A method for determining renal interstitial inflammatory cells in images based on deep learning, a computer device, and a computer-readable storage medium

Through deep learning technology, the problem of low diagnostic efficiency and high false detection rate of renal interstitial inflammatory cells is solved through the use of Unnet network model and density clustering analysis, and more efficient and accurate detection of inflammatory cells and infiltration degree assessment are achieved.

CN114119537BActive Publication Date: 2024-11-26HANGZHOU YIPAI INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111405354.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-11-24
Publication Date
2024-11-26
Estimated Expiration
2041-11-24

AI Technical Summary

Technical Problem

In the prior art, the diagnostic efficiency of renal interstitial inflammatory cells is low, the false detection rate is high, and the cost is high, making it difficult to quickly and accurately determine the existence and infiltration of inflammatory cells.

Method used

Using deep learning-based image processing method, by obtaining pathological images, making masks and cutting them into small pieces, segmenting them using the Unet network model, combining density clustering and convex hull analysis, the area score of inflammatory cells is calculated, and different labels are divided. area to determine the degree of infiltration of inflammatory cells.

Benefits of technology

It improves the detection accuracy and efficiency of renal interstitial inflammatory cells, provides a more reliable diagnostic basis, and has better robustness and generalization capabilities compared with traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114119537B_ABST
    Figure CN114119537B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for determining renal interstitial inflammatory cells in images based on deep learning, a computer device and a computer-readable storage medium. The method of the present invention can be used to determine renal interstitial inflammatory cells, the area and degree of inflammatory cell infiltration, and provide a basis for subsequent diagnosis and treatment. The method proposed by the present invention can also be used to solve the problem of poor timeliness of current traditional image processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention specifically relates to a method for determining renal interstitial inflammatory cells in an image based on deep learning, a computer device, and a computer-readable storage medium. Background Art

[0002] Renal interstitium is the connective tissue between the nephron and the collecting duct, which is composed of interstitial cells, reticular fibers, a small amount of elastic fibers, fibroblasts and macrophages. Damage to the renal interstitium due to inflammation, genetics or other diseases can affect the renal tubules, renal pelvis and calyces, causing urination disorders and electrolyte and acid-base disorders.

[0003] In the prior art, the determination is mainly made through practical experience, which has a low diagnostic efficiency, a high false positive rate, and a high cost. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method for determining renal interstitial inflammatory cells in an image based on deep learning, a computer device and a computer-readable storage medium.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for determining renal interstitial inflammatory cells in an image based on deep learning, comprising the following steps:

[0007] (1) Obtain a pathological strip image, and then create a mask based on the known contours of the glomerulus, renal blood vessels, and renal tubules and the outer contour of the pathological strip, fill the outer contour of the pathological strip with 1, and fill the contours of the glomerulus, renal blood vessels, and renal tubules with 0;

[0008] (2) Cut the image into small pieces according to the obtained mask to obtain an image list;

[0009] (3) Input the image list into the trained segmentation model in batches according to the required batch_size to obtain the segmented results;

[0010] (4) According to the coordinate position of the cropping window, the segmented masks are spliced ​​into a large image; the third channel of the mask is the area of ​​inflammatory cells, and then the positions of non-zero elements in the channel are obtained, the values ​​of the non-zero element positions are added up, and the average is obtained to obtain c1, and then c1 is normalized to obtain score = c1 / 255; in this way, the score and cell outline of each inflammatory cell are obtained.

[0011] Furthermore, it also includes:

[0012] Step (5) obtains the center of gravity position of each inflammatory cell outline based on the obtained inflammatory cell outline, and then clusters the cells together at a radius of 50 using density clustering according to the center of gravity position, and then merges some similar small blocks at a radius of 100 to obtain a relatively large inflammatory cell area.

