CN114359325A - An accurate segmentation method of cell images based on digital holographic imaging technology - Google Patents

Abstract

The invention discloses an accurate segmentation method of a cell image based on digital holographic imaging technology, which includes five steps of acquiring a cell image, preprocessing the image, extracting the cell area in the image, identifying a single cell and segmenting the cell image. The method of the invention can quickly obtain the dynamic information of the cell sample without causing damage to the cell sample; meanwhile, for the phenomenon of cell overlap, cell adhesion and the like in the cell image, the accurate segmentation of a single cell can be realized.

Description

An accurate segmentation method of cell images based on digital holographic imaging technology

technical field

The invention relates to the technical fields of digital holographic imaging and cell segmentation, in particular to an accurate segmentation method of cell images based on digital holographic imaging technology.

Background technique

Since the establishment of cell theory in the early nineteenth century, people have recognized that cells are the basic building blocks of life, and biologists have tried to explain the basic theory that cells make up life. In the past few decades, biologists have made many amazing discoveries and research results, but until today, people are still investing more and developing more complex tools to try to further understand the working mechanism of cells, And study how to improve human health through cells.

Advances in cell biology have largely relied on the development of microscopes, and in the late 16th century, humans invented the first optical microscope. The emergence of this great invention has opened the door to a whole new world for human beings, ending the era of observing the world only with the naked eye. The microscope has been an important scientific instrument since its invention, and has been widely used in many fields such as biology, chemistry, physics, metallurgy, brewing, etc., and has made great and outstanding contributions to the development of human beings. In the years since its invention, with the development of science and technology and the improvement of people's requirements for observation, microscope technology has been continuously developed and improved. Its development can be roughly divided into optical microscopes, electron microscopes, scanning tunneling microscopes, scanning probes Microscope such several processes. Digital holographic microscope is the application of digital holographic technology in the field of microscopy, also known as holographic microscopy. Compared with other microscopy techniques, the digital holographic microscope does not directly record the image of the observed object, but records the hologram containing the wavefront information of the observed object, and then numerically reconstructs the recorded hologram through the computer to obtain the image of the observed object. Measure the phase and amplitude (light intensity) information of the object, and then complete the digital three-dimensional reconstruction. With this technique, dynamic information of the cell sample can be obtained quickly without causing damage to the cell sample.

So far, digital holographic microscopy has been widely used in cytological research, such as the study of cell morphology, and the study of cell behavior. These studies require quantitative analysis of cells, such as cell size measurement, cell counting, and There are cell tracking, etc., and these quantitative analyses are very dependent on accurate segmentation of cell images.

Cell segmentation refers to dividing the cell image into several disjoint regions according to features such as grayscale, color, geometric shape, etc., so that these features in the same region show consistency or similarity, while in different regions Noticeably different. Cell segmentation is one of the most basic and important fields in medical image processing. It is the basic premise of identifying and counting cell images. At the same time, how to improve the accuracy and speed of cell segmentation is a key issue in the field of cell segmentation.

In the research of cell segmentation, according to the different characteristics of cell images, researchers have proposed many segmentation algorithms of cell images. The traditional segmentation methods mainly include the following:

(1) Threshold segmentation

Threshold segmentation is a traditional image segmentation method, which is simple to implement, less computationally expensive, and more stable in performance. It is especially suitable for images with objects and backgrounds occupying different gray levels. In many cases, it is a necessary image preprocessing process before cell segmentation and cell identification. A new method of clustering cells proposed by Shirin Nasr-Isfahani et al. uses a combination of image segmentation algorithm and threshold segmentation to extract foreground objects and convert them into binary images. The main limitation of the threshold segmentation method is that the simplest form of the threshold method can only generate binary images to distinguish two classes, and it only considers the pixels themselves, and generally does not consider the spatial characteristics of the image, so it is very sensitive to noise.

