements performed using the MMMISA and those made with manually selected ROIs. We then demonstrate the use of the program for the measurement of 2-deoxy-2-fluoroglucose (-FDG) uptake by the lungs and subcutaneous adipose tissue using whole-body -FDG-PET/CT scans from the ACRIN-HNSCC-FDG-PET/CT database.

#OSIRIX LITE SEGMENTATION UPDATE#

Therefore, we present herein an update to the DeepImageTranslator software with the addition of a tool for multimodal medical image segmentation analysis (hereby referred to as the MMMISA). We have previously developed a user-friendly software tool for image-to-image translation using deep learning (DeepImageTranslator, described in, released at: ). g., 3D-Slicer, OsiriX Lite, and AMIDE) cannot use these files to generate ROI statistics of multimodal images stored as DICOM files. Nevertheless, most deep learning pipelines for semantic image segmentation generate color-coded segmentation maps stored as image files, while most free software programs for medical image analysis ( e. One possible method is the use of deep learning for automated segmentation. g., the intestines and adipose tissues), manual ROI segmentation is not a scalable approach. However, for organs/tissues with complex shapes ( e. The use of spherical or ellipsoid ROIs may be sufficient for large organs such as the liver and large muscle groups. g., position emission tomography/magnetic resonance imaging and PET/computed tomography ) often requires the selection of one or many anatomical regions of interest (ROIs) for extraction of useful statistics. Analysis of multimodal medical images ( e.

Osirix lite segmentation