Currently, nnUnet tool users need to download pretrained weights for each task and extract (in a specific way) it to RESULTS_FOLDER. The information on available tasks to download, followed by extraction of it to the RESULTS_FOLDER without leaving the MITK is desirable.
Description
Description
Status | Assigned | Task | ||
---|---|---|---|---|
Restricted Maniphest Task | ||||
Resolved | a178n | T29154 Show available Tasks for Download in the nnUnet GUI | ||
Resolved | a178n | T29186 nnUNet Python: Export available model information as JSON file | ||
Resolved | a178n | T29187 Add GUI elements to parse and show available models | ||
Resolved | a178n | T29199 Move model download to a different thread. | ||
Resolved | a178n | T29201 Update checklist for pretrained model download | ||
Resolved | a178n | T29202 Update doku for pretrained model download |
Event Timeline
Comment Actions
One of the ways to accomplish this is to wrap the string output of nnUNet_print_available_pretrained_models command and extract the tasks + respective links. This however is non-trivial.
Comment Actions
The output of the abovementioned command, for reference:
Please cite the following paper when using nnUNet: Isensee, F., Jaeger, P.F., Kohl, S.A.A. et al. "nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation." Nat Methods (2020). https://doi.org/10.1038/s41592-020-01008-z If you have questions or suggestions, feel free to open an issue at https://github.com/MIC-DKFZ/nnUNet The following pretrained models are available: Task001_BrainTumour Brain Tumor Segmentation. Segmentation targets are edema, enhancing tumor and necrosis, Input modalities are 0: FLAIR, 1: T1, 2: T1 with contrast agent, 3: T2. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task002_Heart Left Atrium Segmentation. Segmentation target is the left atrium, Input modalities are 0: MRI. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task003_Liver Liver and Liver Tumor Segmentation. Segmentation targets are liver and tumors, Input modalities are 0: abdominal CT scan. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task004_Hippocampus Hippocampus Segmentation. Segmentation targets posterior and anterior parts of the hippocampus, Input modalities are 0: MRI. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ 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/ Task006_Lung Lung Nodule Segmentation. Segmentation target are lung nodules, Input modalities are 0: abdominal CT scan. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task007_Pancreas Pancreas Segmentation. Segmentation targets are pancras and pancreas tumor, Input modalities are 0: abdominal CT scan. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task008_HepaticVessel Hepatic Vessel Segmentation. Segmentation targets are hepatic vesels and liver tumors, Input modalities are 0: abdominal CT scan. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task009_Spleen Spleen Segmentation. Segmentation target is the spleen, Input modalities are 0: abdominal CT scan. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task010_Colon Colon Cancer Segmentation. Segmentation target are colon caner primaries, Input modalities are 0: CT scan. Also see Medical Segmentation Decathlon, http://medicaldecathlon.com/ Task017_AbdominalOrganSegmentation Multi-Atlas Labeling Beyond the Cranial Vault - Abdomen. Segmentation targets are thirteen different abdominal organs, Input modalities are 0: abdominal CT scan. Also see https://www.synapse.org/#!Synapse:syn3193805/wiki/217754 Task024_Promise Prostate MR Image Segmentation 2012. Segmentation target is the prostate, Input modalities are 0: T2. Also see https://promise12.grand-challenge.org/ Task027_ACDC Automatic Cardiac Diagnosis Challenge. Segmentation targets are right ventricle, left ventricular cavity and left myocardium, Input modalities are 0: cine MRI. Also see https://acdc.creatis.insa-lyon.fr/ Task029_LiTS Liver and Liver Tumor Segmentation Challenge. Segmentation targets are liver and liver tumors, Input modalities are 0: abdominal CT scan. Also see https://competitions.codalab.org/competitions/17094 Task035_ISBILesionSegmentation