Technical Study On The Separation And Segmentation For Cell Overlap And Fusion Image

Research goal and significanceThe method on analyzing the immunohistochemical dyeing result by microscope always plays an important role in the clinical pathology and biomedical engineering. This method has certain subjectivity which affected the accuracy of result and analysis. Along with the computer technology development, computer image process and analysis technology are playing more and more important role in the clinical diagnosis and treatment. Developing the new image analysis system (IAS), automatically processing immunohistochemical cell picture with the computer, carrying on the quantitative analysis, assisting doctor to make the fast and accurate judgment for diagnose have very important application prospect. But the accurate rate of image analysis system is insufficiently high at present. Good classification, effective analysis, high recognition rate, and precise quantitative characteristic parameter rely on accurate, fast and repeatable segmentation technology. In other words, segmentation is essential for IAS. But as a result of question-oriented specialty of image segmentation, there are not yet some universally suitable theory and method until now. Mathematics, computer technology and rapid development of medical correlation domain have laid the foundation for cell automatic division and the quantitative analysis. Studying on the related division and separation technology can lay technological base for the highly effective pathological section assistance analysis system, which become the very popular.Research content and solving questionImmunohistochemical image from the CCD camera output is true colorful image. The count of positive cell and the negative cell as well as proportion between them are important factors to judge immunohistochemical reaction intensity, which have very important value for early diagnosis of tumor. In order to obtain accurate data analysis on immunohistochemical cell picture, the key is to separate positive cell and the negative cell correctly. The positive cell and negative cell all have same values in the RGB three colors spaces, thus, segmentation positive cell and negative cell from one single color space is nearly impossible. Because of slice and the reason of cell itself, overlap or a bigger fusion region will often appears. If overlap cell cannot be effectively separated as a single cell, which will directly affect cell count and other parameter measurement result.On the basis of the characteristic of image, the research proposed the technical study on segmentation of characteristic structure in immunohistochemical fusion image and separation of overlap cell, in order to solving some difficult problem such as division of positive cell and the negative cell, the cell overlaps determination, cell core computation and overlap cell separation.The content of technical study on segmentation of characteristic structure in immunohistochemical fusion image are to design and realize new division methods for the colorful cell picture of closed continual boundary, and make picture division experiment to withdraw positive cell and the negative cell. The goal is to obtain more accurate and faster division method of colored cell picture, which can enhance the specific region recognition and the measuring accuracy.The content of technical study on automatic separation of overlap cell picture is to design and realize new separation algorithm. The goal is to obtain the single cell with boundary integrity and no damages, which are used for quota test on cell count, area, perimeter, shape factor and other parameter.Technical route1. Segmentation technology of characteristic structure in immunohistochemical fusion image(1) Image pretreatmentIt includes the contrast gradient enhancement, the noise filtering, the internal holly fill and so on, which can enhance the quality of the image. It is advantageous for the region recognition.(2) Colorimetric analysisGather many ER/PR immunohistochemical colorful images. Analyze color value in different region. Obtain RGB color space distributed survey of the positive cell and the negative cell. Seek the distributed difference.(3) Rough segmentation of positive cell and negative cellBased on the colorimetric analysis result, propose new colorimetric judgment criterion for immunohistochemical colorful image. Rough divide positive cell and negative cell.(4) Meticulous segmentation of positive cell and negative cellChoose the appropriate division algorithm of C average value. Make the various improvements. Form the new cell division algorithm.(5) Programming under the Matlab6.1 environment to realize this technology.2. Separation of overlap cell(1) Picking out Cell overlap regionGathers some overlap cells with different shape. Measure their parameter. Obtain significant result. Determine whether the cell is overlap.(2) Cell core computationAnalyze the computation of single cell core. Discover core computational method in the overlap cell. (3) The convex -enclosed structure analysis of overlap cellAnalyze concave-convex of the cell overlap region. Calculate convex-enclosed structure. Obtain the different situation of overlap. Discover the separation region.(4) Concave spot searchAnalyze the concave area of overlap cell. Search for the concaves pot from the overlap cell concave area.(5) Overlap cell separationSearch for the precise detachment point from separation area. Realize cell separation.(6) Programming under the Matlab6.1 environment to realize this technology.Innovation spot1. Segmentation technology of characteristic structure in immunohistochemical fusion image(1) Immunohistochemical colorimetric criterionThis technology propose colorimetric criterion for segmentation byanalysis on image:●Scanning the entire picture, subtracts R component and B component for each picture element. According to its value, divide the picture into two big kinds:(R-B)＞=0 kind and (R-B).●In the (R-B)＞=0 kind, every element remains it's original color value if (R-B)＞=0, while element is black if (R-B)＜0. The image A has no negative cell.●In the (R-B)＜0 kind, every element remains it's original color value if (R-B)＜0, while element is black if (R-B)＞=0. The image B has no positive cell.This criterion divides the positive cell and negative cell roughly, which enables two kinds of cells have no similar value in RGB color space. It greatly reduced picture processing Complex.