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.
A ranklet-based image representation for mass classification in digital mammograms
Rent:
Rent this article for
USD
10.1118/1.2351953
/content/aapm/journal/medphys/33/10/10.1118/1.2351953
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/33/10/10.1118/1.2351953

Figures

Image of FIG. 1.
FIG. 1.

Different approaches to computer-aided mass detection. The traditional approach based on (1) ROIs encoding by means of a specific image representation, (2) feature extraction from ROIs as they appear in the image representation domain, and (3) classification is shown in (a). The alternative approach adopted in this work and based on the direct classification of encoded ROIs is shown in (b).

Image of FIG. 2.
FIG. 2.

The three Haar wavelet supports , , and . From left to right, the vertical, horizontal, and diagonal Haar wavelet supports.

Image of FIG. 3.
FIG. 3.

Ranklet transform applied to two synthetic images. The application of the vertical Haar wavelet support results in and for the synthetic images shown in (a) and (b), respectively. Conversely, due to symmetry reasons, the application of the horizontal and diagonal Haar wavelet supports results in and , regardless of the synthetic image analyzed.

Image of FIG. 4.
FIG. 4.

Multiresolution ranklet transform of an image with pixel size at resolutions {8, 4, 2}, namely, using Haar wavelet supports with pixel size , , and .

Image of FIG. 5.
FIG. 5.

ROC curves obtained varying SVM kernels. ( values of ) uses a linear SVM kernel. and use a polynomial SVM kernel with degree 2 and 3, respectively. The difference in the average value of or and that of achieves statistical significance with two-tailed value .

Image of FIG. 6.
FIG. 6.

ROC curves obtained varying ranklet resolutions. ( values of ) and account for low, intermediate, and high ranklet resolutions. for low and high ranklet resolutions. for low ranklet resolutions. The difference in the average value of , or and that of achieves statistical significance with two-tailed value .

Image of FIG. 7.
FIG. 7.

ROC curves obtained applying histogram equalization. ( values of ) deals with crops submitted to histogram equalization before the ranklet transform is performed. The difference in its average value and that of does not achieve statistical significance.

Image of FIG. 8.
FIG. 8.

ROC curves comparison. ( value of ) is one of the best ranklet image representations evaluated. PixHRS , DwtHS3 , and OwtS2 are the best pixel, discrete wavelet, and overcomplete wavelet image representations previously developed and evaluated (Ref. 20). The difference in the value of and that of PixHRS does not achieve statistical significance. Conversely, the difference in the value between and DwtHS3 or OwtS2 is statistically relevant with two-tailed value .

Image of FIG. 9.
FIG. 9.

Gray-level intensity histograms of two crops extracted, as an example, from two different images. Digitized DDSM radiographic image (a). Digital FFDM radiographic image (b).

Image of FIG. 10.
FIG. 10.

Ranklet image representation and overcomplete wavelet image representation in action. SVM classifier tested on (a) and (b). OwtS2 SVM classifier tested on (c) and (d). The small square marks represent the radiologist’s interpretation, whereas the large ones represent the automatic CAD’s analysis.

Image of FIG. 11.
FIG. 11.

FROC curves. SVM classifier tested on and . OwtS2 SVM classifier tested on and .

Tables

Generic image for table
TABLE I.

Multiresolution ranklet transform of an image with pixel size . Number of resulting ranklet coefficients for each different combination of resolutions.

Generic image for table
TABLE II.

Results summary. Ranklet image representations: average values, FPF values at 90% sensitivity, and calculation times for the analysis of an entire radiographic image.

Generic image for table
TABLE III.

Results comparison. Best performing ranklet, pixel, discrete wavelet, and overcomplete wavelet image representations: average values, FPF values of 90% sensitivity, and calculation times for the analysis of an entire radiographic image.

Loading

Article metrics loading...

/content/aapm/journal/medphys/33/10/10.1118/1.2351953
2006-09-27
2014-04-24
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: A ranklet-based image representation for mass classification in digital mammograms
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/33/10/10.1118/1.2351953
10.1118/1.2351953
SEARCH_EXPAND_ITEM