[0013] Furthermore, it also includes:

[0014] Step (6) find the convex hull of each inflammatory cell, count the area of ​​the inflammatory cells in the convex hull, and calculate the area score according to the formula score = 1.0*(area of ​​cells in the non-infiltrated area + area of ​​cells in the infiltrated area) / (area of ​​glomerulus, renal blood vessels, and renal tubules + area of ​​non-cellular area + area of ​​cells in the non-infiltrated area + area of ​​cells in the infiltrated area), and then divide the score into different label areas according to the threshold of [0.05, 0.1, 0.2, 0.3, 1.0] to obtain the area and degree of inflammatory cell infiltration.

[0015] Furthermore, in step (1), the outer contour of the pathological strip is extracted using the findContours function:

[0016] (1.1) First, convert the bgr format image into a grayscale image, then scale the pathological bar, obtain the pixel area with a threshold less than 200 and greater than 10 in the scaled image, convert it to uint8 format, and obtain the image mask;

[0017] (1.2) Then, the closing operation is used to fill the holes in the mask, and then the contour detection function is used to obtain the contour in the image. The contour whose contour area is larger than the set threshold is selected, and finally the outer contour of the pathological strip image is obtained.

[0018] Furthermore, in step (2), the boundary pixels in the cropping process are recorded as boundary_pixels, the image scaling size is img_scale, the scaled boundary pixel point bp = boundary_pixels / img_scale, and the window size is win_size = 640 / img_scale, which also maintains the same scaling ratio as the image; the width and height of the large image are filled with 2*bp size respectively, and then cropped with a window of win_size size, and the overlapping area overlap = 2*bp is cropped.

[0019] Furthermore, in step (4), the findContours function is used to detect the cell contours of all inflammatory cells.

[0020] Furthermore, in step (3), the segmentation model adopts the UNET network model, and the downsampling process of the UNET network model is replaced by EfficientNet_B3.

[0021] Furthermore, in step (3), the parameters set during the segmentation model training process are:

[0022] The number of learning rounds is 200, the initial learning rate is 0.01, the batch_size is 24; weight_decay=0.0001, and the optimizer used is Adam.

[0023] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for determining renal interstitial inflammatory cells in an image based on deep learning can be implemented.

[0024] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the steps in the above-mentioned method for determining renal interstitial inflammatory cells in an image based on deep learning.

[0025] The beneficial effects of the present invention are:

[0026] The method of the present invention can be used to determine the area and degree of renal interstitial inflammatory cells and inflammatory cell infiltration, providing a basis for subsequent diagnosis and treatment. The method proposed by the present invention can also be used to solve the problem of poor timeliness of current traditional image processing. The method based on deep learning in this application makes the model have better robustness and generalization ability, and can obtain better accuracy and improve detection efficiency compared with existing detection methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Flow chart of the method for determining renal interstitial inflammatory cell infiltration.

[0028] Figure 2 It is a schematic diagram of a computer device.

[0029] Figure 3 This is a schematic diagram of the image before and after processing in step (1) of Example 1, wherein (1) is the image of the pathological strip before processing, and (2) is the image obtained after processing.

[0030] Figure 4 This is an example of a small image obtained after cropping in Example 1, where (1) is the small image obtained by cropping, and (2) is the mask image of the small image output by the model.

[0031] Figure 5 It is the output mask map obtained by splicing all the segmented masks together in Example 1.

[0032] Figure 6Schematic diagram of the convex hull of the infiltration area in Example 1, wherein (1) is the convex hull and (2) is the area where inflammatory cells are gathered cut out from the original image.

[0033] Figure 7 This is the final result diagram obtained in Example 1. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings. It should be pointed out that the specific implementation methods are only detailed descriptions of the present invention and should not be regarded as limitations of the present invention.

[0035] A method for determining renal interstitial inflammatory cells in an image based on deep learning, comprising the following steps:

[0036] (1) Obtain a pathological strip image, and then create a mask based on the known contours of the glomerulus, renal blood vessels, and renal tubules and the outer contour of the pathological strip, fill the outer contour of the pathological strip with 1, and fill the contours of the glomerulus, renal blood vessels, and renal tubules with 0. The renal interstitial region can be obtained by step (1).

[0037] Normal images obtained from a pathological image acquisition device (such as a pathological slice scanner) can be applicable to the method of the present invention.

