Artificial intelligence diagnosis assistance system and method based on cell morphology

Abstract

This solution belongs to the field of medical auxiliary diagnosis technology, and specifically relates to an artificial intelligence diagnosis assistance system and method based on cell morphology. The artificial intelligence diagnosis assistance method based on cell morphology includes the following steps: converting the image data into a digital image after amplifying it according to a preset multiple, and grayscale processing the digital image; identifying the red blood cells as the first cells based on the contour features of each cell in the digital image, identifying other cells as positioning objects, and treating the first cell whose contour is enclosed by another cell contour or encloses another cell contour as an overlapping cell, treating the first cell whose contour is connected to another cell contour as an overlapping cell, merging overlapping cells and cells with a closed enclosing relationship or a connected relationship with overlapping cells into cell clusters, and deleting the positioning object marks in the cell clusters. This solution solves the problem of low accuracy in artificial intelligence recognition and division of blood cells by cell morphology.

Description

Technical Field

[0001] This solution belongs to the field of medical auxiliary diagnosis technology, and specifically relates to an artificial intelligence diagnosis auxiliary system and method based on cell morphology. Background Art

[0002] Artificial intelligence diagnosis assistance method is a technical means that combines artificial intelligence technology with cell morphology to assist doctors in diagnosing diseases. In blood tests, artificial intelligence diagnosis assistance method of cell morphology can help doctors distinguish normal blood cells such as red blood cells, white blood cells (such as lymphocytes, granulocytes, etc.), platelets, and abnormal cells such as leukemia cells. Compared with the traditional method of distinguishing various types of cells by relying on doctors' naked eye observation, this method can help doctors quickly process a large number of cell images. Doctors make a comprehensive judgment on the morphological characteristics of red blood cells given by artificial intelligence and other clinical symptoms and examination results of patients, which can reduce the time doctors spend on observing cells, thereby greatly improving doctors' work efficiency.

[0003] At present, when artificial intelligence recognizes and divides blood cells by cell morphology, it usually identifies and marks red blood cells by extracting the edge, texture, and shape features of red blood cells in image data, and then compares the marked red blood cells with normal red blood cells to obtain the changes in the morphology of red blood cells in the image data. However, in most image data, there are multiple overlapping red blood cells. When identifying overlapping red blood cells, it is easy to identify red blood cells with a high degree of overlap as a single red blood cell, resulting in incorrect recognition of the shape of red blood cells, which affects the doctor's judgment of the severity of the disease. Summary of the invention

[0004] The purpose of this program is to provide an artificial intelligence diagnosis assistance system and method based on cell morphology to solve the problem of low accuracy of artificial intelligence in identifying and dividing blood cells by cell morphology.

[0005] In order to achieve the above objectives, this scheme provides an artificial intelligence diagnosis-assisted method based on cell morphology, comprising the following steps:

[0006] The image data is magnified according to a preset multiple and converted into a digital image, and the digital image is gray-scaled;

[0007] Based on the contour features of each cell in the digital image, a red blood cell is identified as a first cell, other cells are identified as positioning objects, a first cell whose contour is enclosed by another cell contour or encloses another cell contour is identified as an overlapping cell, a first cell whose contour is connected to another cell contour is identified as an overlapping cell, overlapping cells and cells that have a closed enclosing relationship or a connected relationship with the overlapping cells are merged into a cell cluster, a positioning object mark existing in the cell cluster is deleted, and the cell cluster is marked as a positioning object;

[0008] Dividing the digital image into a plurality of image blocks, each of which contains at least two positioning objects, and inferring the volume and morphology of the first cell according to the number of pixels and grayscale features of the first cell that does not belong to the positioning object in the image block;

[0009] Abnormal cells are marked from the first cells by the grayscale conditions and cell contours selected by the cell contour frame, and the diagnosis result is output according to the total number of abnormal cells or the proportion of abnormal cells in the first cells.

