Method for processing images of cells in suspension

Abstract

The application provides a processing method of a suspended cell image, which comprises the following steps: obtaining a cell region and a background region after image texture segmentation of the suspended cell image according to the difference between texture features in the cell region and the background region; detecting the suspended cell image by using a Hough transform to obtain a plurality of circles corresponding to a plurality of round suspended cells in the image, wherein the plurality of circles comprise a blank circle, and the blank circle is a false cell detected by the image background noise; judging whether the centers of the plurality of circles are in the cell region after the image texture segmentation, and processing the centers to obtain an image g4(x, y), wherein the processing manner is: if the result is yes, the circle is reserved; if the result is no, the circle is deleted, so that the blank circle is eliminated. The application has the advantages of high processing speed and high processing precision.

Description

Technical Field

[0001] This invention relates to cell analysis, and more particularly to methods and techniques for processing images of suspended cells. Background Technology

[0002] Suspension cells differ from adherent cells in that they grow in suspension in the cell culture medium, such as... Figure 1 As shown, the captured images of suspended cells exhibit partial cell overlap, and are often accompanied by halo artifacts, low contrast between cells and background, background noise, and large-scale cell adhesion in the later stages of culture. Existing segmentation methods mainly include:

[0003] 1. Traditional Otsu thresholding segmentation produces the following results: Figure 2 As shown in (b), or by using the double Gaussian thresholding method, which has good segmentation performance in adherent cells, the results are as follows: Figure 2 As shown in (c) Figure 2 (a) is a cell image. As can be seen from the segmentation results, these methods cannot achieve accurate and effective segmentation. The segmented cell outlines have obvious jagged edges and cannot segment adhered and overlapping cells.

[0004] 2. Based on the region histogram, the processing result is as follows: Figure 2 As shown in (d), or by combining preprocessing, Canny edge blending, and morphological methods, the processing result is as follows: Figure 2 As shown in (e), adhered and overlapping cells cannot be correctly separated. Summary of the Invention

[0005] To address the shortcomings of the existing technical solutions, the present invention provides a method for processing suspended cell images.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A method for processing images of suspended cells, wherein the method for processing images of suspended cells is as follows:

[0008] For the suspended cell image, the cell region and background region after image texture segmentation are obtained based on the different texture features of the cell region and the background region;

[0009] The suspended cell image was detected using Hough transform, and the resulting image contained multiple circles corresponding to multiple round cells, with blank circles among the multiple circles;

[0010] Determine whether the centers of the multiple circles are within the cell regions after image texture segmentation, and process them to obtain image g4(x, y). The processing method is as follows:

[0011] If the result is yes, keep the circle;

[0012] If the result is negative, delete the circle, thereby eliminating the blank circle.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0014] 1. High accuracy;

[0015] By using the texture segmentation results of suspended cell images to filter the center of Hough circle detection, the interference caused by image background noise can be effectively reduced, the occurrence of false cell detection (processing of blank circles) in Hough circle detection results can be significantly reduced, and the accuracy of image processing can be improved.

[0016] By intersecting the binarized results of texture quantization and segmentation using two parameters, the results of different quantization parameters of image texture can be comprehensively considered, reducing the error rate of texture quantization and cell region segmentation by a single parameter.

[0017] 2. Fast processing;

[0018] By performing image texture quantization on the suspended cell image using local standard deviation and local entropy values ​​respectively, and then performing automated thresholding (the threshold automatically obtained when the inter-class variance is at its maximum), this method can easily and quickly classify the two types of pixels based on the characteristics of coarse cell regions and smooth background regions, which can meet the requirements of real-time segmentation and counting of suspended cell images in actual production.