[0038] The pathological strip refers to a valid pathological strip, that is, a pathological strip including a cortical region, and the cortical region includes renal glomeruli, renal blood vessels, and renal tubules.

[0039] The detection of the outer contour of the pathological strip is extracted using opencv:

[0040] (1.1) First, convert the bgr format image into a grayscale image, then scale the pathological bar (generally 1 / 4 or 1 / 8), obtain the pixel area with a threshold less than 200 and greater than 10 in the scaled image, convert it to uint8 format, and obtain the image mask.

[0041] (1.2) Then use the closing operation to fill the holes in the mask, and then use the contour detection function to obtain the contour in the image, select the contour whose area is larger than the set threshold, and finally obtain the outer contour of the pathological strip image. Because the image may have blank spaces when it is converted into a binary image after grayscale conversion, the closing operation function will fill these blank spaces, so that a complete binary image can be obtained, and the error is small when extracting the contour.

[0042] Specifically, the contour detection calls the findContours function defined in opencv, which can detect the outermost contour of the image, and can also detect all contours, establish a hierarchical relationship and is easy to use. In this embodiment, the findContours function is called to detect the outer contour.

[0043] Here, the threshold is set based on the area of ​​the largest contour obtained multiplied by 0.5. However, this coefficient can be appropriately adjusted according to actual conditions.

[0044] (2) The image is cut into small pieces according to the obtained mask. The boundary pixels in the cutting process are recorded as boundary_pixels, the image scaling size is img_scale, the scaled boundary pixel points bp = boundary_pixels / img_scale, and the window size is win_size = 640 / img_scale, which also maintains the same scaling ratio as the image. The width and height of the large image (i.e., the image before cutting) are filled with 2*bp size respectively, and then cut with a win_size window, and the overlapping area overlap = 2*bp is cut to obtain the image list. The picture list is an array with a dimension of (N, C, H, W), where N represents the number of pictures, C represents the channel of the picture, and H and W represent the height and width of the picture respectively, in a format such as np.array([[[[100,255,……]],[[255,120,…]],[[100,90,……]]],[[[0,255,...]],[[3,200,……]],[[4,120,……]]],…]).

[0045] Here, the large image refers to the pathological strip image; the image is generally large, tens of millions of pixels or even larger, and the large image must be cut into small pieces during processing. The window and overlapping area for cropping the image can be set according to the specific task.

[0046] (3) Input the pictures into the trained segmentation model in batches according to the required batch_size, and obtain the segmented results. The batch_size is set according to the maximum batch_size that the network can accept; in some embodiments, the batch_size can be set to 24 or 32, and in other embodiments, it can also be set to other values, which are not limited by this application. Batch processing of pictures has faster processing speed and higher work efficiency; for example, if 32 pictures are input at a time, the results of 32 pictures will be obtained. If the pictures are input separately, it will take 32 times to get the results of 32 pictures.

[0047] The training of the segmentation model mainly includes the dataset construction stage and the model training stage.

[0048] The dataset for the input model is the images in the image list mentioned above. The training phase of the model uses the Unet network model. The downsampling process of the Unet network model is replaced by efficientnet_b3, which makes it more effective to extract edge features and semantic features in the image. The Unet network model has a relatively simple structure, with a low number of parameters and calculations, and is more suitable for this task.

[0049] During the training process, the input to the model is an array image of N, C, H, W dimensions, and the output of the model is a score map of N, C, H, W (i.e., a predicted mask map).

[0050] In this embodiment, the parameters that need to be set during the model training process are:

[0051] The number of learning rounds is 200, the initial learning rate is 0.01, the batch_size is 24; weight_decay=0.0001, the optimizer used is Adam, and the model is trained until convergence;

[0052] (4) According to the coordinate position of the cropping window, the segmented masks are stitched into a large image.

[0053] Because the coordinates of the upper left corner and the lower right corner are obtained when a large image is cut into small images in step (2), the segmented mask is spliced ​​according to the coordinates of the upper left corner and the lower right corner of the current image.