Longitudinal multiple sclerosis lesion segmentation Challenge. Segmentation target is MS lesions, input modalities are 0: FLAIR, 1: MPRAGE, 2: proton density, 3: T2. Also see https://smart-stats-tools.org/lesion-challenge Task038_CHAOS_Task_3_5_Variant2 CHAOS - Combined (CT-MR) Healthy Abdominal Organ Segmentation Challenge (Task 3 & 5). Segmentation targets are left and right kidney, liver, spleen, Input modalities are 0: T1 in-phase, T1 out-phase, T2 (can be any of those) Also see https://chaos.grand-challenge.org/ Task048_KiTS_clean Kidney and Kidney Tumor Segmentation Challenge. Segmentation targets kidney and kidney tumors, Input modalities are 0: abdominal CT scan. Also see https://kits19.grand-challenge.org/ Task055_SegTHOR SegTHOR: Segmentation of THoracic Organs at Risk in CT images. Segmentation targets are aorta, esophagus, heart and trachea, Input modalities are 0: CT scan. Also see https://competitions.codalab.org/competitions/21145 Task061_CREMI MICCAI Challenge on Circuit Reconstruction from Electron Microscopy Images (Synaptic Cleft segmentation task). Segmentation target is synaptic clefts, Input modalities are 0: serial section transmission electron microscopy of neural tissue. Also see https://cremi.org/ Task075_Fluo_C3DH_A549_ManAndSim Fluo-C3DH-A549-SIM and Fluo-C3DH-A549 datasets of the cell tracking challenge. Segmentation target are C3DH cells in fluorescence microscopy images. Input modalities are 0: fluorescence_microscopy Also see http://celltrackingchallenge.net/ Task076_Fluo_N3DH_SIM Fluo-N3DH-SIM dataset of the cell tracking challenge. Segmentation target are N3DH cells and cell borders in fluorescence microscopy images. Input modalities are 0: fluorescence_microscopy Also see http://celltrackingchallenge.net/ Note that the segmentation output of the models are cell center and cell border. These outputs mus tbe converted to an instance segmentation for the challenge. See https://github.com/MIC-DKFZ/nnUNet/blob/master/nnunet/dataset_conversion/Task076_Fluo_N3DH_SIM.py Task082_BraTS2020 Brain tumor segmentation challenge 2020 (BraTS) Segmentation targets are 0: background, 1: edema, 2: enhancing tumor, 3: necrosis Input modalities are 0: T1, 1: T1ce, 2: T2, 3: FLAIR (MRI images) Also see https://www.med.upenn.edu/cbica/brats2020/ Task089_Fluo-N2DH-SIM_thickborder_time Fluo-N2DH-SIM dataset of the cell tracking challenge. Segmentation target are nuclei of N2DH cells and cell borders in fluorescence microscopy images. Input modalities are 0: t minus 4, 0: t minus 3, 0: t minus 2, 0: t minus 1, 0: frame of interest Note that the input channels are different time steps from a time series acquisition Note that the segmentation output of the models are cell center and cell border. These outputs mus tbe converted to an instance segmentation for the challenge. See https://github.com/MIC-DKFZ/nnUNet/blob/master/nnunet/dataset_conversion/Task089_Fluo-N2DH-SIM.py Also see http://celltrackingchallenge.net/ Task114_heart_MNMs Cardiac MRI short axis images from the M&Ms challenge 2020. Input modalities are 0: MRI See also https://www.ub.edu/mnms/ Note: Labels of the M&Ms Challenge are not in the same order as for the ACDC challenge. See https://github.com/MIC-DKFZ/nnUNet/blob/master/nnunet/dataset_conversion/Task114_heart_mnms.py Task115_COVIDSegChallenge Covid lesion segmentation in CT images. Data originates from COVID-19-20 challenge. Predicted labels are 0: background, 1: covid lesion Input modalities are 0: CT See also https://covid-segmentation.grand-challenge.org/ Task135_KiTS2021 Kidney and kidney tumor segmentation in CT images. Data originates from KiTS2021 challenge. Predicted labels are 0: background, 1: kidney, 2: tumor, 3: cyst Input modalities are 0: CT See also https://kits21.kits-challenge.org/
Comment Actions
Instead of command output string parsing, the plan is to export available models' information and place it in a machine-readable format (eg. JSON) in the RESULTS_FOLDER.
This file can be further read by MITK and shown in GUI.
So, this involves making changes on the Python side, as well, for the dumping of the information in a machine-readable format (eg. JSON) natively.