A sophisticated multilevel thresholding optimizer for diagnosing breast cancer disease

Abstract Breast cancer is one of the leading causes of mortality among women worldwide, and early detection is essential to improve survival rates. Among various imaging techniques, thermography is a promising noninvasive, non-ionizing, and cost-effective modality for real-time diagnosis. This study proposes a multilevel threshold segmentation approach based on Enhanced Hippopotamus Optimization (EHO) for breast thermographic images. The proposed method improves the original HO algorithm by strengthening exploitation and enhancing convergence behavior. Its performance has been validated using Otsu’s method and evaluated on the 29 CEC-2017 benchmark functions. Experimental results demonstrate that EHO outperforms several recent optimization algorithms in both qualitative and quantitative metrics, including fitness, PSNR (Peak Signal-to-Noise Ratio), SSIM (Structural Similarity Index Measure), FSIM (Feature Similarity Index), MSE (Mean Squared Error), computation time, precision, sensitivity, specificity, F-measure, and AUC.