1887
banner image
No data available.
Please log in to see this content.
You have no subscription access to this content.
No metrics data to plot.
The attempt to load metrics for this article has failed.
The attempt to plot a graph for these metrics has failed.
Three-dimensional segmentation of three-dimensional ultrasound carotid atherosclerosis using sparse field level sets
Rent:
Rent this article for
USD
10.1118/1.4800797
/content/aapm/journal/medphys/40/5/10.1118/1.4800797
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/40/5/10.1118/1.4800797

Figures

Image of FIG. 1.
FIG. 1.

(a) An example 3D carotid US image. (b) 3D carotid US image with overlaid user-drawn manual contours of the MAB and LIB. Lumen is enclosed within the LIB and vessel wall is enclosed in between the MAB and LIB. (c) Intensity probability density functions (PDF) for lumen, wall, and background regions for the example 3DUS image. Note that the intensity PDFs have high overlap with each other.

Image of FIG. 2.
FIG. 2.

Two example transverse views of the CCA with overlaid manual segmentations of the MAB and LIB from 3DUS images of two subjects with carotid stenosis of more than 60%. Note that although the MAB has a relatively low order convex shape, the LIB has a nonconvex shape.

Image of FIG. 3.
FIG. 3.

Block diagram of the workflow of the algorithm.

Image of FIG. 4.
FIG. 4.

The process of creating a 3D mask from anchor points as the initial surface for the algorithm with an ISD of 2 mm.

Image of FIG. 5.
FIG. 5.

3D image preprocessing steps of the algorithm. The results here are shown for a single slice of the 3DUS image.

Image of FIG. 6.
FIG. 6.

2D slice-by-slice comparisons of algorithm segmentations to manual segmentations for a subject with a mild stenosis. Results for ISD from 1 to 4 mm, and 10 mm are shown. The contours are as follows: Continuous yellow contour-–mean manual MAB and LIB, dashed purple contour-–mean algorithm MAB and LIB, and cyan dashed contour-–one round of algorithm MAB and LIB. Each row corresponds to the distance from the bifurcation (BF) and each column corresponds to the ISD used for initialization.

Image of FIG. 7.
FIG. 7.

2D slice-by-slice comparisons of algorithm segmentations to manual segmentations for a subject with a moderate stenosis (stenosis is between 30% and 70%). Results for ISD from 1 to 4 mm and 10 mm are shown. The accuracy dropped at 4 and 10 mm. The contours are as follows: Continuous yellow contour-–mean manual MAB and LIB, dashed purple contour—mean algorithm MAB and LIB, and cyan dashed contour-–one round of algorithm MAB and LIB. Each row corresponds to the distance from the BF and each column corresponds to the ISD used for initialization.

Image of FIG. 8.
FIG. 8.

Comparison of the MAB and LIB algorithm segmentations to manual segmentations for an ISD of 3 mm for two example 3DUS images that were used for Figs. 6 and 7 . The algorithm-generated surfaces are denoted by the letter ‘A’ and manually generated surfaces are denoted by the letter ‘M’. (a) LIB surface comparison with manual segmentation for a subject with a mild stenosis. (b) MAB surface comparison with manual segmentation of the same subject. (c) LIB surface comparison with manual segmentation for a subject with a moderate stenosis. (d) MAB surface comparison with manual segmentation of the same subject.

Image of FIG. 9.
FIG. 9.

Algorithm and manually generated flattened VWT maps of the surfaces shown in Fig. 8 for the same two example 3DUS images. The first row corresponds to algorithm-generated flattened VWT maps, whereas second row corresponds to manually generated flattened VWT maps. (a) Subject with a mild stenosis. (b) Subject with a moderate stenosis.

Image of FIG. 10.
FIG. 10.

Bland-Altman plot (Ref. ) for comparing algorithm- and manually generated 3DUS VWV, where ISD of 3 mm is used for the algorithm initialization. The red continuous line labeled as mean indicates the bias, the blue dotted lines labeled as 1.96 SD indicate the level of agreement, and the red dashed lines indicate the 95% CI.

Image of FIG. 11.
FIG. 11.

Correlation plot for algorithm- and manually generated 3DUS VWV, where ISD of 3 mm is used for the algorithm initialization. The dashed lines indicate the 95% CI of the best fit line.

Tables

Generic image for table
TABLE I.

Previous papers describing carotid LIB and/or MAB segmentations from 3DUS/B-mode images.

Generic image for table
TABLE II.

Parameters and their optimized values for the preprocessing.

Generic image for table
TABLE III.

Parameters and their optimized values for the MAB and LIB segmentations.

Generic image for table
TABLE IV.

Results for the MAB and LIB segmentation using region-based and distance-based metrics for the 21 3DUS images using the average boundaries. The results of the 3D algorithm are given for ISD of 1, 2, 3, 4, and 10 mm.

Generic image for table
TABLE V.

Results for the VWV, MAB, and LIB using volume-based metrics for the 21 3DUS images. The results of the 3D algorithm are given for ISD of 1, 2, 3, 4, and 10 mm.

Generic image for table
TABLE VI.

Comparison of algorithm- and manually generated VWV for 21 3DUS images using statistical testing and Pearson r. The results of the 3D algorithm are given for ISD of 1, 2, 3, 4, and 10 mm.

Generic image for table
TABLE VII.

Standard deviation (SD), coefficient of variation (CV), and minimum detectable difference (MDD) of volume measurements for 21 3DUS images computed using the repeated measurements of algorithm and manual segmentations. The results of the 3D algorithm are given for ISD of 1, 2, 3, 4, and 10 mm.

Loading

Article metrics loading...

/content/aapm/journal/medphys/40/5/10.1118/1.4800797
2013-04-17
2014-04-19
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: Three-dimensional segmentation of three-dimensional ultrasound carotid atherosclerosis using sparse field level sets
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/40/5/10.1118/1.4800797
10.1118/1.4800797
SEARCH_EXPAND_ITEM