A combined watershed segmentation approach using k-means clustering for mammograms

Identification and segmentation of mass are critical for medical image processing. In this paper a combined approach for image segmentation based on watershed transform and k-means clustering is proposed. A preprocessing step is applied to get an initial region of interest which is enhanced using adaptive histogram equalization. Watershed transform is applied to obtain an initial segmentation of the mammograms. Statistical texture features are also computed for the identified regions. K-Means clustering is then applied to produce foreground markers. These markers are given as input to a second phase of marker controlled watershed segmentation. With this, the unwanted regions are greatly reduced giving the suspicious mass region. The proposed approach is validated on a set of 50 mammograms from DDSM database. These mammograms are selected randomly from malign mass classified images. The identified regions are compared with the ground truth values marked in the database. Results show that the algorithm is more effective for mammogram image segmentation as compared to direct application of watershed segmentation approach.