CN104794684B - A kind of automatic plaque measurement method based on Virus plaque image - Google Patents

Abstract

The invention belongs to the field of medical image processing and application, and relates to an automatic analysis method for virus plaque diameter. The method comprises: first correcting the illumination inhomogeneity of the original image according to the brightness data of the four corners of the obtained plaque image to obtain an enhanced image; Carry out multi-linear regression prediction on the image color space; classify the image into binary images by pixels according to the regression results; after the binary image is smoothed by morphological closing operation, by counting and counting attributes such as the area of connected regions, the phagocytosis is realized. Automatic analysis of speckle images. The method is simple in operation, efficient and fast, and the determination accuracy is similar to the result of manual measurement by experts, and the time for processing a plaque culture dish image is less than 20 seconds. It can effectively improve the working efficiency of experimental operators.

Description

An automatic plaque assay method based on viral plaque images

technical field

The invention belongs to the field of medical image processing and application. The invention relates to a virus plaque assay method, in particular to an automatic plaque assay method based on a virus plaque image, in particular to an automatic plaque assay method based on multi-linear degeneration supervised segmentation of a plaque culture dish image. This method is suitable for qualitative and quantitative analysis of petri dish images obtained from virus plaque experiments in virology or immunology, and then realizes accurate automatic plaque determination, including automatic calculation of statistical data such as area, diameter and number of plaques .

Background technique

Plaque assay is a widely used experimental method in virology. The principle is that after the virus infects cells, due to the limitation of the solid medium, the released virus can only expand from the initially infected cells to the periphery. After several proliferation cycles, a limited area of diseased cells, namely viral plaques, is formed. Theoretically speaking, a plaque is formed by a virus particle, and the variation of the virus itself and environmental factors can affect the size of the virus plaque. Parameters such as density, diameter, and area of plaques are often concerned in the determination of plaques. This technique is commonly used in virus particle technology and isolation of virus clones ^] .

At present, manual measurement is mostly used in the analysis of viral plaque test results in the world. This manual measurement method requires manual direct or indirect measurement of the plaque diameter on the culture dish image. Usually, multiple petri dishes are involved in an experiment, and there are often dozens of plaques on a petri dish, so that researchers or laboratory staff need to manually measure the data of hundreds of plaques in a plaque experiment , and at the same time need to record, enter and count. This manual measurement method not only increases the researcher's work intensity and prolongs the working time, but also is prone to errors in the recording and entry process.

Up to now, image processing technology has rarely been applied in the analysis of plaque experiment results, and automatic plaque determination has not been reported. The inventors of the present application intend to provide an automatic plaque determination method, especially based on multilinear Automated plaque assay method for supervised segmentation of plaque culture dish images to fill this gap in the art.

The prior art relevant to the present invention has:

[1]W.C.Russell, "A sensitive and precise plaque assay for herpesvirus", Nature, vol.195, pp. 1028-1029, 1962.

[2] E.Gronowicz, A.Coutinho, and F.Melchers, "A plaque assay for allcells secreting Ig of a given type or class", European Journal of Immunology, 6(8):pp.588-590,1976.

[3] J.Yang, J.Lv, Y.Wang, S.Gao, q.Yao, D.Qu, and r.Ye, "Replication of murine coronavirus requires multiple cystines in the endodomain of spikeprotein", Virology, 427( 2), pp.98-106, 2012.

[4] D.W.Hosmer Jr, S.Lemeshow, and R.X.Sturdivant, Applied logistic regression, Wiley Series in Probability and Statistics, 2013.

[5] A.E.Hoerl and R.W.Kennard, "Ridge regression: biased estimation for nonorthogonal problems", Technometrics, 12(1), pp.55-67, 1970.

[6] R.M.Haralock and L.G.Shapiro, computer and robot vision, Addison-Wesley Longman Publishing Co., Inc., vol.I, 1992.