(2) Method based on edge detection

The purpose of edge detection is to identify points in a digital image with significant changes in brightness. It can quickly and accurately find the edge, so as to determine the gray or color information in the region through the edge, so as to achieve the rapid segmentation of the image. The determination of edge points is based on the detected point itself and some of its neighboring points, mainly including local differential operators, such as Roberts gradient operator, Sobel gradient operator and Canny operator. The Roberts gradient operator is beneficial to the segmentation of low-noise images with steep edges; the Laplacian operator is isotropic; the Prewitt gradient operator, the Sobel gradient operator, etc. segmentation. For different cell images, there are many other different operators or methods to detect these edge points. For example, Weng Xiumei et al. proposed a method to detect the edge of the image using the phase consistency model and obtain the main geometric structure of the image. Li Tiangang et al. applied multi-scale wavelet transform to edge detection of gastric cancer cell images to solve the segmentation problem of medical pathological cell images with complex textures. A good edge detection operator not only has the differential characteristic to obtain the grayscale change information, it should also be suitable for edge detection at any scale as needed, because the grayscale in the image changes at different scales. The experiment found that the edge information obtained by the edge detection method often produces gaps because the information is not prominent enough, and cannot form a closed curve surrounding the cells. cell boundaries.

In recent years, researchers have conducted in-depth research on various new cell image segmentation algorithms in order to solve the above-mentioned problems in traditional cell segmentation. became mainstream. It can be seen that through the continuous efforts of researchers, cell segmentation technology has achieved very rich scientific research results, and the application of cell segmentation in the field of biological research is becoming more and more extensive, which will undoubtedly make this technology develop in a broader direction .

SUMMARY OF THE INVENTION

The invention discloses an accurate segmentation method of a cell image by digital holographic imaging technology, which is used for improving the segmentation accuracy and segmentation speed of the cell image.

The technical scheme of the present invention is as follows: an accurate segmentation method of cell images based on digital holographic imaging technology, the steps are as follows:

Step 1, acquiring a cell image from a digital holographic imaging system;

Step 2: Preprocess the image and set a threshold to remove image noise points;

Step 3: Extract the cell area in the image, and separate the entire cell area from the background through an adaptive gradient threshold foreground segmentation algorithm;

Step 4: Identify a single cell, use the distance transformation algorithm and the H-minima transformation algorithm to identify the center of the cell, and then complete the identification of the single cell;

Step 5, segment the cell image, and realize the segmentation of the cell image through a marker-controlled watershed algorithm.

Preferably, step 1 uses a three-dimensional dynamic microscopic imaging system based on digital holographic imaging technology, including an optical imaging system and a software algorithm system;

The optical imaging system includes: illumination module, helium-neon laser; imaging module, four microscope objective lenses with different magnifications; beam splitting module, non-polarized beam splitting cube; image acquisition module, CCD industrial camera.

The algorithm of the software algorithm system includes: quantitative phase recovery and three-dimensional reconstruction algorithm based on off-axis holographic interference; phase aberration calibration algorithm based on principal component analysis; spectral sub-pixel displacement aberration compensation algorithm; differential interference phase contrast display algorithm.

Preferably, step 2 is specifically as follows: setting a threshold th, the gray value of the pixel whose gray value is less than the threshold th is set to 0, and the gray value of the pixel whose gray value is greater than or equal to the threshold is unchanged, and th Take 0.45.

Preferably, step 3 is specifically:

Step 3.1: Divide the entire image of size m×n into M×N sub-blocks, where m and n are integer multiples of M and N respectively;

Step 3.2, calculate the gradient histogram of each image sub-block by the Sobel operator, that is, calculate the gradient distribution of the pixels in each sub-image;

Step 3.3, perform Otsu segmentation on each sub-image to obtain the optimal segmentation threshold of each sub-image, and identify cell regions through the threshold.

Preferably, step 4 is specifically:

Step 4.1, the cell image is binarized, the cell area extracted in step 3 is set to white, and the external background area is set to black;

Step 4.2, calculate the binarized cell image through the distance transformation algorithm, identify the initial cell center coordinates and return to obtain the distance image;

Step 4.3, use the H-minima algorithm to suppress the local maximum value in the distance image whose grayscale difference is less than the threshold h, and h is taken as 2.3;

In step 4.4, the coordinates of the cell center are obtained by calculating the centroid of the connected components in the image to complete the identification of a single cell.

Preferably, step 5 is specifically:

Step 5.1, use the morphological opening and closing operation to reconstruct the image of the cell gradient image in step 4, and mark the local minimum value extracted from the reconstructed image, which is an internal marker;

Step 5.2, use the traditional watershed algorithm to segment the distance map obtained by the distance transformation and H-minima algorithm operation in step 4, which is the external marker;

Step 5.3, using the forced minimum technique to generate the local minimum value of the gradient image for the internal marked image and the external marked image generated through step 5.1 and step 5.2;

Step 5.4, perform watershed segmentation on the modified gradient image to obtain the final cell segmentation image.

