Automated Segmentation of Breast Skin for Early Cancer Diagnosis: A Multi-otsu Region Growing Approach for Detecting Skin Thickness Variations

Accurate segmentation of breast skin in mammograms is of great interest for computer-aided diagnosis systems aiming for early cancer detection, as it can highlight contralateral asymmetries linked to the development of different types of carcinomas. We present an automated skin segmentation algorithm based on Otsu multi-thresholding. It was evaluated on a dataset of 102 pairs of mammograms from females of ages ranging from 30 to 74 years. We then proposed two novel characteristics for asymmetry assessment: mean skin thickness, which resulted in 0.88 pixels long, with a standard deviation of 1.16 pixels; and skin area, with an average of 285.79 pixels and a standard deviation of 399.35 pixels. Under these considerations, the algorithm identified 26 out of 102 cases (25.49%) as exhibiting asymmetry in skin thickness, and 24 out of 102 cases (23.52%) as displaying variations in skin area. These results demonstrate the effectiveness of the proposed algorithm in accurately segmenting breast skin and detecting potential asymmetries. Our advanced skin segmentation method enhances breast imaging for cancer detection. Analyzing BI-RADS correlation with skin thickness reveals significant asymmetries, aiding early diagnosis. Our innovative approach outperforms previous techniques, although sample size and subjectivity warrant consideration.