WO2016091016A1

WO2016091016A1 - Nucleus marker watershed transformation-based method for splitting adhered white blood cells

Info

Publication number: WO2016091016A1
Application number: PCT/CN2015/092590
Authority: WO
Inventors: 刘治; 刘晶; 郑成云; 李晓梅; 肖晓燕
Original assignee: 山东大学
Priority date: 2014-12-12
Filing date: 2015-10-23
Publication date: 2016-06-16
Also published as: CN104392460A; CN104392460B

Abstract

Disclosed is a nucleus marker watershed transformation-based method for splitting adhered white blood cells. First, an original RGB image is inputted, a difficult problem common in processing images of peripheral blood leukocytes and bone marrow leukocytes is discovered; then, HIS and LUV color space and grayscale space transformations are performed on the original image and features of component images of each channel are analyzed; then, threshold segmentation and image subtraction are performed on component B and a grayscale image to acquire a white blood cell image containing some impurities; immediately thereafter, a nucleated white blood cell group is acquired by means of an image enhancement technique to serve as a marked target; and then, morphological operations and watershed transformations are performed on the nucleated white blood cell group and the white blood cell image containing the impurities to remove the impurities to produce a precise white blood cell image and to solve the problem of cell adhesion; and finally, target white blood cells are clipped and transformed to a LUV space, the white blood cell image is clustered from space and color perspectives to produce a precise white blood cell nucleus.

Description

Adhesion white blood cell segmentation method based on nuclear marker watershed transformation

Technical field

The invention belongs to the field of biomedical engineering, and particularly relates to a method for blocking adhesion white blood cells based on a nuclear transformation watershed transformation.

Background technique

The examination of white blood cells is an important part of clinical tests. The inflammation or other diseases in the body can cause the total number of white blood cells and the percentage of various white blood cells to change. Therefore, checking the total number of white blood cells and white blood cell classification counts is an important method for assisting diagnosis. The cell image analysis and recognition system has been studied in recent years. Its main task is to segment the collected image by automatic analysis, segment the individual cells, calculate the relevant characteristic parameters of individual cells, and identify the number of different cells. In blood cell recognition, the effect of white blood cell segmentation directly affects the results of the next steps of cell feature extraction and classification. In cell recognition, white blood cell segmentation is the most challenging step.

In the field of image processing and pattern recognition, the traditional white blood cell segmentation methods are roughly divided into the following categories: image segmentation algorithms based on multi-spectral techniques; image segmentation algorithms based on color models, commonly used color models are RGB, HSI, CMYK, etc.; Mathematical morphology algorithm image segmentation, such as the snake algorithm to segment the cytoplasm on the basis of the nucleus, the traditional watershed algorithm to solve the cell adhesion problem; the image segmentation algorithm based on fuzzy mathematics, such as fuzzy C-means algorithm, K-means clustering segmentation of white blood cells Wait. In the process of segmentation, it is difficult to achieve better results by simply using one algorithm, and multiple algorithms combined with segmentation are better.

The shortcomings of the existing major adhesion white blood cell segmentation algorithms:

1. The splitting process takes a long time.

2. The segmented image is limited by the white blood cell bank.

3. The accuracy of white blood cell segmentation is low.

4. The division of bone marrow white blood cells is not precise.

5, can not effectively solve the problem of intercellular adhesions.

Summary of the invention

In this paper, based on the different colors of cell images in different color spaces, the difference in white space and red blood cells in texture space, and the ability of cell nuclei to determine the number and location of white blood cells, a simple algorithm is proposed. The precise segmentation algorithm of blood adhesion leukocytes and white blood cells is completed. The algorithm completes the complete segmentation of white blood cells and effectively solves the problem of adhesion between peripheral blood leukocytes.

The object of the present invention is to overcome the above-mentioned deficiencies of the prior art, and to provide a method for blocking leukocyte differentiation based on nuclear marker watershed transformation, which solves the problem of peripheral blood adhesion leukocyte segmentation. The algorithm is simple and time-consuming, and it has good robustness to different cell bank different forms of adhesion leukocyte segmentation and precise segmentation of white blood cell nuclei.

