Segmentation Approaches for Cancer Detection in Histopathological Images—A Review

Medical image analysis serves as a tool to assist clinicians in diagnosing diseases rather than replacing them. While many imaging modalities, such as mammography, MRI, and CT, have been utilized in cancer diagnosis, histopathological analysis remains the gold standard for cancer detection. However, the accuracy of cancer diagnosis heavily depends on the pathologist's experience, making the results potentially subjective and prone to errors. This highlights the need for developing assistive tools. Segmentation plays a pivotal role in medical image processing, particularly in histological analysis, as irregular changes in cell structures are often indicative of disease. Over the years, numerous segmentation approaches have been developed to identify cell structures, specifically nuclei. Many of these approaches employ deep learning-based methods. However, a trade-off exists among computational complexity, processing time, cost, and accuracy when comparing deep learning-based segmentation methods with conventional approaches, such as thresholding. The paper discusses the current deep learning-based segmentation methods and their evaluation results for histopathological images used in cancer detection and evaluates their respective advantages and disadvantages. Finally, the paper identifies existing gaps in the field and provides suggestions for future improvements.