(2) C-mean algorithm improvementOn the basis of the result, the CMA algorithm is improved from three points:●The CMA is executed apart on two classes pixels in one color space●The original class center of image B is obtained by result of image A;●The next iterative center can be conjectured on iteration change trend, which reduces the times of iteration.divC_2n=C_2n—C_2n-1C_2n+1=C_2n+divC_2nThe improvements lessen sample amounts, reduce algorithm complexity, decrease iteration times and speed up calculation. The results reveal that the technique is effective.2. Separation technology of overlap cellThe separation technology of overlap cell is realized on the following new methods.(1) Automatically discriminating overlap cell based on shape factor analysisAccording to morphological feature of overlap cell, the article proposed a method of discriminating overlap cell by shape factor. Threshold value of shape factor(P0) is 0.5: if PE＜=P0, the target is overlap; else the target is not overlap.(2) Withdrawing overlap cell core based on mathematics morphologyCorrode the cell boundary upon one layer until separating the single cell. Withdraw each cell core as overlap cell core.(3) Withdrawing concave region Based on convex-enclosed structureSearch for all convex spots. Connect convex spots to obtain convex-enclosed structure. Subtract convex-enclosed structure and original image to obtain concave region. The criterion of convex spots lists as following: (a) ify_q1≤yM₁ or x_q1≥x_M2, q₁ is not convex spot.(b) if x_q2=x_q1 and y_q2>y_q1, q₁ is not convex spot, q₂ substitute q₁; ifY_q2＜Y_q1, q₂ is not convex spot.(c) if X_q1＜x_q2＜xM₂ and q₂is on the left or top of line M₁q₁, q₁ is not convex spot, q₂ substitute q₁; ifq₂is on the right of line M₁q₁, and y_q2＞y_q1, q₂ is convex spot;if y_q2＜=y_q1, q₂ is not convex spot。(d) if x_q2＜x_q1, q₂ is on the left of line M₁q₁, q₁ is not convex spot, q₂ substitute q₁; if q₂ is on the right or top of line M₁q₁, q2 is not convex spot。(4) Concave spot search based on concave regionFor series overlap, the couple of concave spots are on the different sides of concave region and their distance is shortest. If there are more than three cells, we should determine which two concave regions are couples. The creation is:|M_k1O_j1—M_k1O_j2|=min{|M_k1O_i—M_k1O_j|}|M_k2O_j1—M_k1O_j2|=min{|M_k2P_i—M_k2O_j|} j1≠j2, i≠j, andjl,j2,i,j≤n_j, n_jis the amount of coresFor parallel overlap, the concave spots are on the concave region, which have the shortest distance between the centers of overlap cell.(5) Overlap cell separation based on concave spotsThis technology proposed a new algorithm for automatically separating overlap cell image. According to the rugged topography of overlap cell, we seek concave spots from the concave regions in the overlap cell image, and judge whether image is cell series or cell parallel by comparing the numbers of concave spots with cores. In the series situation, drawing line by connecting couple of concave spots to separate overlap region; In the parallel situation, drawing line by connecting concave spot with core to separate the overlap region. This algorithm is carried to separate the overlap cell image under the Matlab environment.Results1. Segmentation technology of characteristic structure in immunohistochemical fusion image The positive cell and negative cell can be separately withdrawed if applying this technology under the Matlab6.1 environment. The result is satisfying. This technology can be used in the quota method of immunohistochemical dyeing positive unit. It has provided one valuable auxiliary method for the immunohistochemica colored picture quantitative analysis.The deficiency is: The positive cell and negative cell have overlap situation. This question is solved in the overlap cell separation technology.2. Separation technology of overlap cell(1) Automatically discriminating overlap cell based on shape factor analysisOn matalb6.1 environment, the experimental results indicate this technique is effective. The accurate ratio is 95%.It can also count the amount of cells in overlap region.The deficiency is: If the single cell presents long and narrow, the cell may be sentenced as overlap cell by mistake. This question needs solving by unifying the core withdraw technology.(2) Withdrawing overlap cell core based on mathematics morphologyThis technology can withdraw each cell core for overlap region. Comparing with professional image processing, this method is correct and data reliable. The effect is satisfied.The deficiency is: if overlap cell has little concave, it cannot withdraw all cores in the overlap region. We should use mouse to assign the core.(3) Withdrawing concave region Based on convex-enclosed structureThis technology can withdraw concave region. Comparing with professional image processing, this method is correct and data reliable.The deficiency is: if the cell overlaps has no concave, concave region can not be calculated. This kind of situation does not suit for this research.(4) Concave spot search based on concave regionThis technology has realized searching for concave spots on concave region in the series and parallel cell overlap. The result is good.(5) Overlap cell separation based on concave spotsThis technology has realized separating overlap cells in series and parallel situation. Comparing with IPP6.0 picture processing software, our technology has some advantage: simple realization, speed quick, ideal effect.Deficiency: The efficiency of the algorithm as well as the complex cell picture situation needs more research and exploration.ConclusionThe research firstly proposed the immunohistochemical colorimetric criterion in turn from the image characteristic, which is used for divide the positive area and the negative area roughly. Then we made the improvement to the C-average value algorithm for withdrawing the positive cell and the negative cell. After that ,we proposed the new technology for distinguishing overlap cell on the analysis of shape factor, which realized picking out overlap cell area in the picture; Then we proposed the new technology for withdrawing core of overlap cell based on mathematics morphology, which calculated each core coordinates as well as the amount of cores for the overlap cell; Then we proposed the new technology for withdrawing concave areas based on convex-enclosed structure. Then we proposed new technology for searching for concave spots on concave areas. On above four technical foundations, we finally proposed overlap cell separation technology based on the concave spot search, which has realized separating each kind of overlap cell.