Automatic Landing Zone Plane Detection in Contrast-Enhanced Cardiac CT Volumes

Left atrial appendage closure (LAAC) is a common procedure whereby a device is implanted to prevent blood clots in the heart from entering the bloodstream. Selecting an appropriate occlusion device size is essential for the procedure’s success, which involves detecting the device landing zone from a contrast-enhanced computed tomography angiography (CTA) image, and measuring its diameter for sizing. Automating landing zone contour detection is challenging due to complexity of locating a 2D contour in a 3D volume. In this paper, we automate landing zone plane detection using convolutional neural networks (CNNs). We reformulate plane detection as a volumetric heatmap regression problem, for which CNNs are well-suited. Our proposed approach can accurately detect the center of the landing zone as well as the cross-section orientation necessary for measuring device size. Compared to other segmentation-based methods, our approach removes the need for costly annotations. The proposed approach can be applied to any volumetric plane detection or pose estimation problem, which we also demonstrate in the context of aortic valve plane detection in 4D flow phase-contrast magnetic resonance angiography (PCMRA) volumes.