Fiducial segmentation models UNITY phantom

Trained nnUNet models for segmentation of fiducial spheres in the UNITY Phantom (CaliberMRI Mini-hybrid, model 137), specifically for data acquired with the Hyperfine Swoop, portable MRI system. The three models applies to different orientations. Data should be acquired using the Fast T2w sequences on the Swoop.model237.zip: Axialmodel337.zip: Sagittalmodel437.zip: CoronalFor further details about usage see UNITY ghost repository on github.TrainingThe models were trained using the nnUNet framework. The main steps and options used in the training are listed below. <code>$DATASET_ID</code> is the ID (237 etc) and <code>$fold</code> is either 0, 1, 2, 3, or 4. As recommended by the nnUNet documentation, the 0th fold is first trained and the remaining folds are then trained in parallel. <pre>&gt;&gt; nnUNetv2_plan_and_preprocess -d 237 337 437 -pl nnUNetPlannerResEncM --verify_dataset_integrity<br>&gt;&gt; nnUNetv2_train -device cuda -p nnUNetResEncUNetMPlans $DATASET_ID 3d_fullres $fold --npz<br>&gt;&gt; nnUNetv2_find_best_configuration $DATASET_ID -p nnUNetResEncUNetMPlans -c 3d_fullres <br>&gt;&gt; nnUNetv2_export_model_to_zip -d $DATASET_ID -f 0 1 2 3 4 -tr nnUNetTrainer -c 3d_fullres -p nnUNetResEncUNetMPlans -o $zip_out --exp_cv_preds</pre><br>The training data is composed of data from 17 sites, with one dataset from each. Data was manually labeled using 3D Slicer.