CN110610490A - A method for locating leukocytes in images of diseased cells - Google Patents

Abstract

A method for locating leukocytes in a diseased cell image, which relates to the field of medical diagnosis. In a microscopic image mixed with background, red blood cells, leukocyte plasma and leukocyte nuclei, the nuclei are separated by a threshold segmentation algorithm, and the isolated points are removed by open operation reconstruction. , with expansion treatment to remove irrelevant areas, and isolation of nuclei to achieve preliminary localization of leukocytes. For non-adherent leukocytes, the leukocyte plasma was segmented directly using a threshold segmentation algorithm. For adherent leukocytes, when segmenting leukocyte plasma, first use the maximum inter-class variance algorithm to divide the leukocyte background and the target into two parts to minimize the probability of misclassification. Second, use Sobel to obtain the gradient image, and perform watershed transformation according to the obtained gradient amplitude. , and finally perform the reconstruction-based open-close operation on the obtained grayscale image, because without affecting the target image, the effect of removing image fragments is better, and white blood cell segmentation can be achieved. The invention is used for medical clinical diagnosis and improves the diagnosis efficiency of medical personnel.

Description

A method for locating leukocytes in images of diseased cells

technical field

The invention relates to the field of medical image processing, in particular to a method for locating leukocytes in a diseased cell image.

Background technique

The role and imaging of medical image processing technology in medical research and clinical practice is increasing day by day, and as a result, clinicians can observe the internal lesions of the human body more directly, clearly, and with higher accuracy. Therefore, medical image processing technology has always been highly valued by relevant experts at home and abroad. In this paper, an automatic leukocyte localization method is also proposed in medical processing. The location of the leukocytes was initially obtained, and the specific location of the leukocytes was obtained according to the largest area and the location of the nucleus, and the leukocytes were segmented.

The traditional leukocyte segmentation method adopts artificially designed features, which are affected by external factors such as illumination, and the methods used for different leukocyte images are relatively limited. In some scenarios, traditional methods cannot obtain satisfactory results and have poor robustness.

The present invention proposes a method for locating leukocytes in medical diseased cell images. The method can separate the nuclei of adherent cells and non-adherent cells from the microscopic image through threshold segmentation, and uses the maximum inter-class variance algorithm for adherent leukocytes. , the image is divided into two parts, and then the leukocyte plasma is segmented by the watershed algorithm. For the adhering cells, it is simpler and can be directly separated by the threshold segmentation algorithm. The main purpose of the method for locating leukocytes in diseased cell images is to make leukocytes more accurate in clinical diagnosis, so that clinicians can observe the diseased parts in the human body more directly, clearly, and with higher accuracy, thereby making the medical staff more accurate. Work efficiency is improved.

SUMMARY OF THE INVENTION

In order to solve the problems existing in the background art, the purpose of the present invention is to provide a method for locating leukocytes in an image of diseased cells.

The invention separates cell nuclei through a threshold segmentation algorithm, uses open operation to reconstruct and removes isolated points, and uses expansion processing to remove irrelevant areas, and the separation of cell nuclei realizes the preliminary positioning of white blood cells. For non-adherent leukocytes, the leukocyte plasma was segmented directly using a threshold segmentation algorithm. For adherent leukocytes, when segmenting leukocyte plasma, first use the maximum inter-class variance algorithm to divide the leukocyte background and the target into two parts to minimize the probability of misclassification. Second, use Sobel to obtain the gradient image, and perform watershed transformation according to the obtained gradient amplitude. , and finally perform the reconstruction-based open-close operation on the acquired grayscale image, so that the image fragment removal effect is better, and the white blood cell segmentation is realized.

The technical scheme adopted in the present invention comprises the following steps:

(1) Acquire 100 leukocyte image data sets, and convert the collected microscopic images from color images to grayscale images;

(2) performing image enhancement on the acquired grayscale image to highlight the features of the image;

(3) Separating the nuclei of leukocytes from the obtained image after image enhancement by a method based on threshold segmentation, so as to realize the preliminary positioning of leukocytes;

(4) removing the isolated points where the image of leukocyte nuclei is obtained;

(5) Remove the irrelevant area of the isolated point image of the nucleus image, so as to obtain the complete leukocyte nucleus, the above steps are also suitable for the process of extracting the nucleus of the adherent cells and non-adherent cells;

(6) For the microscopic images of non-adherent cells, according to the initial location of the leukocyte nucleus and the largest cell area, the leukocyte plasma is directly segmented by the threshold segmentation method;

(7) For the microscopic image of the adherent cells, according to the grayscale characteristics of the image, the image is divided into two parts: background and target, so as to minimize the probability of misclassification;

(8) Obtain the gradient image of the image obtained in step 7, and simultaneously perform watershed transformation on the gradient magnitude of the obtained gradient image;

(9) Perform the reconstruction-based open-close operation on the grayscale image obtained in step 8, the purpose is to remove the debris in the image better without affecting the target image, and then extract the cytoplasm of the adherent cells, To achieve the segmentation of leukocytes.

