DESIGN AND DEVELOPMENT OF EFFICIENT METHODS FOR HISTOPATHOLOGICAL IMAGE ANALYSIS

Abstract

Histopathological image classification is a vital component in disease diagnosis and treat- ment, particularly for cancer. This thesis focuses on designing efficient methods for segmen- tation, feature selection, and classification of histopathological images using enhanced meta- heuristic algorithms. An Enhanced Multi-Objective Grey Wolf Optimization (EMOGWO) algorithm was devel- oped for segmentation, achieving a mean Dice coefficient of 0.964 and a segmentation accu- racy of 96.4% on H&E-stained ER+ breast cancer images. Compared with baseline methods (K-means-SC and MOGWO-SC), the proposed EMOGWO-SC improved boundary detection accuracy by 3.2% and reduced computation time by 22%. For feature selection, an Improved Multi-Objective Whale Optimization Algorithm (IMO- WOA) was proposed. IMOWOA selected an optimal subset of features, reducing feature dimen- sionality by 25–35% while maintaining high discriminative power. When applied to multiple benchmark histopathological datasets such as BreakHis and BACH, the IMOWOA-based fea- ture selection achieved an average classification accuracy of 98.1%, outperforming existing techniques including DE, Jaya, and Adaptive Jaya by up to 4.5%. The framework also reduced processing time by approximately 30%. Comprehensive statistical analysis using IGD, SP, MS, and t-tests confirmed that the im- provements were significant at a 95% confidence level (p < 0.05). The overall framework demonstrates competitive accuracy, robustness, and computational efficiency, offering strong potential for computer-aided diagnostic applications.