Automatic Assessment of Transcatheter Aortic Valve Implantation Results on Four-Dimensional Computed Tomography Images Using Artificial Intelligence

Figure 1. Schema of the pipeline for automatic extraction of TAVI devices and implants proposed in this work.

Figure 2. Original and orthogonal views of the CT images captured from a patient with an Allegra of 27 mm implanted. (a) Axial view of the original image slices, (b) prosthesis 3D reconstruction, with the axial view of the original images, (c) orthogonal view of the prosthesis, after the image transformation, (d) prosthesis 3D reconstruction, with a plane orthogonal to the prosthesis principal axis.

Figure 3. 3D reconstruction of the results of the intermediate steps of the prosthesis segmentation process. (a) Results after the image thresholding, (b) results after the removal of the small islands in the thresholding, (c) final prosthesis segmentation, after removing the large structures.

Figure 4. Reformat of the images to obtain orthogonal views along the prosthesis long axis. The images include the 3D reconstruction of the prosthesis frame segment (red) and the filling segment (blue), as well as the planes. (a) Original axial views; (b) orthogonal plane, obtained after the reformatting; (c) orthogonal plane, but rotated, aligning the prosthesis long axis with Y axis; (d) orthogonal plane view from the top.

Figure 5. Different views of the prosthesis segmentation (red) and the seeds used to initialize the GrowCut algorithm for the aortic root segmentation: foreground seed (yellow) and background seed (green). (a) Axial view, (b) 3D reconstruction, (c) sagittal view, (d) coronal view.

Figure 6. Aortic root segmentation. (a) Segmentation obtained by the GrowCut algorithm, using the seeds defined in Figure 5, (b) post-processed version of the aortic root segmentation after removing noise and the coronary arteries as well as flattening the surface.

Figure 7. Schema of the prosthesis parameters studied in this work for a given time instant t. On the left, the figure shows the 3D reconstruction of the TAVI frame (red) and its filling (blue) segments, along with the coordinate axes [x,y,z], the prosthesis principal axis (red line), and an example of a region of interest (ROI) of an orthogonal plane. The variables V(t) and c(t) refer to the prosthesis volume and center of mass in instant t, respectively. The variables αx, αy, and αz represent the angles between the principal axis and the coordinate axes. The orthogonal plane located at height h, shown in the 3D view, is also displayed in the right part of the figure. The 2D view on the right shows the cross-sections of the segments at h and the parameter CSA(t,h), which refers to the CSA of the filling segment at time instant t and height h.

Figure 8. Example of the 3D reconstruction of different prosthesis models with the specific planes h1, h2, and h3, defined for the prosthesis characterization. (a) Allegra 27 mm, (b) CoreValve 26 mm (h2 coincides with h1).

Figure 9. Cross-section of the TAVI frame and the aortic root segmentations, orthogonal to the prosthesis axis. The CSAs of the prosthesis (blue) and the aortic root (yellow) are defined for a given plane h and time instant t.

Figure 10. Results obtained for the prosthesis characterization as a 3D object for eight patients with different prosthesis models (Allegra 23 mm in blue, Allegra 27 mm in red, Allegra 31 mm in green, CoreValve 26 mm in purple, CoreValve 34 mm in black). (a) Evolution of the inner volume of the prosthesis over the cardiac cycle for the patients with the minimum and maximum mean volume (computed over all the time volumes) for each prosthesis model, (b) Displacement of the prosthesis center of mass over the cardiac cycle for the patients with the minimum and maximum mean displacement (computed over all the time volumes) for each prosthesis model.

Figure 11. Analysis of the CSA of the prosthesis along the principal axis and over the cardiac cycle for four patients, with different prosthesis models implanted. (a) CSA evolution along the principal axis for a patient with an Allegra 23 mm, where each curve corresponds to a different time stamp, besides three vertical lines determining the planes h1, h2, and h3; (b) CSA evolution over the cardiac cycle for the same patient in (a), for the three planes h1, h2, and h3. The next pairs of graphs replicate the same analysis for patients with other prosthesis models: (c,d) Allegra 27 mm, (e,f) Allegra 31 mm, and (g,h) CoreValve 26 mm.

Figure 12. CSAs of the prosthesis and the aorta and dCSA for a single patient (Allegra 27 mm), besides the planes h1, h2, and h3. (a) CSAp(t,h) and CSAa(t,h) along the principal axis for each time stamp; (b) dCSA(t,h) along the principal axis for each time stamp; (c) dCSA(t,h) evolution over the cardiac cycle for each of the specific planes h1, h2, and h3.