[0054] The third channel of the mask is the area of ​​inflammatory cells. Then, the position of the non-zero elements in the channel is obtained, and the values ​​of the non-zero element positions are added up and the average value is obtained to obtain c1. Then, c1 is normalized to obtain score = c1 / 255. In this way, the score and cell contour of each inflammatory cell can be obtained. The contour of the inflammatory cell is obtained based on the mask image. Each cell is represented by two values ​​0 and 1 in the mask. The findContours function can be used to detect all cell contours.

[0055] (5) Based on the obtained inflammatory cell contours, the centroid position of each inflammatory cell contour is obtained, and then the cells are clustered together using density clustering at a radius of 50 according to the centroid position, and then some similar small blocks are merged at a radius of 100 to obtain a relatively large inflammatory cell area;

[0056] The conventional density clustering method in the prior art is adopted here, and the present invention does not improve it.

[0057] The radius 50,100 is not the radius described in conventional mathematics. It represents the maximum distance between two samples here, which can be understood as the distance between pixels. The radius refers to a relative relationship, which is used as a parameter of the density clustering function. In this embodiment, the radius 50,100 is selected; mainly to merge some nearby inflammatory cells together first, and to think that such inflammatory cells are the same within a distance of 50, so they are merged (i.e., clustered). Then a larger distance of 100 is selected, and clustering is performed again, so that the small inflammatory cell area is clustered into a larger block, which is convenient for counting the area of ​​inflammatory cells, and then to determine whether inflammatory cells are infiltrated. In this embodiment, the area of ​​inflammatory cells obtained by clustering after selecting a radius of 50,100 is more in line with the situation in clinical diagnosis. In practice, it can be selected according to the actual situation.

[0058] (6) Finally, the convex hull of each inflammatory cell is calculated, and the area of ​​the inflammatory cells in the convex hull is counted. The area score is calculated according to the formula score = 1.0*(area of ​​inflammatory cells in non-infiltrated area + area of ​​inflammatory cells in infiltrated area) / (area of ​​glomerulus, renal blood vessels, and renal tubules + area of ​​non-cellular area + area of ​​inflammatory cells in non-infiltrated area + area of ​​inflammatory cells in infiltrated area). The score is then divided into different label areas according to the threshold, thereby obtaining the area and degree of inflammatory cell infiltration.

[0059] The infiltration area refers to the area where inflammatory cells gather; the sum of the inflammatory cell area of ​​the non-infiltration area and the inflammatory cell area of ​​the infiltration area is equal to the cell area of ​​the inflammatory cell area obtained in step (5). The area of ​​glomerulus, renal blood vessels, and renal tubules + the area of ​​non-cell area + the cell area of ​​non-infiltration area + the cell area of ​​infiltration area = the outer contour area of ​​the entire pathological strip. The area calculation can be obtained by using the opencv function contourArea().

[0060] In some preferred methods, the convex hull can be obtained by using the convexHull function in opencv.

[0061] Example 1

[0062] Specifically, first obtain the renal pathology image of patient Zhang San, such as Figure 3 As shown in (1), the contours of glomeruli, renal tubules, and renal blood vessels are obtained (when judging renal interstitial inflammatory cells, the contours of glomeruli, renal tubules, and renal blood vessels are known and have been saved in the json file. Here, we only need to read the json file of the corresponding pathology strip). Then, the outer contour of the image is obtained, and a mask image with the same size as the pathology strip image is generated based on the outer contour. The outer contour part of the mask is filled with 1, and the positions of glomeruli, renal blood vessels, and renal tubules are filled with 0. In this way, the final mask image of the pathology image is obtained, as shown in Figure 3 (2) as shown.

[0063] Then the pathological image is cropped according to the preset window size to obtain the input image list of the segmentation model. The cropped image list is batch-input according to the batch_size required by the segmentation model. After the input, the model prediction result is obtained, and then the prediction result is post-processed. During the post-processing process, the predicted mask image is placed on a sheet with the same size as the image according to the coordinates of the upper left and lower right corners of the image obtained during cropping. Figure 1 The mask image is the same as the one above (here is the process of merging small images into a large image).

