An analytical method for automatically identifying fat vacuoles in bone marrow sections

Abstract

The present invention relates to an analysis method for automatically identifying fat vesicles in bone marrow slices, comprising the steps of reading bone marrow cell images, calculating image histograms, histogram equalization, image filtering, image threshold segmentation, hole seed filling, and fat vesicle statistical analysis. In the fat vesicle extraction step, the present invention extracts fat vesicles through two processes, firstly, the first threshold processing is performed on the whole image to obtain a preliminary target, and the second time, the fat vesicle boundary is determined by hole filling and gradient extraction. Experiments have shown that the method can also complete the fat vesicle extraction work well under complex backgrounds. Compared with traditional manual recognition, a large amount of workload and recognition difficulty are saved, and the accuracy and consistency of recognition are guaranteed.

Description

Technical Field

[0001] The embodiments of the present invention relate to a computer analysis method, and more particularly to an analysis method for automatically identifying fat vesicles in bone marrow slices. Background Art

[0002] Tumor drug resistance is a bottleneck problem that has yet to be overcome in the treatment of acute leukemia. Abnormal body fat metabolism is closely related to leukemia chemotherapy resistance. However, little is known about how the in situ adipocytes, which are an important component of the bone marrow microenvironment and are significantly reduced in the bone marrow of acute myeloid leukemia (AML), play a pathological role. Adipocytes, as components of the bone marrow stroma, occupy about 20%-40% of the normal bone marrow space, and together with blood vessels, endosteum and reticular fiber scaffolds, constitute the hematopoietic microenvironment. A large number of studies have shown that the participation of peripheral adipose tissue in leukemia cell chemotherapy resistance further affects the prognosis of leukemia patients. As an important component of the bone marrow microenvironment, in situ adipocytes in the bone marrow cavity provide more direct protection against chemotherapy. Analysis of pathological sections in recent years has found that the number of adipocytes in the bone marrow is significantly reduced when AML develops, and they are mainly small adipocytes. However, the current counting of adipocytes is still at the level of manual counting. For example, Figure 1 The manual counting method shown in the figure uses an eyepiece micrometer grid for positioning and calculation. Manual counting involves a large amount of workload and difficulty in recognition. The accuracy and consistency of recognition are difficult to guarantee, and an analysis method based on computer technology is needed. Summary of the invention

[0003] The purpose of the embodiments of the present invention is to provide an analysis method for automatically identifying fat vesicles in bone marrow slices based on computer technology, mainly to solve the technical problems that the counting of fat cells in the prior art is still at the level of manual counting, manual counting has a large workload and recognition difficulty, and the accuracy and consistency of recognition are difficult to guarantee.

[0004] In order to achieve the above object, the embodiment of the present invention designs an analysis method for automatically identifying fat vesicles in bone marrow sections, characterized in that it comprises the following steps:

[0005] Step S10: reading a bone marrow cell image, transmitting an image of a bone marrow slice from a B-ultrasound or MRI device; transmitting the bone marrow cell image to an image storage device in a cavity filling system that automatically identifies fat bubble images of bone marrow slices; and proceeding to step S20;

[0006] Step S20: Calculating the image histogram: sending the bone marrow cell image in step S10 to the image histogram recognition device for image histogram conversion to form the image histogram of the bone marrow cell image, and performing image histogram calculation on the image histogram of the bone marrow cell image by RGB three-color averaging to form the image histogram of the bone marrow cell image after three-color averaging, and saving the image histogram of the bone marrow cell image after three-color averaging to the image memory, and then proceeding to step S30;

[0007] Step S30: Histogram equalization: In the image histogram recognition device, the image histogram of the bone marrow cell image after the three-color averaging described in step 20 is counted by the number of gray levels, and 256 gray levels are set; the number of corresponding gray value pixels is increased by 1; the gray distribution density is calculated; the cumulative histogram distribution is calculated; the cumulative distribution is rounded; the gray value is equalized; and the image histogram of the bone marrow cell image after equalization is saved in the image memory, and then the process proceeds to step S40;