The data set is an image data set mixed with background, leukocyte plasma, erythrocytes and leukocyte nuclei at the same time.

Described step (1) is specifically:

(1.1) Convert the acquired white blood cell image into a gray image by means of color conversion.

Described step (2) is specifically:

(2.1) The method of using linear transformation to obtain the grayscale image in step (1) adopts the following formula to enhance the image to highlight the features of the image:

G=F(g)=k·g+b

Among them, when k>1, it means that the contrast of the output image increases; when k<1, it means that the contrast of the output image decreases; when k=1, b≠0, it means that only the gray value of the output image is moved up or down Shift, which has the effect of making the entire image lighter or darker. In general, linear transformation stretches a smaller grayscale range to a larger grayscale range, so it is often called grayscale stretching. Let the grayscale variation range of the image be [g _min , g _max ], and through a function F to extend the range to [G _min , G _max ], there is a linear transformation:

That is, the transformation function is the equation of a straight line passing through two points (g _min , G _min ) and (g _max , G _max ).

Described step (3) is specifically:

(3.1) Select a suitable threshold for the image after image enhancement obtained in step (2), and use the method based on threshold segmentation to separate the nuclei of leukocytes using the following formula to achieve the initial positioning of leukocytes:

Among them, T represents the threshold, g(x, y) represents the original gray value of the pixel (x, y), and G(x, y) represents the transformed (enhanced) gray value of the pixel (x, y). The threshold is used as the threshold to distinguish the target and background pixels during segmentation. The pixels greater than or equal to the threshold belong to the target, while the others belong to the background, and vice versa.

Described step (4) is specifically:

(4.1) The cell nucleus extracted in step (3) is reconstructed by open operation to remove the isolated points of the cell nucleus image.

Described step (5) is specifically:

(5.1) The isolated point image obtained in step (4) is subjected to expansion processing to remove irrelevant regions with the following formula:

in, means expansion, Represents an empty set, and B is a structuring element.

Described step (6) is specifically:

(6.1) When there is no leukocyte adhesion, the leukocyte plasma can be directly segmented by the threshold segmentation algorithm according to the initial location of the nucleus and the principle of maximum area.

Described step (7) is specifically:

(7.1) For the microscopic image of adherent cells, first use the maximum inter-class variance algorithm, and use the following formula to divide the image into two parts: background and target according to the grayscale characteristics of the image, so as to minimize the probability of misclassification:

g=w0(μ0-μ)^2+w1(μ1-μ)^2

Among them, for the image I(x, y), the segmentation threshold of the foreground (and the target) and the background is denoted as T, the proportion of the pixels belonging to the foreground to the entire image is denoted as w0, and the average gray level μ0; the number of background pixels accounts for The scale of the whole image is w1, its average gray level is μ1, the total average gray level of the image is recorded as μ, and the inter-class variance is recorded as g.

Described step (8) is specifically:

(8.1) Use Sobel to obtain the gradient image of the image obtained in step (7), and perform watershed transformation on the gradient magnitude of the obtained gradient image.

Described step (9) is specifically:

(9.1) Perform an open-close operation based on reconstruction on the grayscale image obtained in step (8) to remove debris in the image, and then extract the cytoplasm of adherent cells to achieve leukocyte segmentation.

Beneficial effects of the present invention:

The present invention and the traditional leukocyte localization method overcome the previous feature of slow operation efficiency and have stronger robustness. It has good advantages in blood cell segmentation, which can reduce the cost of automatic blood analyzers and assist clinical diagnosis, so it has important clinical significance and broad development prospects.

Description of drawings

Schematic diagram of the segmentation result of the white blood cell image of the present invention: Figure 1 is a color image of non-adherent cells, Figure 8 is a color image of adherent cells, Figure 2 is the result of the grayscale image of Figure 1, and Figure 3 is the image of Figure 2 after image enhancement Figure 4 is the result of the image after threshold segmentation in Figure 3, Figure 5 is the result of the image reconstructed by the open operation in Figure 4, Figure 6 is the result of the nucleus image after the expansion processing of Figure 5, and Figure 12 is Figure 8 Figure 7 is the result of the leukocyte plasma image of the non-adherent cells after threshold segmentation in Figure 1; Figure 9 is the image result of the grayscale image of Figure 8 after the Sobel gradient algorithm, and Figure 10 is Figure 9 The image result after the watershed algorithm, Figure 11 is the image result after the reconstruction of the open-close operation in Figure 10.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and embodiments.

