A Novel Tet-LDN Based Feature Extraction and Parallel Convolutional LeNet for Breast Cancer Detection and Classification Using Histopathological Images

One of the main reasons for the high death rate in today's world is Breast Cancer (BC), which affects women all over the world. This work proposes a novel Parallel Convolutional-LeNet (PConv-LeNet) technique for BC detection and classification employing Histopathological images. Originally, the input histopathological image is obtained from a particular dataset, and image denoising is performed to remove the noise in the images using the Medav Filter. After image denoising, blood cell segmentation is carried out using the Parallel Reverse Attention Network (PraNet). Afterwards, feature extraction is done using shape features, including area, solitary, perimeter, eccentricity, and major axis length, and the proposed Tetrolet-Local Direction Number (Tet-LDN), which is developed using Tetrolet features and Local Direction Number pattern (LDN). Finally, BC detection is carried out using a Parallel Convolutional Neural Network (PCNN), which is tuned by Remora Kill Herd Optimization (RKHO), developed by the combination of the Remora Optimization Algorithm (ROA) and Kill Herd Algorithm (KH). Afterwards, BC classification is carried out using the proposed PConv-LeNet, developed by PCNN and LeNet. Further, analysis of the PConv-LeNet reveals that 91.78% accuracy, 91.88% specificity, and 92.38% sensitivity are attained.