[0008] Step S40: Image filtering: In the image histogram recognition device, a local threshold is calculated for the bone marrow cell image after being equalized in step 30, and the image histogram of the bone marrow cell image that is less than the local threshold is deleted to form a filtered bone marrow cell image, which is then stored in the image memory and the process proceeds to step S50;

[0009] Step S50: Image threshold segmentation: In the image histogram recognition device, the grayscale values ​​of adjacent pixels in the target area or the background are equalized by the threshold segmentation method, and the grayscales of pixels in different target areas or backgrounds are unequal. Different target areas and backgrounds correspond to different peaks on the image histogram. The selected threshold is located at the valley between the two peaks to separate the peaks, forming the bone marrow cell image after the image threshold segmentation, which is saved in the image memory and then goes to step S60;

[0010] Step S60: Hole seed filling: In the image histogram recognition device, it includes access coordinates and mark coordinates. When the pixel point is greater than the threshold, the mark coordinates are set. When the pixel point is less than the threshold, the mark coordinates are also set until all non-hole pixel points in the area to be marked are found and the mark coordinates are set. When all non-hole pixel points are accessed and all points within the threshold are marked, the areas of all connected non-marked areas are counted; and the gradient is used to extract the boundaries of fat vesicles until the non-connected areas inside the non-cells are distinguished and the boundaries of the non-connected areas inside the non-cells are divided out; the non-connected areas less than the threshold are stored in an array, and the pixels with the mark coordinates in the array are restored to the background color; after being restored to the background color, the Determine the first current point on the bone marrow cell image, use the Sobel template to search the bone marrow cell image after the first current point is restored to the background color, use the horizontal and vertical templates to perform weighted calculations on at least 9 points around the current point, and finally accumulate the absolute values ​​of the gradients of the current point in the x and y directions; compare the result values ​​of the absolute value sum 255*8, and determine the current point greater than the result value 255*8 as a boundary point, otherwise take the average value of the surrounding pixels, for the small non-connected area in the array, the gradient value of the pixel is greater than the threshold, then the marking coordinate fails, and the background color value is returned; form the bone marrow cell image after the hole seed is filled, save it to the image memory, and enter step S70;

[0011] Step S70: Statistical analysis of fat bubbles. In the cavity filling system for automatically identifying fat bubble images of bone marrow slices, the bone marrow cell image after the hole seed filling in step S60 is called, and the mark coordinates set on the bone marrow cell image after the hole seed filling and the access coordinates are found, and are set as the points to be counted, and a recursive function is called; and it is judged that the point to be counted is not a boundary point, the fat bubble area is incremented by 1, and the point to be counted is set as the mark access point; the recursive function is used to process the four directions of the point to be counted in the up, down, left and right directions until it is a boundary point; the number of fat bubbles is incremented by 1; until the bone marrow cell image after the entire hole seed filling is completed, the diameter of the fat bubble is calculated, and the statistical analysis of the fat bubble is completed.

[0012] Furthermore, the diameter of the fat bubble is calculated by using the minimum circumscribed circle method as a measurement method to calculate the roundness, and at the same time the diameter of the fat bubble is measured.

[0013] Furthermore, the cavity filling system for automatically identifying fat bubble images of bone marrow slices has a computer system with Graph PAD software and Mann-Whitney U algorithm.

[0014] Furthermore, the image histogram recognition device is a computer image histogram recognition system with an untitled function.

[0015] Furthermore, the image storage is a hard disk or a server with a hard disk storage function.

[0016] Furthermore, the threshold segmentation method is a single threshold segmentation method or a multi-threshold segmentation method.

[0017] Furthermore, the single threshold segmentation method divides the filtered bone marrow cell image in step S40 into a target image and a background image.

