Research On Improved Watershed Algorithm For Pancreatic Neuroendocrine Tumor Cells Image Segmentation

The main content of this paper is how to apply Morphological Watershed Algorithm to cell image segmentation.In this paper,a marker based watershed cell segmentation method is proposed.The core idea of this method is to use fast radial symmetry(FRS)transform to extract the nucleus center and use it as a marker for watershed segmentation.Based on the affine transformation from circle to ellipse,the fast radial symmetry transformation is extended to the generalized fast radial symmetry(GFRS)transformation,which improves the aforementioned marker based watershed cell segmentation method.Finally,the experimental analysis shows that the watershed algorithm based on FRS transform has achieved high accuracy,accuracy and Dice coefficient value,which can better complete the cell segmentation.Compared with the watershed algorithm based on FRS transform,the improved watershed algorithm based on GFRS transform has a certain improvement in segmentation accuracy and Dice coefficient.This paper aims at the H&E staining of pancreatic neuroendocrine tumor cells.The basic steps of the above-mentioned marker based watershed cell segmentation method are as follows:firstly,the RGB color image is analyzed by the color deconvolution technology to obtain the hematoxylin channel image;secondly,the filter denoising and morphological processing of the hematoxylin channel image are carried out;and then the fast radial symmetry transformation is carried out to obtain the hematoxylin channel image Then,the seed points are extracted by solving the local extremum of the transformation result graph.Finally,the extracted seed points are used as foreground markers for watershed segmentation.The basic steps of the improved marker based watershed cell segmentation method are the same as before,only FRS transformation needs to be replaced by GFRS transformation.The experimental results show that the watershed cell segmentation method based on FRS transform has achieved good segmentation results,but some over segmentation phenomenon is produced for oval cells.The improved watershed cell segmentation method based on GFRS transform solves this problem well.In view of the problem that morphological watershed algorithm is sensitive to noise and easy to form over segmentation,this paper discusses several different technical routes.Including:image preprocessing before watershed segmentation,using prior knowledge to guide segmentation;or after watershed segmentation,merging the segmented small areas through certain criteria.