Examples of the present invention are as follows:

Step 1: Select 100 image datasets mixed with background, leukocyte plasma, erythrocytes and leukocyte nuclei, and convert the collected dataset images from color images to grayscale images.

(1.1) Convert the acquired white blood cell image into a gray image by means of color conversion.

Step 2: Perform image enhancement on the acquired grayscale image to highlight the features of the image.

(2.1) Using the method of linear transformation to obtain the grayscale image in step 1, the following formula is used to enhance the image to highlight the features of the image:

G=F(g)=k·g+b

Among them, when k>1, it means that the contrast of the output image increases; when k<1, it means that the contrast of the output image decreases; when k=1, b≠0, it means that only the gray value of the output image is moved up or down Shift, which has the effect of making the entire image lighter or darker. Let the grayscale variation range of the image be [g _min , g _max ], and through a function F to extend the range to [G _min , G _max ], there is a linear transformation:

That is, the transformation function is the equation of a straight line passing through two points (g _min , G _min ) and (g _max , G _max ).

Step 3: The obtained image after image enhancement is used to separate the nuclei of the leukocytes by a method based on threshold segmentation, so as to realize the preliminary positioning of the leukocytes.

(3.1) Select the appropriate threshold for the enhanced image obtained in step 2, and use the threshold-based segmentation method to separate the nuclei of the leukocytes using the following formula to achieve the initial positioning of the leukocytes:

Among them, T represents the threshold, g(x, y) represents the original gray value of the pixel (x, y), and G(x, y) represents the transformed (enhanced) gray value of the pixel (x, y). The threshold is used as the threshold to distinguish the target and background pixels during segmentation. The pixels greater than or equal to the threshold belong to the target, while the others belong to the background, and vice versa.

Step 4: Remove isolated spots where images of leukocyte nuclei were obtained.

(4.1) The cell nucleus extracted in step 3 is reconstructed by open operation to remove the isolated points of the cell nucleus image.

Step 5: Remove the irrelevant area of the isolated point image of the nucleus image, so as to obtain the complete leukocyte nucleus. The above steps are also applicable to the process of extracting the nucleus of the adherent cells and non-adherent cells.

The isolated point image obtained in step (4) is expanded to remove irrelevant regions, and the formula is:

in, means expansion, Represents an empty set, and B is a structuring element.

Step 6: For the microscopic images of non-adherent cells, according to the initial location of the leukocyte nucleus and the largest cell area, the leukocyte plasma is directly segmented by the threshold segmentation method.

(6.1) When there is no leukocyte adhesion, the leukocyte plasma can be directly segmented by the threshold segmentation algorithm according to the initial location of the nucleus and the principle of maximum area.

Step 7: For the microscopic image of the adherent cells, according to the grayscale characteristics of the image, the image is divided into two parts, the background and the target, so as to minimize the probability of misclassification.

(7.1) For the microscopic image of adherent cells, first use the maximum inter-class variance algorithm, and use the following formula to divide the image into two parts: background and target according to the grayscale characteristics of the image, so as to minimize the probability of misclassification:

g=w0(μ0-μ)^2+w1(μ1-μ)^2

Among them, for the image I(x, y), the segmentation threshold of the foreground (and the target) and the background is denoted as T, the proportion of the pixels belonging to the foreground to the entire image is denoted as w0, and the average gray level μ0; the number of background pixels accounts for The scale of the whole image is w1, its average gray level is μ1, the total average gray level of the image is recorded as μ, and the inter-class variance is recorded as g.

Step 8: Obtain the gradient image of the image obtained in step 7, and perform watershed transformation on the gradient magnitude of the obtained gradient image.

Step 9: Perform the reconstruction-based open-close operation on the grayscale image obtained in Step 8 to remove debris in the image, and then extract the cytoplasm of the adherent cells to achieve the segmentation of white blood cells.