[0018] Furthermore, the multi-threshold segmentation method is directed to multiple threshold segmentation, and the filtered bone marrow cell image in step S40 is segmented into multiple target regions and backgrounds, and each of the target regions is marked.

[0019] Furthermore, the Sobel template is a Pascal triangle Sobel template.

[0020] Furthermore, the analysis method for automatically identifying fat vesicles in bone marrow sections is applied to the statistical analysis of the number of fat vesicles, the diameter of fat vesicles and the area of ​​a single fat vesicle in bone marrow sections of acute myeloid leukemia.

[0021] Compared with the prior art, the present invention extracts fat bubbles through two processes in the fat bubble extraction step. First, the whole image is subjected to the first threshold processing to obtain the preliminary target, and the second time, the fat bubble boundary is determined by hole filling and gradient extraction. Experiments have shown that this method can also complete the fat bubble extraction work well under complex backgrounds. Compared with traditional manual recognition, a lot of workload and recognition difficulty are saved, and the accuracy and consistency of recognition are guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The counting method in the prior art of the present invention;

[0023] Figure 2 Schematic diagram of the analytical method of the present invention;

[0024] Figure 3 A schematic diagram of the threshold segmentation of the present invention;

[0025] Figure 4 is a schematic diagram of cavity filling of a fat bubble image of the present invention; DETAILED DESCRIPTION

[0026] In order to make the purpose, technical scheme and advantages of the present invention clearer, the following will be described in detail with reference to the accompanying drawings. However, it will be appreciated by those skilled in the art that in the various embodiments of the present invention, many technical details are provided in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed for protection in the claims of the present application can be implemented.

[0027] The first embodiment of the present invention relates to an analysis method for automatically identifying fat vesicles in bone marrow sections, such as Figure 2 As shown, the following steps are included:

[0028] Step S10: read the bone marrow cell image, and transmit the image of the bone marrow slice from the B-ultrasound or MRI device; transmit the bone marrow cell image to the image memory of the cavity filling system that automatically identifies the fat bubble image of the bone marrow slice; and proceed to step S20;

[0029] Step S20: Calculating the image histogram: sending the bone marrow cell image in step S10 to the image histogram recognition device for image histogram conversion to form an image histogram of the bone marrow cell image, and performing image histogram calculation on the image histogram of the bone marrow cell image by RGB three-color averaging to form an image histogram of the bone marrow cell image after three-color averaging, and saving the image histogram of the bone marrow cell image after three-color averaging to the image memory, and then proceeding to step S30;

[0030] The display mode of the image histogram is dark on the left and bright on the right. The left side is used to describe the darkness of the image, and the right side is used to describe the brightness of the image. When the histogram displays the image data, it will be presented in the form of a distribution curve with dark on the left and bright on the right, rather than displaying the original image data. In addition, the image can be scaled down by an algorithm, and it has many advantages such as image translation, rotation, and scaling invariance. The histogram is less expensive when performing image calculations, so it is often used in image processing. The RGB three-color averaging, if you want them to average to less than 255, you have to weaken their values. If you process them one by one, it is obvious that the operation unit is not brought into full play. If the three RGB values ​​are added together and then the average value is calculated through a specific algorithm and then split, the calculation speed is obviously faster than a single one. Therefore, the RGB three-color averaging method is used to calculate the image histogram. Based on the above reasons, the RGB three-color averaging method is used in this embodiment to calculate the image histogram.