[0064] In this embodiment, the preset window size is 800*800. In other embodiments, preset windows of other sizes may be used, such as 1024*1024, which is determined according to the size of the picture finally input to the network model. If the size of the picture input to the network is 512*512, then both window sizes can be used. If the input picture is smaller than 512, then a window of 800*800 is used. In addition, the size of this window should also be cut to the renal interstitial part as much as possible. Generally, the overall pathological picture is first reduced by 1 / 4 or 1 / 8, and then a window of 1024*1024 or 800*800 is selected, or a larger window such as 2048*2048 is selected.

[0065] Finally, the predicted mask of the large image is obtained, such as Figure 5 As shown; the contours of the inflammatory cells in the renal interstitial area are further obtained. According to the predicted mask map, the contours of the inflammatory cells can be obtained using the findContours function of opencv.

[0066] After obtaining the outline of the inflammatory cells, the inflammatory cells were first merged at a radius of 50 to obtain a small inflammatory cell area. Then, the small inflammatory cell areas were merged at a radius of 100 to obtain a relatively large inflammatory cell area.

[0067] Find the convex hull of each inflammatory cell, such as Figure 6 As shown, Figure 6 (1) is an example of a convex hull (in some embodiments, there may be many convex hulls). The area of ​​the cells in the convex hull is counted, and then according to the formula score = 1.0*(area of ​​cells in non-infiltrated area + area of ​​cells in infiltrated area) / (area of ​​glomerulus, renal blood vessels, and renal tubules + area of ​​non-cellular area + area of ​​cells in non-infiltrated area + area of ​​cells in infiltrated area), according to the threshold [0.05, 0.1, 0.2, 0.3, 1.0], the degree of inflammatory cell infiltration can be obtained, which is divided into no inflammatory cell infiltration, focal, multifocal, large sheets, and diffuse.

[0068] The threshold value can be obtained by clustering using a conventional kmeans algorithm.

[0069] In this embodiment, if the score is less than 0.05, it is judged that there is no inflammatory cell infiltration; if the score is greater than or equal to 0.05 and less than 0.1, it is focal inflammatory cell infiltration; if the score is greater than or equal to 0.1 and less than 0.2, it is multifocal inflammatory cell infiltration; if the score is greater than or equal to 0.2 and less than 0.3, it is large-scale inflammatory cell infiltration; if the score is greater than or equal to 0.3 and less than 1.0, it is diffuse inflammatory cell infiltration.

[0070] The threshold selected in this embodiment is a relatively appropriate threshold at present. The score is compared with the threshold to determine the degree of inflammatory cell infiltration (no infiltration, focal, multifocal, large sheets, diffuse).

[0071] In this embodiment, the method of the present invention is used to process specific pictures, and the average processing time for each picture is 0.43 ms, with an accuracy rate of 98.52%. Figure 7 Here is an example of the final image: Figure 7 In the picture, we can see the severity of inflammatory cell infiltration; the bright red dots represent the areas of inflammatory cell infiltration, from which we can see which areas have more serious inflammatory cell infiltration (darker areas indicate more serious inflammatory cell infiltration) and which areas have less serious inflammatory cell infiltration.

[0072] Example 2, refer to the attached Figure 2 .

[0073] In this embodiment, a computer device 100 is provided, including a memory 102, a processor 101, and a computer program 103 stored in the memory 102 and executable on the processor 101. When the processor 101 executes the computer program 103, the steps of a method for determining renal interstitial inflammatory cells in an image based on deep learning provided in the above-mentioned embodiment 1 can be implemented.

[0074] Example 3

[0075] In this embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for determining renal interstitial inflammatory cells in an image based on deep learning provided in the above-mentioned embodiments can be implemented.

[0076] In this embodiment, the computer program may be the computer program in Embodiment 2.

[0077] In this embodiment, the computer-readable storage medium can be executed by the computer device in Embodiment 2.

[0078] Those skilled in the art can understand that all or part of the processes in the above embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0079] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of protection. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the scope of protection of the present invention.

