Deep Features with Improved Extreme Learning Machine for Breast Cancer Classification

Abstract

Breast cancer classification problem is receiving more attention among researchers due to its global impact on women's healthcare. There is always a demand for research analysis in the earlier diagnosis of breast cancer. The paper proposes a new computer-aided diagnosis (CAD) framework which integrates deep learning and Extreme Learning Machine (ELM) for feature extrication and classification of breast cancer. The proposed CAD tool is very much helpful for radiologists in the earlier diagnosis of breast cancer using digital mammograms. Herein, the research uses the Sine-Cosine Crow-Search Optimization Algorithm (SC-CSOA) for improving the ELM’s classification performance. And to extricate the robust features from the input mammograms, the concept of transfer learning is applied. For that, the work adopts the three most efficient Residual Network (ResNet) families of CNN, namely ResNet18, ResNet50, and ResNet101 architectures. The input database used for evaluation is the INbreast dataset which comprises Full-Field Digital Mammogram (FFDM) images. At this point, the research compares the results obtained with the existing ELM and K-NN algorithms where it is found that the performance of the proposed framework provides the supreme classification (95.811% of accuracy) over others.