[0031] Step S30: histogram equalization: in the image histogram recognition device, the image histogram of the bone marrow cell image after the three-color averaging in step 20 is counted by the number of gray levels, and 256 gray levels are set; the number of pixels corresponding to the gray value is increased by 1; the gray distribution density is calculated; the cumulative histogram distribution is calculated; the cumulative distribution is rounded; the gray value is equalized; and the image histogram of the bone marrow cell image after the equalization is saved in the image memory, and then the process proceeds to step S40;

[0032] In this embodiment, according to the distribution density curve principle in probability theory, in order to facilitate digital image processing, a discrete form must be introduced. In the discrete form, represents the discrete gray level, represents, and the following formula is established:

[0033] p r (r k )=n k / n 0≤r k ≤1,k=0,1,2,…,L-1 (1)

[0034] Where is the number of pixels with this grayscale in the image, is the total number of pixels in the image, and is what is called frequency in probability theory. A graph of the relationship between and is made in a rectangular coordinate system. This graph is called a histogram. The grayscale level of a given image is distributed in the range. Any value in the interval [0,1] can be transformed as follows:

[0035] s=T(r) (2)

[0036] That is to say, through the above transformation, each pixel gray value of the original image generates a corresponding value. The transformation function should meet the following conditions:

[0037] 1) In the interval 0≤r≤1, it is a single value and monotonically increasing.

[0038] 2) For 0≤r≤1, 0≤T(r)≤1.

[0039] The first condition here ensures that the grayscale order from white to black remains unchanged. The second condition ensures that the pixel grayscale value after mapping is within the allowed range. The inverse transformation relationship from to can be expressed as:

[0040] r=T -1 (s) (3)

[0041] From probability theory, we know that if the probability density of a random variable is known to be , and the random variable is a function of , that is, the probability density of can be calculated by .

[0042] Because it is monotonically increasing, mathematical analysis shows that its inverse function is also a monotonous function. In this case, it occurs when and only when, so the distribution function of the random variable can be obtained as:

[0043]

[0044] By taking the derivative of both sides of the above formula, we can get the distribution density function of the random variable, which is:

[0045]

[0046] It can be seen that the probability density function of the gray level of the original image can be controlled by the transformation function, thereby improving the gray level of the original image. Histogram equalization is a histogram correction method based on the cumulative distribution function transformation method. Assume that the transformation function is:

[0047]

[0048] Where is the integral variable, and is the cumulative distribution function (CDF) of. Here, the cumulative distribution function is a function of, and increases monotonically from 0 to 1, so this transformation function satisfies the single-valued single-increasing in the interval 0≤r≤1, and for 0≤r≤1, there are two conditions of 0≤T(r)≤1.

[0049] Taking the derivative of formula (6), we get:

[0050]

[0051] Substituting the result into formula (7), we get:

[0052]

[0053] From the above derivation, it can be seen that the probability density in the domain of the transformed variable is uniformly distributed. It can be seen that using the cumulative distribution function as a transformation function can produce an image with a uniform probability density of grayscale distribution. As a result, the dynamic range of pixel values ​​is expanded.

[0054] The above correction method is discussed based on continuous random variables. In order to process digital images, discrete formulas must be introduced. When the gray level is a discrete value, the frequency can be used to approximate the probability value, that is,

[0055] p r (r k )=n k / n 0≤r k ≤1,k=0,1,2,…,L-1 (9)

[0056] Where is the total number of gray levels, is the probability of taking this gray level, is the number of times this gray level appears in the image, and is the total number of pixels in the image. The image enhancement technology to obtain a uniform histogram is usually called histogram equalization. The discrete form of formula (6) can be expressed as:

[0057]

[0058] Its inverse transformation is:

[0059] r k =T -1 (s k ) (11)

[0060] T is Figure 3 The cycle shown in , then, according to the above theory, an image histogram of the bone marrow cell image after equalization is formed.

