An Improved Sine-Cosine Algorithm with Diversity Control Based on Nonlinearity for Medical Image Segmentation

The Sine-Cosine Algorithm (SCA), though recognized for its global search capability, exhibits notable deficiencies in convergence speed and local exploitation during later optimization stages. To overcome these drawbacks, this study introduces an Improved Sine-Cosine Algorithm (ISCA) based on nonlinearity, incorporating three strategic enhancements: a Tent mapping-based initialization to increase population diversity and ergodicity, an elite-guided mechanism to intensify local search, and a diversity-driven evolutionary strategy to stabilize convergence. In terms of nonlinear feature extraction, extensive experiments were conducted to assess ISCA’s performance across two domains: numerical optimization and medical image segmentation. On the CEC2022 benchmark suite, ISCA demonstrated superior convergence speed and optimization accuracy over several classical and contemporary metaheuristic algorithms, including GWO, WOA, SWO, and OOA. This nonlinear feature extraction method not only improves the robustness of the algorithm when processing complex medical images, but also significantly improves the segmentation accuracy. When applied to COVID-19 chest X-ray segmentation through integration with the K-Means algorithm, ISCA-K-Means showed improved segmentation performance compared to several baseline methods, including K-Means, K-Means[Formula: see text], SCA-K-Means, PSO-K-Means, and Otsu thresholding. It achieved an MSE of 0.0062, a PSNR of 23.40, an SSIM of 0.854, a Dice coefficient of 0.784, and an IoU of 0.716, reflecting consistent improvements in pixel-level accuracy and structural consistency across multiple evaluation metrics. These findings highlight ISCA’s effectiveness as a robust and adaptive optimization framework, with strong potential for advancing clustering-based medical image analysis and solving complex real-world optimization tasks.