Multi-Objective Feature Selection Method by Using ACO with PSO Algorithm for Breast Cancer Detection
Rajesh Saturi, Parvataneni Premchand
Abstract
Breast cancer is the second most cause of cancer deaths in women after lung cancer, and the treatment and diagnosis of breast cancer at an earlier stage is an important requirement for the reduction of the mortality rate. The detection of cell nuclei is the core operation involved in Computer-Aided Diagnosis (CAD) that is used in breast cancer detection. The existing deep learning methods used contour information for accurate detection. But, the existing cancer detection method extracted only the semantic information from the former layer and was not provided the details about the semantic shallow layers in detecting cancer. To solve such an issue, a multi-objective feature selection method was proposed by using Ant Colony Optimization (ACO) with Particle Swarm Optimization (PSO) for the detection of breast cancer. The proposed method utilized the BreakHis dataset and data augmentation was applied to analyze the internal morphological features of the images. The features were extracted by using Convolutional Neural Network (CNN), Histogram of Gradients (HoG) and Local Ternary Pattern (LTP). Then, the features are selected by the proposed multi-objective feature selection method by using ACO with PSO. The selected features are forwarded to the Long Short Term Memory (LSTM) network for the process of classification. The proposed multi-objective feature selection method has the advantage of minimizing the features and reduces the rate of error classification, through a selection of optimal features to increase the rate of breast cancer detection. The proposed multi-objective feature selection method achieved a higher accuracy rate of 95.72% in the detection of breast cancer. Whereas, the existing method showed an accuracy of 93.68% in the detection of breast cancer.