Multi-myeloma myeloid image recognition system based on computer vision

Abstract

The invention discloses a multiple myeloma bone marrow image recognition system based on computer vision, relates to the technical field of cell recognition, solves the problem that traditional manual microscopic examination is large in limitation, and accurately recognizes burr change segments by conducting angle analysis on the edge contour of a feature area, so that the accuracy of recognition is improved. According to the comprehensive proportion of the burr change section on the overall edge contour, reliably judging whether the to-be-detected cell is a tumor cell or not, and calibrating and displaying the tumor cell in a gray image; according to the mode, the obvious difference of the tumor cells and normal cells in cell nucleus morphology is effectively utilized, rapid and accurate identification of the tumor cells is achieved, a visual and clear result is provided for clinical diagnosis, doctors are assisted to accurately judge multiple myeloma in time, the diagnosis efficiency and the treatment effect are improved, mean value processing and radial ratio analysis are conducted on contour area gray values, and the diagnosis accuracy is improved. The contour region associated with the cell nucleus can be accurately locked from the to-be-detected cell and is calibrated as the feature region, and a key cell nucleus feature basis is provided for tumor cell recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of cell recognition, and specifically to a multiple myeloma bone marrow image recognition system based on computer vision. Background Technique

[0002] Multiple myeloma (MM), as a common malignant tumor in the hematological system, seriously threatens human health. According to the statistical data of the World Health Organization, the incidence of MM shows an increasing trend year by year globally, and the age of onset tends to be younger.

[0003] With the rapid development of computer technology, computer vision technology has been increasingly widely applied in the medical field; computer vision can quickly process a large amount of image data, accurately extract image features through algorithms, and show great potential in disease diagnosis; in the recognition of MM bone marrow images, using computer vision technology to construct an automated recognition system is expected to break through the limitations of traditional manual microscopy, realize rapid and accurate recognition of myeloma cells, improve the diagnosis efficiency and accuracy of MM, gain precious time for clinical treatment, and improve the prognosis of patients. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a multiple myeloma bone marrow image recognition system based on computer vision, which solves the problem of large limitations in traditional manual microscopy.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A multiple myeloma bone marrow image recognition system based on computer vision, including:

[0006] A high-definition microscopic image conversion terminal that converts the obtained high-definition microscopic image into a grayscale image by converting the grayscale value;

[0007] A grayscale feature processing terminal that combines the SobeL algorithm to confirm the gradient features associated with different pixel points in the grayscale image, and according to the gradient features corresponding to different pixel points in the grayscale image, confirms the cell contour and selects cells to confirm cell monomers. The specific method is:

[0008] Use the SobeL algorithm to confirm the horizontal gradient X i and the vertical gradient S i associated with different pixel points in the grayscale image, where i represents different pixel points, and use: to confirm the gradient feature ZH i corresponding to this pixel point;

[0009] Pixels that satisfy ZH i ≥Y1 are calibrated as gradient pixels, otherwise, no calibration is performed, where Y1 is a preset value;

[0010] According to the gradient pixel points sequentially identified in the grayscale image, several adjacent gradient pixel points are connected to confirm the connection contour, and the individual associated with this connection contour is denoted as a cell monomer, and this connection contour belongs to the edge contour of this cell monomer;

[0011] The polymorphic cell analysis terminal, based on the different cell monomers calibrated in the grayscale image, confirms the contour areas existing inside the cell monomers, calibrates the cell monomers as monomorphic cells or polymorphic cells based on the number of contour areas, and determines the cells to be inspected from the monomorphic cells. The specific method is as follows:

[0012] The Sobel algorithm is used to confirm the horizontal gradient X associated with different pixel points inside different cell monomers k and the vertical gradient S k , where k represents different pixel points inside the cell monomer, and the following is used: To confirm the gradient feature ZH corresponding to this pixel point k ;

[0013] The pixel points that satisfy ZH k ≥Y2 are calibrated as gradient pixel points, otherwise, no calibration is performed. Y2 is a preset value, and Y2 < Y1;

