Automated Detection and Classification of Breast Cancer Nuclei with Deep Convolutional Neural Network

Abstract

Heterogeneous regions present in tissue with respect to cancer cells are of various types. This study aimed to analyze and classify the morphological features of the nucleus and cytoplasm regions of tumor cells. This tissue morphology study was established through invasive ductal breast cancer histopathology images accessed from the Databiox public dataset. Automatic detection and classification was carried out by means of the computer analytical tool of deep learning algorithm. Residual blocks with short skip were employed with hidden layers of preserved spatial information. A ResNet-based convolutional neural network was adapted to perform end-to-end segmentation of breast cancer nuclei. Nuclei regions were identified through color and tubular structure morphological features. Based on the segmented and extracted images, classification of benign and malignant breast cancer cells was done to identify tumors. The results indicated that the proposed method could successfully segment and classify breast tumors with an average Dice score of 90.68%, sensitivity = 98.64, specificity = 98.68, and accuracy = 98.82.