Development of a semi-automated method for mitral valve modeling with medial axis representation using 3D ultrasound
Schematic of semi-automated image analysis. First, a deformable medial template of the mitral valve is generated from 3D ultrasound image data. This template generation step is performed once. Then, for each subject in the study, a 3D ultrasound image of the mitral valve is segmented using the level set method. The template is then deformed to fit each segmented image. Finally, morphological features are automatically computed from the fitted medial representation of each valve.
Outline of semi-automated segmentation. (a) The user initializes two points in a long-axis cross-section of the 3D US image volume, identifying an ROI (red) containing the valve along the axial dimension. (b) The user initializes a series of annular points in an enhanced projection image depicting the valve from an atrial perspective. (c) The user shifts posterior annular points into the coaptation zone, forming an outline of the anterior leaflet in the enhanced projection image. (d) A 3D point cloud delineating the valve is automatically generated. (e) The 3D point cloud is morphologically dilated with a spherical structuring element to obtain an ROI containing the valve. (f) A final segmentation of the valve is obtained by thresholding and active contour evolution. (LA = left atrium, LV = left ventricle, AL = anterior leaflet, PL = posterior leaflet, RO = regurgitant orifice).
The local geometry of a point on the medial manifold. The ball represents a medial node m(u 1,u 2), and the vectors ∂m/∂u1 and ∂m/∂u2 span the tangent plane to m. The vector ∇ m R lies in the tangent plane and points in the direction of maximal change of the radial thickness field R on m. The points of tangency between a ball centered at m(u 1,u 2) with radius R and the boundary of the object are b + and b −. Unit vectors U + and U − point from m(u 1,u 2) toward b + and b − and are perpendicular to the object boundary at these points.
Schematic of the template generation process. (a) A 3D US image volume is segmented to obtain binary images of the anterior (top) and posterior (bottom) mitral leaflets. (b) The binary images are skeletonized in 3D. (c) A 2D scatter plot of the skeletons’ vertices in u 1,u 2 space. (d) The constrained conforming Delaunay triangulation of the region containing the skeletons’ vertices. (e) The combined medial manifolds of the anterior and posterior leaflets used for deformable modeling.
A medial representation of the valve is obtained by fitting a valve template to a binary image of the valve produced by segmentation.
Leaflet overlap at the coaptation line (shown on left) is corrected during the final model fitting stage (shown on right). Magnifications of leaflet overlap and leaflet overlap correction are shown below the valves.
Automated extraction of annular geometry. (a) The annulus (bold curve) is identified as points on the valve boundary mapped to the outer edges of the leaflets’ medial manifolds. Annular landmarks include: (1) the anterior aortic peak, (2) the anterior commissure, (3) the midpoint on the posterior annulus, and (4) the posterior commissure. The line from (1) to (3) represents the septolateral diameter, and the line from (2) to (4) represents the intercommissural width. (b) Annular height is plotted as a function of rotation angle, where 0° corresponds to the anterior aortic peak of the annulus.
Automated reconstructions of the valves of six patients. All valves are depicted at midsystole. For each valve, an atrial view is shown on top and a side view is shown on the bottom. The anterior leaflet is on the right, the posterist leaflet on the left. Clinical assessments of MR severity, based on Doppler imaging, are indicated.
Bland-Alman plots showing the difference between manual and automated measurements as a function of the mean measurement for each of 14 patients. (AA = annular area, AC = annular circumference, AH = annular height, IW = intercommissural width, SL = septolateral length, TTV = total tenting volume, PATV = percent anterior tenting volume).
Semi-automated and manual reconstructions of a mitral valve. (a) A fitted cm-rep of the valve. (b) Manual tracing of the atrial surface of the valve in the same image data. (b) Manual tracing (blue) overlaid on the fitted cm-rep.
Automated measurement of the regurgitant orifice area (ROA) is shown as a function of MR severity assessed by intraoperative Doppler imaging.
Probable foci of anterior chordal tethering are identified on the ventricular side of the anterior leaflet. These foci, marked by black arrows, are identified as (a) bulges on the valve reconstruction, (b) local areas of thickening on the medial manifold, (c) regions of high mean curvature convex toward the left ventricle.
The range in valve measurements derived from semi-automated analysis, and the average difference ± standard deviation in valve measurements derived from semi-automated and manual 3D US image analysis (shown in italics). The values are shown separately for all subjects, subjects with trace to mild MR, and subjects with moderate to severe MR. (AA = annular area, AC = annular circumference, AH = annular height, IW = intercommissural width, SL = septolateral length, TTV = total tenting volume, PATV = percent anterior tenting volume).
Average percent difference in the regurgitant orifice area (ROA) when noise is applied to the user-initialized points in semi-automated segmentation. The p-values obtained by single-factor ANOVA indicate a statistically significant difference in the ROA measurements of three categories of MR severity (trace, mild, and moderate to severe).
Dice similarity coefficient measuring overlap between the binary segmentation and fitted medial model using three different templates. The Dice similarities for the anterior and posterior leaflets are separately computed. S1, S3, and S5 refer to the valve templates generated from image data acquired from Subjects 1 (severe MR), 3 (mild MR), and 5 (trace MR), respectively.
The terms and weights used in the objective function during each stage of model fitting.
Article metrics loading...
Full text loading...