[0061] Step S40: Image filtering: In the image histogram recognition device, a local threshold is calculated for the bone marrow cell image after being equalized in step 30, and the image histogram of the bone marrow cell image that is less than the local threshold is deleted to form a filtered bone marrow cell image, which is then stored in the image memory and enters step S50;

[0062] like Figure 3 As shown, step S50: image threshold segmentation: in the image histogram recognition device, the threshold segmentation method is a single threshold segmentation method or a multi-threshold segmentation method. The single threshold segmentation method divides the filtered bone marrow cell image described in step S40 into a target image and a background image. The multi-threshold segmentation method divides the filtered bone marrow cell image described in step S40 into multiple target areas and backgrounds for multiple threshold segmentation, and marks each of the target areas. The threshold segmentation method is used to make the grayscale values ​​between adjacent pixels in the target area or the background equal, and the pixels in different target areas or backgrounds are not equal in grayscale. On the image histogram, different target areas and backgrounds correspond to different peaks. The selected threshold is located at the valley between the two peaks to separate the peaks, forming a bone marrow cell image after image threshold segmentation, which is saved in the image memory and enters step S60;

[0063] like Figure 4As shown, step S60: hole seed 1 filling: in the image histogram recognition device, including access coordinates and mark coordinates, when the pixel point is greater than the threshold, the mark coordinates are set, and when the pixel point is less than the threshold, the mark coordinates are also set until all non-hole pixel points in the area to be marked are found and the mark coordinates are set. When all non-hole pixel points are accessed and all points within the threshold are marked, the areas of all connected non-marked areas are counted; and the gradient is used to extract the boundaries of fat vesicles until the non-connected areas inside the non-cells are distinguished and the boundaries of the non-connected areas inside the non-cells are divided out; the non-connected areas less than the threshold are stored in an array, and the pixels with mark coordinates in the array are restored to the background color; the first current point is determined on the bone marrow cell image after being restored to the background color, so that Using the Sobel template in this embodiment, the Sobel template is based on the Pascal triangle Sobel template. The Sobel template is used to search for the bone marrow cell image after being restored to the background color from the first current point, and weighted calculations are performed on at least 9 points around the current point using two templates in the horizontal direction and the vertical direction respectively. Finally, the sum of the absolute values ​​of the gradients in the x direction and the y direction of the current point is accumulated; the result value of the absolute value sum 255*8 is compared, and the current point greater than the result value of 255*8 is determined as a boundary point, otherwise the average value of the surrounding pixels is taken. For the small non-connected area in the array, if the gradient value of the pixel is greater than the threshold, the coordinate marking fails and the background color value is returned; the bone marrow cell image after the hole seed 1 is filled is formed, and is saved in the image memory, and the process goes to step S70;

[0064] Step S70: Statistical analysis of fat bubbles. In the cavity filling system for automatically identifying fat bubble images of bone marrow slices, the bone marrow cell image after the hole seed 1 is filled in step S60 is called, and the mark coordinates set and the access coordinates are not found on the bone marrow cell image after the hole seed 1 is filled, and the points to be counted are set, and the recursive function is called; and it is judged that the point to be counted is not a boundary point, the fat bubble area is incremented by 1, and the point to be counted is set as a mark access point; the recursive function is used to process the four directions of the point to be counted in the up, down, left and right directions until it is a boundary point; the number of fat bubbles is incremented by 1; until the bone marrow cell image after the entire hole seed 1 is filled is completed, the diameter of the fat bubble is calculated, and the statistical analysis of the fat bubble is completed.

[0065] In the above embodiment, fat bubble extraction is performed by two processes in the fat bubble extraction step. First, the whole image is processed by threshold processing to obtain the preliminary target, and the fat bubble boundary is determined by hole filling and gradient extraction. Experiments have shown that this method can also complete the fat bubble extraction work well under complex backgrounds. Compared with traditional manual recognition, it saves a lot of workload and recognition difficulty, and the accuracy and consistency of recognition are guaranteed.

[0066] In this embodiment, the diameter of the fat bubble is calculated by using the minimum circumscribed circle method as a measurement method to calculate the roundness, and at the same time, the diameter of the fat bubble is measured. The cavity filling system for automatically identifying bone marrow slice fat bubble images has a computer system with Graph PAD software and Mann-Whitney U algorithm; the image histogram recognition device is a computer image histogram recognition system with an untitled function, and the image storage is a hard disk or a server with a hard disk storage function. The above hardware equipment ensures that the embodiment of the present invention can be implemented.