[0010] And, the principle and technical effect of the present scheme are: first, the present scheme amplifies the image data by a preset multiple and converts it into a digital image and then performs grayscale processing, thereby enhancing the contrast between the cell and the background, making the key morphological features such as the cell's outline and internal structure more prominent, laying a good foundation for the subsequent accurate identification of various types of cells, reducing recognition errors caused by poor initial image quality, and thus improving the overall recognition and classification accuracy.

[0011] Secondly, after preliminarily classifying red blood cells (first cells) and other cells (positioning objects) through cell contours, this scheme carefully sorts out the complex spatial relationship between cells through the contour relationship between the first cell and other first cells or positioning objects, and well copes with the overlap and aggregation of blood cells that may appear in the image, that is, clearly and accurately identifies cell clusters. In actual blood cell images, cell overlap (cell clusters) is a common factor that increases the difficulty of identification and division. Through such processing, the classification of cells in these special cases can be clearly defined, avoiding the misjudgment and omission of single cells caused by cell overlap, and improving the accuracy of cell morphology recognition and division.

[0012] Furthermore, the morphology of non-red blood cells (white blood cells, platelets, etc.) is different from that of red blood cells. The outer contours of cell clusters are random and different due to their overlapping conditions, resulting in the positioning objects having different morphological characteristics. When the digital image is cut into image blocks, the position of the image block in the digital image can be quickly and accurately determined according to the positions of the two positioning objects with different shapes in the digital image; at the same time, the relative positions of the two positioning objects and the first cells in the image block can be used to quickly distinguish and locate each first cell in the image block. Compared with the method of locating the first cell by relying on pixel points, the positioning method of this scheme only needs to refer to two relative positions, and does not need to search for pixel points one by one. This scheme can locate the first cell more quickly, thereby improving the division speed and efficiency.

[0013] In addition, by dividing the image blocks and ensuring a certain number of positioning objects in each image block, the model can quickly screen out those areas with high diagnostic value (i.e., image blocks containing more positioning objects) for key analysis during the analysis process. For example, in a blood sample, if there are a large number of positioning objects (such as white blood cells) or a high proportion of abnormal cell clusters in a certain image block, a detailed morphological analysis of this area can be prioritized to speed up the diagnosis. For example, in an inflammatory response, white blood cells may gather around red blood cells. This positioning method can well capture this cell aggregation phenomenon and provide an intuitive morphological basis for studying the mechanism of inflammation. For example, in some blood diseases, the formation of abnormal cell clusters may be a key diagnostic indicator. This positioning method can better help identify and study such cell clusters.

[0014] In summary, this solution solves the problem of low accuracy of artificial intelligence in identifying and dividing blood cells by cell morphology.

[0015] Furthermore, when identifying the first cell and locating the object, the Sobel operator or the Prewitt operator is used to calculate the horizontal and vertical grayscale gradients of each pixel in the image to detect edge contour features, wherein the gradient calculation formula (1) is as follows:

[0016] (1)

[0017] in, Used to calculate the horizontal gradient, It is used to calculate the gradient in the vertical direction, (i, j) is the position of a pixel point, when the gradient amplitude exceeds the preset threshold, the pixel point is marked as a pixel point located on the outline of the cell, and the closed outline is judged as a complete cell outline by the position of the pixel point on the cell outline; then the cell image dataset is annotated using the cell recognition model constructed by the convolutional neural network through the pixel points on the cell outline, and the cell image dataset is input into the cell recognition model for training, and the cell recognition model is used to annotate the red blood cells with complete cell outlines in the digital image as the first cell, and other cells with complete cell outlines are annotated as positioning objects.

[0018] The contour recognition method can not only accurately identify red blood cells with clear boundaries (can be observed normally), but also exclude broken red blood cells. It also helps to exclude incomplete contours caused by noise or partial cell overlap, and only conduct subsequent identification and analysis on cells with complete contours, thereby improving the accuracy of recognition.

