Cell recognition method based on self-adaptive ROI (Region of Interest) feature image extraction

Abstract

The invention relates to cell detection, and particularly provides a cell recognition method based on adaptive ROI (Region of Interest) feature image extraction, which comprises the following steps of: (S1) adaptively extracting a cell feature image to obtain a cell ROI feature image; (S2) the quality of the ROI cell feature image is evaluated, if an evaluation value is smaller than a first threshold value, the cell ROI feature image is reserved, and if the evaluation value is not smaller than the first threshold value, the cell ROI feature image is abandoned; and (S3) carrying out naive Bayesian classification on the characteristics of the reserved ROI cell image, and identifying cells and positions thereof. The method has the advantages of accurate recognition and the like, and is applied to various cell detection.

Description

Technical Field

[0001] The invention relates to cell detection, and in particular to a cell recognition method based on adaptive ROI feature image extraction. Background Art

[0002] Cells are the basic structural and functional units of organisms, and are also important research objects for people to conquer diseases and explore the mysteries of life. Traditional methods of analyzing cell groups make it difficult for people to find differences between cells. Some studies have tested the DNA content of tissues and found that there is a large difference between the average signal obtained from tissue measurements and the signal expressed by individual cells. This kind of tissue experiment may be misleading and cover up many important weak signals, and fail to expose the heterogeneity between single cells. This "tumor heterogeneity" has brought great trouble to clinical drug use. In order to study the physical and chemical properties of cells, further distinguish the differences between cells, and primary tumors, it is impossible to detect and fully explore through tissue-level analysis methods. Therefore, cell characteristic analysis with single-cell resolution is crucial.

[0003] Single-cell detection tools can elucidate cell heterogeneity, facilitate in-depth understanding of related cell behaviors and functions, and can effectively help people explore the mysteries of life and explain disease mechanisms. Currently, in the process of single-cell mass spectrometry analysis, professional experimenters usually manually control the manipulator in front of the microscope to complete the operation. This semi-automatic operation method is inefficient, subject to a lot of human intervention, and there are differences in the analysis results. Therefore, current researchers are more keen on the research of ultra-trace, high-sensitivity, and fully automated single-cell analysis systems. However, the identification and positioning of single cells is the key to fully automated detection equipment, and this technology lays the foundation for the realization of fully automated single-cell analysis. The automatic identification and positioning technology of cells can provide strong support for single-cell detection tools and has great value in life science research and other aspects.

[0004] In recent years, with the development of micromanipulation technology, there are many methods for cell identification, such as: 1. Cell identification is performed through fluorescent labeling, which changes the physiological state of cells. However, this method affects the physiological state of cells and requires a dedicated microscope and analysis, so the analysis and detection equipment is limited.

[0005] 2. Recognition is performed through a network model algorithm that combines the spatial attention module and the channel attention module. However, this method has large computational complexity, high equipment performance requirements, low recognition efficiency, and high algorithm complexity. It requires a large amount of data images for training, otherwise it is difficult to achieve good results.

[0006] 3. A colorimetric characteristic equation is established based on quantitative analysis for identification, but this method only supports cells stained by immunochemical methods and has poor universality. Summary of the invention

[0007] In order to solve the deficiencies in the above-mentioned prior art solutions, the present invention provides a cell recognition method based on adaptive ROI feature image extraction.

[0008] The objective of the present invention is achieved through the following technical solutions: A cell recognition method based on adaptive ROI feature image extraction includes the following steps: (S1) Adaptively extracting cell feature images to obtain cell ROI feature images; (S2) evaluating the quality of the ROI cell feature image, if the evaluation value is less than a first threshold, retaining the cell ROI feature image, if the evaluation value is not less than the first threshold, discarding the cell ROI feature image; (S3) Perform naive Bayes classification on the features of the retained ROI cell images to identify the cells and their locations.