[0014] According to the gradient pixel points sequentially identified inside the cell monomer, several adjacent gradient pixel points are connected to confirm the corresponding contour area. If there is only one set of contour areas inside this cell monomer, this cell monomer is calibrated as a monomorphic cell. If there are multiple sets of contour areas in this cell monomer, it is calibrated as a polymorphic cell;

[0015] Based on the edge contour associated with the monomorphic cell, first confirm the cell center point belonging to this monomorphic cell. Randomly select a set of contour points on the edge contour and denote them as the main points. Connect the main points with the cell center point to confirm a set of connection lines, and extend this connection line to the other side of the edge contour to confirm the first set of dividing lines. Then construct a perpendicular line belonging to this dividing line, and the two end points of the perpendicular line are also located on the edge contour. Confirm the evaluation ratio of the dividing line to the perpendicular line: the evaluation ratio = the length of the long line segment ÷ the length of the short line segment. The long line segment is the longest line segment among the dividing line and the perpendicular line, and the short line segment is the same. Rotate the dividing line according to this cell center point, and according to the rotation process, confirm the maximum evaluation ratio. If the maximum evaluation ratio ≥ 1.5, then this monomorphic cell is calibrated as the cell to be inspected, otherwise, no calibration is performed;

[0016] The feature area calibration terminal uses the Sobel algorithm to sequentially confirm the other contour areas existing inside the cell to be inspected, lock the contour area associated with the cell nucleus, and perform the specific calibration of the feature area again. The specific method is as follows:

[0017] The Sobel algorithm and the set threshold are used to sequentially confirm other contour regions existing in the cell to be detected, and different contour regions correspond to different thresholds;

[0018] The average value processing is performed on several groups of gray values associated with different pixel points in different contour regions to confirm the regional features associated with the corresponding contour regions. The contour regions whose regional features belong to the set interval are marked as suspected regions, and the set interval is the preset interval;

[0019] If there is only one group of suspected regions, this suspected region is marked as the characteristic region;

[0020] If there are multiple groups of suspected regions, the radial ratio of multiple groups of suspected regions is confirmed by the verification unit, and then the feature analysis is carried out to lock the characteristic region:

[0021] Based on the edge contour of the suspected region, the region center point belonging to this suspected region is locked. Then, the same processing method for confirming the cell to be detected for single-state cells is adopted. A dividing line and a perpendicular line are constructed in the suspected region, and the evaluation ratio belonging to this suspected region is confirmed. From several groups of evaluation ratios, the maximum evaluation ratio is selected and recorded as the standard feature of this suspected region. The value closest to 1 is selected from several groups of standard features, and the suspected region associated with the selected value is recorded as the characteristic region;

[0022] The cancer cell recognition and calibration end, based on the characteristic region calibrated in the cell to be detected, confirms the burr change segments on the edge contour of the characteristic region, and based on the comprehensive proportion of the burr change segments on the overall edge contour, confirms whether this cell to be detected is a cancer cell and conducts calibration and display. The specific method is as follows:

[0023] Based on the edge contour of the characteristic region, the characteristic center point belonging to this characteristic region is confirmed, and a group of contour points is randomly selected from the edge contour and recorded as the starting point. According to the clockwise direction, the angle of the contour segment between the starting point and the adjacent point is confirmed:

[0024] The starting point is connected to the characteristic center point to confirm the angle line 1, and then the starting point is connected to the adjacent point to confirm the angle line 2. The included angle J between the angle line 1 and the angle line 2 is confirmed. If 85°≤J≤95°, the partial contour segment between the starting point and the adjacent point is marked as the normal segment, otherwise, it is marked as the burr change segment;

[0025] Then, the same processing method is used to sequentially mark the subsequent partial contour segments as normal segments or burr change segments until the entire edge contour of the characteristic region is completely marked;

[0026] Confirm the total length L1 of several burr change segments, and then confirm the total length L2 of the edge contour of the feature region. If (L1÷L2) ≤ 0.2, no calibration is performed. Otherwise, the cell to be detected is calibrated as a tumor cell, and the calibration is synchronized in the grayscale image.

