Best-parameterized Sigmoid ELM for benign and malignant breast cancer detection

Breast cancer is one of the leading deadly causes for women. Detection of this disease in the early stages, especially before spreading to other organs, can significantly improve survival patients' rates. Medical decision support systems with intelligent classification systems can give second opinion and help reducing possible error because of inexperienced experts, fatigued or improper time limit medical data examination. Backpropagation Artificial Neural Networks (BP ANN) has been extensively used in intelligent breast cancer diagnosis. Nevertheless, the standard gradient-based learning algorithm has several drawbacks such as probability to reach local minima, inefficient training time, and too many setting parameters. Recent studies proved that Extreme Learning Machine (ELM), mathematically and experimentally, could solve BP ANN limitations in several problems. In this paper, we implemented best-parameterized ELM for benign and malignant breast cancer detection. Results showed that Sigmoid ELM generally gave the best performances compared to widely used classification methods such as Decision Tree, BP ANN, and other ELM models with 96% accuracy. This method deployment is promising in medical decision support systems as its intelligent component.