[0019] Furthermore, pixel points on the edge contour of the first cell are obtained, a curve fitting algorithm is used to fit the shape of the first cell, the shape of the first cell is compared with a circle or an ellipse to obtain an approximation, and the first cell whose approximation is less than a preset approximation is judged as an abnormal cell; the grayscale value of each pixel point in the edge contour of the first cell is obtained, and based on constructing a basic plane parallel to the digital image, the standard three-dimensional model of the red blood cell and the edge contour of the first cell are combined to infer the posture of the first cell, and the posture of the first cell is restored by combining the position of the first cell with the position of the pixel point, the spatial mapping of the pixel point in the standard three-dimensional model of the red blood cell and the grayscale value of the pixel point, and whether the first cell is an abnormal cell is judged based on the morphology and volume reflected by the three-dimensional model of the first cell.

[0020] Using the Sobel or Prewitt operator can more accurately calculate the grayscale gradient of the pixel point. This not only helps to determine the edge contour of the cell, but also captures the tiny details of the cell edge. The captured precise data can not only more accurately identify the cell edge features (cell shape), but also more accurately restore the three-dimensional shape of the red blood cells in combination with the identified shape, thereby making a more accurate judgment on the shape and volume of the red blood cells.

[0021] Further, a set M is generated according to the positioning objects, two positioning objects with the shortest straight-line distance are taken out from the set, the minimum rectangular range that simultaneously selects the two positioning objects is used as the division range of the image block, and the image block is divided from the digital image according to the division range of the image block; the above operation is repeated until all the positioning objects in the set M are completely taken out, and no more positioning objects are added to the set M; the undivided area in the digital image is divided into a number of image blocks using a preset division area.

[0022] The minimum rectangular range of the two positioning objects with the shortest straight-line distance is used as the division range of the image block, which can avoid including too many irrelevant background areas to a certain extent. At the same time, this way of dividing (image blocks) (according to the cell adhesion situation) helps to place cells with similar characteristics or interrelated cells in the same image block, which is convenient for subsequent analysis of the relationship between these cells, thereby reducing the loss of feature information caused by random cell division, and helping to improve the recognition accuracy of abnormal cell morphology. Furthermore, this flexible division method can also be highly adaptable according to different (patient) image data.

[0023] Furthermore, when dividing the image blocks, the pixel points selected within the division range are marked as divided pixel points in the digital image; when dividing the image blocks again, the pixel points within the division range that are not marked as divided are divided into the image blocks in the digital image, and after the division, the pixel points selected within the division range are marked as divided pixel points in the digital image.

[0024] The above method can avoid the situation where the same pixel is repeatedly divided into different image blocks. This accurate division method ensures the integrity and accuracy of cell feature data and improves the accuracy of artificial intelligence in blood cell morphology recognition and division. In addition, this method avoids unnecessary repeated judgment and calculation of processed pixels. When processing a large amount of blood cell image data, this screening mechanism can reduce the amount of calculation.

[0025] Furthermore, it is also used to obtain the distribution of each positioning object in the set M in the digital image. After dividing the image block from the digital image according to the division range of the image block, the positioning object with a sparser distribution than the positioning objects in the set M is selected from the two taken out positioning objects and put back into the set M.

[0026] By replacing the positioning objects that are more sparsely distributed with the positioning objects in the set M, the cell density in the divided image blocks can be avoided to be too high. For the image blocks with dense cells, too many cells overlap and block each other, which will increase the difficulty of identification and division; and the image blocks with too few cells may not provide enough information to accurately judge the morphological characteristics of the cells. This balancing operation makes the number of cells in each image block relatively reasonable, thereby improving the accuracy of artificial intelligence in morphological identification and division of blood cells in each image block. At the same time, the replacement operation takes into account the overall distribution of the positioning objects (cells) in the set M, which helps to discover abnormal aggregation or sparseness of cells, provides more valuable clues for disease diagnosis, and enables the artificial intelligence model to better understand the relationship between cells during the learning and recognition process, thereby improving the accuracy of morphological identification and division of blood cells. When dividing the image blocks, there may be a situation where a first cell is divided into two different image blocks. The above division method makes adjacent image blocks overlap to a certain extent, avoids the situation where a first cell is divided into two different image blocks, and improves the accuracy of morphological identification and division of blood cells.