[0027] The present invention provides a multiple myeloma bone marrow image recognition system based on computer vision. Compared with the prior art, it has the following beneficial effects:

[0028] The polymorphic cell analysis terminal can accurately distinguish monomorphic cells and polymorphic cells based on the number of internal contour regions of a single cell, and effectively screen out the cells to be detected suspected of tumor cells from monomorphic cells through a unique division line and perpendicular line evaluation ratio method, providing a precise target for subsequent in-depth analysis and helping to improve the pertinence of tumor cell diagnosis;

[0029] By using the Sobel algorithm and empirical threshold setting, combined with the processing of the average gray value of the contour region and radial ratio analysis, the contour region associated with the cell nucleus can be accurately locked from the cells to be detected and calibrated as the feature region, providing a key nuclear feature basis for identifying tumor cells and improving the accuracy and reliability of tumor cell recognition;

[0030] By analyzing the angle of the edge contour of the feature region, the burr change segments are accurately identified. Based on the comprehensive proportion of the burr change segments on the overall edge contour, it is reliably determined whether the cell to be detected is a tumor cell, and it is calibrated and displayed in the grayscale image; this method effectively utilizes the significant difference in the nuclear morphology between tumor cells and normal cells to achieve rapid and accurate identification of tumor cells, providing an intuitive and clear result for clinical diagnosis, helping doctors to timely and accurately judge multiple myeloma, and improving the diagnosis efficiency and treatment effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the principle framework of the present invention;

[0032] Figure 2 It is a schematic diagram for determining the included angle between angle line 1 and angle line 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] Please refer to Figure 1, this application provides a multiple myeloma bone marrow image recognition system based on computer vision, including a high-definition microscopic image conversion terminal, a grayscale feature processing terminal, a polymorphic cell analysis terminal, a feature area calibration terminal, a verification unit, and a tumor cell recognition and calibration terminal;

[0035] Among them, the high-definition microscopic image conversion terminal, the grayscale feature processing terminal, the polymorphic cell analysis terminal, and the feature area calibration terminal are electrically connected in sequence from the output node to the input node, and the verification unit is bidirectionally connected to the feature area calibration terminal, and the feature area calibration terminal is electrically connected to the input node of the tumor cell recognition and calibration terminal;

[0036] Among them, the high-definition microscopic image conversion terminal converts the obtained high-definition microscopic image into a grayscale image by converting the grayscale value. Based on the preset weights and the RGB values of the corresponding points, the grayscale value of the corresponding point is confirmed. Specifically, the high-definition microscopic image generally uses a high-resolution microscope imaging system equipped with a 50 million-pixel professional microscope camera, which can clearly capture the details of the bone marrow smear. The automatic focus and intelligent exposure adjustment functions are set to ensure stable and high-quality images are obtained at different magnifications (such as 400X, 1000X);

[0037] The specific method for converting the grayscale image is as follows:

[0038] According to the confirmed high-definition microscopic image, the RGB values associated with the corresponding pixel points are confirmed, and then: HD = 0.299R + 0.587G + 0.114B is used to confirm the grayscale value HD associated with the corresponding pixel point. Based on the grayscale values associated with different pixel points, the grayscale value is adjusted to convert the high-definition microscopic image into a grayscale image.

[0039] Specifically, the method for converting the grayscale value is relatively common in the prior art, so it will not be elaborated here. The RGB value is a set of three values, and each value is given a different weight, so that the grayscale values of pixel points with different characteristics can be adjusted.

[0040] Among them, the grayscale feature processing end, based on the grayscale values associated with different pixel points in the grayscale image, combines the Sobel algorithm to confirm the gradient features associated with the corresponding pixel points. According to the gradient features corresponding to different pixel points in the grayscale image, the cell contours are confirmed and cells are selected. Specifically, when sorting pixel points, they are in the sorting state of a nine-square grid. Therefore, there are eight groups of pixel points around each pixel point, and each pixel point has a corresponding grayscale value. By using the Sobel algorithm to assign different weight factors to different pixel points, the associated weight factors are different when confirming the horizontal gradient and the vertical gradient. Since the method of confirming the corresponding gradient value of a pixel point is relatively common in the prior art, it will not be elaborated here. The Sobel algorithm uses two 3×3 convolution kernels. One is used to calculate the horizontal (horizontal) gradient, denoted as Gx; the other is used to calculate the vertical (vertical) gradient, denoted as Gy. Different gradients correspond to different convolution kernels. Based on different convolution kernels and the grayscale values associated with different pixel points, the horizontal gradient and the vertical gradient associated with the central pixel point can be confirmed;