[0009] Compared with the prior art, the present invention has the following beneficial effects: 1. Accurately identify cell locations and have a wide range of applications; The present invention realizes the rapid identification, location and counting of cells under complex backgrounds, successfully realizes the automatic location of cells with high precision, and lays the foundation for automated biological detection equipment. This technology lays the foundation for the realization of fully automatic single-cell analysis and plays an important role in the field of micromanipulation. There is no need to set a fixed window to segment the original image of the cell, and then judge each small window one by one; not only does it greatly reduce the time of cell recognition, but the recognition accuracy is no longer affected by the size of the image segmentation window; It is no longer limited to cell recognition in clean backgrounds such as suspended cells, and has strong robustness; 2. Good stability; Independent of external conditions, improving system efficiency and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The disclosure of the present invention will become easier to understand with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only used to illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention. In the drawings: Figure 1 is a schematic flow chart of a cell identification method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the mouse spleen cell identification results; Figure 3 This is a schematic diagram of the bladder cancer cell identification results. DETAILED DESCRIPTION

[0011] Figure 1-Figure 3 The following description describes optional specific embodiments of the present invention to teach those skilled in the art how to implement and reproduce the present invention. In order to teach the technical solution of the present invention, some conventional aspects have been simplified or omitted. Those skilled in the art will understand that variations or replacements derived from these specific embodiments will be within the scope of the present invention. Those skilled in the art will understand that the following features can be combined in various ways to form multiple variations of the present invention. Thus, the present invention is not limited to the following optional specific embodiments, but is only limited by the claims and their equivalents.

[0012] Example 1

[0013] A cell recognition method based on adaptive ROI feature image extraction, such as Figure 1 As shown, the following steps are included: (S1) Adaptively extracting cell feature images to obtain cell ROI feature images.

[0014] (S2) evaluating the quality of the ROI cell feature image, if the evaluation value is less than a first threshold, retaining the cell ROI feature image, and if the evaluation value is not less than the first threshold, discarding the cell ROI feature image.

[0015] (S3) Perform naive Bayes classification on the features of the retained ROI cell images to identify the cells and their locations.

[0016] In order to improve the efficiency and accuracy of extraction, further, in step (S1), the method of extracting cell feature images is as follows: After the cell surface is focused and clear, the cell image is captured, the channels of the original RGB image are split, and the R channel of the image is extracted.

[0017] The OSTU method is used to segment the image, remove noise and then search for contours to obtain the minimum bounding rectangle of cells in the image and filter the cell contour area.

[0018] In order to improve the screening accuracy, the cell contour area is further screened as follows: If the area of ​​the circumscribed rectangle is higher than the second threshold, it is confirmed that the material in the region is suspected to be a cell region. If it is not higher than the second threshold, the region is discarded.

[0019] Remove the contours contained in or intersecting the contour circumscribed rectangle in the image and extract independent contours.

[0020] In order to improve the accuracy of screening, further, the method for judging whether the independent outline rectangles are contained or intersected is as follows: Suspected cell outline image ; , (x1, y1) and (x2, y2) are the center points of the contours L1 and L2 respectively, Δx and Δy are the x-axis component and y-axis component of the distance between the centers of the two contours respectively, and width and height are the sum of the length and width of the contours L1 and L2 respectively.

[0021] In order to improve the screening accuracy, the grayscale co-occurrence matrix contrast value D of the contour area after screening is further calculated to determine whether the value D is greater than the third threshold. If the result is yes, the contour is the ROI cell characteristic contour. If the result is no, it is confirmed as an impurity contour and no subsequent processing is performed.

[0022] In order to improve the accuracy of the third threshold, further, the method for obtaining the third threshold is: The contour image f(x) includes a cell image I(x, y) and an impurity image N(x, y).

[0023] , P1+P2=1.

[0024] K= P1·K1+P2·K2, σ=P1(K1-K) 2 +P2(K2-K) 2 .

[0025] P1 is the ratio of the cell image to the contour image f(x), and the mean value of the gray-level co-occurrence matrix contrast of the cell image is K1; P2 is the ratio of the impurity image to the contour image f(x), and the mean value of the gray-level co-occurrence matrix contrast of the impurity image is K2. The mean value of the gray-level co-occurrence matrix contrast of the contour image f(x) is K, and the number of gray-level co-occurrence matrix contrasts of the contour image f(x) greater than the threshold σ is I. CELL , the number of values ​​less than the threshold σ is I Noise .

[0026] All images are traversed according to the above calculation to obtain the maximum inter-class variance σ as the third threshold.

[0027] In order to improve the accuracy of the evaluation, the evaluation method is further as follows: OSTU was used to segment the cell ROI feature image, mark the connected domain and sort the size of the connected domain, obtain the image edge with a larger connected domain as the candidate arc for least squares algorithm ellipse fitting, determine the focus, calculate the distance FD from the remaining point set to the high ellipse focus, iterate, and the ellipse fitted by the arc segment with the most equal FD is determined as the cell edge; Use Hough transform to fit the ellipse and calculate the similarity between the two ellipses; Normalize the two ellipses, calculate the variance of each point, and find the sum of their variances as the evaluation value.