[0019] Meanwhile, the threshold automatically selected based on the rule of maximizing inter-class variance avoids the need for manual adjustment of segmentation threshold parameters, thus improving image processing speed. Attached Figure Description

[0020] The disclosure of this invention will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are merely illustrative of the technical solutions of this invention and are not intended to limit the scope of protection of this invention. In the drawings:

[0021] Figure 1 This is a schematic diagram of a suspended cell image;

[0022] Figure 2 This is a schematic diagram of various image processing results in existing technologies;

[0023] Figure 3 This is a schematic diagram of the image texture feature segmentation results;

[0024] Figure 4 It is a schematic diagram of the processing results at each stage of the image processing process;

[0025] Figure 5 This is a diagram showing the comparison before and after image processing;

[0026] Figure 6 It is the cell growth curve corresponding to the dataset. Detailed Implementation

[0027] Figure 1-6 The following description illustrates optional embodiments of the invention to teach those skilled in the art how to implement and reproduce the invention. Some conventional aspects have been simplified or omitted to explain the technical solutions of the invention. Those skilled in the art should understand that variations or substitutions derived from these embodiments will be within the scope of the invention. Those skilled in the art should understand that the following features can be combined in various ways to form multiple variations of the invention. Therefore, the invention is not limited to the following optional embodiments, but is defined only by the claims and their equivalents.

[0028] Example 1:

[0029] A method for processing images of suspended cells, wherein the method for processing images of suspended cells is as follows:

[0030] For the suspended cell image, the cell region and background region after image texture segmentation are obtained based on the different texture features of the cell region and the background region;

[0031] The suspended cell image was detected using Hough transform, and the resulting image contained multiple circles corresponding to multiple round cells, with blank circles among the multiple circles;

[0032] Determine whether the centers of the multiple circles are within the cell regions after image texture segmentation, and process them to obtain image g4(x, y). The processing method is as follows:

[0033] If the result is yes, keep the circle;

[0034] If the result is negative, delete the circle, thereby eliminating the blank circle.

[0035] To improve processing accuracy, the image texture segmentation method is further as follows:

[0036] The texture of the suspended cell image is quantized using local standard deviation to obtain a first image f1(x, y), which is then thresholded to obtain an image g1(x, y). In this image, areas larger than the first threshold T1 are designated as cell regions, and areas not larger than the first threshold T1 are designated as background regions; (x, y) represents the coordinates of a pixel in the image.

[0037] The local entropy value is used to quantize the texture of the suspended cell image to obtain a second image f2(x,y) and then threshold segmentation to obtain an image g2(x,y). In this image, the regions greater than the second threshold T2 are taken as cell regions, and the regions not greater than the second threshold T2 are taken as background regions.

[0038] By performing an intersection operation on images g1(x,y) and g2(x,y), a third image g3(x,y) is obtained.

[0039] To improve processing accuracy, morphological operations are further performed on the third image to obtain cell regions and background regions after image texture segmentation. The morphological operations include morphological opening operations, morphological closing operations, and area constraints.

[0040] To automate the acquisition of the threshold and improve processing speed, the first threshold is further obtained in the following way:

[0041] In the first image, the threshold T is obtained when the inter-class variance between the cell class and the background class is the largest, and the first threshold T1 is 0.5T.

[0042] To automate the acquisition of the threshold and improve processing speed, the second threshold T2 is further obtained as follows:

[0043] In the second image, the threshold T is obtained when the inter-class variance between the cell class and the background class is the largest, and the second threshold T2 is T.

[0044] To improve processing accuracy, further,

[0045]

[0046] To effectively remove blank circles, further, the center positions (X...) of the plurality of circles... i ,Y i ), corresponding to radius R i , i = 1, 2, ..., N; the processing method is as follows:

[0047] The position of the center of the circle in the image g4(x,y) radius

[0048] Example 2:

[0049] An application example of the method for processing suspended cell images according to Embodiment 1 of the present invention.

[0050] In this application example, the method for processing the suspended cell image is as follows:

[0051] For the suspended cell image, the cell region and background region after image texture segmentation are obtained based on the different texture features of the cell region and the background region. The specific method is as follows:

[0052] Local standard deviation (LSD) and local entropy (LE) were used to quantify the texture features of the suspended cell image. When using LSD to quantize the image texture, a larger LSD indicates a less smooth region, i.e., a cellular region; a smaller LSD indicates a smoother region, i.e., a background region. When using LE to quantize the image texture, a larger LE indicates a region containing more information, i.e., a less smooth region, i.e., a cellular region; a smaller LE indicates a region containing less information, i.e., a smoother region, i.e., a background region.