[0041] Among them, the specific method of cell selection is as follows:

[0042] Use the Sobel algorithm to confirm the horizontal gradient X associated with different pixel points in the grayscale image i and the vertical gradient S i , where i represents different pixel points, and use: to confirm the gradient feature ZH corresponding to this pixel point i ;

[0043] Mark the pixel points that satisfy ZH i ≥Y1 as gradient pixel points. Otherwise, no marking is performed. Y1 is a preset value, and its specific value is determined by the operator according to experience. Its general value is between 150 and 200 and is determined in advance by the operator;

[0044] According to the gradient pixel points sequentially confirmed in the grayscale image, connect several adjacent gradient pixel points to confirm a set of connection contours, and record the individual associated with this connection contour as a cell monomer. This connection contour belongs to the edge contour of this cell monomer;

[0045] Specifically, on the contour of each cell, the associated contour points are all gradient points. There is an obvious grayscale change feature between the cell and the bone marrow fluid, and the corresponding gradient pixel change is also relatively obvious. Then there are corresponding gradient contour points outside the corresponding cell, that is, the associated edge contour.

[0046] The polymorphic cell analysis terminal, based on different cell monomers calibrated in the grayscale image, confirms the contour regions (i.e., vacuole regions) existing inside the cell monomers, calibrates the cell monomers as monomorphic cells or polymorphic cells based on the number of contour regions, and determines the cells to be inspected from the monomorphic cells. The monomorphic cell can be understood as a single cell, that is, a complete cell. The polymorphic cell can be understood as multiple cells gathered together, resulting in overlapping edge contours, obtaining a large contour, but there are corresponding multiple vacuoles in multiple cells, so there are multiple contour regions. The specific method for calibrating cell monomers is as follows:

[0047] Use the Sobel algorithm to confirm the horizontal gradient X associated with different pixel points inside different cell monomers k and the vertical gradient S k , where k represents different pixel points inside the cell monomer, and use: to confirm the gradient feature ZH corresponding to this pixel point k ;

[0048] Calibrate the pixel points that satisfy ZH k ≥Y2 as gradient pixel points, otherwise, no calibration is performed. Y2 is a preset value, and Y2 < Y1. Since the gray value evaluation benchmark for Y1 is the gray feature between the cell and the medulla, and its feature is strong, while Y2 is the gray feature between the vacuole and the cytoplasm, and its feature is weak, so Y2 < Y1;

[0049] According to the gradient pixel points sequentially confirmed inside the cell monomer, connect several adjacent gradient pixel points to confirm the corresponding contour region. If there is only one set of contour regions in this cell monomer, then this cell monomer is calibrated as a monomorphic cell. If there are multiple sets of contour regions in this cell monomer, then it is calibrated as a polymorphic cell;

[0050] Based on the edge contour associated with the singlet cell, first confirm the cell center point belonging to this singlet cell (combine the edge contour with the two-dimensional coordinate system, confirm the two-dimensional coordinates of different points within the contour, then perform mean processing on several two-dimensional coordinates to confirm the mean coordinate, and mark the confirmed mean coordinate within the singlet cell to confirm the cell center point). Randomly select a set of contour points on the edge contour and mark them as main points. Connect the main points with the cell center point to confirm a set of connecting lines, and extend this connecting line to the other side of the edge contour to confirm the first set of dividing lines. Then construct a perpendicular line belonging to this dividing line, and both endpoints of the perpendicular line are also located on the edge contour. Confirm the evaluation ratio of the dividing line to the perpendicular line: the evaluation ratio = the length of the long line segment ÷ the length of the short line segment. The long line segment is the longest line segment among the dividing line and the perpendicular line, and the same applies to the short line segment. Rotate the dividing line according to this cell center point. According to the rotation process, confirm the maximum evaluation ratio. If the maximum evaluation ratio ≥ 1.5, then mark this singlet cell as a cell to be detected; otherwise, do not perform any marking. For example, the obvious difference between normal cells and tumor cells in appearance is that normal cells are in a relatively circular state, while tumor cells are in a relatively elliptical state, and the degree of ellipse is relatively large. Therefore, by confirming the ratio of the internal line segment lengths, the maximum evaluation ratio is determined. According to the rotation process of the corresponding dividing line, the perpendicular line rotates accordingly. Then the dividing line and the perpendicular line can confirm the specific elliptical state of the cell monomer to specifically confirm the corresponding evaluation ratio, so as to initially screen out the corresponding cells to be detected. Then analyze the cells to be detected to determine whether such cells to be detected are tumor cells, and it is necessary to check and analyze the internal cell nucleus.

