FmRMR-OBL-ChefNet for Breast Cancer Histopathology Image Classification

Deep learning models can produce very large feature representations, making it difficult to analyze the detailed cellular and tissue information found in breast cancer histopathology images, which can aid in computer-aided diagnosis. These representations could have redundant or poorly informative features, which would raise the computational cost and complicate the classification step that follows. The hybrid deep learning and metaheuristic framework FmRMR-OBL-ChefNet is presented in this paper with the goal of obtaining a compact and discriminative feature representation for the classification of breast cancer histopathology. The suggested framework learns complementary deep features by utilizing pretrained ResNet, ResNeXt, and NASNet architectures. Fuzzy minimum redundancy maximum relevance (FmRMR), chaotic weighting, and adaptive relevance–redundancy balancing are used to refine these features. An efficient subset of the refined features is then found using an opposition-based Chef Optimization Algorithm (OBL-COA). The chosen representation is assessed using area under the curve (AUC), F1-score, recall, accuracy, precision, and specificity. The framework's goal is to minimize computational complexity and feature redundancy while preserving information that can be used to differentiate between benign and malignant breast tissue.