The model was trained using the Niftynet framework; instructions are provided below to allow researchers to utilize the trained DL model. The instructions given here will largely follow the instructions given at https://github.com/NifTK/NiftyNet.
pip install “tensorflow==1.15.*”
pip install niftynet
We have provided a configuration file that can be used for inference on 1H MRI scans. We highlight the main parameters in the configuration file required for inference that are specific to the primary user.
The user must provide paths to folders containing the 1H MRI scans and a path to the location to save the outputted segmentations.
[Proton]
path_to_search = #path to 1H MRI scans
save_seg_dir = #path to output location
Depending on the computational resources available to the individual researcher, the following parameters will need to be amended:
num_threads = 0
num_gpus = 0
In our investigations, we used a single GPU with two threads.
The following command can be used to run inference by indicating the path to the configuration file:
net_segment inference -c /path_to_config/config_inference.ini