Bilkent University
Department of Computer Engineering


Cell Segmentation in Cancerous H&E Tissue Images


Deniz Doğan
MS Student
Computer Engineering Department
Bilkent University

Automatic identification of nucleus regions in cancerous tissues is still an ill-posed problem. Due to various reasons (bad staining, different cell shapes in different tissues, etc.), identification of nuclei regions can be difficult and cells may not be discerned very easily from other tissue components. In our research, we use some rule based techniques for the identification of those nuclei regions in liver and lung H&E images. We first do k-means clustering to find the connected components (nucleus, background, etc.). Then, we apply sobel operation on the images in order to find the edges (left, right, top and bottom) of the cell regions. By combining clustering and sobel edges, we have constructed an adjacency matrix to identify which sobel regions are neighbor to which nuclei regions. In this way, non-identifiable nuclei regions in k-means clustering can also be detected.


DATE: 07 November, 2016, Monday @ 17:00