Computer Vision News - January 2021

The modalities matching the identifiers and other specs are described in the dataset.json file. An example of how to create a suitable json file for new data is contained in the folder nnUnet- >dataset_conversion . Have a look there if you need to create this file for your task. Here I used the Prostate dataset from the MSD challenge, and the nnU- Net package has a dedicated script to convert the folder downloaded from the decathlon directly into the right format and this can be run with the following line: 17 nnU-Net After making sure everything is correctly set up, all is left to do is see what the nnU-Net tool has to offer in terms of pre-trained models, download the one we are interested in and use it to predict our data. nnUNet_convert_decathlon_task -I ~/Downloads/Task05_Prostate nnUNet_print_available_pretrained_models Among the others, the above command prints the Prostate segmentation which I will use for my inference: Task005_Prostate Prostate Segmentation. Segmentation targets are peripheral and central zone, input modalities are 0: T2, 1: ADC. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ nnUNet_download_pretrained_model Task005_Prostate nnUNet_predict -i $nnUNet_raw_data_base/nnUNet_raw_data/Task005_Prostate/imagesTs/ -o OUTPUT_DIRECTORY -t 5 -m 3d_fullres When the inference is finished, a new directory called OUTPUT_DIRECTORY should be found inside the main data folder. This contains the predictions, which can be visualised and overlaid on top of the corresponding original imaging files using any biomedical imaging visualisation tool like below.