[0027] Further, when calculating the proportion of abnormal cells in the first cells, if the proportion is greater than a preset lesion proportion, the number of first cells in the positioning object is obtained as the first adhesion count, the number of positioning objects containing the first cells is obtained as the second adhesion count, and the ratio of the first adhesion count to the second adhesion count is added to the diagnosis result as the adhesion situation.

[0028] Adding adhesion to the diagnosis results can provide doctors with more detailed information, helping them to more accurately determine whether the disease is in the early, middle or late stages, so as to make more accurate treatment decisions. In addition, the proportion of abnormal cells alone can only reflect the pathological conditions of the first cell itself, but after adding the adhesion between cells, the diagnosis results contain more information about the relationship between cell groups. This information can verify and supplement each other, making the diagnosis more complete.

[0029] Further, when the ratio of the number of first cells to the positioning object is less than a preset sampling ratio, and the ratio of abnormal cells in the first cells is not greater than a preset lesion ratio, insufficient reference of image data is added to the diagnosis result.

[0030] This prompt can make relevant personnel aware of the insufficiency of the current samples. In clinical testing, accurate diagnosis often requires sufficient sample data support. When it is prompted that the image data reference is insufficient, further action needs to be considered. Compared with conventional observation methods, the positioning object will be excluded as a factor affecting the observation results or diagnosis results. This scheme uses the appearance characteristics of the positioning object to locate observable red blood cells, flexibly and reasonably divide image blocks, and judge whether the data collected by the image data is valid. It is also used to judge the patient's special blood condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flowchart of the artificial intelligence diagnosis-assisted method based on cell morphology. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the concept and technical effects of the present invention in combination with the embodiments, so as to fully understand the purpose, features and effects of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention:

[0033] like Figure 1 As shown, the artificial intelligence diagnosis-assisted method based on cell morphology includes the following steps:

[0034] The image data is magnified according to a preset multiple (usually 40-100 times, which can be set by the doctor according to observation requirements) and converted into a digital image, and the digital image is gray-processed;

[0035] Based on the contour features of each cell in the digital image, a red blood cell is identified as a first cell, other cells are identified as positioning objects, a first cell whose contour is enclosed by another cell contour or encloses another cell contour is identified as an overlapping cell, a first cell whose contour is connected to another cell contour is identified as an overlapping cell, overlapping cells and cells that have a closed enclosing relationship or a connected relationship with the overlapping cells are merged into a cell cluster, a positioning object mark existing in the cell cluster is deleted, and the cell cluster is marked as a positioning object;

[0036] Dividing the digital image into a plurality of image blocks, each of which contains at least two positioning objects, and inferring the volume and morphology of the first cell according to the number of pixels and grayscale features of the first cell that does not belong to the positioning object in the image block;

[0037] Abnormal cells are marked from the first cells by the grayscale conditions and cell contours selected by the cell contour frame, and the diagnosis result is output according to the total number of abnormal cells or the proportion of abnormal cells in the first cells.

[0038] When identifying the first cell and locating the object, the Sobel operator or the Prewitt operator is used to calculate the horizontal and vertical grayscale gradients of each pixel in the image to detect edge contour features, wherein the gradient calculation formula (1) is as follows:

[0039] (1)

[0040] in, Used to calculate the horizontal gradient, Used to calculate the vertical gradient, (i, j) is the position of a pixel point. When the gradient amplitude exceeds the preset threshold (usually 30-60, which can be set by the doctor according to the light and dark contrast of the image data), the pixel point is marked as a pixel point located on the cell contour. The closed contour is judged as a complete cell contour by the position of the pixel point on the cell contour.

