Improved Three-Dimensional Reconstruction of Patient-Specific Carotid Bifurcation Using Deep Learning Based Segmentation of Ultrasound Images

Clinical examination is crucial during diagnostics of many diseases, including carotid artery disease. One of the most commonly used imaging techniques is the ultrasound (US) examination. However, the main drawback of US examination is that only two-dimensional (2D) cross-sectional images are obtained. For a more detailed analysis of the state of the patient’s carotid bifurcation it would be very useful to analyze a three-dimensional (3D) model. Within this study, an improved methodology for the 3D reconstruction is proposed. US images were segmented by using deep convolutional neural networks, and lumen and arterial wall regions are extracted. Instead of using a generic model of the carotid artery as the basis that is further adapted to the particular patient with individual US cross-sectional images, in the presented approach the longitudinal cross-sectional US image of the whole carotid bifurcation is used to extract the shape of the whole geometry, which ensures more realistic 3D model. Computer AI-based 3D reconstruction of patient-specific geometry could ensure more complete view of the carotid bifurcation, but also this geometry could be further used within numerical simulations such as blood flow simulation or simulation of plaque progression, that could provide additional quantitative information useful for clinical diagnostics and treatment planning.