In order to achieve the above object, the present invention adopts the following technical solutions:

A method for segmentation of adhesion white blood cells based on watershed transformation of nuclear markers, comprising the following steps:

(1) processing the original color blood cell image to obtain a binary image I containing only the white blood cell nuclear region, a white blood cell and red blood cell binary map II, and a binary image III containing only the white blood cell nucleus and the red blood cell region;

(2) Subtracting the binary image I from the binary image III to obtain a new binary image IV containing only the red blood cell region,

(3) subtracting the binary image II from the binary image IV to obtain a binary image V containing the complete white blood cell contour region;

(4) Enhancing the nucleus region of white blood cells in the original color blood cell image, obtaining the enhanced image N, and performing morphological processing to obtain a precise white blood cell nuclear image X;

(5) analyzing the binary image X, judging whether the white blood cell nucleus is the lobulated nucleus, judging whether the white blood cell nucleus is stuck, and performing the corresponding operation on the binary image X to obtain the internal seed binary image VI;

(6) Using the internal seed image VI as a marker image and the binary image V as a mask, performing logical and morphological reconstruction operations on the two to obtain a binary image Y of the white blood cell region as an external seed;

(7) judging the adhesion condition, and performing a two-step watershed transformation on the binary image Y to obtain a separated white blood cell binary image VII-2;

(8) logically and operation the binary image X and the binary image VII-2 to obtain an accurate separated white blood cell binary image Z1; the binary image Z1 is used as a marker image, and VII-2 is used as a mask. By performing a morphological reconstruction operation on both, an accurate white blood cell binary image Z2 can be obtained.

The specific method of the step (1) is: observing the grayscale image in the matlab, determining the gray value of the white blood cell nucleus, the red blood cell and the background; calculating the global threshold T of the gray histogram by the graythresh function, and then using the im2bw function pair The grayscale image is thresholded to obtain the binogram of white blood cells and red blood cells. The gray histogram is analyzed. The empirical threshold T2=0.5 of the segmented nuclei is selected. The fraction of the white blood cells is segmented by the im2bw function threshold to obtain the white matter nuclei only. Binary image I; analyzing the B component image, analyzing the pixel values of the red blood cell and white blood cell nuclear color in the image, and obtaining the binary image III containing only the white blood cell core and the red blood cell region by global threshold segmentation technology;

The specific method of the step (4) is: analyzing the G channel component and the S channel component image, normalizing the two, respectively obtaining the normalized matrices Ig and Is, and performing pixel values on the two normalized matrices. Subtraction yields an enhanced image N, where N = 2Ig-Is; then N is binarized and morphologically manipulated to obtain an accurate white blood cell nuclear map X.

The specific method of the step (5) is:

(i) Determining the nucleus as the lobular nucleus condition: when the centroid distance between the two target nuclei in the nuclear binary map X is less than 25, and the area of the two targets is greater than 30 and less than 150, then the target cell nucleus is determined to be a nucleus nucleus, wherein the area is represented by the number of pixels in the white blood cell area; at this time, the centroid coordinate position of the two target nucleus is obtained, and the two targets are Perform centroid-linked operation to form a nuclear nucleus of the lobulated nucleus as an internal seed point;

(ii) determining the nuclear adhesion condition: when the binary image X contains the target nuclear area greater than 1000, it is determined that the binary image X contains the target adhesion nuclei; wherein the roundness value = circumference ² / 4π * area, circumference The number of pixels in the white blood cell boundary is represented by the number and the number of pixels in the white blood cell area. Create a flat disc structural element with a radius of 1, and perform three morphological corrosion operations on the target cell nucleus to obtain an internal seed point of the target cell nucleus;

(iii) Directly acting as an internal seed point for nuclear targets in X that are not lobulated or not adherent.

(iv) Combining the internal seed points obtained in (i) to (iii) to obtain a nuclear nuclei of the binary image X, which is an internal seed, and is referred to as VI.

The specific method of the step (7) is:

(i) judging the target cell adhesion condition: when the binary image Y contains the target cell area greater than 2000 or the roundness value is greater than 2, it is judged that the binary image Y contains the target adhesion cell; at this time, the internal seed VI A distance-based watershed transformation is performed, and the obtained watershed ridge is displayed on the binary image Y, and the obtained image is recorded as VII-1. This process is called a watershed transformation process based on nuclear labeling. It mainly relies on the nucleus of the peripheral blood or the non-adhesion of seed points to solve the problem of cytoplasmic adhesion of leukocytes.

