A classification detection system based on cell fluorescence images

Abstract

The application discloses a kind of classification detection systems based on cell fluorescence image, including: image acquisition module, image processing module and type labeling module;First, image acquisition module generates four kinds of fluorescence characteristic images and Merge characteristic images with four kinds of fluorescence characteristics through orthosteric microscope;Second, image processing module is preprocessed, positioning, cell cluster segmentation, cell feature extraction and classification to fluorescence image;Finally, image type labeling module labels the type of cell area image classified by image processing module, obtains the cell area image with annotation.The application carries out classification detection to the fluorescence image collected, can accurately locate cell position and effectively identify different cell types, can accurately segment for the adherent cell existing in image, and also can effectively identify for the boundary cell in field of view, fluorescence brightness too dark cell.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cell type detection, and particularly relates to a classification detection system based on cell fluorescence images. TECHNICAL BACKGROUND

[0002] At present, the classification of fluorescent cells is in the form of artificial detection. Firstly, the method of artificially identifying the type of cell fluorescence images needs to analyze a large number of images, which is time-consuming and labor-intensive. Secondly, the recognition efficiency of the method of artificially identifying the type of cell fluorescence images is relatively low, and it is difficult to meet the requirements of rapid and accurate recognition. Most importantly, artificial recognition is affected by the environment, so it is easy to be disturbed and prone to errors, which affects the recognition accuracy. SUMMARY

[0003] The present application aims to solve the problems of low artificial detection efficiency and unguaranteed detection accuracy, and provides a classification detection system based on cell fluorescence images, so as to process a large number of fluorescence images and classify them, thereby shortening the detection time and improving the accuracy of cell recognition.

[0004] In order to achieve the above-mentioned application purposes, the present application adopts the following technical solutions:

[0005] The classification detection system based on cell fluorescence images comprises an image acquisition module, an image processing module and a type labeling module.

[0006] The image acquisition module generates four fluorescence feature images, including a DAPI feature image, a CD45 feature image, a Vimentin feature image and an EpCAM feature image, through an upright microscope, and fuses the four fluorescence feature images generated by the upright microscope to obtain a Merge feature image with four fluorescence features.

[0007] The image processing module comprises a fluorescence image preprocessing unit, a fluorescence cell cluster segmentation unit, a fluorescence cell positioning unit, a fluorescence cell feature extraction unit and a fluorescence cell classification unit.

[0008] The fluorescence image preprocessing unit performs preprocessing operations on the Merge feature image, including converting the three-channel Merge image into a single-channel gray image, then using a Gaussian filtering method to filter the noise of the gray image to obtain a filtered single-channel gray image, and then using an OTSU adaptive threshold segmentation method to perform thresholding processing on the filtered single-channel gray image to obtain a binary image of the fluorescence image, wherein the white area in the binary image is a cell area and the black area is a background area.

[0009] The fluorescent cell cluster segmentation unit segments the cell clusters in the binary image to obtain a plurality of cell region images;

[0010] The fluorescent cell positioning unit extracts the cell region edges in the cell region image to obtain a cell contour edge point set, and then calculates the center point of the cell using a statistical method;

[0011] The fluorescent cell feature extraction unit converts the four fluorescent feature images into four HSV images respectively, sets the cell region of interest of each HSV image according to the center point of the cell, and then calculates the brightness feature of the cell region of interest on each HSV image to obtain the brightness feature BF d of the fluorescent region of interest in the DAPI feature image, the brightness feature BF c of the fluorescent region of interest in the CD45 feature image, the brightness feature BF v of the fluorescent region of interest in the Vimentin feature image, and the brightness feature BF e of the fluorescent region of interest in the EpCAM feature image.

[0012] The cell classification unit classifies the cell region image according to the four brightness features BF d , BF c , BF v , and BF e .

[0013] Step a, if BF d > Δ1, the cell in the cell region image is classified as a live cell, and step b is performed; otherwise, the cell in the cell region image is classified as a dead cell, and the classification is ended.

[0014] Step b, if BF c < Δ2, the cell in the cell region image is classified as an abnormal cell, and step d is performed; otherwise, step c is performed; wherein Δ1 and Δ2 represent two brightness thresholds, and Δ2 < Δ1.

[0015] Step c, it is judged whether BF v / BF c ≤ δ and BF e / BF c ≤ δ are true or not, if true, the cell in the cell region image is classified as a normal cell, and the classification is ended; otherwise, step d is performed; wherein δ represents a proportional brightness threshold.

[0016] Step d, if BF v / BF c ≤ δ and BF e / BF c> delta, then the cell in the cell region image is classified as an epithelial type abnormal cell;

[0017] If BF v BF c > delta and BF e BF c > delta, then the cell in the cell region image is classified as an epithelial type abnormal cell;

[0018] If BF v BF c > delta and BF e BF c > delta, then the cell in the cell region image is classified as an epithelial type abnormal cell;

[0019] The image type labeling module labels the type of the cell region image classified by the image processing module, and obtains the cell region image with the label.