[0041] The cell image dataset is then annotated using a cell recognition model constructed using a convolutional neural network through the pixel points on the cell contour, and the cell image dataset is input into the cell recognition model for training. The cell recognition model is used to annotate the red blood cells with complete cell contours in the digital image as the first cell, and other cells with complete cell contours as positioning objects.

[0042] wherein, pixel points on the edge contour of the first cell are obtained, a curve fitting algorithm is used to fit the shape of the first cell, the shape of the first cell is compared with a circle or an ellipse to obtain a degree of similarity, and the first cell whose degree of similarity is less than a preset degree of similarity (generally 80%, which can be set by a doctor) is determined to be an abnormal cell;

[0043] The grayscale value of each pixel point within the edge contour of the first cell is obtained, and based on constructing a basic plane parallel to the digital image, the standard three-dimensional model of the red blood cell and the edge contour of the first cell are combined to infer the posture of the first cell, and the three-dimensional model of the first cell is restored by combining the posture of the first cell with the position of the pixel point, the spatial mapping of the pixel point in the standard three-dimensional model of the red blood cell and the grayscale value of the pixel point, and whether the first cell is an abnormal cell is determined based on the morphology and volume reflected by the three-dimensional model of the first cell.

[0044] Wherein, a set M is generated according to the positioning objects, two positioning objects with the shortest straight-line distance are taken from the set, the minimum rectangular range that simultaneously selects the two positioning objects is used as the division range of the image block, and the image block is divided from the digital image according to the division range of the image block;

[0045] Repeat the above operation until all positioning objects in the set M are completely taken out, and no more positioning objects are added to the set M; use the preset division area (determined by the doctor's input of the specific division area, if the doctor does not input it, it is determined by the average area of ​​the divided image blocks) to divide the undivided area in the digital image into several image blocks.

[0046] Among them, it is also used to obtain the distribution of each positioning object in the set M in the digital image. After the image block is divided from the digital image according to the division range of the image block, the positioning object with a sparser distribution than the positioning objects in the set M is selected from the two taken out positioning objects and put back into the set M.

[0047] Among them, when dividing the image blocks, the pixel points selected in the division range are marked as divided pixel points in the digital image; when dividing the image blocks again, the pixel points in the division range that are not marked as divided are divided into the image blocks in the digital image, and after the division, the pixel points selected in the division range are marked as divided pixel points in the digital image.

[0048] Among them, when calculating the proportion of abnormal cells in the first cells, if the proportion is greater than the preset lesion proportion (specifically set by the doctor), the number of first cells in the positioning object is obtained as the first adhesion count, the number of positioning objects containing the first cells is obtained as the second adhesion count, and the ratio of the first adhesion count to the second adhesion count is added to the diagnosis result as the adhesion situation.

[0049] Among them, when the ratio of the number of first cells to the positioning object is less than a preset sampling ratio (specifically set by a doctor), and the ratio of abnormal cells in the first cells is not greater than a preset lesion ratio, insufficient image data reference content is added to the diagnosis result.

[0050] This solution also includes a cell morphology-based artificial intelligence diagnosis assistance system that uses a cell morphology-based artificial intelligence diagnosis assistance method.

[0051] During the specific implementation, the doctor imports the patient's human blood slide image data into the artificial intelligence diagnosis assistance system based on cell morphology.

[0052] The artificial intelligence diagnosis assistance system based on cell morphology first magnifies the image data according to the preset magnification (for example, the doctor sets it to 80 times according to the observation requirements this time) and converts it into a digital image, and then performs grayscale processing. In the stage of identifying cell contour features, the Sobel operator is used to calculate the grayscale gradient of the image pixels. The light and dark contrast of this image is relatively moderate, and the doctor sets the gradient threshold to 40. After calculation and judgment, the pixels on the cell contour are successfully marked and the complete cell contour is determined.

