Segmentation Of Pectoral Muscle From Digital Mammograms With Breadth -First Search Algorithm Towards Breast Cancer Detection

Abstract

Digital mammography, a powerful tool for early breast cancer detection, can face challenges when the pectoral muscle is present in the breast area, making accurate classification of breast density and diagnosis difficult. This article introduces a robust methodology that effectively segments the pectoral muscle from digital mammographic images by applying the breadth-first search (BFS) algorithm with a heuristic approach. The proposed methodology includes preprocessing digital mammograms using the SAGCWD algorithm to enhance image quality. The BFS algorithm then detects the entire pectoral muscle as a single connected component and removes artifacts and tags from the background region. Finally, the BFS algorithm is used with and without a heuristic technique to eliminate the pectoral muscle from digital mammographic images. The proposed algorithm is subjected to rigorous testing on 2500 images from the DDSM dataset, 322 images from the MIAS dataset, and 194 images from the In-Breast dataset. Expert Radiologists help subjectively calculate the Segmentation Accuracy of the proposed algorithm while quantitative metrics like Jaccard Index and Dice Coefficient are useful in calculating the same objectively. The overall segmentation accuracy subjectively is reported to be an impressive 85.40%, with mean values of the Jaccard index and the Dice similarity coefficient at 0.96 and 0.97, respectively. These metrics highlight the high level of agreement between the algorithm's output and the ground truth, confirming its effective segmentation performance. The successful integration of the proposed algorithm into the preprocessing stage of digital mammograms underscores its potential to enhance the accuracy of pectoral muscle removal, a crucial step in breast cancer classification.