Deep learning based breast cancer detection and classification using fuzzy merging techniques

Abstract

Automatic identification of abnormal and normal cells is a critical step in computer-assisted pathology, owing to certain heterogeneous characteristics of cancer cells. However, automated nuclei detection is problematic in unevenly shaped, overlapping and touching nuclei. It is, consequently, essential to detect single and overlapping nuclei and distinguish them from single ones for a reasonable quantitative analysis. Diagnosis is improved by introducing a computer-aided diagnosis system to automatically detect breast cancer tissue nuclei from whole slide images of hematoxylin and eosin stains. We propose a method for the automatic cell nuclei detection, segmentation, and classification of breast cancer using a deep convolutional neural network (Deep-CNN) approach. The main contribution of this work is the detection of nuclei using anisotropic diffusion in a filter and applying a novel multilevel saliency nuclei detection model in ductal carcinoma of breast cancer tissue. The detected nuclei are classified into benign and malignant cells by applying the new Deep-CNN model. Finally, the novel multilevel saliency nuclei detection technique is integrated with the Deep-CNN to produce an nMSDeep-CNN model that turns out to be the most accurate results with very less computation time. The accuracy, sensitivity and specificity of the proposed system are 98.62%, 0.947 and 0.964, respectively. The classification for benign and malignant cells is evaluated by applying 10 fold cross-validation. Thus, the system can be clinically used for an objective, accurate, and rapid diagnosis of abnormal tissue. The effectiveness of the suggested framework is demonstrated through experiments on several datasets.