Abstract
Magnetic resonance imaging (MRI) is an effective imaging modality foridentifying and localizing breast lesions in women. Accurate and precise lesionsegmentation using a computer-aided-diagnosis (CAD) system, is a crucial stepin evaluating tumor volume and in the quantification of tumor characteristics.However, this is a challenging task, since breast lesions have sophisticatedshape, topological structure, and high variance in their intensity distributionacross patients. In this paper, we propose a novel marker-controlled watershedtransformation-based approach, which uses the brightest pixels in a region ofinterest (determined by experts) as markers to overcome this challenge, andaccurately segment lesions in breast MRI. The proposed approach was evaluatedon 106 lesions, which includes 64 malignant and 42 benign cases. Segmentationresults were quantified by comparison with ground truth labels, using the Dicesimilarity coefficient (DSC) and Jaccard index (JI) metrics. The proposedmethod achieved an average Dice coefficient of 0.7808$\pm$0.1729 and Jaccardindex of 0.6704$\pm$0.2167. These results illustrate that the proposed methodshows promise for future work related to the segmentation and classification ofbenign and malignant breast lesions.