In medical imaging, image segmentation enables accurate diagnostics, surgical guidance, disease monitoring, and facilitates remote healthcare services. Thresholding is a commonly employed method in image segmentation due to its straightforward approach and effectiveness in distinguishing objects from backgrounds using pixel intensity levels. Thresholding can be bi-level, separating pixels into two classes, or multilevel, categorizing them into more than two classes based on intensity values. This paper centers its attention on the concept of multilevel thresholding. An effective segmentation technique employing multilevel thresholding pinpoints appropriate threshold values to enhance between-class variance or adhere to the entropy criterion. Nevertheless, the task of determining these optimal thresholds during the preprocessing phase becomes time-consuming when conventional methods are applied. However, these challenges can be overcome through the utilization of metaheuristic algorithms. In this study, Kapoor entropy is employed in conjunction with the Chaos Game Optimization algorithm, which is inspired by principles from chaos theory. This algorithm utilizes the arrangement of fractals through the concept of a chaos game and addresses the challenge of fractal self-similarity to determine the optimal values of multilevel thresholds. The Chaos Game Optimization performs effectively by minimizing parameters, achieving a balance between exploration and exploitation, and avoiding premature convergence. The study quantitatively examined outcomes using established evaluation methods (PSNR, SSIM, FSIM) on eight chest X-ray images from COVID-19 patients, employing various threshold values (2, 4, 6, 8 thresholds). These results were then compared with six well-known metaheuristic optimization methods. The experimental findings demonstrate that the proposed method outperformed in both quality and consistency.