Comprehensive evaluation of optimization algorithms for medical image segmentation

Medical image segmentation algorithms play a crucial role in assisting healthcare professionals with disease identification, research, and diagnosis. Numerous digital image segmentation methods have been developed, with multilevel thresholding techniques consistently outperforming others in terms of evaluation metrics. The standard algorithms include classical statistical methods, such as the Otsu and Kapur methods, which yield highly accurate results. However, when applied to multilevel thresholding, these methods incur significant computational costs, presenting an optimization challenge. In this work, a set of well-known optimization algorithms is integrated with Otsu's method to assess their effectiveness in reducing computational demands while preserving optimal segmentation quality. Experiments are conducted on publicly available datasets, including the TCIA dataset, particularly the COVID-19-AR collection. This work evaluates the performance of each optimization algorithm in combination with Otsu's method, highlighting those that achieve substantial reductions in computational cost and convergence time while maintaining a competitive level of segmentation quality.