[7] J.Yang, J.Lv, Y.Wang, S.Gao, q.Yao, D.Qu, and r.Ye, "Replication of murine coronavirus requires multiple cystines in the endodomain of spikeprotein", Virology, 427( 2), pp.98-106, 2012.

Contents of the invention

The purpose of the present invention is to overcome the defects of image processing technology in the application of virus plaque analysis in the prior art, to provide a virus plaque assay method, in particular to an automatic plaque assay method based on virus plaque images, especially An automated plaque assay method involving supervised segmentation of plaque culture dish images based on multilinear degradation.

This method is suitable for qualitative and quantitative analysis of petri dish images obtained from virus plaque experiments in virology or immunology, and then realizes accurate automatic plaque determination, including automatic calculation of statistical data such as area, diameter and number of plaques . The method can make the analysis of virus plaque test results more efficient and fast.

The inventive method is realized through the following technical solutions (as shown in Figure 1):

Firstly, after obtaining the petri dish image of the virus plaque experiment in the laboratory, it is used as the original input image for uneven illumination correction to obtain an enhanced uniform background image; secondly, a small number of plaque pixels and background pixels are manually marked through the interactive interface points, so that the RGB color data of two groups of pixels can be collected, and they will be used as training data to estimate the degradation parameters in the segmentation method based on multi-linear degradation. Here, the Logistic and Ridge degradation techniques have been tested respectively; once the degradation is estimated parameter, it can be applied to each pixel of the entire enhanced image, and the pixels are classified to obtain a binary image marked as a plaque and a non-plaque area; finally, the binary image is squared with a size of 5 pixels The morphological closed operation of the structural elements is smooth, and the attributes of the connected blocks on the obtained image can be statistically analyzed to obtain initial statistical data, and then the plaque measurement data can be obtained by screening these data, including the number of plaques, diameter and area etc.

More specifically, a kind of automatic plaque determination method based on virus plaque image of this method is characterized in that, it comprises steps:

Firstly, the illumination inhomogeneity of the original image is corrected according to the brightness data of the four corners of the obtained plaque image, and the enhanced image is obtained; then the plaque and the background are sampled by manually marking a small number of points on the enhanced image, and the image color space is performed on the enhanced image. Multi-linear regression prediction on the Internet; the regression results are used to classify the image pixel by pixel into a binary image; after the binary image is smoothed by morphological closing operation, the automatic analysis of the plaque image is realized by counting and counting attributes such as the area of connected regions .

In the present invention, the non-uniform brightness correction method based on bilinear interpolation is applied to each plaque culture dish image to enhance the surface information of the image; when the original image is corrected for uneven illumination, the four-corner blank area pixels are sampled, and the The mean value of the blank area uses bilinear interpolation to estimate the illumination bias field, and the original image is subtracted from the estimated bias field to obtain the enhanced image;

In this method, 20 marker points are randomly selected from the image, among which 10 points are taken from the plaque and 10 points are taken from the medium background, and their brightness attributes are used as the training set to estimate the parameters of the multi-linear degradation, The probability distribution of the entire image is further estimated, so that the plaques in the final segmented image; in the embodiment of the present invention, 20 plaques and pixels on the background are artificially sampled on the enhanced image, and their (color) The brightness gray value can be used as a training sample, which can be substituted into the multi-linear regression to estimate the regression parameters. It can map each pixel in the image to the probability classification of plaque pixels and non-plaque pixels. The probability value of plaque pixels is set to 0.75. The probability value for non-plaque background pixels was set to 0.25.

In this method, the Logistic regression method and the Ridge regression method are used; when using the Logistic regression, set the probability threshold to 0.5, and when using the Ridge regression, the threshold is the mean value of the results of the training sample calculation (see step 2 of the inventive method), by Therefore, the probability image is converted into a binary image.

In this method, a plurality of plaques that are incomplete or combined on the binary image are eliminated by using the noise removal technology; Spots such as incomplete plaques, combined superimposed plaques, and the background of non-petri dish areas, and then count the number, diameter, and area of the remaining connected areas, so as to realize the automatic determination of viral plaques.