Claims (10)

1. A method for locating white blood cells in a lesion cell image is characterized by comprising the following steps:

(1) acquiring 100 white blood cell image data sets, selecting the image data sets mixed with background, white blood cell plasma, red blood cells and white cell nuclei at the same time, and converting the acquired microscopic images into gray images from color images;

(2) carrying out image enhancement on the acquired gray level image for highlighting the characteristics of the image;

(3) separating the cell nucleus of the white blood cell from the obtained image after the image enhancement by using a threshold segmentation method to realize the primary positioning of the white blood cell;

(4) removing isolated points for obtaining the image of the cell nucleus of the white blood cell;

(5) removing the independent area of the isolated point image of the cell nucleus image so as to obtain the complete cell nucleus of the white cell, wherein the steps are simultaneously suitable for the process of extracting the cell nucleus from the adherent cell and the non-adherent cell;

(6) for the microscopic image without the adherent cells, directly segmenting the white cell pulp by using a threshold segmentation method according to the primary positioning of white cell nuclei and the maximum cell area;

(7) for the microscopic image of the adhesion cells, dividing the image into a background part and a target part according to the gray characteristic of the image, so that the probability of wrong division is minimum;

(8) solving a gradient image of the image obtained in the step 7, and performing watershed transformation on the gradient amplitude of the obtained gradient image;

(9) and (3) carrying out reconstruction-based opening-closing operation on the gray level image obtained in the step (8), so that the fragment removing effect in the image is better on the premise of not influencing the target image, and then the cytoplasm with adhered cells is extracted, and the segmentation of white blood cells is realized.

2. The method of claim 1, wherein the method comprises: the step (1) is specifically as follows:

and (1.1) converting the acquired white blood cell image into a gray image by using a color conversion mode.

3. The method of claim 1, wherein the method comprises: the step (2) is specifically as follows:

(2.1) the method of obtaining the gray-scale image in the step (1) uses a linear transformation to enhance the image by adopting the following formula for highlighting the characteristics of the image:

wherein whenWhen, the output image contrast is increased; when in useWhen, it represents a decrease in output image contrast; when in useIn general, linear transformation is to stretch a small gray scale range to a large gray scale range, and is often called gray scale stretching, where the gray scale variation range of an image is set to beBy a functionExtend the range toThen, linear transformation:

i.e. the transformation function is passed through two pointsAndthe equation of the straight line of (c).

4. The method of claim 1, wherein the method comprises: the step (3) is specifically as follows:

(3.1) selecting a proper threshold value from the image obtained in the step (2) after image enhancement, and separating the cell nucleus of the white blood cell by using a threshold value segmentation-based method and using the following formula to realize the primary positioning of the white blood cell:

wherein, representing a threshold, G (x, y) represents the original gray value of the pixel (x, y), G (x, y) represents the gray value of the pixel (x, y) after transformation (after enhancement), the threshold is used as a threshold for distinguishing the target from the background pixel during segmentation, the pixel larger than or equal to the threshold belongs to the target, and the other pixels belong to the background, or vice versa.

5. The method of claim 1, wherein the method comprises: the step (4) is specifically as follows:

and (4.1) removing isolated points of the cell nucleus image by using a method of open operation reconstruction of the cell nucleus extracted in the step (3).

6. The method of claim 1, wherein the method comprises: the step (5) is specifically as follows:

(5.1) carrying out expansion processing on the isolated point removed image obtained in the step (4) to remove the irrelevant area by using the following formula:

in which, in the representation of swelling,representing an empty set, B being a structural element.

7. The method of claim 1, wherein the method comprises: the step (6) is specifically as follows:

(6.1) when no white blood cells are adhered, the white blood cell pulp can be directly segmented by a threshold segmentation algorithm according to the preliminary location and the area maximization principle of cell nuclei.

8. The method of claim 1, wherein the method comprises: the step (7) is specifically as follows:

(7.1) for the microscopic image of the adherent cells, firstly, dividing the image into a background part and a target part according to the gray level characteristics of the image by using a maximum between-class variance algorithm and the following formula so as to minimize the probability of wrong division:

wherein, for image I (x, y), the segmentation threshold of foreground (and target) and background is marked as T, the proportion of the number of pixels belonging to foreground in the whole image is marked as w0, and the average gray level is mu 0; the proportion of the number of background pixels in the whole image is w1, the average gray scale is μ 1, the total average gray scale of the image is μ, and the inter-class variance is g.

9. The method of claim 1, wherein the method comprises: the step (8) is specifically as follows:

and (8.1) solving a gradient image from the image obtained in the step (7) by using Sobel, and performing watershed transformation on the gradient amplitude of the obtained gradient image.

10. The method of claim 1, wherein the method comprises: the step (9) is specifically as follows:

and (9.1) performing on-off operation based on reconstruction on the gray image obtained in the step (8), removing fragments in the image, further extracting cytoplasm with adhered cells, and realizing the segmentation of the white blood cells.