Segmentation methods of H&E-stained histological images of lymphoma: A review

Abstract

Image processing techniques are being widely developed for helping specialists in analysis of histological images obtained from biopsies for diagnoses and prognoses determination. Several types of cancer can be diagnosed using segmentation methods that are capable to identify specific neoplastic regions. The use of these computational methods makes the analysis of experts more objective and less time-consuming. Thus, the progressive development of histological images segmentation is an important step for modern medicine. This study presents the progress of recent advances in methods for segmentation of chronic lymphocytic leukemia, follicular lymphoma and mantle cell lymphoma images. The paper shows the main techniques of image processing employed in the stages of preprocessing, detection/segmentation and post-processing of published approaches and discusses their advantages and disadvantages. This study presents the most often used segmentation techniques for these images segmentation, such as thresholding, region-based methods and K-means clustering algorithm. The addressed cancers are also described in histological details as well as possible variations in the tissue preparation and its digitization. Besides, it includes a review of validation techniques and discusses the potential future directions of research in the segmentation of these neoplasias.