Automated Estimation of Cardiac Stroke Volumes from Computed Tomography

Ana Teodora Radutoiu, Yasmin W.E. Youssef, Paula López Diez, Kristine A. Sørensen, Mathias B. Lindeskog, Michael H.C. Pham, Jonas J. Pedersen, Andreas Fuchs, Klaus F. Kofoed, Rasmus R. Paulsen

Abstract: Accurate assessment of left and right ventricular stroke volume is crucial for diagnosing and managing cardiac diseases such as heart failure, mitral regurgitation, and pulmonary embolism. The ventricular stroke volumes can be assessed from accurate ventricle segmentations in temporal 4D computed tomography (CT) images. Due to limited contrast and complicated sub-structures, the challenge of segmenting precise heart chamber volumes from CT images makes the estimation of both left and right ventricles a complicated task. Manual segmentation is time-consuming, necessitating the development of high-performing automatic segmentation methods for clinical use and large-scale studies. Existing models offer robust whole-heart segmentations but often include non-blood pool structures, leading to discrepancies in stroke volume calculations. This study proposes a data-driven method to correct left and right ventricle segmentations by accurately delineating the pure blood volume, leveraging Hounsfield Unit distributions. Our approach validates segmentation accuracy beyond traditional manual annotations by focusing on achieving equilibrium between left and right ventricular stroke volumes in healthy individuals. This method demonstrates the potential to improve cardiac function assessments from CT, facilitating more reliable and efficient clinical evaluations.

External IDs:doi:10.1007/978-3-031-87756-8_3