Breast Mass Classification in Mammograms Based on the Fusion of Traditional and Deep Features

Breast cancer is a disease that affects the life and health of women worldwide. Early detection and treatment of the disease can effectively prolong patients' life and improve their quality of life. Mammography is one of the most commonly used examination methods for breast cancer screening and early diagnosis. Radiologists can differentiate breast masses as benign or malignant according to their shape and appearance. However, the large variation in X-ray appearances of breast masses brings great diagnostic difficulties to doctors. Therefore, this paper proposes a breast mass classification model that fuses traditional and deep features to assist doctors in classifying and diagnosing breast masses. A variety of handcrafted traditional features and the deep features learned from the improved SE-DenseNet model based on DenseNet are integrated. An improved genetic algorithm based on mutual information is used to select the most valuable subset of features in the feature selection stage. The benign and malignant classification of breast masses is finally accomplished through the fusion of multi-classifier decisions. The experimental results show that the classification accuracy of the proposed method for feature fusion has better classification performance compared to using traditional features or deep features separately, indicating that the two types of features contain complementary information, and the fusion of the two can achieve the complementary advantages to a certain extent.