Segmentation and Enhancement of Mammograms for the Detection of Cancer Using Gradient Weight Map and Decorrelation Stretch

Abstract

To deal with computer aided detection of breast cancer in mammograms, an efficient automated preprocessing stage is the most vital step which assists radiologist’s decision. The approach proposes a model for the segmentation and enhancement of the breast region. The proposed method includes gradient weight map followed by region property-based extraction and morphological operations for background suppression and artifacts removal. Intensity adjustment and Otsu’s thresholding methods were adopted for breast region extraction. Finally, enhancement of the breast region is accomplished by Contrast-Limited Adaptive Histogram Equalization (CLAHE) and Decorrelation stretch. Experimentation conducted on mini Mammographic Image Analysis Society (miniMIAS) dataset shows segmentation accuracy of about 97.64%. Enhancement contributes with better feature discrimination with high Peak Signal Noise Ratio (PSNR) and low Root Mean Square Error (RMSE).