Compared with the prior art, the present invention has the following advantages: (1) the dynamic information of the cell sample can be quickly obtained without causing damage to the cell sample; (2) for the phenomenon of cell overlap and cell adhesion in the cell image, Accurate segmentation of single cells can be achieved.

Description of drawings

FIG. 1 is a flow chart of an accurate segmentation method of a cell image based on digital holographic imaging technology according to the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application.

As shown in FIG. 1 , it is a flowchart of an accurate segmentation method of a cell image based on the digital holographic imaging technology according to this embodiment. Specific steps are as follows.

Step 1, first acquire cell images from a digital holographic imaging system.

The digital holographic imaging system used in this embodiment is mainly composed of two core parts, an optical imaging system and a software algorithm system. The optical imaging system is mainly responsible for acquiring accurate and stable holographic images, which can be subdivided into: (1) Illumination module: He-Ne laser; (2) Imaging module: four microscope objectives with different magnifications (Olympus RMS4X, RMS10X, RMS20X , RMS40X); (3) Spectral module: non-polarized beam splitter cube (BS019); (4) Image acquisition module: CCD industrial camera. The software algorithm system is the key to the whole system, which is responsible for the restoration, reconstruction and analysis of image information. The specific algorithms include: (1) quantitative phase recovery and three-dimensional reconstruction algorithms based on off-axis holographic interference; (2) based on principal components Analyzed phase aberration calibration algorithm; (3) spectral sub-pixel displacement aberration compensation algorithm; (4) differential interference contrast display algorithm.

The specific operation process of using the digital holographic imaging system can be divided into the following two steps: (1) Connect and debug the device: connect the microscope part to the computer, place the sample close to the camera, and after the camera captures the image of the image plane, connect the The collected image is transmitted to the computer; (2) Start the imaging algorithm: the collected image is displayed by DIC (Digital Differential Interference Contrast Display) through the digital holographic quantitative phase recovery algorithm, and then the acquisition of the cell image can be realized;

The second step is to preprocess the image.

Considering that in order to eliminate the influence of other factors such as uneven brightness and dust in the air that may be caused by poor sealing of the illumination module of the digital holographic imaging system, the imaging information is mixed with extra noise. The image is denoised. In order to protect the information of the original image, the present invention will perform noise reduction processing on the acquired cell image by setting a threshold. Threshold processing is similar to piecewise function processing. A threshold th is set. The gray value of the pixel whose gray value is less than the threshold th is set to 0, and the gray value of the pixel whose gray value is greater than or equal to the threshold is unchanged. The present invention finds that the threshold th is 0.45 is more suitable through a large number of experiments.

Step 3, extract the cell area in the image.

This embodiment separates the entire cell area from the background through an adaptive gradient threshold foreground segmentation algorithm, which mainly relies on generating an appropriate threshold for the different distribution of image gradient features to perform foreground segmentation. The specific steps are as follows: (1) The size of the entire image The image of m×n is divided into M×N sub-blocks, and m and n are integer multiples of M and N respectively; (2) Calculate the gradient histogram of each image sub-block by the Sobel operator, that is, calculate the gradient histogram of each sub-image. Gradient distribution of pixels. The purpose of this is to only consider the pixel points related to the edge information in the sub-image, so that compared with the original histogram, the double peaks remain basically unchanged, and the valleys become deeper; (3) For each The sub-images are subjected to Otsu segmentation. Since in step (1), the valley bottom of the histogram has been strengthened, and by continuing to use the Otsu segmentation method, the optimal segmentation threshold of each sub-image can be easily obtained, and finally the cell region is identified by the threshold.

Step four, identify single cells.

In this embodiment, the distance transformation algorithm and the H-minima transformation algorithm are used to identify the cell center, and then the identification of a single cell is completed. The specific steps are as follows: (1) Binarize the cell image, and set the cell region extracted in step 3 as white , the external background area is set to black; (2) The binarized cell image is calculated by the distance transformation algorithm, and the initial cell center coordinates are identified and returned to obtain the distance image. The basic meaning of distance transformation is to calculate the distance from a non-zero pixel in an image to the nearest zero pixel, that is, the shortest distance to the zero pixel, find the maximum value of the shortest distance, and record the position to get the cell center. Coordinate; (3) suppress the local maximum value in the distance image with the grayscale difference less than the threshold h by the H-minima algorithm, the present invention finds that the threshold h is 2.3 is more suitable through a large number of experiments; (4) by calculating the connected components in the image. The centroid obtains the coordinates of the center of the cell, that is, the identification of a single cell is completed.