[0051] Feature region calibration end. Use the SobeL algorithm to sequentially confirm other contour regions existing in the cell to be detected, and lock the contour region associated with the cell nucleus from the confirmed contour regions, and perform specific calibration of the feature region. The specific calibration method is as follows:

[0052] Use the SobeL algorithm and the set threshold to sequentially confirm other contour regions existing in the cell to be detected. Different contour regions correspond to different thresholds, and the value of the threshold is determined by the operator according to experience.

[0053] Perform mean processing on several groups of gray values associated with different pixel points in different contour regions to confirm the region features associated with the corresponding contour regions. Mark the contour regions whose region features belong to the set interval as suspected regions. The set interval is a preset interval, which is determined in advance based on the gray feature of the cell nucleus, and generally takes values in the range of [80, 100].

[0054] If there is only one group of suspected regions, then mark this suspected region as the feature region.

[0055] If there are multiple groups of suspected regions, the radial ratio of the multiple groups of suspected regions is confirmed by the verification unit, and then feature analysis is performed to lock the feature region. The specific method of locking is as follows:

[0056] Based on the edge contour of the suspected region, lock the region center point belonging to this suspected region, and then adopt the same processing method as that for single-state cell confirmation of the cell to be detected. Construct a dividing line and a perpendicular line within the suspected region, and confirm the evaluation ratio belonging to this suspected region. Select the maximum evaluation ratio from several groups of evaluation ratios, and record the maximum evaluation ratio as the standard feature of this suspected region. Select the value closest to 1 from several groups of standard features, and record the suspected region associated with the selected value as the feature region. Specifically, its corresponding cell nucleus belongs to a spherical body. In most biological cells, the cell nucleus presents a spherical or approximately spherical shape; this is because the spherical structure enables the cell nucleus to have the largest volume within a limited space, which is conducive to the uniform distribution of genetic material (such as chromosomes) in the cell nucleus and provides a relatively stable environment for various physiological activities in the nucleus; for example, the cell nuclei of human liver cells, nerve cells, etc. are mostly spherical, and the evaluation ratio associated with their spherical shape is generally the value closest to 1. Except for the cell nucleus, which is the closest to a circle within the cell body, other contour regions are irregular regions. Therefore, 1 can be directly used as the associated numerical feature for selection.

[0057] Among them, for the cancer cell recognition and calibration end, according to the feature region calibrated in the cell to be detected, confirm the burr change segment on the edge contour of the feature region, and based on the comprehensive proportion of the burr change segment on the overall edge contour, confirm whether this cell to be detected is a cancer cell and perform calibration and display. The specific method of confirmation is as follows:

[0058] Combined with Figure 2 , based on the edge contour of the feature region, confirm the feature center point belonging to this feature region (combine the edge contour with the two-dimensional coordinate system, confirm the two-dimensional coordinates of different points within the contour, then perform mean processing on several two-dimensional coordinates, confirm the mean coordinate, and calibrate the confirmed mean coordinate within the feature region to confirm the feature center point), and randomly select a group of contour points from the edge contour and record them as the starting point. Then, according to the clockwise direction, confirm the angle of the contour segment between the starting point and the adjacent point:

[0059] Connect the starting point and the feature center point to confirm the first angle line, and then connect the starting point and the adjacent point to confirm the second angle line. Confirm the included angle J between the first angle line and the second angle line. If 85° ≤ J ≤ 95°, then calibrate the partial contour segment between the starting point and the adjacent point as the normal segment; otherwise, calibrate it as the burr change segment.