(ii) Continue to judge the adhesion condition. When the binary image VII-1 contains the target cell area greater than 2000 or the roundness value is greater than 2, it is judged that the binary image VII-1 contains the target adhesion cell to the target adhesion cell. Do adaptive corrosion operation until the number of target cells increases or disappears;

(iii) If the target cell disappears, the cell is not treated; if the target number is increased, the multiple targets after corrosion are used as a new internal seed point to perform a watershed transformation based on distance transformation, and the watershed ridge is displayed on the target adhesion cell, adhesion The cells can be separated;

(iv) Continue to judge the cell adhesion conditions until the end of the cycle to obtain the isolated cell binary image VII-2. The (ii) to (iv) processes are recorded as the second watershed segmentation process.

The present invention has the following advantages due to the above technical solutions:

1, the method is simple to operate, and it takes a short time.

2. An image subtraction method is proposed to obtain a complete white blood cell contour. Ensure the integrity of white blood cell division.

The watershed transformation based on the inner nucleus of the nucleus is proposed, which avoids the occurrence of over-segmentation and improves the stability of the watershed transformation adhesion segmentation.

4. The idea of using the number of cells to mark the number of white blood cells is proposed, which avoids errors in the segmentation process. Red blood cells appear, which reduces the rate of miscuting and improves the accuracy of segmentation.

A new method for segmentation of peripheral blood adhesion leukocytes is proposed. The algorithm has high segmentation precision and strong stability, which is superior to traditional algorithms.

DRAWINGS

Figure 1 is a flow chart of peripheral blood adhesion leukocyte segmentation system

Figure 2 shows images of peripheral blood cells from two different illuminations

Figure 3 shows the white matter nuclear I segmentation binary map

Figure 4 shows the effect of white blood cell and red blood cell II segmentation

Figure 5 shows the effect of white blood cell nucleus and red blood cell III segmentation

Figure 6 shows the red blood cell IV segmentation effect

Figure 7 shows the inaccurate white blood cell V segmentation effect map

Figure 8 shows the segmented nuclear X-value map

Figure 9 shows the internal seed point VI binary map

Figure 10 shows the preliminary segmentation of the white blood cell Y binary map.

Figure 11 shows the watershed ridge obtained by X

Figure 12 shows the result of the first watershed transformation VII-1

Figure 13 shows the result of the first watershed transformation VII-2

Figure 14 shows the separation of the precise white blood cell core Z1

Figure 15 shows the separation of precise white blood cells Z2

The segmentation effect of the white blood cells and white blood cells nucleus isolated in Figure 16 is shown on the original image (the nucleus edge is shown by a red line and the white blood cell edge is shown by a green line)

detailed description

The invention is further illustrated by the following examples, which are not intended to limit the invention. The experimental methods in the following examples which do not specify the specific conditions are selected in accordance with conventional methods and conditions.

Example 1:

As shown in FIG. 1 , a specific implementation process of a peripheral blood adhesion white blood cell image segmentation algorithm based on nuclear labeling according to the present invention is as follows:

In white blood cell recognition, medical experts usually distinguish white blood cells and red blood cells according to their characteristics such as color and shape, and discriminate white blood cell types based on information such as texture and space. In this paper, a multi-user peripheral blood cell formation cell bank was collected, and the white blood cells were segmented from the perspective of color and space. The characteristics of the color channel of the cell bank image were analyzed. White space is precisely segmented by color space and morphological manipulation. Adhesion problems present in leukocytes can be accurately and stably differentiated by leukocyte adhesion by improved nuclear marker-based watershed transformation. The algorithm is simple and easy to operate, and can effectively solve the problem of cell adhesion. The cell images under different illuminations of different cell banks are shown in Fig. 2, which has higher segmentation rate and good robustness.

(1) Inaccurate division of white blood cell V. In this process, the white blood cell binarized images may have some red blood cells and other impurities or cell adhesions, but the number and morphology of white blood cells remain intact, there is no problem of leukocyte missing or white blood cell defects, and inaccurate white blood cell segmentation. It is realized by binarized image subtraction technology. The segmentation method is: firstly, the original RGB color image is converted into a gray space with a pixel range of 0 to 1. By setting different thresholds, the threshold value of the gray image is 0.5 and the threshold segmentation based on the ostu adaptive threshold segmentation technique is respectively performed. Each of them obtained two binarized images of white blood cell nucleus I (as shown in Figure 3) and white blood cells and red blood cells II (shown in Figure 4). Then extract the B component of the RGB space and find that the pixel values of the white blood cell nucleus and the red blood cell are lower in this channel component. According to this feature, the ostu-based adaptive threshold segmentation can be used to obtain the binary image of the white blood cell nucleus and the red blood cell III, as shown in Fig. 5. Shown. As shown in Fig. 6, the image of erythrocyte IV can be obtained by subtracting the white blood cell nuclear map I from the binary image of the white blood cell nucleus and the red blood cell. The white blood cell and red blood cell binary map II minus the red blood cell binary map V can obtain an inaccurate white blood cell binary image V, as shown in FIG.

