Limits...
Quantification of fibrous cap thickness in intracoronary optical coherence tomography with a contour segmentation method based on dynamic programming.

Zahnd G, Karanasos A, van Soest G, Regar E, Niessen W, Gijsen F, van Walsum T - Int J Comput Assist Radiol Surg (2015)

Bottom Line: In the context of preoperative planning and perioperative decision making, intracoronary optical coherence tomography imaging can provide a very detailed characterization of the arterial wall structure.However, visual interpretation of the images is laborious, subject to variability, and therefore not always sufficiently reliable for immediate decision of treatment.Validated on a dataset of 179 images from 21 patients, the present framework could successfully extract the fibrous cap contours.

View Article: PubMed Central - PubMed

Affiliation: Biomedical Imaging Group Rotterdam, Departments of Radiology and Medical Informatics, Erasmus Medical Center, P.O. Box 2040, 3000 CA, Rotterdam, The Netherlands, g.zahnd@erasmusmc.nl.

ABSTRACT

Objectives: Fibrous cap thickness is the most critical component of plaque stability. Therefore, in vivo quantification of cap thickness could yield valuable information for estimating the risk of plaque rupture. In the context of preoperative planning and perioperative decision making, intracoronary optical coherence tomography imaging can provide a very detailed characterization of the arterial wall structure. However, visual interpretation of the images is laborious, subject to variability, and therefore not always sufficiently reliable for immediate decision of treatment.

Methods: A novel semiautomatic segmentation method to quantify coronary fibrous cap thickness in optical coherence tomography is introduced. To cope with the most challenging issue when estimating cap thickness (namely the diffuse appearance of the anatomical abluminal interface to be detected), the proposed method is based on a robust dynamic programming framework using a geometrical a priori. To determine the optimal parameter settings, a training phase was conducted on 10 patients.

Results: Validated on a dataset of 179 images from 21 patients, the present framework could successfully extract the fibrous cap contours. When assessing minimal cap thickness, segmentation results from the proposed method were in good agreement with the reference tracings performed by a medical expert (mean absolute error and standard deviation of 22 ± 18 μm) and were similar to inter-observer reproducibility (21 ± 19 μm, R = .74), while being significantly faster and fully reproducible.

Conclusion: The proposed framework demonstrated promising performances and could potentially be used for online identification of high-risk plaques.

No MeSH data available.


Related in: MedlinePlus

Schematic representation of the front propagation scheme, corresponding to the panel (f) in Fig. 1. a Original polar image, with the pixels represented by the nodes (in this example, the layout is coarse for improved visibility). b Left-to-right front propagation, with the current location of the front indicated by the vertical dashed line. The current node is indicated by an asterisk, and the connected gray nodes correspond to the set of potential neighbors. In this example, the number 2N + 1 of horizontally reachable neighbors is equal to 3. The black lines connecting the nodes represent the successive backtracking steps from a given node to the left border of the image. Please note that in the case of segmenting the luminal interface, the nodes of the upper row correspond to the top of the polar image (as shown in this example), whereas in the case of the segmenting the abluminal interface, the nodes of the upper row correspond to the luminal interface
© Copyright Policy - OpenAccess
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4563002&req=5

Fig2: Schematic representation of the front propagation scheme, corresponding to the panel (f) in Fig. 1. a Original polar image, with the pixels represented by the nodes (in this example, the layout is coarse for improved visibility). b Left-to-right front propagation, with the current location of the front indicated by the vertical dashed line. The current node is indicated by an asterisk, and the connected gray nodes correspond to the set of potential neighbors. In this example, the number 2N + 1 of horizontally reachable neighbors is equal to 3. The black lines connecting the nodes represent the successive backtracking steps from a given node to the left border of the image. Please note that in the case of segmenting the luminal interface, the nodes of the upper row correspond to the top of the polar image (as shown in this example), whereas in the case of the segmenting the abluminal interface, the nodes of the upper row correspond to the luminal interface

Mentions: We now present a dynamic programming strategy to determine the path that runs in the cost image  from left-to-right with the minimum cumulated cost . A schematic representation of this front propagation approach is displayed in Fig. 2. The proposed approach extends a previously proposed method [23, 24] and takes into account both the image feature (i.e., strong intensity gradient locally corresponding to a low cost in ) and a geometrical constraint (i.e., the shape a priori that describes a smooth structure). Therefore, high cost values as well as vertical displacement are penalized when generating the cumulated cost function , as detailed in Eq. 3 (Fig. 1f).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathbb {C}}(r, \theta +1)= & {} \min _{d_r \in \{ -N,\dots 0,\dots N \}} \Big \{ {\mathbb {C}}(r+d_r, \theta ) + \left( {\mathcal {C}}(r,\theta +1)\right. \nonumber \\&\left. +\, {\mathcal {C}}(r+d_r,\theta ) \right) \cdot \left( 1+ \alpha \cdot d_r^{\beta } \right) \Big \}, \end{aligned}$$\end{document}C(r,θ+1)=mindr∈{-N,⋯0,⋯N}{C(r+dr,θ)+C(r,θ+1)+C(r+dr,θ)·1+α·drβ},with  the vertical and horizontal coordinates, the vertical displacement of the path between two consecutive points, and 2N + 1 the number of reachable neighbors. The smoothness of the path is ruled by the positive parameters and . More specifically, the overall flexibility of the path is controlled by (i.e., small values enable vertical transitions of the path, and large values favor long horizontal plateaus), and the roughness of the path is controlled by (i.e., small values yield contours that are locally spiky, and large values impose smooth contours).Fig. 2


Quantification of fibrous cap thickness in intracoronary optical coherence tomography with a contour segmentation method based on dynamic programming.

