Improved Segmentation of Suspicious Regions of Masses in Mammograms by Watershed Transform

摘要

Screening mammograms are powerful aids in early detection of breast cancer. This work deals with segmentation of suspicious region of anomalies known as masses in mammogram. The proposed method is based on watershed transform of morphologically reconstructed image. The suspicious mass regions are isolated and the results are compared with two current methods: thresholding of graph based saliency map of preprocessed mammogram and morphological extraction from saliency. Quantitative analysis and comparison of results are performed using similarity measures and classifier performance and found that the proposed method gives better results achieving 0.95 similarity measure and 83% of ROC area.