(2) The internal seeds are also obtained from the nuclear nuclei. The cell nucleus is taken as an internal seed to determine the number of white blood cells and to resolve the adhesion problems of white blood cells. The method of obtaining the nucleus is to first convert the original RGB color image into the HSI space, and extract the G and S channel components of the two spaces respectively. Observe the G component and find that the white blood cell and platelet pixel values are small, and other components have larger pixels. Value, observe the S component can be found, white blood cells and platelets have larger pixel values, other components have smaller pixel values, normalize the two-channel components, and then subtract the pixel values to enhance the nuclear image. . The binarization and morphological processing of the obtained enhanced image can obtain a nuclear binary image X, and the segmentation effect is shown in FIG. In the case where the nucleus of the nucleus in the leukocyte is multinuclear, in order to make it a nucleus, the centroid is connected to each other, and the multinucleus becomes a nucleus. Not only can the number of white blood cells be determined, but also the internal seed of the watershed transformation. Solve cell adhesion problems. For the adhesion of leukocyte nucleus in leukocytes, a flat disc structural element with a radius of 1 is created, and a morphological corrosion operation is performed on the target nuclei to obtain a seed point of the adherent nuclei. The internal seed image obtained by the method is as shown in FIG.

(3) Acquisition of external seeds. A cell nucleus corresponds to a white blood cell, and an internal seed binary image VI determines the number of white blood cells. By logically ANDing the internal seed binary image VI and the inexact white blood cell binary image V, a large area of impurity such as excess red blood cells in the image V can be removed. The logical and subsequent images of the two are used as the marker images, and the inaccurate white blood cell image V is used as a mask to perform a morphological reconstruction operation to obtain an accurate white blood cell binarization map. Like Y is the external seed, as shown in Figure 10.

(4) Watershed transformation to resolve cell adhesion. In peripheral blood cells, there is generally no adhesion in the nucleus. In the case of non-adhesion of the nucleus and adhesion of the cytoplasm, a distance-based watershed transformation is performed on the internal seed VI, and the watershed ridge is displayed on the external seed Y to obtain a white blood cell binary image. -1, the watershed ridge line is shown in Figure 11, the first watershed transformation result binary value VII-1 is shown in Figure 12; when there is cytoplasm and erythrocyte adhesion or cell nucleus adhesion in the cell image, continue to judge the target adhesion A watershed transformation of adaptive corrosion of VII-1 yields a separate cell binary image VII-2, as shown in FIG.

(5) Separation of precise white blood cells and precise white blood cell nuclei. In the binary image X, for the case of nuclear adhesion, the binary image Z1 can be solved by the binary image VII-2 and the binary image X, and the separated nuclear binary image Z1 is shown in FIG. The binary image VII-2 may contain impurities such as red blood cells, and the binary image Z1 is used as a marker image, and VII-2 is used as a mask to perform a morphological reconstruction operation on the two to obtain a precise white blood cell binary value. Image Z2, as shown in FIG.

(6) The divided edge lines of the blood cell image white blood cells and white blood cell cores are sequentially displayed on the original image as shown in FIG. It is found that this method divides white blood cells, which is short in time, high in precision, good in segmentation effect and superior to traditional peripheral blood cell adhesion segmentation algorithm.

The above description of the specific embodiments of the present invention has been described with reference to the accompanying drawings, but it is not intended to limit the scope of the present invention. Those skilled in the art should understand that the skilled in the art does not require the creative work on the basis of the technical solutions of the present invention. Various modifications or variations that can be made are still within the scope of the invention.

Claims

A method for segmentation of adhesion white blood cells based on watershed transformation of nuclear markers, comprising the following steps:

(1) processing the original color blood cell image to obtain a binary image I containing only the white blood cell nuclear region, a white blood cell and red blood cell binary map II, and a binary image III containing only the white blood cell nucleus and the red blood cell region;

(2) Subtracting the binary image I from the binary image III to obtain a new binary image IV containing only the red blood cell region,

(3) subtracting the binary image II from the binary image IV to obtain a binary image V containing the complete white blood cell contour region;

(4) Enhancing the nucleus region of white blood cells in the original color blood cell image, obtaining the enhanced image N, and performing morphological processing to obtain a precise white blood cell nuclear image X;

