Breast Cancer Semantic Segmentation for Accurate Breast Cancer Detection with an Ensemble Deep Neural Network

Abstract

Breast tumors are the major malignancy in females and diagnostic systems using artificial intelligence algorithms for breast imaging have shown promising results. Among many algorithms, a deep convolutional neural network (DCNN) using K-means clustering and a multiclass support vector machine model enhance the precision of categorizing breast tumors from mammogram images. Nonetheless, effective breast tumor identification is still difficult without partitioning the pectoral muscle (PM) boundary from the remaining breast area. Therefore, this article proposes an Ensemble-Net model by ensembling the transfer learning model with different pre-trained CNN structures for partitioning the PM boundary from the remaining breast area in the mammographic scans. The segmentation process has 2 phases. In the initial phase, different region-of-interests are generated that include the object according to the input images. In the secondary phase, the object class is predicted after the areas of bounding boxes are refined and a pixel-range mask is created for the entity. These 2 different phases are associated with the backbone structure which creates the pyramid hierarchy of DCNN to acquire the features from the raw images. Moreover, it employs global average pooling followed by the softmax classification to recognize the normal, benign and malignant cases. Finally, the experimental outcomes demonstrate that the Ensemble-Net achieves 96.72% accuracy than the other classical classifiers.