[0060] Then, use the same processing method to calibrate the subsequent partial contour segments as normal segments or burr change segments in sequence, and stop when the overall edge contour of the feature region is completely calibrated;

[0061] Confirm the total length L1 of several burr change segments, and then confirm the total length L2 of the edge contour of the feature region. If (L1÷L2)≤0.2, no calibration is performed. Otherwise, this cell to be detected is calibrated as a tumor cell, and the calibration is synchronously performed in the grayscale image.

[0062] Specifically, the specific difference between tumor cells and normal cells is that there are obvious regions in their cell nuclei. In normal cells, the cell nuclei are relatively round and smooth, without any bulge or burr-related features. However, there are a large number of burr bulge features on the surface of the cell nuclei associated with tumor cells. Such features are significantly different from normal cells, so tumor cells can be quickly calibrated.

[0063] Some of the data in the above formula are numerically calculated after removing their dimensions, and the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0064] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A multiple myeloma bone marrow image recognition system based on computer vision, characterized in that Including: A high-definition microscopic image conversion terminal that converts the obtained high-definition microscopic image into a grayscale image by converting the grayscale value; A grayscale feature processing terminal that combines the Sobel algorithm to confirm the gradient features associated with different pixel points in the grayscale image, and based on the gradient features corresponding to different pixel points in the grayscale image, confirms the cell contour and selects cells to confirm cell monomers; A polymorphic cell analysis terminal that, based on different cell monomers calibrated in the grayscale image, confirms the contour regions existing inside the cell monomers, calibrates the cell monomers as monomorphic cells or polymorphic cells based on the number of contour regions, and determines the cells to be tested from the monomorphic cells; A feature region calibration terminal that uses the Sobel algorithm to sequentially confirm other contour regions existing in the cells to be tested, locks the contour region associated with the cell nucleus, and performs specific calibration of the feature region again; A cancer cell recognition and calibration terminal that, based on the feature regions calibrated in the cells to be tested, confirms the burr change segments on the edge contour of the feature regions, and based on the comprehensive proportion of the burr change segments on the overall edge contour, confirms whether the cells to be tested are cancer cells and performs calibration and display; 2. The myeloma bone marrow image recognition system based on computer vision according to claim 1, characterized in that, For the high-definition microscopic image conversion terminal, the specific method for converting the grayscale image is as follows: Based on the confirmed high-definition microscopic image, confirm the RGB values associated with the corresponding pixel positions, and then use: HD = 0.299R + 0.587G + 0.114B to confirm the grayscale value HD associated with the corresponding pixel positions, and adjust the grayscale values based on the grayscale values associated with different pixel positions to convert the high-definition microscopic image into a grayscale image.

3. The myeloma bone marrow image recognition system based on computer vision according to claim 1, characterized in that, For the grayscale feature processing terminal, the specific method for cell selection is as follows: Use the Sobel algorithm to confirm the horizontal gradient X associated with different pixel points in the grayscale image i and the vertical gradient S i , where i represents different pixel points, and use: to confirm the gradient feature ZH corresponding to this pixel point i ; Pixels that meet ZH i are calibrated as gradient pixels. Otherwise, no calibration is performed, where Y1 is a preset value; Based on the sequentially confirmed gradient pixel points in the grayscale image, connect several adjacent gradient pixel points to confirm the connection contour, and record the individual associated with this connection contour as a cell monomer, and this connection contour belongs to the edge contour of this cell monomer.

4. A myeloma bone marrow image recognition system based on computer vision according to claim 1, characterized in that, For the polymorphic cell analysis terminal, the specific method for calibrating cell monomers as monomorphic cells or polymorphic cells is as follows: Use the Sobel algorithm to confirm the horizontal gradient X associated with different pixel points within different cell monomers k and the vertical gradient S k , where k represents different pixel points within the cell monomer, and use: to confirm the gradient feature ZH corresponding to this pixel point k ; Pixels that satisfy ZH k ≥ Y2 are calibrated as gradient pixels. Otherwise, no calibration is performed. Here, Y2 is a preset value and Y2 < Y1; Based on the sequentially confirmed gradient pixel points in the cell monomer, connect several adjacent gradient pixel points to confirm the corresponding contour regions. If there is only one set of contour regions in this cell monomer, then calibrate this cell monomer as a monomorphic cell. If there are multiple sets of contour regions in this cell monomer, then calibrate it as a polymorphic cell.

