Algorithm Design And Implementation For Detecting Cell And Vascular With In Vivo Image Flow Cytometry

Circulating tumor cells are closely related to the tumor.It is an important biomarker for determining the malignancy of tumor and prognosis of antitumor therapy.Therefore,non-invasive counting of circulating tumor cell populations in mice plays a very important role in many areas of biomedical research.In vivo image flow cytometry(IVIFC)is an emerging live-assay technology that aims to perform real-time monitoring of cells and dynamic monitoring with noninvasive characterization.By analyzing the captured cells,the movement status and changes of circulating tumor cells(CTCs)and dendritic cells(DCs)in the blood flow can be observed.In this work,we mainly describe a new fluorescence imaging system with algorithms that aim at cell detecting-based on the Codebook background modeling algorithm.Then the segmentation algorithm based on the filtering method to identify the blood vessels.At last,co-localization analysis is performed on the segmented target cells as well as the relationship of target cells(CTCs and DCs)with the vessels.The imaging system combines confocal microscopy and other optical imaging techniques to dynamically monitor blood vessels and cells in mice.Two types of target cells,CTC cells and DC cells,can be captured by the in vivo image flow cytometry(IVIFC).The green signal,fluorescence of the GFP detected by the 488-nm channel,represents circulating tumor cells(CTCs);the red signal,fluorescence of the DiD detected at the 635-nm channel,represents dendritic immune cells(DCs).The detected data of the circulating target cells are used as verification data of the algorithm herein to identify the presence of CTCs and DCs,the relationship between the two target cells and the relationship between the target cells and the vessel wall.This work aims to solve the challenges by developing an automated computer vision algorithm that simultaneously performs cell counting and segmentation.In this paper,the cell recognition consists of two parts: the first one is based on Codebook background modeling algorithm which is in fact a compressed background image sequence sampling.The background modeling process is a statistical process that sampling background information of the pixels showing the motion and stores the values into the Codebook.Then,it is used to obtain the structured background changes to verify the pixels belongs to the target cells or background.On the other hand,the recognition of tiny fluorescent cells based on the region search method.The algorithm first needs to give the initial pixel of the target cell in the first frame,and then search the current frame for the pixels that can be identified as the foreground information in the current search frame area.Then the cone-like search area is determined according to the movement speed and state of the cell.Cell trajectory is determined by connecting cells in two adjacent image frames.The filter-based vessel segmentation algorithm is mainly through the pre-processing and filtering method of which the input information is difference of R-channel and G-channel.The proposed algorithms are then verified on the experiment data demonstrating the function of identifying cells and vessels.