Fig 1.
Schematic pipeline of the reconstruction algorithm.
Fig 2.
From the left: 2D angiographic image (pixel resolution of 512x512 and image pixel spacing of 0.278mm); vesselness image (red colour indicates higher response); extracted coronary tree; labelled segments (white) and nodes (red) overlaid the original image.
Fig 3.
Intersection relationship of two corresponding surfaces for 3D reconstruction of an arterial segment centreline.
The two triangulated surfaces Sn,m intersect at C.
Fig 4.
Intersection relationship of the reference plane with an additional plane
.
Diameter vector at (xk,yk,zk)′ is stored at (xk,yk,zk).
Fig 5.
3D luminal cross-section reconstruction.
a) Schematic view of a projection plane , correspondence of a 3D centreline point; definition of cross-sectional plane
. b) 3D reconstruction of a diameter vector d′k,n; magnification, followed by projection onto the cross-sectional plane
.
Fig 6.
3D luminal contour reconstruction.
An initial circular luminal contour on a cross-sectional plane (diameter d′k,N) and its control points; four (top) and six (bottom) interpolatory control points Qk are reconstructed; the initial control points are adjusted as described in the text; the final contour is obtained.
Fig 7.
3D luminal surface reconstruction.
A subset of contours (green) is shown; red surface is the lofted surface. Alignment of the first control points (blue) was required to avoid torsions.
Fig 8.
Reduction in diameter, comparison between manual and semi-automatic detection (n = 24 data).
On the left, the regression line between the two methods. Regression line has a slope of 0.869 and an intercept of 0.0388. Pearson r-value squared is r2 = 0.9691. Sum of the squared errors is SSE = 0.021. On the right, the Bland-Altman plot. Reproducibility coefficient and % of mean values RPC (%) = 0.047 (9%); coefficient of variation CV = 4.6%. The solid line represents the mean of the differences; dotted lines define the interval mean difference ± 1.96 SD.
Fig 9.
Luminal volume, comparison between the proposed method (PM) and OCT.
n = 6 patients. On the left, the regression line between the two methods. Regression line has a slope of 0.929 and an intercept of 0.286. Pearson r-value squared is r2 = 0.9863. Sum of the squared errors is SSE = 3.2 mm3. On the right, the Bland-Altman plot. Reproducibility coefficient and % of mean values RPC (%) = 6.8(19%). The solid line represents the mean of the differences; dotted lines define the interval mean difference ± 1.96 SD.
Table 1.
Results of the analysis with OCT and proposed method (PM).
Fig 10.
3D reconstruction of Patient 1 and Patient 2 (Table 2).
Top: 2D angiographic images and their 3D relationship. Bottom: an example of 3D reconstructed luminal cross-section (green), circular fitting (blue), and elliptical fitting (red); corresponding OCT frame.
Table 2.
Comparison of proposed method (PM), circular and elliptical fit.
Fig 11.
2D angiographic images and their 3D relationship; 3D reconstructed model and interpolated model (shaded grey region). The illustrated cases reported an FFR value equal to 0.75 and 0.89, respectively.
Fig 12.
Volume reduction as from proposed method (PM) and measured pressure gradient (FFR measurement).
On the left, the regression line between the two methods. Dashed circle encloses positive FFR clinical outcomes; solid circle encloses negative FFR clinical outcomes. Dashed lines indicate that the cut-off of 0.8 in FFR corresponds to a cut-off of 0.77 in volume drop. On the right, the Bland-Altman plot. Reproducibility coefficient and % of mean values RPC (%) = 0.16% (20%); coefficient of variation CV = 10%. The solid line represents the mean of the differences.
Fig 13.
RMSE with subsets of the projection planes.
Patient 5, Table 2.