Segmentation of Different Human Organs on 3D Computer Tomography and Magnetic Resonance Imaging using an Open Source 3D U-Net Framework

Medical segmentation represents a method to delineate the borders of internal human organs on acquired 3D Computer Tomography or MRI images, which are the most currently used imaging modalities. The U-net represents a deep neural network architecture that has the basic structure consisting in two paths and can be used for medical image segmentation. For U-Net Framework implementation we used Monai Toolkit from Kitware Inc, which is a PyTorch-based, open-source framework for deep learning in healthcare imaging used to create advanced training workflows and provides deep learning models. As input data we used two types of internal organs acquired with two image modalities:3D CT and MRI.For visualization we used an in-house VTK based 3D Volume Visualization Application CardioCTNav.For the manual segmentation of the training datasets we used an open source segmentation application ITK-SNAP. Our results shows that the U-Net segmentation method delivers good final results and that is represents an robust method to segment internal organs.