Claims

1. A method for determining renal interstitial inflammatory cells in images based on deep learning, characterized in that: The following steps are involved: (1) Obtain a pathological strip image, and then create a mask based on the known contours of the glomerulus, renal blood vessels, and renal tubules and the outer contour of the pathological strip, fill the outer contour of the pathological strip with 1, and fill the contours of the glomerulus, renal blood vessels, and renal tubules with 0; (2) Cut the image into small pieces according to the obtained mask to obtain an image list; (3) Input the image list into the trained segmentation model in batches according to the required batch_size to obtain the segmented results; (4) According to the coordinate position of the cropping window, the segmented masks are spliced ​​into a large image; the third channel of the mask is the area of ​​inflammatory cells, and then the positions of non-zero elements in the channel are obtained, the values ​​of the non-zero element positions are added up, and the average is obtained to obtain c1, and then c1 is normalized to obtain score = c1 / 255; in this way, the score and cell outline of each inflammatory cell are obtained.

2. The method for determining renal interstitial inflammatory cells in an image based on deep learning according to claim 1, characterized in that: include: Step (5) obtains the center of gravity position of each inflammatory cell outline based on the obtained inflammatory cell outline, and then clusters the cells together at a radius of 50 using density clustering according to the center of gravity position, and then merges some similar small blocks at a radius of 100 to obtain a relatively large inflammatory cell area.

3. The method for determining renal interstitial inflammatory cells in an image based on deep learning according to claim 1, characterized in that: include: Step (6) find the convex hull of each inflammatory cell, count the area of ​​the inflammatory cells in the convex hull, and calculate the area score according to the formula score = 1.0*(area of ​​cells in the non-infiltrated area + area of ​​cells in the infiltrated area) / (area of ​​glomerulus, renal blood vessels, and renal tubules + area of ​​non-cellular area + area of ​​cells in the non-infiltrated area + area of ​​cells in the infiltrated area), and then divide the score into different label areas according to the threshold of [0.05, 0.1, 0.2, 0.3, 1.0] to obtain the area and degree of inflammatory cell infiltration.

4. The method for determining renal interstitial inflammatory cells in an image based on deep learning according to claim 1, characterized in that: In step (1), the outer contour of the pathological strip is extracted using the findContours function: (1.1) First, convert the bgr format image into a grayscale image, then scale the pathological bar, obtain the pixel area with a threshold less than 200 and greater than 10 in the scaled image, convert it to uint8 format, and obtain the image mask; (1.2) Then, the closing operation is used to fill the holes in the mask, and then the contour detection function is used to obtain the contour in the image. The contour whose contour area is larger than the set threshold is selected, and finally the outer contour of the pathological strip image is obtained.

5. The method for determining renal interstitial inflammatory cells in an image based on deep learning according to claim 1, characterized in that: In step (2), the boundary pixels in the cropping process are recorded as boundary_pixels, the image scaling size is img_scale, the scaled boundary pixel point bp = boundary_pixels / img_scale, and the window size is win_size = 640 / img_scale, which also maintains the same scaling ratio as the image; the width and height of the large image are filled with 2*bp size respectively, and then cropped with a window of win_size size, and the overlapping area overlap = 2*bp is cropped.

6. The method for determining renal interstitial inflammatory cells in an image based on deep learning according to claim 1, characterized in that: In step (4), the findContours function is used to detect the cell contours of all inflammatory cells.

7. The method for determining renal interstitial inflammatory cells in an image based on deep learning according to claim 1, characterized in that: In step (3), the segmentation model uses the UNET network model, and the downsampling process of the UNET network model is replaced by EfficientNet_B3.

8. The method for determining renal interstitial inflammatory cells in an image based on deep learning according to claim 1, characterized in that: In step (3), the parameters set during the segmentation model training process are: The number of learning rounds is 200, the initial learning rate is 0.01, the batch_size is 24; weight_decay=0.0001, and the optimizer used is Adam.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, it can implement the steps in the method for determining renal interstitial inflammatory cells in an image based on deep learning as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it can implement the steps in the method for determining renal interstitial inflammatory cells in an image based on deep learning as described in any one of claims 1 to 8.