5. The myeloma bone marrow image recognition system based on computer vision according to claim 4, characterized in that, For the polymorphic cell analysis terminal, the specific method for determining the cells to be tested from the monomorphic cells is as follows: Based on the edge contour associated with the singlet cell, first confirm the cell center point belonging to this singlet cell. Randomly select a set of contour points on the edge contour and denote them as the main points. Connect the main points with the cell center point to confirm a set of connecting lines, and extend this connecting line to the other side of the edge contour to confirm the first set of dividing lines. Then construct the perpendicular lines belonging to this dividing line, and the two end points of the perpendicular line are also located on the edge contour. Confirm the evaluation ratio of the dividing line to the perpendicular line: the evaluation ratio = the length of the long line segment ÷ the length of the short line segment, where the long line segment is the longest line segment among the dividing line and the perpendicular line, and the short line segment is the same. Rotate the dividing line according to this cell center point. According to the rotation process, confirm the maximum evaluation ratio. If the maximum evaluation ratio ≥ 1.5, then label this singlet cell as a cell to be detected; otherwise, do not perform any labeling.

6. The myeloma bone marrow image recognition system based on computer vision according to claim 5, wherein The specific method for calibrating the characteristic area by the characteristic area calibration end is as follows: Use the SobeL algorithm and the set threshold to sequentially confirm the other contour areas existing in the cell to be detected. Different contour areas correspond to different thresholds; Perform mean processing on several groups of gray values associated with different pixel points in different contour areas to confirm the area characteristics associated with the corresponding contour areas. Label the contour areas whose area characteristics belong to the set interval as suspected areas, and the set interval is the preset interval; If there is only one group of suspected areas, then label this suspected area as the characteristic area; If there are multiple groups of suspected areas, then the radial ratio of the multiple groups of suspected areas is confirmed by the verification unit, and then feature analysis is performed to lock the characteristic area.

7. The myeloma bone marrow image recognition system based on computer vision according to claim 6, characterized in that, The specific method for the verification unit to perform radial ratio confirmation is as follows: Based on the edge contour of the suspected area, lock the area center point belonging to this suspected area, and then use the same processing method for confirming the cell to be detected for the singlet cell. Construct dividing lines and perpendicular lines in the suspected area, and confirm the evaluation ratio belonging to this suspected area. Select the maximum evaluation ratio from several groups of evaluation ratios, and denote the maximum evaluation ratio as the standard feature of this suspected area. Select the value closest to 1 from several groups of standard features, and denote the suspected area associated with the selected value as the characteristic area.

8. The myeloma bone marrow image recognition system based on computer vision according to claim 1, wherein, The specific method for the cancer cell recognition and calibration end to confirm whether this cell to be detected is a cancer cell is as follows: Based on the edge contour of the characteristic area, confirm the characteristic center point belonging to this characteristic area, and randomly select a set of contour points from the edge contour and denote them as the starting points. According to the clockwise direction, confirm the angle of the contour segment between the starting point and the adjacent point: Connect the starting point with the characteristic center point to confirm the first angle line, and then connect the starting point with the adjacent point to confirm the second angle line. Confirm the included angle J between the first angle line and the second angle line. If 85° ≤ J ≤ 95°, then label the partial contour segment between the starting point and the adjacent point as the normal segment; otherwise, label it as the burr change segment; Then use the same processing method to sequentially label the subsequent partial contour segments as normal segments or burr change segments until the overall edge contour of the characteristic area is completely labeled and then stop; Confirm the total length L1 of several burr change segments, and then confirm the total length L2 of the edge contour of the feature region. If (L1÷L2) ≤ 0.2, no calibration is performed. Otherwise, this cell to be detected is calibrated as a tumor cell, and the calibration is synchronized in the grayscale image.

Images