| مستخلص: |
Automated detection of breast boundary is an important and fundamental step involved in the identification and classification of benign and malignant tissues of the breast from analysis of Digital Mammograms. In this paper, Preprocessing of the mammogram dataset is presented in detail to obtain a processed mammogram image with clear demarcation between breast and background. Most of the breast cancer detection and location identification algorithms developed for diagnosis from digital mammograms start with an input image, properly enclosing only the breast region, not covering the pectoral muscle, and a clear background image with no artifacts. In practice, digital mammograms, used by radiologists have a background marked by varying intensity levels and have labels to indicate details of the patient's id and breast. Removal of the scanning artifacts, labels, and varying background intensity presents a complex problem in presenting a clear demarcation between breast and background image, drawing a clear boundary on the Skin air interface. This paper approaches the problem of extracting skin-air interface employing size and rotation-invariant template matching to remove labels, and an advanced lee filter for suppressing speckle noise, while preserving texture information to obtain mammogram image, cleared of scanning artifacts. Then, the Local Binary Pattern method is applied to compute cross-correlation analysis. Then, the resultant image is given to the Support Vector Machine (SVM) classifier. SVM is employed to classify breast and background. From the classified image, the breast boundary is extracted and overlaid on the mammogram image. Results were compared with actual breast boundary using the AMinDist procedure and are found to be 96% accurate with an error distance of fewer than 2 pixels. [ABSTRACT FROM AUTHOR] |