Detection of centroblasts in H&E stained images of follicular lymphoma
Emmanouil Michail, Evgenios Kornaropoulos, Kosmas Dimitropoulos, Nikos Grammalidis, Triantafyllia Κoletsa, Ioannis S. Kostopoulos · 2014
This paper presents a complete framework for automatic detection of malignant cells in microscopic images acquired from tissue biopsies of follicular lymphoma. After pre-processing to remove noise and suppress small details, images are segmented by using intensity thresholding, in order to detect the cell nuclei. Subsequently, touching cells are being separated using Expectation Maximization algorithm. Candidate centroblasts are then selected for classification by using size, shape and intensity histogram criteria. Finally, candidates are classified by using a Linear Discriminant Analysis classifier. The application of the methodology in a generated dataset of microscopic images, stained with Hematoxylin and Eosin, showed promising results by detecting in average 82.58% of the annotated malignant cells.