[0067] In this embodiment, the analysis method for automatically identifying fat bubbles in bone marrow sections is applied to the statistical analysis of the number of fat bubbles, the diameter of fat bubbles and the area of ​​a single fat bubble in the bone marrow sections of acute myeloid leukemia. The analysis method of the present invention is used to collect 30 bone marrow and paraffin-embedded tissue samples of AML patients admitted to the applicant for initial and post-remission. All diagnoses meet the 1991 FAB classification standards, and chemotherapy adopts international standard programs. 30 normal subjects (including orthopedic hip replacement, anemia and other bone marrow diagnosis normal subjects) are used as controls. The collected AML and normal bone marrow paraffin tissue sections are routinely dewaxed, dehydrated, and stained for standby use. The above-mentioned analysis method for automatically identifying fat bubbles in bone marrow sections is used to count the number of fat cells, diameter and single fat cell area in the bone marrow sections of AML cases and normal control groups. According to the grouping of fat cell morphological characteristics, the correlation between the morphological characteristics of each group of fat cells and the patient's prognosis is further analyzed. The content and size of bone marrow fat cells between AML cases and normal groups are verified, and the abnormal cell differentiation of AML fat bubbles is verified. The Kaplan-Meier survival function was used to analyze the impact of the morphological characteristics of AML adipocytes on the survival of AML patients and to evaluate their significance in the prognosis of AML cases, both of which achieved good technical results.

[0068] Those skilled in the art will appreciate that the above-mentioned embodiments are specific examples for implementing the present invention, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method for automatically identifying fat vesicles in bone marrow sections, characterized in that: The following steps are involved: Step S10: reading bone marrow cell images, transmitting bone marrow slice images from B-ultrasound or MRI equipment; transmitting the bone marrow cell images to an image memory in the cavity filling system that automatically identifies bone marrow slice fat bubble images; proceeding to step S20; Step S20: Calculating the image histogram: sending the bone marrow cell image in step S10 to the image histogram recognition device for image histogram conversion to form the image histogram of the bone marrow cell image, and performing image histogram calculation on the image histogram of the bone marrow cell image by RGB three-color averaging to form the image histogram of the bone marrow cell image after three-color averaging, and saving the image histogram of the bone marrow cell image after three-color averaging to the image memory, and then proceeding to step S30; Step S30: Histogram equalization: In the image histogram recognition device, the image histogram of the bone marrow cell image after the three-color averaging described in step S20 is set to 256 gray levels by counting the number of gray levels; the number of pixels corresponding to the gray value is increased by 1; and the gray distribution density is calculated; Calculate the cumulative histogram distribution; round the cumulative distribution; equalize the gray value; and save the image histogram of the bone marrow cell image after equalization into the image memory, and enter step S40; Step S40: Image filtering: In the image histogram recognition device, a local threshold is calculated for the bone marrow cell image after being equalized in step S30, and the image histogram of the bone marrow cell image that is less than the local threshold is deleted to form a filtered bone marrow cell image, which is then stored in the image memory and the process proceeds to step S50; Step S50: Image threshold segmentation: In the image histogram recognition device, the grayscale values ​​of adjacent pixels in the target area or background are equalized by the threshold segmentation method, and the grayscale values ​​of pixels in different target areas or backgrounds are unequal. Different target areas and backgrounds correspond to different peaks on the image histogram. The selected threshold is located at the valley between the two peaks to separate the peaks, forming the bone marrow cell image after the image threshold segmentation, which is saved in the image memory and then goes to step S60; Step S60: hole seed filling: in the image histogram recognition device, including access coordinates and mark coordinates, when the pixel point is greater than the threshold, the mark coordinates are set, when the pixel point is less than the threshold, the mark coordinates are also set, until all non-hole pixel points in the marked area are found to set the mark coordinates, when all non-hole pixel points are accessed and all points within the threshold are marked, the areas of all connected non-marked areas are counted; and the fat bubble boundary is extracted by gradient until the non-connected area inside the non-cell is distinguished and the boundary of the non-connected area inside the non-cell is divided out; the non-connected area less than the threshold is stored in an array, and the pixels with the mark coordinates in the array are restored to the background color; Determine the first current point on the bone marrow cell image after being restored to the background color, use the Sobel template to search the bone marrow cell image after being restored to the background color from the first current point, use the horizontal and vertical templates to perform weighted calculations on at least 9 points around the current point, and finally accumulate the sum of the absolute values ​​of the gradients in the x and y directions of the current point; compare the sum of the absolute values ​​with the result value of 255 * 8, and determine the current point with a result value greater than 255 * 8 as a boundary point, otherwise take the average value of the surrounding pixels, for the small non-connected area in the array, if the gradient value of the pixel is greater than the threshold, the marking coordinate fails, and its background color value is returned; The bone marrow cell image after the hole seed filling is formed and saved in the image memory, and then the process goes to step S70; Step S70: Fat bubble statistical analysis, in which the bone marrow cell image after hole seed filling in step S60 is called in the hole filling system for automatically identifying bone marrow slice fat bubble images, the bone marrow cell image after hole seed filling is found on the bone marrow cell image with set marker coordinates and no access coordinates, and the coordinates are set as points to be counted, and a recursive function is called; It is determined that the point to be counted is not a boundary point, the area of ​​the fat bubble is incremented by 1, and the point to be counted is set as a landmark access point; a recursive function is used to process the point to be counted in four directions, up, down, left, and right, until it is a boundary point; The number of fat bubbles increases by 1; The bone marrow cell image is obtained after the entire hole is filled with seeds, the diameter of the fat bubble is calculated, and the statistical analysis of the fat bubble is completed.

