Toward breast cancer diagnosis based on automated segmentation of masses in mammograms

Toward breast cancer diagnosis based on automated segmentation of masses in mammograms,10.1016/j.patcog.2008.08.006,Pattern Recognition,Alfonso Rojas

Toward breast cancer diagnosis based on automated segmentation of masses in mammograms   (Citations: 12)
BibTex | RIS | RefWorks Download
This work explores the use of characterization features extracted based on breast-mass contours obtained by automated segmentation methods, for the classification of masses in mammograms according to their diagnosis (benign or malignant). Two sets of mass contours were obtained via two segmentation methods (a dynamic-programming-based method and a constrained region-growing method), and simplified versions of these contours (modeling the contours as ellipses) were employed to extract a set of six features designed for characterization of mass margins (contrast between foreground region and background region, coefficient of variation of edge strength, two measures of the fuzziness of mass margins, a measure of spiculation based on relative gradient orientation, and a measure of spiculation based on edge-signature information). Three popular classifiers (Bayesian classifier, Fisher's linear discriminant, and a support vector machine) were then used to predict the diagnosis of a set of 349 masses based on each of said features and some combinations of these. The systems (each system consists of a segmentation method, a featureset, and a classifier) were compared with each other in terms of their performance on the diagnosis of the set of breast masses. It was found that, although there was a percent difference of about 14% in the average segmentation quality between methods, this was translated into an average percent difference of only 4% in the classification performance. It was also observed that the spiculation feature based on edge-signature information was distinctly better than the rest of the features, although it is not very robust to changes in the quality of the segmentation. All systems were more efficient in predicting the diagnosis of benign masses than that of the malignant masses, resulting in low sensitivity and high specificity values (e.g. 0.6 and 0.8, respectively) since the positive class in the classification experiments is the set of malignant masses. It was concluded that features extracted from automated contours can contribute to the diagnosis of breast masses in screening programs by correctly identifying a majority of benign masses.
Journal: Pattern Recognition - PR , vol. 42, no. 6, pp. 1138-1148, 2009
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
Sort by: