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.
Segmentation of ultrasonic breast tumors based on homogeneous patch
Rent:
Rent this article for
USD
10.1118/1.4718565
/content/aapm/journal/medphys/39/6/10.1118/1.4718565
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/39/6/10.1118/1.4718565

Figures

Image of FIG. 1.
FIG. 1.

Flowchart of the HP-NCut algorithm.

Image of FIG. 2.
FIG. 2.

The illustration of filter bank and texton map (a) first to sixth rows: even and odd filter bank consisting of one scale and six orientations; seventh row: center-surround filter f DoG. (b) The original image. (c) The texton map derived from a universal dictionary. (d) The texton map created using the proposed approach. (e) Zoom-in view of the region inside the white square outlined in (d). The ground truth tumor boundary indicated in dark color (green in online version), is overlaid to all texton maps.

Image of FIG. 3.
FIG. 3.

An illustration of partitioning neighborhood.

Image of FIG. 4.
FIG. 4.

Boundary map comparison of different methods. (a) Original image. (b) Boundary map obtained by Sobel operator. (c) Boundary map obtained using the texton histograms. (d) Boundary map obtained by the proposed approach. The ground truth tumor boundary indicated in light color (green in online version), is overlaid to all boundary maps. Three dark color (red in online version) rectangles are drawn to indicate the differences among the detected boundary maps.

Image of FIG. 5.
FIG. 5.

Examples of HPs on a breast US image. Eight search windows located in smooth region (outlined with light color (yellow in online version) squares) or nonsmooth region (outlined with dark color (blue in online version) squares) are listed in a test image. The specified area inside each search window is zoomed and displayed above the images. The corresponding HPs are displayed on the right and shown in color-coding. Along with the increase of the membership, the color varies from blue to darker red as shown by the color bar displayed at the top of the figure (available in online version).

Image of FIG. 6.
FIG. 6.

An illustration of the major challenge of using textures for successful segmentation of tumor from breast US images. (a) The two white arrows indicate two points located in the tumor and normal tissue regions, respectively. The white circles represent their round neighborhoods. (b) The HP of point p. (c) The HP of point q.

Image of FIG. 7.
FIG. 7.

For the points p and q in Fig. 6, two texton histograms were computed over: (a) the fixed round neighborhoods and (b) the HPs.

Image of FIG. 8.
FIG. 8.

The segmentation of a breast US image with irregular shape and posterior acoustic shadowing. (a) The original image with a rectangular ROI placed by user. (b) Manual segmentation. The segmentation results using (c) HP-NCut, (d) Interactive-NCut, and (e) PBLS after 380 iterations. (f) The edge detection result of PBLS (the intensity value ranges from 0 to 1, with the values close to one indicating an edge pixel and the values close to zero indicating a background pixel).

Image of FIG. 9.
FIG. 9.

The segmentation of a breast tumor with similarity to the surrounding normal tissue. (a) The original image with a rectangular ROI placed by user. (b) Manual segmentation. The segmentation results using (c) HP-NCut, (d) Interactive-NCut, and (e) PBLS after 240 iterations.

Image of FIG. 10.
FIG. 10.

The segmentation of a breast tumor with regular shape. (a) The original image with a rectangular ROI placed by user. (b) The manual segmentation. The segmentation results using (c) HP-NCut, (d) Interactive-NCut, and (e) PBLS after 300 iterations.

Image of FIG. 11.
FIG. 11.

The worst segmentation result. (a) The original image with a rectangular ROI placed by the user. (b) Manual segmentation. The segmentation results using (c) HP-NCut, (d) Interactive-NCut, and (e) PBLS.

Image of FIG. 12.
FIG. 12.

Examples of the boundary-detection function using three different radius values. From left to right r 1 = {1%, 2%, 3%} of the image diagonal.

Image of FIG. 13.
FIG. 13.

(a) Original images. (b) The boundary maps. (c) The histograms of the boundary energies over the entire image domain.

Image of FIG. 14.
FIG. 14.

Influence of the energy threshold τ for HP construction. The areas enclosed by light color (yellow in online version) contours represent HPs in the smooth regions, whereas those enclosed by dark color (blue in online version) contours are HPs in the nonsmooth regions. Gray (red in online version) dots are the central pixels. From left to right threshold τ = {0.86, 0.76, 0.66} for row 1 and τ = {0.95, 0.9, 0.8} for row 2. The original images can be seen in Fig. 9(a) in row 1 and Fig. 2(a) in row 2.

Image of FIG. 15.
FIG. 15.

The average minimum Euclidean error in the two methods after repeating 50 experiments with different sampling points selected randomly.

Image of FIG. 16.
FIG. 16.

Ten segmentation results (a)–(j) on a breast US image using Interactive-NCut with different sampling points selected randomly.

Image of FIG. 17.
FIG. 17.

Ten segmentation results (a)–(j) on a breast US image using HP-NCut with different sampling points selected randomly.

Tables

Generic image for table
TABLE I.

The pathological type of breast tumors.

Generic image for table
TABLE II.

The summary of parameter setting of HP-NCut.

Generic image for table
TABLE III.

Distance error metrics of three different image segmentation methods.

Generic image for table
TABLE IV.

Overlapping area error metric of three different image segmentation methods.

Generic image for table
TABLE V.

Average run-time of three different image segmentation methods.

Loading

Article metrics loading...

/content/aapm/journal/medphys/39/6/10.1118/1.4718565
2012-05-22
2014-04-20
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: Segmentation of ultrasonic breast tumors based on homogeneous patch
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/39/6/10.1118/1.4718565
10.1118/1.4718565
SEARCH_EXPAND_ITEM