Pectoral muscle line detection in mammogram images

Abstract

Mammography is an important instrument in national breast screening programs and helps in early detection of breast cancer. Mammography is an imaging tool which is used to examine a patient’s breast by using low doses of X-rays. In Medio-Lateral Oblique (MLO) view of a mammogram, the pectoral muscle may affect the outcome of the mammography test results, due to the similarity between the pectoral muscle and any abnormal tissues close to the muscle. Pre-processing of the MLO image can provide further tools to aid diagnostics. In this work we apply image processing techniques to detect the edge of the pectoral muscle which can then be used as a reference in subsequent screening, thus helping in detecting breast cancer. The detection of the pectoral muscle is achieved by first selecting the Region of Interest (ROI) from the whole image which is then passed through a median filter, to smoothen the image. Edge detection is then applied and the edges found are passed through the random sample consensus (RANSAC) algorithm, which chooses the best fitting lines. The line with the steepest gradient is chosen and plotted on the ROI filtered image. The edges of that specific line are re-arranged in order to be divided into two, three and four clusters. Furthermore, these clusters are used to achieve an improved estimated line of the pectoral muscle. The RANSAC algorithm is executed on each cluster and the same process of choosing the steepest gradient is repeated. The lines generated are stitched together to form one continuous line. Finally, the estimated line for each cluster is stored and compared to the ground truth. The mean error and standard derivation of each line is calculated and a plot of accuracy vs. time is achieved.