Developing an Evolutionary-based Algorithm for the Multi-Level Thresholding Problem in Medical Image Segmentation

Abstract

Image segmentation is a critical operation in medical image processing, enabling precise diagnosis, treatment planning, and patient follow-up. However, the complex nature and heterogeneity of tumor diagnosis render segmentation a difficult process, particularly for multi-modal scans. This paper explores the integration of optimization techniques into the segmentation process to enhance accuracy and robustness. It begins with an intro- ductory account of optimization principles, including classical and modern methods, and their application to high-dimensional, non-linear situations like medical image analysis. The focus then turns to brain tumor segmentation, where the role of imaging modalities, segmentation techniques like classical and optimization-based. A number of measures of evaluation are used to find segmentation quality objectively. Experimental findings indi- cate that optimization-augmented segmentation methods improve accuracy, consistency, especially with the inclusion of additional clinical ground truth data. Finally, the paper demonstrates the potential of optimization-based methods in reducing manual workload and assisting AI-enabled radiology systems in the detection of brain tumors.