Mayfly-Algorithm Selected Features for Classification of Breast Histology Images into Benign/Malignant Class

Abstract

Incidence rate of Breast Cancer (BC) is rising globally and the early detection is important to cure the disease. The detection of BC consist different phases from verification to clinical level diagnosis. Confirmation of the cancer and its stage is performed normally with breast biopsy. This research aims to develop a framework to identify Benign/Malignant class images from the Breast Histology Slide (BHS). This technique consist the following phases; (i) Cropping and resizing the image slice, (ii) Deep-feature extraction using pre-trained network, (iii) Discrete Wavelet Transform (DWT) feature mining, (iv) Optimal feature selection with Mayfly algorithm, (v) Serial feature concatenation, and (vi) Binary classification and validation. This work considered the test image with dimension $896 \times 768 \times 3$ pixels. During the investigation, every picture is cropped into 25 slices and resized to $224 \times 224 \times 3$ pixels. This work implements the following stages; (i) BC detection with deep-features and (ii) BC recognition with concatenated features. In both the cases, a 5-fold cross validation is employed and the experimental investigation of this research confirms that the proposed work helped to achieve an accuracy of 91.39% with deep-feature and 95.56% with concatenation features.