2. The method for automatically identifying fat vesicles in bone marrow slices according to claim 1, characterized in that: The diameter of the fat bubble is calculated by using the minimum circumscribed circle method as a measurement method to calculate the roundness and simultaneously measure the diameter of the fat bubble.

3. The method for automatically identifying fat vesicles in bone marrow slices according to claim 1, characterized in that: The cavity filling system for automatically identifying fat bubble images of bone marrow slices has a computer system with Graph PAD software and Mann-Whitney U algorithm.

4. The method for automatically identifying fat vesicles in bone marrow slices according to claim 1, characterized in that: The image histogram recognition device is a computer image histogram recognition system with an untitled function.

5. The method for automatically identifying fat vesicles in bone marrow slices according to claim 1, characterized in that: The image storage device is a hard disk or a server with a hard disk storage function.

6. The method for automatically identifying fat vesicles in bone marrow slices according to claim 1, characterized in that: The threshold segmentation method is a single threshold segmentation method or a multi-threshold segmentation method.

7. The method for automatically identifying fat vesicles in bone marrow slices according to claim 6, characterized in that: The single threshold segmentation method divides the filtered bone marrow cell image in step S40 into a target image and a background image.

8. The method for automatically identifying fat vesicles in bone marrow slices according to claim 6, characterized in that: The multi-threshold segmentation method is directed to multiple threshold segmentation, and the filtered bone marrow cell image in step S40 is segmented into multiple target regions and backgrounds, and each of the target regions is marked.

9. The method for automatically identifying fat vesicles in bone marrow slices according to claim 1, characterized in that: The Sobel template is a Pascal triangle Sobel template.

10. The method for automatically identifying fat vesicles in bone marrow sections according to any one of claims 1 to 9, characterized in that: The analysis method for automatically identifying fat vesicles in bone marrow sections is applied to the statistical analysis of the number of fat vesicles, the diameter of fat vesicles and the area of ​​a single fat vesicle in bone marrow sections of acute myeloid leukemia.

Images