Bi-Atrial Model Repository (ImageCAS)

This dataset is linked to publication: "Beyond the Left Atrium: Joint Generative Modeling of Bi-Atrial Anatomy" presented at STACOM 2026. Preprint: https://papers.miccai.org/miccai-2026-sat/paper/STACOM2026_016.pdf Code for training a generative model for bi-atrial anatomy to create synthetic cohorts based on this dataset is available here. Contact: Axel Loewe - [e-mail] Dataset This dataset builds on the ImageCAS dataset (Zeng et al. 2023), data can be downloaded here: Kaggle ImageCAS.Atrial labels for this dataset by Hansen et al., 2026 are openly available here and described here Please note that the dataset by Hansen et al. is only a subset of the ImageCAS dataset that excludes scans with incomplete left atrial appendage (LAA).We refined the provided labels by resolving intersections between LAAs and pulmonary veins (PVs).Three-dimensional surface models were extracted using 3D Slicer. All surfaces were remeshed to a target edge length of 1 mm. The mitral valve was opened by removing points close to the provided left ventricle segmentations. PVs were automatically extracted and used for vein clipping with DIVAID (Goetz et al., 2026), which can be downloaded from https://gitlab.kit.edu/kit/ibt-public/divaid. The automatic extraction was visually inspected by the authors and only geometries with correctly clipped veins were kept. These geometries are provided in the folder geometries_la_open. Additionally, the geometries were subsequently closed and aligned at the septum. Left atrium is labeled as atrium with tag 1 and right atrium as atrium with tag 0. These geometries are provided in the folder geometries_closed_aligned_at_septum. File ids are the same as in the original ImageCAS dataset and the refined version by Hansen et al. Note RA data fidelity is comparatively lower than LA data fidelity (e.g., due to absence of vena cavae).We are working on improving this and will update this dataset accordingly.Missing geometries due to automatic preprocessing are not included in the dataset.We are working on improving this and will update this dataset accordingly. References Fedorov, A., Beichel, R., Kalpathy-Cramer, J., Finet, J., Fillion-Robin, J.C., Pujol, S., et al.: 3d slicer as an image computing platform for the quantitative imaging network. Magnetic resonance imaging 30(9), 1323–1341 (2012), https://doi.org/10.1016/j.mri.2012.05.001.Goetz, C., Eichenlaub, M., Schmidt, K., Wiedmann, F., Rubio, E.I., Díaz, P.M., et al.: DIVAID: Consistent division of atrial geometries from multimodal imaging according to the EHRA/EACVI 15-segment bi-atrial model (Apr 2026), https://doi.org/10.64898/2026.04.22.26351448).Hansen, B., Pedersen, J., Kofoed, K.F., Camara, O., Paulsen, R.R., Sørensen, K.: A public cardiac CT dataset featuring the left atrial appendage. In: Statistical Atlases and Computational Models of the Heart. Regular and CMRxRecon Challenge Papers. pp. 171–180. Springer Nature Switzerland, Cham (2026), https://doi.org/10.48550/arXiv.2510.06090.Wasserthal, J., Breit, H. C., Meyer, M. T., Pradella, M., Hinck, D., Sauter, A. W., ... & Segeroth, M. (2023). TotalSegmentator: robust segmentation of 104 anatomic structures in CT images. Radiol Artif Intell 5: e230024, https://doi.org/10.1148/ryai.230024.Zeng, A., Wu, C., Lin, G., Xie, W., Hong, J., Huang, M., ... & Xu, X. (2023). ImageCAS: A large-scale dataset and benchmark for coronary artery segmentation based on computed tomography angiography images. Computerized Medical Imaging and Graphics, 109, 102287, https://doi.org/10.1016/j.compmedimag.2023.102287.