Zahnd G, Karanasos A, van Soest G, Regar E, Niessen W, Gijsen F, van Walsum T - Int J Comput Assist Radiol Surg (2015)

Schematic representation of the front propagation scheme, corresponding to the panel (f) in Fig. 1. a Original polar image, with the pixels represented by the nodes (in this example, the layout is coarse for improved visibility). b Left-to-right front propagation, with the current location of the front indicated by the vertical dashed line. The current node is indicated by an asterisk, and the connected gray nodes correspond to the set of potential neighbors. In this example, the number 2N + 1 of horizontally reachable neighbors is equal to 3. The black lines connecting the nodes represent the successive backtracking steps from a given node to the left border of the image. Please note that in the case of segmenting the luminal interface, the nodes of the upper row correspond to the top of the polar image (as shown in this example), whereas in the case of the segmenting the abluminal interface, the nodes of the upper row correspond to the luminal interface
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4563002&req=5

Fig2: Schematic representation of the front propagation scheme, corresponding to the panel (f) in Fig. 1. a Original polar image, with the pixels represented by the nodes (in this example, the layout is coarse for improved visibility). b Left-to-right front propagation, with the current location of the front indicated by the vertical dashed line. The current node is indicated by an asterisk, and the connected gray nodes correspond to the set of potential neighbors. In this example, the number 2N + 1 of horizontally reachable neighbors is equal to 3. The black lines connecting the nodes represent the successive backtracking steps from a given node to the left border of the image. Please note that in the case of segmenting the luminal interface, the nodes of the upper row correspond to the top of the polar image (as shown in this example), whereas in the case of the segmenting the abluminal interface, the nodes of the upper row correspond to the luminal interface
Mentions: We now present a dynamic programming strategy to determine the path that runs in the cost image  from left-to-right with the minimum cumulated cost . A schematic representation of this front propagation approach is displayed in Fig. 2. The proposed approach extends a previously proposed method [23, 24] and takes into account both the image feature (i.e., strong intensity gradient locally corresponding to a low cost in ) and a geometrical constraint (i.e., the shape a priori that describes a smooth structure). Therefore, high cost values as well as vertical displacement are penalized when generating the cumulated cost function , as detailed in Eq. 3 (Fig. 1f).3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathbb {C}}(r, \theta +1)= & {} \min _{d_r \in \{ -N,\dots 0,\dots N \}} \Big \{ {\mathbb {C}}(r+d_r, \theta ) + \left( {\mathcal {C}}(r,\theta +1)\right. \nonumber \\&\left. +\, {\mathcal {C}}(r+d_r,\theta ) \right) \cdot \left( 1+ \alpha \cdot d_r^{\beta } \right) \Big \}, \end{aligned}$$\end{document}C(r,θ+1)=mindr∈{-N,⋯0,⋯N}{C(r+dr,θ)+C(r,θ+1)+C(r+dr,θ)·1+α·drβ},with  the vertical and horizontal coordinates, the vertical displacement of the path between two consecutive points, and 2N + 1 the number of reachable neighbors. The smoothness of the path is ruled by the positive parameters and . More specifically, the overall flexibility of the path is controlled by (i.e., small values enable vertical transitions of the path, and large values favor long horizontal plateaus), and the roughness of the path is controlled by (i.e., small values yield contours that are locally spiky, and large values impose smooth contours).Fig. 2

Bottom Line: In the context of preoperative planning and perioperative decision making, intracoronary optical coherence tomography imaging can provide a very detailed characterization of the arterial wall structure.However, visual interpretation of the images is laborious, subject to variability, and therefore not always sufficiently reliable for immediate decision of treatment.Validated on a dataset of 179 images from 21 patients, the present framework could successfully extract the fibrous cap contours.

View Article: PubMed Central - PubMed

Affiliation: Biomedical Imaging Group Rotterdam, Departments of Radiology and Medical Informatics, Erasmus Medical Center, P.O. Box 2040, 3000 CA, Rotterdam, The Netherlands, g.zahnd@erasmusmc.nl.

ABSTRACT

Objectives: Fibrous cap thickness is the most critical component of plaque stability. Therefore, in vivo quantification of cap thickness could yield valuable information for estimating the risk of plaque rupture. In the context of preoperative planning and perioperative decision making, intracoronary optical coherence tomography imaging can provide a very detailed characterization of the arterial wall structure. However, visual interpretation of the images is laborious, subject to variability, and therefore not always sufficiently reliable for immediate decision of treatment.

Methods: A novel semiautomatic segmentation method to quantify coronary fibrous cap thickness in optical coherence tomography is introduced. To cope with the most challenging issue when estimating cap thickness (namely the diffuse appearance of the anatomical abluminal interface to be detected), the proposed method is based on a robust dynamic programming framework using a geometrical a priori. To determine the optimal parameter settings, a training phase was conducted on 10 patients.

Results: Validated on a dataset of 179 images from 21 patients, the present framework could successfully extract the fibrous cap contours. When assessing minimal cap thickness, segmentation results from the proposed method were in good agreement with the reference tracings performed by a medical expert (mean absolute error and standard deviation of 22 ± 18 μm) and were similar to inter-observer reproducibility (21 ± 19 μm, R = .74), while being significantly faster and fully reproducible.

Conclusion: The proposed framework demonstrated promising performances and could potentially be used for online identification of high-risk plaques.

No MeSH data available.


Related in: MedlinePlus