GSRPSO: A multi-strategy integrated particle swarm algorithm for multi-threshold segmentation of real cervical cancer images

• Published in: Swarm and evolutionary computation Vol. 91; p. 101766
• Main Authors: Hu, Gang, Zheng, Yixuan, Houssein, Essam H., Wei, Guo
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2024
• Subjects: Cervical cancer; Global optimization; Medical image segmentation; Multi-level thresholds; Particle swarm optimization; Global optimization; Particle swarm optimization; Cervical cancer; Medical image segmentation; Multi-level thresholds
• ISSN: 2210-6502
• DOI: 10.1016/j.swevo.2024.101766

Cervical cancer is one of the most common gynecological malignant tumors and the fourth largest malignant tumor endangering the life and health of women worldwide. The lesion areas of cervical cancer usually show complexity and diversity. To distinguish different cervical lesion sites and assist doctors in diagnosis and decision-making, this paper first proposes a multi-strategy integrated particle swarm optimization algorithm (GSRPSO, for short). The four strategies in GSRPSO work together to improve its optimization capabilities. Among them, the dynamic parameters balance the exploration and exploitation phases. The gaining sharing strategy and the random position updating strategy accelerate the convergence process while enhancing the diversity of the population. The vertical crossover mutation strategy improves local exploitation and avoids premature stagnation of the algorithm. The optimization performance of GSRPSO is validated by comparison experiments with 15 state-of-the-art algorithms on the CEC2020 test set. In addition, we established a MIS method based on the GSRPSO algorithm by combining the non-local mean algorithm and 2D Kapur entropy. This method is compared with the MIS methods combining 2D Renyi entropy, 2D Tsallis entropy, and 2D Masi entropy in a comparison experiment with four sets of thresholds on six BSDS500 images. The experimental results show that this MIS method exhibits obvious advantages in segmentation quality and stability. Finally, nine cervical cancer images are segmented using the GSRPSO-based MIS method, and experiments are conducted with nine excellent optimization algorithms under six different sets of thresholds. The experimental results show that this MIS method demonstrates better segmentation quality and accuracy than similar methods with the optimal evaluation indexes PSNR=28.3645, SSIM=0.8996, FSIM=0.9494, AD=8.1939, and NAE=0.0710. In conclusion, the GSRPSO-based MIS method is one of a class of highly promising methods to assist physicians in accurately diagnosing cervical cancer.