https://doi.org/10.1007/s00392-025-02625-4
1Universitätsklinikum Heidelberg Klinik für Innere Med. III, Kardiologie, Angiologie u. Pneumologie Heidelberg, Deutschland; 2Karlsruher Institut für Technologie (KIT) Institut für Biomedizinische Technik Karlsruhe, Deutschland; 3Hospital Clinic University of Barcelona Arrhythmia Section Barcelona, Spanien
Introduction
Cardiovascular diseases are the leading cause of death globally. Recent technological advances in cardiac imaging, e.g. cardiac computed tomography, magnetic resonance imaging and electroanatomical mapping enable detailed analyses of cardiac morphology and function. However, to perform regional quantitative intra- and inter-individual, as well as cross-modality comparisons, a standardized regionalization of the heart is required. While there are multiple software solutions realizing the 17-segment AHA model of the left ventricle, there are currently no automated algorithms for a standardized regionalization of the atria.
Cardiovascular diseases are the leading cause of death globally. Recent technological advances in cardiac imaging, e.g. cardiac computed tomography, magnetic resonance imaging and electroanatomical mapping enable detailed analyses of cardiac morphology and function. However, to perform regional quantitative intra- and inter-individual, as well as cross-modality comparisons, a standardized regionalization of the heart is required. While there are multiple software solutions realizing the 17-segment AHA model of the left ventricle, there are currently no automated algorithms for a standardized regionalization of the atria.
Methods
Here, we introduce an open-source algorithm (DIVAID, https://gitlab.kit.edu/kit/ibt-public/divaid) that automatically divides any bi-atrial geometry according to previously specified definitions, which are based on anatomical, electrophysiological and clinical considerations. The regionalization is performed in three steps. First, the input geometry is standardized by remeshing and semi-automatically clipping the veins. Second, the orifices are extracted and annotated automatically. Then, anatomical landmarks are derived by using the annotated ostia and the attitudinal anatomical orientation of the atria, i.e. the body coordinate system. Lastly, the boundary of each region is determined by computing the shortest geodesic path between two corresponding points. To validate the accuracy of the automatic division, we compared the regionalization results to blinded manual expert annotations in 20 bi-atrial geometries from multiple acquisition modalities (CT and MRI).
Here, we introduce an open-source algorithm (DIVAID, https://gitlab.kit.edu/kit/ibt-public/divaid) that automatically divides any bi-atrial geometry according to previously specified definitions, which are based on anatomical, electrophysiological and clinical considerations. The regionalization is performed in three steps. First, the input geometry is standardized by remeshing and semi-automatically clipping the veins. Second, the orifices are extracted and annotated automatically. Then, anatomical landmarks are derived by using the annotated ostia and the attitudinal anatomical orientation of the atria, i.e. the body coordinate system. Lastly, the boundary of each region is determined by computing the shortest geodesic path between two corresponding points. To validate the accuracy of the automatic division, we compared the regionalization results to blinded manual expert annotations in 20 bi-atrial geometries from multiple acquisition modalities (CT and MRI).
Results
The mean regional overlap (Dice score) between both divisions across all geometries for both atria combined was 0.92. In the LA, the Dice score was 0.95, whereas in the RA, it was 0.88. The mean Euclidean distance between regional boundaries derived from either manual or automatic division in both atria was 1.70mm. In the LA, the mean Euclidean distance between regional boundaries was 1.16mm, while in the RA it was 2.25mm. Given a mesh element edge length of 1mm, this implies a mean deviation between manual and automatic division of less than 2 mesh elements. The higher agreement between manual and automatic regionalization in the LA compared to the RA can be explained by the symmetric shape of the LA and the presence of multiple significant landmarks. In contrast, the RA shows a higher anatomical variability and asymmetry with fewer prominent landmarks.
The mean regional overlap (Dice score) between both divisions across all geometries for both atria combined was 0.92. In the LA, the Dice score was 0.95, whereas in the RA, it was 0.88. The mean Euclidean distance between regional boundaries derived from either manual or automatic division in both atria was 1.70mm. In the LA, the mean Euclidean distance between regional boundaries was 1.16mm, while in the RA it was 2.25mm. Given a mesh element edge length of 1mm, this implies a mean deviation between manual and automatic division of less than 2 mesh elements. The higher agreement between manual and automatic regionalization in the LA compared to the RA can be explained by the symmetric shape of the LA and the presence of multiple significant landmarks. In contrast, the RA shows a higher anatomical variability and asymmetry with fewer prominent landmarks.
Conclusion
This open-source algorithm enables standardized, consistent, reproducible and operator-independent regional quantitative comparisons of the atria between multiple modalities, patients and centers which may facilitate cardiac research towards personalized treatment approaches.
This open-source algorithm enables standardized, consistent, reproducible and operator-independent regional quantitative comparisons of the atria between multiple modalities, patients and centers which may facilitate cardiac research towards personalized treatment approaches.