基于改进正余弦算法的多阈值图像分割：乳腺癌显微镜的案例研究
Multi-Threshold Image Segmentation Based on Improved Sine-Cosine Algorithm: Case Study of Breast Cancer Microscopy

医学图像分割是辅助诊断的重要基础，它能够识别医学影像中的不同组织和特征，实现对病灶区域的定位。分割质量将直接影响疾病诊断与治疗效果。多阈值图像分割因其高效性和实用性在该领域备受关注，其关键在于阈值的选择。但现存的阈值优化策略易陷入局部最优。因此，本研究提出一种改进的正弦余弦算法(ISCA)来增强搜索能力，从而找到更好的阈值。具体的，该方法引入的基于引导的搜索策略和动态调整交叉率机制有效地平衡了探索与开发；提出的进化方向采样策略进一步探索更多有前途的解，从而增强解质量。在IEEE CEC2017基准函数上的实验结果表明，该算法比先进算法具有更快的收敛速度和更高的求解精度。此外，我们基于ISCA、非局部均值二维直方图和Rényi熵构建的多阈值图像分割框架对乳腺癌图像进行图像分割实验，并通过PSNR、FSIM和SSIM指标进行定量评估。实验结果表明基于ISCA的分割方法能选择更好的阈值，在分割效果方面有显著提升。
Medical image segmentation is a foundation for computer-aided diagnosis, enabling the identification of different tissues and features within medical images and facilitating precise localization of lesion areas. The quality of segmentation directly impacts the accuracy of disease diagnosis and treatment outcomes. Among various techniques, multi-threshold image segmentation has garnered significant attention in this field due to its efficiency and practicality, with threshold selection being the key challenge. However, existing threshold optimization strategies often suffer from premature convergence to local optima. To address this issue, this study proposes an Improved Sine Cosine Algorithm (ISCA) to enhance the search capability and identify optimal thresholds. Specifically, the proposed ISCA incorporates a guided search strategy and a dynamically adjusted crossover rate to effectively balance exploration and exploitation. In addition, an evolutionary direction sampling strategy is introduced to explore more promising solutions, improving solution quality. Experimental results on the IEEE CEC2017 benchmark functions demonstrate that ISCA achieves faster convergence and higher solution accuracy compared to state-of-the-art algorithms. Furthermore, we construct a multi-threshold image segmentation framework based on ISCA, non-local means two-dimensional histograms, and Rényi entropy. This framework is evaluated on the breast cancer images. Quantitative assessments using PSNR, FSIM, and SSIM metrics show that the ISCA-based segmentation method consistently selects superior thresholds and significantly improves segmentation performance.