Detection of lobular carcinoma in situ(LCIS) by image analysis

Abstract

In this study, we aimed to develop a quantitative image analysis method that may enhance the detection of the lobular carcinoma in-situ (LCIS) in breast cancer specimens. Glandular areas were segmented by using mathematical morphology from 5X histologic images of breast cancer cases (n=213). Computational features such as shape, intensity, and texture were extracted from glandular areas. Segmented glandular areas of LCIS were significantly larger, thicker, lower and less variant in intensity, compared to normal areas (Mann-Whitney test, p<0.01). Our models based on data mining algorithms detected LCIS frames at the accuracy rate close to 99%. Our proposed methods may be well incorporated into a further development of computer aided detection (CAD) software for the screening of whole slide images to locate the LCIS areas.