Version update (12-sept-2024):Now using .zip files for the models, making it easier to install them using existing nnUNet functionality.Note: The remainder of the text here has been adapted to refer to the correct files/models.Please note that these networks are for research only!The development of this work was part of the research within the STOPSTORM consortium. STOPSTORM is funded through Horizon 2020 EU (grant agreement No 945119). More information is available at: https://stopstorm.eu/enThis upload contains four different auto-contouring models, each of them with the aim to automatically delineate cardiac substructures on CT scans. All of them are nnUNet's (Isensee 2021) trained on different input data and thus with different performance. A working installation of nnUNet v2 is required to run these networks, please find the instructions online (https://github.com/MIC-DKFZ/nnUNet). Note that all files have an extension of "_NetworkX" to their names, which needs to be removed before the files can be used to run the trained network.Network A (504) was trained on lung cancer patient data and is able to create contours for large cardiac substructures and vessels (four chamber + Aorta, superior/inferior vena cava and pulmonary artery). This was trained only on CT data (without cardiac leads) from 70 lung cancer patients.Network B (603) was trained on VT patient data and is able to contour small cardiac substructures (The four valves and four coronary artery segments). The training set consisted of 55 VT patient data, also including 21 contrast-enhanced CT (CECT) scans. Due to the limited amount of CECT scans, these were added to the training set as different patients. The total training set consisted of 52 CT and 21 CECT scans of 55 different patients.Network B large (604) was similar to network B (603), but focussed on only the large cardiac substructures, such that it could be compared to network 1 and 4.Network C (605) was trained on the combination of lung cancer patient and VT patient data (70 lung + 55 VT, total of 143 scans for training). This network is able to create large cardiac substructures on both CT as well as contrast-enhanced CT scans (although, it has seen limited contrast-enhanced CT's during training).Networks B (603 and 604) and C had cardiac leads in their training data, therefore they can be used to create cardiac substructures for VT patients in the research setting.