(5) analyzing the binary image X, judging whether the white blood cell nucleus is the lobulated nucleus, judging whether the white blood cell nucleus is stuck, and performing the corresponding operation on the binary image X to obtain the internal seed binary image VI;

(6) Using the internal seed image VI as a marker image and the binary image V as a mask, performing logical and morphological reconstruction operations on the two to obtain a binary image Y of the white blood cell region as an external seed;

(7) judging the adhesion condition, and performing a two-step watershed transformation on the binary image Y to obtain a separated white blood cell binary image VII-2;

(8) logically and operation the binary image X and the binary image VII-2 to obtain an accurate separated white blood cell binary image Z1; the binary image Z1 is used as a marker image, and VII-2 is used as a mask. By performing a morphological reconstruction operation on both, an accurate white blood cell binary image Z2 can be obtained.
The segmentation method according to claim 1, wherein the specific method of the step (1) is: observing the grayscale image in the matlab, determining the gray value of the white blood cell nucleus, the red blood cell and the background; calculating by the graythresh function The global threshold T of the gray histogram, and the threshold image segmentation of the gray image by the im2bw function, the binary image of the white blood cell and the red blood cell region is obtained; the gray histogram is analyzed, and the empirical threshold T2=0.5 of the segmented cell core is selected, and the im2bw function is used. Threshold segmentation of the white blood cell fraction, obtaining a binary image I containing only the white blood cell nucleus; analyzing the B component image, analyzing the pixel values of the red blood cell and white blood cell nuclear color in the image, and obtaining only the white blood cell nucleus and the red blood cell by the global threshold segmentation technique The binary image of the region III.
The segmentation method according to claim 1, wherein the specific method of the step (4) is: analyzing the G channel component and the S channel component image, normalizing the two, and respectively obtaining the normalized The matrix Ig and Is, the pixel values of the two normalized matrices are subtracted to obtain an enhanced image N, where N=2Ig-Is; then N is binarized and morphologically operated to obtain an accurate white blood cell nuclear map X .
The segmentation method according to claim 1, wherein the specific method of the step (5) is:

(i) determining that the nucleus is a nucleus nucleus condition: when the centroid distance between the two target nuclei in the nuclear binary map X is less than 25, and the area of the two targets is greater than 30 and less than 150, the target nucleus is determined to be a lobular nucleus, wherein The area is represented by the number of pixels in the white blood cell area; at this time, the centroid coordinate positions of the two target nuclei are obtained, and the centroids are connected to each other to form a cell nuclei as internal seed points;

(ii) determining the nuclear adhesion condition: when the binary image X contains the target nuclear area greater than 1000, it is determined that the binary image X contains the target adhesion nuclei; wherein the roundness value = circumference 2 / 4π * area, circumference Using the number of pixels of the white blood cell boundary, the area is represented by the number of pixels in the white blood cell area; creating a flat disc structure element with a radius of 1, performing three morphological corrosion operations on the target cell nucleus to obtain the interior of the target cell nucleus Seed point

(iii) directly acting as an internal seed point for a nuclear target in the binary image X that is not a lobulated nucleus or is not a ligated;

(iv) Combining the internal seed points obtained in (i) to (iii) to obtain a nuclear nuclei of the binary image X, which is an internal seed, and is referred to as VI.
The segmentation method according to claim 1, wherein the specific method of the step (7) is:

(i) Judging the target cell adhesion condition: when the binary image Y contains the target cell area greater than 2000 or the roundness value is greater than 2, it is judged that the binary image Y contains the target adhesion cell; the internal seed VI is based on the distance-based watershed Transform, the obtained watershed ridge line is displayed on the binary image Y, and the obtained image is recorded as VII-1;

(ii) Continue to judge the adhesion condition. When the binary image VII-1 contains the target cell area greater than 2000 or the roundness value is greater than 2, it is judged that the binary image VII-1 contains the target adhesion cell, and the target adhesion cell is self-made. Adapt to corrosion operations until the number of target cells increases or disappears;

(iii) If the target cell disappears, the cell is not treated; if the target number is increased, the multiple targets after corrosion are used as a new internal seed point to perform a watershed transformation based on distance transformation, and the watershed ridge is displayed on the target adhesion cell, adhesion The cells can be separated;

(iv) Continue to determine the cell adhesion conditions until the end of the cycle to obtain the isolated cell binary image VII-2; (ii) to (iv) the process is recorded as the second watershed segmentation process.