[0053] The local standard deviation (LSD) and local entropy (LE) are calculated as follows:

[0054] For a local region centered at (x, y) with a window size of (2n+1)*(2n+1), where p(x, y) is the gray value at point (x, y), then the average gray value m within this local region... p (x, y) is:

[0055]

[0056] The local standard deviation (LSD) is:

[0057] If p h p represents the proportion of pixels with a gray value of h in an image, where p h The local entropy value LE of this region can be obtained by using the normalized histogram of the local region.

[0058] After quantizing the texture of the suspended cell image using local standard deviation and local entropy values, the following was obtained: Figure 3 (b) Figure 3 The image texture shown in (d) is a texture image. Figure 3 (a) is the original cell image. From the texture quantization maps of the two different images, it can be seen that the brighter parts of the image, i.e., the parts with higher gray values, represent the cell regions, while the darker parts, i.e., the parts with lower gray values, represent the background. This is consistent with the characteristic analyzed earlier that cell regions are not smooth and contain more information, while background regions are smoother and contain less information.

[0059] To classify cellular and background regions, a thresholding method is used to classify the parameter-quantized texture image; regions below the threshold are classified as background, and regions above the threshold are classified as cellular. To avoid manual adjustment of the threshold parameters, an automatic thresholding method is proposed for texture image segmentation, based on the principle that the optimal threshold T is found when the inter-class variance between the two classes (cellular pixels and background pixels) in the image reaches its maximum. After multiple experimental verifications, the threshold T1 = T / 2 for texture images obtained using the LSD parameter and T2 = T for texture images obtained using the LE parameter were determined. Using these two automatic thresholds to segment the texture image yielded the following results: Figure 3 Two binarization results are shown ( Figure 3 (c) Figure 3 (e)).

[0060] After obtaining the thresholds T1 and T2, threshold segmentation can be performed on the corresponding texture images f1(x,y) and f2(x,y) to obtain the corresponding binary images g1(x,y) and g2(x,y). The portion larger than the threshold is considered a cell region and represented as white in the image; the portion smaller than the threshold is considered a background region and represented as black in the image. That is:

[0061]

[0062] To improve the accuracy of image segmentation, the two different threshold segmentation results are intersected to obtain image g3(x, y), as shown below. Figure 4 As shown in (b). Figure 4 (a) is the original image.

[0063] The method for handling intersections is as follows:

[0064]

[0065] Simultaneously, a series of morphological operations were performed on the binarized result to optimize the segmented cell contours. The morphological opening operation was used to break some fine connections; the morphological closing operation was used to fill some small holes; and the area constraint was used to eliminate some small impurities in the image. The final result is as follows: Figure 4 As shown in (c), the cell outlines are extracted and mapped back to the original image, as shown. Figure 4 As shown in (d), it can be seen that the image texture segmentation method accurately distinguishes the cell region from the background region.

[0066] Using the Hough circular transform to detect suspended cell images yields multiple circles corresponding to each roundish cell, and the center position (X) of these multiple circles is determined. i ,Y i ), corresponding to radius Ri , i = 1, 2, ..., N; there are blank circles among the multiple circles.

[0067] To eliminate these blank circles, this method uses the cell regions segmented by texture as a reference and combines them with the Hough circular transform. The center of the circle detected by Hough is filtered based on whether it is within a cell region. If the center is within a cell region, the center and its corresponding radius are retained; if the center is not within a cell region, the center and its corresponding radius are removed. This process eliminates false cell detections caused by background noise, thus removing blank circles from the Hough detection results and obtaining image g4(x, y). The filtering method is as follows:

[0068] The position of the center of the circle in the image g4(x,y) radius

[0069] The final segmentation result is as follows Figure 5 As shown in the third column, Figure 5 The first column is the original image, the second column is the image texture feature segmentation result, and the contour curve of the cell region is mapped onto the original image. The third column is the final segmentation result of this method. Figure 5 The cell density varies in each row, demonstrating that this method achieves good segmentation results when processing suspended cell images with sparse, dense, overlapping, or adherent structures.

