br ROI in mammogram image is segmented
ROI in mammogram image is segmented into maximum possi-ble number of non-overlapping small squared shape region of fixed size to acquire a large dataset for the further studies. A typical mammogram classification system generally consists of three sequential steps: (1) Extraction of region of interest, (2) features extraction from selected ROI, and (3) classification of mammogram based on extracted features. background noise. Intensity is the parameter measured here. Pre-processing is done because of the low Taxol of mammographic images that is hard to interrupt masses in the mammograms. Generally there is no much variation in the intensities of pec-toral muscle when compared to the tumor intensity. So we should remove pectoral muscle region before feature extraction. Prepro-cessing stage is very much used to remove the labels and back-ground noise in the digital mammograms. After removing all unwanted labels and noise in the mammogram Hough transform is applied to this processed image which is similar to random transform. It is mainly used to detect arbitrary shapes and straight lines. Hough transform is tolerant of gaps noise and occlusion in the mammograms. The feature should be extracted effectively and it should be separated. The Fig. 1 shows the proposed block diagram Figs. 2–9.
This section explains the techniques used to classify digital mammogram which has the process like,
Image acquisition Pre-processing
Feature extraction using Hough transform Classification using SVM
In the early stage mammogram images were collected and pro-cessed below.
Feature extraction using Hough
3. Proposed methodology Normal Abnormal
The mammograms are preprocessed in the early stage and this
is increase the difference between needed objects and unwanted Fig. 1. Proposed method.
Fig. 2. Elements of mammogram image.
Fig. 3. Label removed binary image – using gradient based threshold.
Fig. 4. Label removed mammogram image.
Fig. 5. Mammogram image – estimation maximization applied.
Fig. 6. Pre-processed mammogram image without edge detection.
Fig. 7. Pre-processed mammogram image with canny edge detection.
Mammogram images of type normal, benign and malignant for fatty-glandular breast types are collected form the MIAS database-mammographic image analysis database society. It has mammo-gram images digitized at 50 mm pixel edge. It has been reduced and clipped so that every image is of 1024 * 1024 pixels. There are 322 mammograms images from pseudocoelomates 95 images taken for this work. Cancer detection in the dense breast cancer is difficult to find thorough mammogram. So dense images was not taken in this.
The images are collected from database will have unwanted information and background noise. Pre-processing stage is mainly
Fig. 8. Mammogram image with edge detection.
Fig. 9. Wrongly classified image after maximization estimation.
used to remove these from the mammogram and make image more suitable for further process. The figure below represents the unwanted region of breast cancer and tumor.
The region in yellow is the breast region, unwanted label and background is represented using the blue circle and red circle – which represents as tumor. The unwanted label should be removed first by using gradient based threshold method. Several morpho-logical operations was carried out to generate a mask. Dilation and hole-filling are the main operations used here and by utilizing some reference we generated a binary edge map of the image using gradient based threshold method.
Here GT is a gradient threshold which is found using Otsu’s adaptive method. After this the binary image is dilated using dia-mond structuring element. Now this mask is multiplied with orig-inal image. The below two images shows mask generated through gradient method and other is label removed image. The main intention of the dilation operation in the binary image is to bring