Computer Vision Segmentation of Nucleolar Organizer Regions by Means of High Dynamic Range Cell Imaging and Analysis

Active nucleolar organizer regions (NORs) are affine to silver (Ag) and therefore visible in silver stained cell specimens. These so called AgNORs are of diagnostic relevance for early cancer diagnosis. Therefore we seek to automatically detect and segment these dark, spot-like regions. This is challenging, since the contrast between the AgNORs and the nucleus background varies heavily from cell to cell due to the not standardized staining process. Besides those cell images where AgNORs can be easily detected, there are cell images which appear to be too dark or too bright. Hence, the imaging itself has to be improved for an accurate segmentation of the AgNORs. We show that high dynamic range (HDR) imaging can be used as a basis for AgNOR segmentation. Exploiting the greater amount of information in such HDR images we normalize the cell images, such that a mean shift segmentation provides promising results independent of the staining intensity.