[0070] The segmentation results can be quantitatively analyzed using cell counting, and the effectiveness of image segmentation can be measured using precision, recall, and F-score. The expressions for precision, recall, and F-score are as follows:

[0071]

[0072] The cell counting results for different suspended cell images are shown in the table below. The manual counting result GT is the cell count obtained by experts using specialized annotation software to label the cell images, and this is considered the true cell count. EN represents the number of cells identified by the algorithm, TP represents the number of cells correctly identified by the algorithm, FP represents the number of incorrectly identified cells, and FN represents the number of cells not identified by the algorithm. Higher precision indicates a lower probability of false positives; higher recall indicates a lower probability of false negatives; the F-value is the harmonic mean of precision and recall, which measures the overall performance of the algorithm. A higher F-value indicates better detection performance.

[0073]

[0074] Because this method offers fast segmentation and counting speeds, it can meet the real-time segmentation and real-time cell counting requirements in actual cell culture. It segments cell images and plots real-time cell growth curves, such as... Figure 6 As shown, the timing of cell passage can be determined based on the curve's trend, slope changes, and cell count, among other relevant parameters.

Claims

1. A method for processing images of suspended cells, wherein the method for processing images of suspended cells is as follows: For the suspended cell image, the cell region and background region after image texture segmentation are obtained based on the different texture features of the cell region and the background region; The suspended cell image was detected using Hough transform, and the resulting image contained multiple circles corresponding to multiple round cells, with blank circles among the multiple circles; Determine whether the centers of the multiple circles are within the cell regions after image texture segmentation, and process them to obtain image g4(x, y). The processing method is as follows: If the result is yes, keep the circle; If the result is negative, delete the circle, thereby eliminating the blank circle.

2. The method for processing suspended cell images according to claim 1, characterized in that, The image texture segmentation method is as follows: The local standard deviation is used to quantize the texture of the suspended cell image to obtain the first image f1(x,y) and threshold segmentation to obtain the image g1(x,y). In this image, the regions larger than the first threshold T1 are taken as cell regions, and the regions not larger than the first threshold T1 are taken as background regions. (x, y) represents the coordinates of a pixel in the image; The local entropy value is used to quantize the texture of the suspended cell image to obtain a second image f2(x,y) and then threshold segmentation to obtain an image g2(x,y). In this image, the regions greater than the second threshold T2 are taken as cell regions, and the regions not greater than the second threshold T2 are taken as background regions. By performing an intersection operation on images g1(x,y) and g2(x,y), a third image g3(x,y) is obtained.

3. The method for processing suspended cell images according to claim 2, characterized in that, Morphological operations are performed on the third image to obtain the cell region and background region after image texture segmentation.

4. The method for processing suspended cell images according to claim 3, characterized in that, The morphological operations include morphological opening operations, morphological closing operations, and area constraints.

5. The method for processing suspended cell images according to claim 2, characterized in that, The first threshold is obtained as follows: In the first image, the threshold T is obtained when the inter-class variance between the cell class and the background class is the largest, and the first threshold T1 is 0.5T.

6. The method for processing suspended cell images according to claim 2, characterized in that, The second threshold T2 is obtained as follows: In the second image, the threshold T is obtained when the inter-class variance between the cell class and the background class is the largest, and the second threshold T2 is T.

7. The method for processing suspended cell images according to claim 2, characterized in that, 8. The method for processing suspended cell images according to claim 2, characterized in that, The center positions of the plurality of circles (X) i ,Y i ), corresponding to radius R i , i = 1, 2, ..., N; the processing method is as follows: The position of the center of the circle in the image g4(x,y) radius

Images