| Abstrakt: |
The breast cancer is a major health issue worldwide, so the early detection of abnormalities decreases the mortality rate. For the early detection of breast cancer, a new model is proposed in this research using mammogram images which is an effective technique used for screening and detecting the breast cancer. At first, the images are acquired from the digital database for screening mammography and mammographic image analysis society datasets. Then, the visual quality of the images is improved by using normalization, contrast limited adaptive histogram equalization and median filter. Further, multi-level multi objective electro magnetism like optimization algorithm is proposed to segment the non-cancer and cancer regions from the enhanced images. Additionally, feature vectors are extracted from the segmented regions using local directional ternary pattern, histogram of oriented gradients and Haralick texture features. Next, infinite feature selection with genetic algorithm is applied to select the active or relevant features for breast cancer classification. The genetic algorithm is applied with entropy value to measure the homogeneity to find the mutual information between the extracted features that helps in provide prominent feature values. The genetic algorithm has the advantage of probabilistic transition rule that helps to analysis the associate neighbour features. The selected feature vectors are fed to deep neural network to classify the mammogram images as malignant and benign classes. From the simulation result, the proposed infinite feature selection with genetic algorithm and deep neural network showed 0. 10–7% improvement in accuracy related to the existing models such as an extreme learning machine, conditional generative adversarial network with convolutional neural network, etc. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink