A Computer-Aided Diagnosis System of Breast Intraductal Lesion Using Histopathological Images

The diagnosis of intraductal proliferation poses a big challenge to pathologists or clinical surgeons in examinations of breast carcinoma. Clinically, there are three types of intraductal breast lesions including the usual ductal hyperplasia (UDH), atypical hyperplasia (ADH), and ductal carcinoma in situ (DCIS) for this examination. This study aims to analyze H&E stained breast biopsy samples in order to further validate the classes of breast intraductal hyperplasia using computer aided diagnosis so that unnecessary surgeries or under treatments leading to invasive carcinoma can be avoided. To this end, our evaluation will emphasize on the classification of action or non-action type using intraductal images such as DCIS or UDH. The entire system was demonstrated on these stained images using a series of processing operations in order to detect intraductal regions and nuclei from a tissue specimen. These automatic processing operations contain color space transform, nucleus segmentation, watershed operation, feature extraction and dimensionality reduction were performed. Classification based on support vector machine was further performed to identify the lesion types of the breast lesions. Experimental result demonstrates that the performance measure based on the ROC curve reaches 0.894 for the testing dataset.