[0028] In order to improve the accuracy of the first threshold, further, the first threshold is: The variance value of the cell image was obtained by calculating the Hough transform of the actual ROI template image and iterative direct least squares fitting.

[0029] Example 2

[0030] An example of application of the cell recognition method according to Example 1 of the present invention in T cell recognition in mouse spleen tissue.

[0031] Specifically to this embodiment, Figure 1 As shown, the cell recognition method comprises the following steps: (S1) According to the method of adaptively extracting cell feature images under complex backgrounds, the cell ROI feature image is automatically obtained: after the cell surface is adjusted to be clear, the cell image is captured, the image is split into channels, and the R channel of the image is extracted to obtain the image R(x, y); the image R(x, y) is binarized to obtain the image R bin (x,y), remove the noise image by the maximum connected domain detection method, and perform morphological closing operations. Find the contour and obtain the minimum bounding rectangle to remove the area that is obviously abnormal to the cell. In this example, the second threshold is 1000; perform pairwise calculations on all contour center points to determine whether the contours intersect or contain each other. For the intersecting or contained contours, select the cell contour according to the following method.

[0032] .

[0033] An independent bounding rectangle is obtained, and the image within the outline is automatically segmented by threshold. The contrast of the segmented image is calculated. When the contrast is greater than the threshold σ, it is determined to be a cell outline. In this example, the contrast threshold σ (i.e., the third threshold) is 42. The threshold is determined based on the pre-collected mouse spleen tissue T cell data set. In this example, P1=3 / 4, P2=1 / 4, K1=60, K2=48, and K=57. When the contrast is greater than the threshold σ, it is determined to be a cell outline, otherwise it is an impurity outline.

[0034] (S2) According to the cell ROI image quality assessment method, the quality of the ROI cell feature image is assessed. The original image corresponding to the cell contour is cropped into the ROI region. After OSTU segmentation, the connected region is marked. The edges of the first 1 / 3 of the marked region are directly fitted by the least squares method. The distance FD from the point set on the remaining edge to the focus of the ellipse is calculated. After iteration, the ellipse fitted by the arc segment with the most equal FD is determined to be the cell edge. The Hough transform ellipse is also fitted to the ROI region after OSTU segmentation, and the sum of the variances of the two cell edges is calculated. When the sum of the variances is less than the first threshold, it is regarded as a high-quality cell ROI image, otherwise it is an impurity or low-quality cell image. In this case, the first threshold is 500, which is obtained by calculating the sum of the variances of the cell edges fitted by the template image.

[0035] (S3) The contrast of the ROI cell image after automatic evaluation is input as a feature value into the naive Bayes classifier to identify and locate the cells.

[0036] The recognition results of this embodiment are as follows Figure 2 As shown, this method can accurately identify localized cells.

[0037] Example 3

[0038] An example of application of the cell recognition method according to Example 1 of the present invention in identifying bladder cancer cells.

[0039] Specifically to this embodiment, Figure 1 As shown, the cell recognition method comprises the following steps: (S1) According to the method of adaptively extracting cell feature images under complex background, the cell ROI feature image is automatically obtained: after the cell surface is adjusted to be clear, the cell image is captured, the image is split into channels, and the R channel of the image is extracted to obtain the image R(x,y); the image R(x,y) is binarized to obtain the image R bin (x,y), remove the noise image by the maximum connected domain detection method, and perform morphological closing operations. Find the contour and obtain the minimum bounding rectangle to remove the area that is obviously abnormal to the cell. In this example, the second threshold is 2500; perform pairwise calculations on all contour center points to determine whether the contours intersect or contain each other. For the intersecting or contained contours, select the cell contour according to the following method.

[0040] .

[0041] An independent circumscribed rectangle is obtained, and the image within the outline is automatically segmented by threshold. The contrast of the segmented image is calculated. When the contrast is greater than the threshold σ, it is determined to be a cell outline. In this example, the contrast threshold σ (i.e., the third threshold) is 100. The threshold is determined based on the pre-collected bladder cancer cell data set. In this example, P1=1 / 2, P2=1 / 2, K1=80, K2=60, K=70. When the contrast is greater than the threshold σ, it is determined to be a cell outline, otherwise it is an impurity outline.

