Pectoral and Breast Segmentation Technique Based on Texture Information

Khamsa Djaroudib , Pascal Lorenz , Abdelmalik Taleb-Ahmed (1) , Abdelmadjid Zidani

1 LAMIH - Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201

Pectoral and breast segmentation is necessary and cumbersome step for the Computer Aided Diagnosis systems (CAD). This paper presents new pectoral and breast segmentation technique based on texture information from Gray Level Co-occurrence Matrix (GLCM). It showed good results to solve certain problems not yet resolved until presents, such as the presence of anomaly of mass or microcalcification in the pectoral borders, omitted in breast segmentation step, and the confusion between the pectoral line and the pectoral border. First, we applied smoothing and enhancing techniques to enhance breast image. Second, we compute textural images representing statistics parameters from GLCM in any pixel of the breast image, to detect breast and pectoral borders. These techniques have been applied to the MIAS database, consisting of MLO mammograms. The results were evaluated by expert radiologists and are promising, compared to other related works. © Springer International Publishing Switzerland 2015.

Mots clés: Breast; Gray level co-occurrence matrices (GLCM); Mammography; Pectoral muscle; Segmentation

Domaines: Sciences de l'ingénieur [physics]

Khamsa Djaroudib, Pascal Lorenz, Abdelmalik Taleb-Ahmed, Abdelmadjid Zidani. Pectoral and Breast Segmentation Technique Based on Texture Information. Lecture Notes in Computational Vision and Biomechanics, 21, pp.219-228, 2015, Lecture Notes in Computational Vision and Biomechanics, ⟨10.1007/978-3-319-15799-3_16⟩. ⟨hal-03659717⟩