CardioVision: A fully automated deep learning package for medical image segmentation and reconstruction generating digital twins for patients with aortic stenosis

Abstract

Aortic stenosis (AS) is the most prevalent heart valve disease in western countries that poses a significant public health challenge due to the lack of a medical treatment to prevent valve calcification. Given the aging population demographic, the prevalence of AS is projected to rise, resulting in a progressively significant healthcare and economic burden. While surgical aortic valve replacement (SAVR) has been the gold standard approach, the less invasive transcatheter aortic valve replacement (TAVR) is poised to become the dominant method for high- and medium-risk interventions. Computational simulations using patient-specific models, have opened new research avenues for optimizing emerging devices and predicting clinical outcomes. The traditional techniques of generating digital replicas of patients' aortic root, native valve, and calcification are time-consuming and labor-intensive processes requiring specialized tools and expertise in anatomy. Alternatively, deep learning models, such as the U-Net architecture, have emerged as reliable and fully automated methods for medical image segmentation. Two-dimensional U-Nets have been shown to produce comparable or more accurate results than trained clinicians' manual segmentation while significantly reducing computational costs. In this study, we have developed a fully automatic AI tool capable of reconstructing the digital twin geometry and analyzing the calcification distribution on the aortic valve. The developed automatic segmentation package enables the modeling of patient-specific anatomies, which can then be used to simulate virtual interventional procedures, optimize emerging prosthetic devices, and predict clinical outcomes.

Keywords: Aortic stenosis; Automated AI platform; Calcium distribution; Digital twin.

Fig. 1:

Deep learning powered and fully automatic pipeline to generate patient-specific models of aortic stenosis cases. The blue and red arrows show the training and prediction processes, respectively.

Fig. 2:

Cropping procedure used to export the patient CT images, focusing on the aortic root and LVOT regions.

Fig. 3:

Deep convolutional neural network (DCNN) architecture used for semantic image segmentation.

Fig. 4:

Enumeration of the aortic valve landmarks, illustrating the basal (left) and the commissural attachment points (middle), along with the point of coaptation on the average plane (right).

Fig. 5:

Landmark Detection. (a) Bottom landmark guidelines connecting bottom enclose triangle vertices to the triangle centroid; (b) Top landmark guidelines connecting centroid of the bottom enclosed triangle transferred to the commissural plane to the mid points of the opposite edges; (c) Fitting ellipse calculated from the aortic cross-section on coronary plane; (d) Fitted plane transferred to the bottom landmark for landmark assignment; (e) Calculating the landmark distance to the major axis of fitted ellipse to assign the landmarks.

Fig. 6:

(a) Side view of the design with the anatomical landmarks and derived construction points marked in blue, construction points in red, and parameters in green (b) Free edge curves and guide curves; (c) The spline guiding points on the cross splines resulting in a homogenous point cloud; (d) Offsetting the leaflet point cloud along the normals, resulting in a solid body point cloud.

Fig. 7:

Calcium detection using thresholding techniques

Fig. 8:

Steps taken to produce the STL geometry for the aorta (orange), calcifications (blue), and leaflets (green).

Fig. 9:

Automatic generation of the patient-specific aortic root and LVOT (left) compared to the ground truth obtained from manual segmentation conducted by experienced clinicians (right)

Fig. 10:

(a) Detected commissural landmarks identified on the top plane; (b) Detected basal landmarks identified on the bottom plane; (c) 3D representation of all the landmarks; (d) Final reconstructed model of the aortic valve.

Fig. 11:

The process of calcification adjustment and segregation; (a) All the detected calcium islands within the defined mask; (b) Isolation and adjustment of the detected calcifications on the aortic valve leaflets (dark green). A mask, representing the cross-sectional area of the generated leaflets, is created below the sinotubular junction (yellow). Calcifications (orange in a) whose center of mass falls within the mask are translated onto the leaflet, resulting in improved visualization (purple); (c) Patient-specific calcium distribution on individual leaflets extracted automatically. Left, right, and non-coronary calcifications are shown in yellow, blue, and red.

Fig. 12:

Representative calcium distribution analysis using the automatically reconstructed digital twin. The blue plane includes the top landmarks of the aortic valve. The distribution of the selected, isolated calcium island (red) is characterized where R represents the radial position of its center of mass (dark sphere) in relation to the aortic orifice center (white sphere). The projection of the R vector on the landmark plane is measured as r. The location of each calcium island is determined using two parameters (r and α) where α is the minimum angular distance between the r and any of the given boundaries (dashed lines) or an individual leaflet. The arc angle (β) is defined to quantify the angular stretch of the isolated calcium island.

Fig. 13:

a) Automated extraction of the aortic root and coronaries; b) Detection of coronary artery ostia, individual valve leaflets, and calcifications; c) Pearson correlation factors for calcification metrics and the outcome of permanent pacemaker placement installation (PPM); d) Principal component analysis with support vector machine decision boundary for PPM and no PPM classes