Breast Mass Segmentation in Mammography Using Plane Fitting and Dynamic Programming
Received 12 October 2008; accepted 25 November 2008. published online 13 April 2009.
Rationale and Objectives
Segmentation is an important and challenging task in a computer-aided diagnosis (CAD) system. Accurate segmentation could improve the accuracy in lesion detection and characterization. The objective of this study is to develop and test a new segmentation method that aims at improving the performance level of breast mass segmentation in mammography, which could be used to provide accurate features for classification.
Materials and Methods
This automated segmentation method consists of two main steps and combines the edge gradient, the pixel intensity, as well as the shape characteristics of the lesions to achieve good segmentation results. First, a plane fitting method was applied to a background-trend corrected region-of-interest (ROI) of a mass to obtain the edge candidate points. Second, dynamic programming technique was used to find the “optimal” contour of the mass from the edge candidate points. Area-based similarity measures based on the radiologist's manually marked annotation and the segmented region were employed as criteria to evaluate the performance level of the segmentation method. With the evaluation criteria, the new method was compared with 1) the dynamic programming method developed by Timp and Karssemeijer, and 2) the normalized cut segmentation method, based on 337 ROIs extracted from a publicly available image database.
Results
The experimental results indicate that our segmentation method can achieve a higher performance level than the other two methods, and the improvements in segmentation performance level were statistically significant. For instance, the mean overlap percentage for the new algorithm was 0.71, whereas those for Timp's dynamic programming method and the normalized cut segmentation method were 0.63 (P < .001) and 0.61 (P < .001), respectively.
Conclusions
We developed a new segmentation method by use of plane fitting and dynamic programming, which achieved a relatively high performance level. The new segmentation method would be useful for improving the accuracy of computerized detection and classification of breast cancer in mammography.
Center for Biomedical Imaging and Bioinformatics, School of Computer Science and Technology (E.S., L.J., R.J.) and the Department of Surgery, Tongji Hospital, Tongji Medical College (L.Z., Y.Y.), Huazhong University of Science and Technology, Luo Yu Road No. 1037, Wuhan 430074, China; Department of Radiology, Duke University Medical Center, (Q.L.)
Address correspondence to: E.S.
This research was supported by the National 863 High Technology Research and Development Program of China (grant no. 2006AA02Z347).
ABSTRACT