[0053] Next, the artificial intelligence diagnosis assistance system based on cell morphology uses a convolutional neural network cell recognition model built based on pixel points on the cell contour to perform cell labeling. The model has been pre-trained using a large number of labeled cell image data sets, and can accurately label red blood cells as the first cell and other cells as positioning objects. In the abnormal judgment of the first cell, the artificial intelligence diagnosis assistance system based on cell morphology obtains the pixel points on the edge contour of the first cell and uses a curve fitting algorithm to fit its shape. After comparing with the circle, it is found that the approximation of some red blood cells is only 70%, which is lower than the preset approximation of 80%. These cells are preliminarily judged as abnormal cells. At the same time, the artificial intelligence diagnosis assistance system based on cell morphology also combines the standard three-dimensional model of red blood cells and the grayscale value of the pixel points in the edge contour of the first cell to restore the three-dimensional model of the first cell, and further evaluate whether the cell is abnormal in terms of three-dimensional morphology and volume. It is found that some red blood cells have local depressions or protrusions in the three-dimensional morphology, and the volume is also biased, which is confirmed as abnormal cells again.

[0054] In the process of image block division, the artificial intelligence diagnosis assistance system based on cell morphology generates a set M according to the positioning objects, selects the two positioning objects with the shortest straight-line distance, and divides the image blocks by simultaneously selecting their minimum rectangular range. When dividing, carefully mark the divided pixels to ensure that there is no repeated division. After dividing some image blocks, it is found that one of the two positioning objects taken out is extremely sparsely distributed with other positioning objects in set M, so it is put back into set M to participate in the division again. For the undivided area, since the doctor did not enter the preset division area, the artificial intelligence diagnosis assistance system based on cell morphology divides it according to the average area of ​​the divided image blocks.

[0055] When calculating the proportion of abnormal cells in the first cell, it was found that the proportion was greater than the preset lesion proportion (this time the doctor set it to 20%). The artificial intelligence diagnosis assistance system based on cell morphology then obtained the number of first cells in the positioning object as the first adhesion count, and the number of positioning objects containing the first cells as the second adhesion count, and calculated the ratio of the two and added them to the diagnosis result. For example, the first adhesion count is 30, the second adhesion count is 50, and the ratio is 0.6, indicating that there is a certain degree of abnormal cell adhesion.

[0056] However, when checking the ratio of the number of first cells to the located object, it was found that it was less than the preset sampling ratio (set by the doctor to 0.3), and the ratio of abnormal cells in the first cells was not greater than the preset lesion ratio. At this time, the artificial intelligence diagnostic assistance system based on cell morphology added insufficient image data reference to the diagnosis results, prompting the doctor that the diagnosis results may have limitations, and it is recommended to combine other detection methods or re-collect samples for further analysis.

