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.
Eigendetection of masses considering false positive reduction and breast density information
Rent:
Rent this article for
USD
10.1118/1.2897950
/content/aapm/journal/medphys/35/5/10.1118/1.2897950
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/35/5/10.1118/1.2897950

Figures

Image of FIG. 1.
FIG. 1.

Proposed CAD system for mammographic mass detection, including false positive reduction.

Image of FIG. 2.
FIG. 2.

Proposed CAD system for mammographic mass detection including false positive reduction and breast density information.

Image of FIG. 3.
FIG. 3.

The upper row shows four ROIs corresponding to manually detected masses of similar size, while the lower row shows the corresponding eigenmasses with the greatest eigenvalues obtained using the cluster of masses of the same size.

Image of FIG. 4.
FIG. 4.

The probabilistic templates obtained by clustering the dataset in four clusters; with from left to right increasing sizes. Brighter pixels represent a higher probability of a mass boundary.

Image of FIG. 5.
FIG. 5.

Intensity profile of the templates as a function of the distance to the center (in pixels).

Image of FIG. 6.
FIG. 6.

Center and size of the suspicious regions found in three mammograms of the MIAS database with clear masses (marked in white). These results are obtained thresholding the posterior probability at the same threshold. Note that the number of false positive obtained is different in each image, but also note that the masses have been identified.

Image of FIG. 7.
FIG. 7.

Center and size of the suspicious regions found in the same mammograms of Fig. 6. Note that almost all false positive regions have been removed.

Image of FIG. 8.
FIG. 8.

FROC analysis of the algorithm using the MIAS database for both training and testing and compared with algorithms and .

Image of FIG. 9.
FIG. 9.

FROC analysis of the algorithm (including false positive reduction) using the MIAS database for both training and testing. The results are separated according to the size of the lesions.

Image of FIG. 10.
FIG. 10.

FROC analysis of the proposed algorithms using the DDSM database for training and the MIAS database for testing the system.

Image of FIG. 11.
FIG. 11.

FROC analysis of the proposed algorithms using the DDSM database for training and the MIAS database for testing the system. Comparison with algorithms and .

Image of FIG. 12.
FIG. 12.

Statistical comparison of the algorithms performance using a box-plot representation: the central line indicates the median value, the boxes represent the boundaries for the first and third quartiles, while the dashed lines indicate the 1.5 standard deviation boundary, and outlier values are shown as plus signs.

Image of FIG. 13.
FIG. 13.

FROC analysis of the proposed algorithms (Eig) using the DDSM database for training and the MIAS database for testing the system. DDSM1 uses a mass size distribution of four clusters (the same as used for the MIAS data) and DDSM2 uses six clusters (more evenly distributed according to the DDSM database).

Tables

Generic image for table
TABLE I.

MIAS set of mammograms (and ROIs in parenthesis). Details for breast density (rows) and size lesion (columns). , , , and . B-I: BIRADS I, B-II: BIRADS II, B-III: BIRADS III, and B-IV: BIRADS IV.

Generic image for table
TABLE II.

First DDSM training set of mammograms, with the size clustered as in the MIAS database. Details for breast density (rows) and size lesion (columns). , , , and . B-I: BIRADS I, B-II: BIRADS II, B-III: BIRADS III, and B-IV: BIRADS IV.

Generic image for table
TABLE III.

Second DDSM training set of mammograms, representing a more general distribution of mass sizes for the database. Details for breast density (rows) and size lesion (columns). , , , , , and . B-I: BIRADS I, B-II: BIRADS II, B-III: BIRADS III, and B-IV: BIRADS IV.

Generic image for table
TABLE IV.

Influence of the lesion size (in ) for algorithms , and our proposal . The subscripts MIAS and DDSM refer to the training database, and to the training taking the breast density information into account. The results show the mean and the standard deviation of values, all obtained using the MIAS database for testing. The last column shows the false positives per image at a sensitivity of 0.8.

Generic image for table
TABLE V.

Influence of the number of training images for each size cluster. Respectively, 5, 10, 15, 20, and all the images have been used to create the templates.

Generic image for table
TABLE VI.

Computational timing for training and application. FPRed and BDI refers to taking false positive reduction and breast density information into account. All times are in seconds. The set of values in brackets are timings for the individual BIRADS classes within the BDI modeling.

Loading

Article metrics loading...

/content/aapm/journal/medphys/35/5/10.1118/1.2897950
2008-04-16
2014-04-16
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: Eigendetection of masses considering false positive reduction and breast density information
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/35/5/10.1118/1.2897950
10.1118/1.2897950
SEARCH_EXPAND_ITEM