The experimental results show that this method can effectively remove the heterogeneous spots, combined plaques, incomplete plaques and irregular background outside the petri dish, and the remaining connected blocks are the isolated plaques detected automatically Connected blocks, which will be used to count the determination parameters such as diameter, area and density of plaques.

The precision of this method is similar to the result of manual measurement by experts, but the time spent by this method to process a plaque culture dish image is not more than 20 seconds.

The present invention has the following advantages:

(1) The operation is simple, efficient and fast. Only a few calibration points are used as the estimated parameters of the prior training set to quickly segment any petri dish images obtained by different operators and under different conditions;

(2) The accuracy of the results is very good, basically similar to the results of manual measurement by different experts;

(3) High stability and repeatability;

(4) Improve the work efficiency of pathological experiment operators;

(5) The present invention has good adaptability to plaques in different medium backgrounds, and can provide a good interaction mechanism through a graphical user interface.

Description of drawings

Fig. 1 is a flow chart of the data collection operation of the method of the present invention.

Figure 2 is the correction of non-uniform brightness field.

Figure 3 is an example of plaque segmentation results, where (a) is the original image, and (b) is the segmented image, where the black areas are plaques.

detailed description

Example 1

In this method, it is assumed that the color image of the petri dish as a result of the plaque experiment represents I(x, y, k), and its size is M×N×K, where K is the total number of channels, x and y are the pixel positions, and k is the The number of channels, then, through the following three steps to realize the details of the specific method, automatic plaque determination can be carried out:

1) Correction of image inconsistent brightness illumination,

Since I(x, y, k) has inconsistent brightness illumination, and the four corners of the image, i.e. upper left, upper right, lower left, and lower right, can reflect changes in brightness, therefore, it is necessary to use the blank parts of the four corners to estimate the uneven illumination field , so as to correct the original image illumination;

In this method, for each channel image, its brightness values on the upper left, upper right, lower left, and lower right are obtained, that is, I(0,0,k), I(0,N-1,k), I(M- 1, 0, k) and I(M-1, N-1, k), after which a bilinear interpolation algorithm is used to estimate the luminance field F(x, y, k), namely:

This will be used to correct the original image I to obtain the enhanced image I ^E , namely:

The resulting enhanced image is shown in Figure 2.

2) Supervised plaque segmentation based on multi-linear degradation,

2P plaque pixels and background pixels (P = 10) were manually selected on the enhanced image through an interactive interface, and the attributes of these points were {I ^E ( _xi , y, k), _i = 1, 2, . .., 2P; k=1, 2, ..., K} is used as the training set to estimate the parameters of the multi-linear regression, wherein, this method uses Logistic regression and Ridge regression;

The Logistic regression: suitable for predicting binary responses, in this method, the probability that any pixel with K attributes (brightness) in the enhanced image belongs to a specific class can be calculated as follows:

Logit(i)＝b ₀ +b ₁ I ^E ( _xi , y _i , 1)+b ₂ I ^E ( _xi , y _i , 2)...+b _K I ^E ( _xi , y _i , K), (4)

Then the classification probability of 2P points can be expressed in matrix form:

This can be further reduced to:

Y=XB, (6)

Then the degradation parameter B can be unbiasedly estimated using the following linear model:

B=(X'X ^-1 )X'Y. (7)

The Ridge degradation: when the image or point set information is incomplete, the Ridge degradation can replace the unbiased estimate for the best estimate, and the Ridge degradation introduces α times of the identity matrix, that is, αE, to minimize the mean square error, thus, the degradation parameter B can be calculated as follows,

B=(X'X+αE) ^-1 X'Y. (8)

Wherein, α=0.10.

In these two degradations, setting the plaque prediction probability to 0.75 and the background medium probability to 0.25 means that the response vector Y = [0.75,...,0.25,...] ^T , thus, one can estimate Parameter B, which can be used to estimate the probability of all pixels on the enhanced image; for Logistic degeneration, the threshold for transforming the resulting probability image into a binary image is 0.5; for Ridge degeneration, the binary threshold is the mean value of the XB operation results.

