Computer Vision News - August 2018

How to install NiftyNet The NiftyNet platform comes with an easy installation via the pip installation. All you need to do is type “ pip install niftynet ”. NiftyNet use case Of all the available applications, we will demonstrate the work with the config file for U-Net. U-Net is one of the most popular (possibly the most popular) deep learning architectures for medical image segmentation. Its publication in 2015 demonstrated significant medical segmentation results across a wide variety of different scenarios. We will demonstrate how you can work with U-Net to segment medical images using the NiftyNet software package. After installation, all you need to do in order to work with NiftyNet is to set a config file appropriate to the network implemented on NiftyNet. Once you set your config file, you run NiftyNet simply by running: NiftyNet configuration files have a number of sections, two of which -- SYSTEM and NETWORK -- are mandatory. The configuration file sections are listed below: • [INPUT DATA SPECIFICATIONS] -- defines the input for the network. • [SYSTEM] -- defines what computer resources the network will use. • [NETWORK] -- defines the network architecture. • [TRAINING] -- defines the training parameters. • [INFERENCE] -- defines the size/resolution of the final layer of the network, which inference will be performed on, and how many times to run inference. • [EVALUATION] -- defines the method used for evaluation (dice, jaccard, etc.) and in what form to save the results. We will proceed now step by step through the U-Net configuration file. [T1] Images will be read from ./example_volumes/monomodal_parcellation , with filenames contain T1 and not containing T2. These images will be read into 17 Tool Computer Vision News NiftyNet