[0057] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. An artificial intelligence diagnosis-assisted method based on cell morphology, characterized in that: The following steps are involved: The image data is magnified according to a preset multiple and converted into a digital image, and the digital image is gray-scaled; Based on the contour features of each cell in the digital image, a red blood cell is identified as a first cell, other cells are identified as positioning objects, a first cell whose contour is enclosed by another cell contour or encloses another cell contour is identified as an overlapping cell, a first cell whose contour is connected to another cell contour is identified as an overlapping cell, overlapping cells and cells that have a closed enclosing relationship or a connected relationship with the overlapping cells are merged into a cell cluster, a positioning object mark existing in the cell cluster is deleted, and the cell cluster is marked as a positioning object; Dividing the digital image into a plurality of image blocks, each of which contains at least two positioning objects, and inferring the volume and morphology of the first cell according to the number of pixels and grayscale features of the first cell that does not belong to the positioning object in the image block; Mark abnormal cells from the first cell by using the grayscale conditions and cell contours selected by the cell contour frame, and output a diagnosis result according to the total number of abnormal cells or the proportion of abnormal cells in the first cell; When identifying the first cell and locating the object, the Sobel operator or Prewitt operator is used to calculate the horizontal and vertical grayscale gradients of each pixel in the image to detect edge contour features, where the gradient calculation formula (1) is as follows: (1)， in, Used to calculate the horizontal gradient, Used to calculate the vertical gradient, (i, j) is the position of a pixel point. When the gradient amplitude exceeds the preset threshold, the pixel point is marked as a pixel point located on the cell contour. The closed contour is judged as a complete cell contour by the position of the pixel point on the cell contour. Then, the cell image dataset is annotated using the cell recognition model constructed by the convolutional neural network through the pixel points on the cell contour, and the cell image dataset is input into the cell recognition model for training. The cell recognition model is used to annotate the red blood cells with complete cell contours in the digital image as the first cell, and other cells with complete cell contours are annotated as positioning objects; Obtaining pixel points on the edge contour of the first cell, fitting the shape of the first cell using a curve fitting algorithm, comparing the shape of the first cell with a circle or an ellipse to obtain a degree of approximation, and determining the first cell whose degree of approximation is less than a preset degree of approximation as an abnormal cell; The grayscale value of each pixel point within the edge contour of the first cell is obtained, and based on constructing a basic plane parallel to the digital image, the standard three-dimensional model of the red blood cell and the edge contour of the first cell are combined to infer the posture of the first cell, and the three-dimensional model of the first cell is restored by combining the posture of the first cell with the position of the pixel point, the spatial mapping of the pixel point in the standard three-dimensional model of the red blood cell and the grayscale value of the pixel point, and whether the first cell is an abnormal cell is determined based on the morphology and volume reflected by the three-dimensional model of the first cell.

2. The artificial intelligence diagnosis-assisted method based on cell morphology according to claim 1, characterized in that: A set M is generated according to the positioning objects, two positioning objects with the shortest straight-line distance are taken from the set, a minimum rectangular range that simultaneously selects the two positioning objects is used as a division range of the image block, and an image block is divided from the digital image according to the division range of the image block; The above operation is repeated until all positioning objects in the set M are completely taken out, and no more positioning objects are added to the set M; the undivided area in the digital image is divided into a number of image blocks using a preset division area.

3. The artificial intelligence diagnosis-assisted method based on cell morphology according to claim 2, characterized in that: It is also used to obtain the distribution of each positioning object in the set M in the digital image. After dividing the image block from the digital image according to the division range of the image block, the positioning object with a sparser distribution than the positioning objects in the set M is selected from the two taken out positioning objects and put back into the set M.

4. The artificial intelligence diagnosis-assisted method based on cell morphology according to claim 3 is characterized in that: When dividing the image blocks, the pixel points selected within the division range are marked as divided pixel points in the digital image; When the image blocks are divided again, the pixels in the divided range without the divided mark are divided into the image blocks in the digital image, and after the division, the pixels selected in the divided range are marked as divided pixels in the digital image.

5. The artificial intelligence diagnosis-assisted method based on cell morphology according to claim 4, characterized in that: When calculating the proportion of abnormal cells in the first cells, if the proportion is greater than the preset lesion proportion, the number of first cells in the positioning object is obtained as the first adhesion count, the number of positioning objects containing the first cells is obtained as the second adhesion count, and the ratio of the first adhesion count to the second adhesion count is added to the diagnosis result as the adhesion situation.

6. The artificial intelligence diagnosis-assisted method based on cell morphology according to claim 5, characterized in that: When the ratio of the number of first cells to the positioning object is less than a preset sampling ratio, and the ratio of abnormal cells to the first cells is not greater than a preset lesion ratio, insufficient reference of image data is added to the diagnosis result.

7. An artificial intelligence diagnosis assistance system based on cell morphology, characterized in that: The artificial intelligence diagnosis-assisted method based on cell morphology as described in any one of claims 1-6 is used.

Images