3) Elimination of binary image noise and calculation of plaque determination parameters,

When there are holes and noise points on the resulting binary image, the square structure element with a size of 5 pixels is used to perform morphological closing operation on the binary image to eliminate these interference noises;

Finally, apply the Flood-fill algorithm to the obtained binary image to perform statistics on each connected region, such as recording the center of the connected region, the length of the major axis, and the length of the minor axis. From these measured data, perform the following steps in sequence :

(1) Remove the connected blocks outside the petri dish;

(2) Remove connected blocks whose major axis is greater than 1.5 compared with the minor axis;

(3) Remove connected point sets whose area is greater than 2.5 times the average area obtained when connected blocks are not excluded.

The results show that this method can effectively remove the heterogeneous spots, combined plaques, incomplete plaques and irregular background outside the culture dish, and the remaining connected blocks are the connected blocks of each isolated plaque automatically detected. Blocks, which will be used to count the measurement parameters such as diameter, area and density of plaques.

Example 2

In this example, the average plaque diameter calculated for each plaque assay image provided by three different virologists was compared with the manual results of experts, the results of Logistic regression and Ridge regression, and the results showed that, The accuracy of this method is similar to the result of manual measurement by experts, but the time spent by this method to process a plaque culture dish image is not more than 20 seconds.

Table 1 is the average diameter of plaques calculated on each plaque assay image provided by three different virologists. The manual results of experts, the results of Logistic regression and Ridge regression are compared here, (unit: pixel ).

Table 1.

Claims (5)

1. A kind of 1. automatic plaque measurement method based on Virus plaque image, it is characterised in that it includes step:

   The even property of corner regional luminance data correction original image uneven illumination of gained plaque image is first depending on, wherein, sampling Corner white space pixel, illumination deviation field is estimated using bilinear interpolation by the average of corner white space, original image with Estimated bias field subtract each other strengthened after image；

   Then by the way that a small amount of mark point samples to plaque and background by hand on image after enhancing, wherein, in enhanced image Pixel on upper artificial sample 20 plaques and background, its luminance grayscale values are substituted into Multilinear Regression and estimated as training sample Regression parameter is counted, by the probabilistic classification of each pixel-map in image to plaque pixel and non-plaque pixel；To scheming after enhancing As carrying out the Multilinear Regression prediction on image color space,

   Image is pressed as bianry image by pixel classifications using regression result, wherein, use Logistic homing methods and Ridge to return Method；When being returned using Logistic, probability threshold value is set as 0.5, and when being returned using Ridge, threshold value is training sample The result average of computing, thus, probabilistic image is switched into bianry image；

   After taking closing operation of mathematical morphology smooth bianry image, by the quantity of plaque in connected region, diameter, area system Meter, realizes plaque automated image analysis.
2. 2. the automatic plaque measurement method based on Virus plaque image as described in claim 1, it is characterised in that described bites The probable value of spot pixel is set to 0.75, and the probable value of non-plaque background pixel is set to 0.25.
3. 3. the automatic plaque measurement method based on Virus plaque image as described in claim 1, it is characterised in that described side In method, incomplete on bianry image or combination multiple plaques are eliminated using miscellaneous removal technology.
4. 4. the automatic plaque measurement method based on Virus plaque image as described in claim 3, it is characterised in that described side In method, closing operation of mathematical morphology is carried out to bianry image, and after discharging the miscellaneous point of the noise in connected region, to remaining connected region The statistics of quantity, diameter and area is carried out, so as to realize automatically determining for Virus plaque.
5. 5. the automatic plaque measurement method based on Virus plaque image as described in claim 4, it is characterised in that described makes an uproar The miscellaneous point of sound is imperfect plaque, the superposition plaque of combination, miscellaneous background of non-culture dish region.