[0042] (S2) According to the cell ROI image quality assessment method, the quality of the ROI cell feature image is assessed. The original image corresponding to the cell contour is cropped into the ROI region. After OSTU segmentation, the connected region is marked. The edges of the first 1 / 3 of the marked region are directly fitted by the least square method. The distance FD from the point set on the remaining edge to the focus of the ellipse is calculated. After iteration, the ellipse fitted by the arc segment with the most equal FD is determined to be the cell edge. The Hough transform ellipse is also fitted to the ROI region after OSTU segmentation, and the sum of the variances of the two cell edges is calculated. When the sum of the variances is less than the first threshold, it is regarded as a high-quality cell ROI image, otherwise it is an impurity or low-quality cell image. In this case, the first threshold is 900, which is obtained by calculating the sum of the variances of the cell edges fitted by the template image.

[0043] (S3) The contrast of the ROI cell image after automatic evaluation is input as a feature value into the naive Bayes classifier to identify and locate the cells.

[0044] The recognition results of this embodiment are as follows Figure 3 As shown, this method can accurately identify localized cells.

Claims

1. A cell recognition method based on adaptive ROI feature image extraction, comprising the following steps: (S1) Adaptively extracting cell feature images to obtain cell ROI feature images; (S2) evaluating the quality of the ROI cell feature image, if the evaluation value is less than a first threshold, retaining the cell ROI feature image, if the evaluation value is not less than the first threshold, discarding the cell ROI feature image; (S3) Perform naive Bayes classification on the features of the retained ROI cell images to identify the cells and their locations.

2. The cell identification method according to claim 1, characterized in that: In step (S1), the method of extracting the cell feature image is as follows: After the cell surface is focused to be clear, the cell image is captured, the channels of the original RGB image are split, and the R channel of the image is extracted; The OSTU method is used to segment the image, remove noise and then search for contours to obtain the minimum bounding rectangle of cells in the image and filter the cell contour area.

3. The cell identification method according to claim 2, characterized in that: The method of filtering the cell outline area is: If the area of ​​the circumscribed rectangle is higher than the second threshold, it is confirmed that the material in the region is suspected to be a cell region; if it is not higher than the second threshold, the region is discarded; Remove the contours contained in or intersecting the contour circumscribed rectangle in the image and extract independent contours.

4. The cell identification method according to claim 3, characterized in that: The method for determining whether the independent outline rectangle is contained or intersected is as follows: Suspected cell outline image ; , (x1, y1) and (x2, y2) are the center points of the contours L1 and L2 respectively, Δx and Δy are the x-axis component and y-axis component of the distance between the centers of the two contours respectively, and width and height are the sum of the length and width of the contours L1 and L2 respectively.

5. The cell identification method according to claim 4, characterized in that: Calculate the grayscale co-occurrence matrix contrast value D of the contour area after screening, and judge whether the value D is greater than the third threshold. If the result is yes, the contour is the ROI cell feature contour. If the result is no, it is confirmed as an impurity contour and no subsequent processing is performed.

6. The cell identification method according to claim 5, characterized in that: The method for obtaining the third threshold is: The contour image f(x) includes the cell image I(x,y) and the impurity image N(x,y); ,P1+P2=1; K= P1·K1+P2·K2, σ=P1(K1-K) 2 +P2(K2-K) 2 ; P1 is the ratio of the cell image to the contour image f(x), and the mean value of the gray-level co-occurrence matrix contrast of the cell image is K1; P2 is the ratio of the impurity image to the contour image f(x), and the mean value of the gray-level co-occurrence matrix contrast of the impurity image is K2. The mean value of the gray-level co-occurrence matrix contrast of the contour image f(x) is K, and the number of gray-level co-occurrence matrix contrasts of the contour image f(x) greater than the threshold σ is I. CELL , the number of values ​​less than the threshold σ is I Noise ; All images are traversed according to the above calculation to obtain the maximum inter-class variance σ as the third threshold.

7. The cell identification method according to claim 1, characterized in that: The evaluation method is: OSTU was used to segment the cell ROI feature image, mark the connected domain and sort the size of the connected domain, obtain the image edge with a larger connected domain as the candidate arc for least squares algorithm ellipse fitting, determine the focus, calculate the distance FD from the remaining point set to the high ellipse focus, iterate, and the ellipse fitted by the arc segment with the most equal FD is determined as the cell edge; The sum of the variance values ​​of the corresponding points on the edges of two cells is obtained as the evaluation value.

8. The cell identification method according to claim 7, characterized in that: The first threshold is: The variance value of the cell image was obtained by calculating the Hough transform of the actual ROI template image and iterative direct least squares fitting.

9. The sewage detection method according to claim 1, characterized in that: In step (S1), noise is removed by maximum connected component detection and morphological closing operation.

Images