Research On The Medical Microscopic Cell Image Segmentation Algorithms And Fluorescence Intensity Extraction

Nowadays, with the improvement of living standard, people begin to pay muchmore attentions to their health, thus to do more physical examination. The cell featureextraction is an important part of the biomedical examinations, and differentphysiological states would lead to cellular changes which are directly related to thetypes and severities of diseases. Meanwhile, the rapid development of computertechnology makes it possible to process medical cell images based on computervision and provides much convenience to observe cell characteristics only through themicroscopic observation. As some characteristic parameters, such as morphology andfluorescence intensity of cell, should be extracted to achieve quantitative analysis, inthis case, segmenting cell images is necessary. However, the irregularity andcomplexity of adhesion bring great difficulty to cell segmentation. Hence, thesegmentation of two different medical microscopic cells—macrophages and humanumbilical vein endothelial cell are discussed in this paper, specific studies aresummarized below:For the segmentation of strong adhesion macrophages, a novel advancedwatershed algorithm combining the gray-scale morphological erosion and dilationoperations, top-hat transform and bottom-hat transformation, H-minima transformand watershed algorithm is proposed. Compared to watershed algorithm based ondistance transform and seed-watershed algorithm, the proposed algorithm cansegment the strong adhesion macrophages images effectively with lowerover-segmentation and down-segmentation ratio.For the segmentation of human umbilical vein endothelial cells with differentshapes and uneven brightness, two novel cell image segmentation techniquescombining cell image with its corresponding nucleus image are proposed. Onemethod is that find the number of seed points of object region on the cell imageusing corresponding nucleus seed point map, and judge the object region and pair theseed points, segment the object region using dividing line obtained by given dividingrules. The other method is that find the center point of cell using corresponding nuclear seed point map and regard the center point as the marker, and segment thecell image using label-controlled watershed. Performance and result analysisdemonstrate that these two algorithms are valid, precise, and very helpful to furthercell feature extraction.In order to extract the fluorescence intensity of cell/nucleus/cytoplasm of humanumbilical vein endothelial cell, a new mark correction method is presented, which isimplemented and verified on VC++platform. What’s more, we construct a simplefluorescence analysis system and some data obtained on the system are adopted bythe First Affiliated Hospital of Nanchang University.