Breast Cancer Diagnosis from Histopathology Images using Supervised Algorithms

Abstract

Breast cancer is the most common cancer type worldwide. In cancer studies, histopathological breast images are used in the process of diagnosis. In this paper, we defined three sets of features to represent the characteristics of the cell nuclei to detect malignant cases. Geometric, directional, and intensity-based features, a total of 33, are derived and evaluated using breast cancer histopathological images from the BreaKHis database. Four machine learning algorithms, including Decision Tree, Support Vector Machines, K-Nearest Neighbor, and Narrow Neural Networks (NNN), are designed to assess the efficiency of the sets. The preliminary results showed that the proposed methodology achieved high performance in classifying cancerous cells as the directional feature set was the most effective set among the three sets. The combination of the sets achieved the best performance by the NNN, which reached an accuracy, recall, precision, AUC, and F1 score of 96.9%, 97.4%, 98%, 98.8%, and 97.7%, respectively.