In VivoSupervised Analysis of Stent Reendothelialization From Optical Coherence Tomography

Abstract

The aim of this study is to interactively assess reendothelialization of stents at an accuracy of down to a few micrometer by analyzing endovascular optical coherence tomography (OCT) sequences. Vessel wall and stent struts are automatically detected by using morphological, gradient, and symmetry operators coupled with active contour models; alerts are issued to ask for user supervision over some extreme irregular geometries caused by thrombotic lesions or dissections. A complete distance map is then computed from sparse distances measured between wall and struts. Missing values are interpolated by thin-plate spline (TPS) functions. Accuracy and robustness are increased by taking into account the inhomogeneity of data points and integrating in the same framework orthogonalized forward selection of support points, optimal selection of regularization parameters by generalized cross-validation, and rejection of detection outliers. Validation is performed on simulated data, phantom acquisitions and 11 typical in vivo OCT sequences. The comparison against manual expert measurements demonstrates a bias of the order of OCT resolution (less than 10 microm) and a standard deviation of the order of the strut width (less than 150 microm).

Automated peroperative assessment of stents apposition from OCT pullbacks

“…The detection of struts allows us to initialize the distance map and to minimize the number of false positive at the same time. Comparing with the method already published [19], the TPS reflects the angular continuity considering periodic angles, takes into account the presence of a small proportion of false positive during the interpolation and rejects them. The longitudinal resolution allows us to have a good initialization of the distance map with our detected struts.…”

“…To segment the border of the lumen, some studies used deformable model like a Catmull-Rom spline [12,14,19]. The initialization of the spline is usually the inner edge of a region of interest which isolates the high reflective echoes in the slice.…”

“…In our algorithm, the detection of the probe's echo and the rail before searching the struts allows to avoid detecting these elements as struts. Similarly to [16,19], we also use both the intensity profile of the rays and the position of the shadow zone to detect the exact position of struts.…”

“…Moreover, an M-estimator stage is introduced in order to generate a robust regression and to cause the rejection of the false positive detected as struts. The details of the resolution of this problem and the summary of the criteria used are given in [19].…”

“…This distance is imperfectly known at the detected struts and is extended to the whole stent using a thin-plate spline model, resulting in a distance map expressed in cylindrical coordinates [18]. This metric distance map is converted into a color score map to differentiate types of struts found in OCT sequences [19]. Fig.…”