Step 5, segment the cell image.

This embodiment realizes the segmentation of the cell image by controlling the watershed algorithm of the markers. The specific steps are as follows: (1) The cell gradient image in step 4 is reconstructed by morphological opening and closing operation, and the reconstructed image is extracted locally. Mark the minimum value, which is the internal marker; (2) Use the traditional watershed algorithm to segment the distance map obtained by the distance transformation and H-minima algorithm in step 4, which is the external marker; The internal label image and the external label image generated in steps (1) and (2) use forced minimum technology to generate the local minimum value of the gradient image; (4) perform watershed segmentation on the modified gradient image to obtain the final cell segmentation image.

The above description of the disclosed embodiments enables any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, this application is not intended to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (6)

1. A cell image accurate segmentation method based on a digital holographic imaging technology is characterized by comprising the following steps:

firstly, acquiring a cell image from a digital holographic imaging system;

preprocessing the image, and setting a threshold value to remove image noise points;

extracting a cell region in the image, and separating the whole cell region from the background through a self-adaptive gradient threshold foreground segmentation algorithm;

identifying single cells, and identifying cell centers by adopting a distance transformation algorithm and an H-minima transformation algorithm so as to complete the identification of the single cells;

and step five, segmenting the cell image, and realizing segmentation of the cell image through a watershed algorithm controlled by the marker.

2. The method for accurately segmenting the cell image based on the digital holographic imaging technology according to claim 1, wherein the step one uses a three-dimensional dynamic microscopic imaging system based on the digital holographic imaging technology, which comprises an optical imaging system and a software algorithm system;

the optical imaging system includes: an illumination module, a helium-neon laser; the imaging module is used for imaging the object; a beam splitting module, a non-polarizing beam splitting cube; an image acquisition module, a CCD industrial camera;

the algorithm of the software algorithm system comprises the following steps: quantitative phase recovery and three-dimensional reconstruction algorithm based on off-axis holographic interference; a phase aberration calibration algorithm based on principal component analysis; a spectral sub-pixel displacement aberration compensation algorithm; differential interference phase contrast display algorithm.

3. The method for accurately segmenting the cellular image based on the digital holography imaging technology according to the claim 1, wherein the second step is specifically as follows: setting a threshold th, setting the gray value of the pixel point with the gray value smaller than the threshold th as 0, and keeping the gray value of the pixel point with the gray value larger than or equal to the threshold unchanged, wherein th is 0.45.

4. The method for accurately segmenting the cellular image based on the digital holography imaging technology according to the claim 1, wherein the third step is specifically as follows:

step 3.1, dividing the whole image with the size of M multiplied by N into M multiplied by N subblocks, wherein M and N are integral multiples of M and N respectively;

step 3.2, calculating a gradient histogram of each image sub-block through a Sobel operator, namely calculating the gradient distribution of pixel points in each sub-image;

and 3.3, performing Otsu segmentation on each subimage to obtain the optimal segmentation threshold of each subimage, and identifying the cell area through the threshold.

5. The method for accurately segmenting the cellular image based on the digital holography imaging technology according to the claim 1, wherein the step four is specifically as follows:

step 4.1, carrying out binarization on the cell image, setting the cell area extracted in the step three as white, and setting the external background area as black;

step 4.2, calculating the binarized cell image through a distance transformation algorithm, identifying an initial cell center coordinate and returning to obtain a distance image;

step 4.3, restraining a local maximum value with a gray difference value smaller than a threshold H in the distance image through an H-minima algorithm, wherein H is 2.3;

and 4.4, obtaining the coordinates of the cell center by calculating the mass center of the connected component in the image, and finishing the identification of the single cell.

6. The method for accurately segmenting the cellular image based on the digital holography imaging technology according to the claim 1, wherein the step five is specifically as follows:

step 5.1, reconstructing the cell gradient image in the step four by using morphological open-close operation, extracting a local minimum value from the reconstructed image and marking the local minimum value to obtain an internal marker;

step 5.2, segmenting a distance map obtained by distance conversion and H-minima algorithm operation in the step four by using a traditional watershed algorithm to obtain an external marker;

step 5.3, generating local minimum values of the gradient images by using a forced minimum technology for the internal marker images and the external marker images generated in the step 5.1 and the step 5.2;

and 5.4, performing watershed segmentation on the modified gradient image to obtain a final cell segmentation image.

Images