[0020] Compared with the prior art, the present application has the following advantages:

[0021] 1. The present application classifies the fluorescence cell image by using the image processing method, which can greatly improve the efficiency of image processing and complete a large number of cell fluorescence image classification tasks in a short time. In addition, it can cope with various environmental changes and is not easily affected by external interference, so that the recognition of cell fluorescence image type and characteristics is more accurate, which is of great help to the subsequent auxiliary judgment.

[0022] 2. The present application classifies the cell image by using the image processing technology based on the cell fluorescence image collected by the upright microscope, greatly shortens the time of artificial classification of cells, and has certain guarantee for the accuracy rate. At the same time, a kind of automatic cell classification system is disclosed, which is used for medical automatic detection industry and has certain reference function for initial screening of pathological conditions. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is the flow chart of the fluorescence cell image classification method of the present application.

[0024] Figure 2 is the schematic diagram of the adhesion cell segmentation method of the present application.

[0025] Figure 3 is the decision tree for cell classification of the present application. DETAILED DESCRIPTION

[0026] In this embodiment, as Figure 1As shown, a classification detection system based on cell fluorescence image is proposed to solve the problems of low efficiency and low classification accuracy caused by manual detection. The collected fluorescence image is preprocessed to remove the influence of interfering fluorescent point noise. The preprocessed fluorescent cells are located to determine the specific position of the disordered distributed cells. At the same time, the multiple adherent cells in the field of view are segmented to prevent the adherent cells from being treated as one cell. The segmented adherent cells can be effectively located, and the accuracy of cell classification is improved. Then, four brightness features BF d , BF c , BF v , BF e of the cell fluorescence image are extracted for cell region image classification. Finally, the classification results are automatically labeled. Specifically, the system includes an image acquisition module, an image processing module, and a type labeling module.

[0027] The image acquisition module generates four fluorescence feature images, including DAPI feature image, CD45 feature image, Vimentin feature image, and EpCAM feature image, through an upright microscope. The four fluorescence feature images are fused through the upright microscope to obtain a Merge feature image with four fluorescence features. In a specific implementation, the cell fluorescence image is obtained by injecting different cell membrane fluorescent markers into cells. The substance can be absorbed by specific substances in the cells, thereby forming a fluorescence image with different fluorescence reactions. By enlarging or reducing the fluorescence image, the position and characteristics of the cells can be observed, thereby realizing the recognition and classification of the cells. In this embodiment, the fluorescence image is collected by the upright microscope to obtain the cell fluorescence image to be classified.

[0028] The image processing module includes a fluorescence image preprocessing unit, a fluorescence cell cluster segmentation unit, a fluorescence cell positioning unit, a fluorescence cell feature extraction unit, and a fluorescence cell classification unit.

[0029] The fluorescence image preprocessing unit performs preprocessing operations on the Merge feature image, including converting the three-channel Merge image into a single-channel grayscale image, then using a Gaussian filter method with a Gaussian kernel size of 9x9 to filter the noise of the grayscale image to obtain a filtered single-channel grayscale image, and then using an OTSU adaptive threshold segmentation method to perform thresholding processing on the filtered single-channel grayscale image to obtain a binary image of the fluorescence image. The white area in the binary image is the cell region, and the black area is the background region.

[0030] The fluorescence cell cluster segmentation unit segments the cell clusters in the binary image to obtain a plurality of cell region images. In a specific implementation, the fluorescence cell segmentation method includes the following steps:

[0031] S1: finding the minimum convex hull of the cell binary image;

[0032] S2: finding the maximum concave figure of the cell binary image, i.e. the region with the gray value of 255 inside the cell;

[0033] S3: subtracting the minimum convex hull of S1 from the maximum concave figure of S2 to obtain the concave region on both sides of the adherent cell cluster, at this time, the gray value of the concave region on both sides is set to 255 white region, refer to (b) in Figure 2 ; the contour extraction is performed on the concave region, and the extracted contour point set is stored in a point set container. The two contour containers are iterated, and the distance between the two contour point sets is calculated, and the two points with the shortest distance are saved, i.e. the two concave points of the adherent contour, refer to (c) in Figure 2 .

[0034] S4: for the two concave points of the adherent contour in S3, by iterating each pixel point of the image, the pixel value of 10 pixels near the straight line determined by the concave points is modified to 0, at this time, the pixel feature in the straight line field between the two concave points is changed to black, so as to achieve the effect of adherent segmentation. After that, the adherent cell contour is effectively segmented, refer to (d) in Figure 2 ; the fluorescent cell positioning unit extracts the cell region edge in the cell region image to obtain the cell contour edge point set P i (X i ,Y i ), the center point P(X,Y) is obtained according to the edge point set, wherein:

[0035] The fluorescent cell feature extraction unit changes the four kinds of fluorescent feature images into four kinds of HSV images respectively, and sets the cell region of interest of the four kinds of HSV images according to the center point of the cell. The region of interest is a circular fluorescent cell region, and the specific range of the region is (X-a) 2 +(Y-b) 2 =R 2 , wherein a and b are the coordinate values of the X dimension and the Y dimension of the center point, and R is the radius of the fluorescent region; and the brightness features of the cell region of interest on the four kinds of fluorescent cell images are counted, and the corresponding brightness features BF d of the fluorescent region of interest in the DAPI feature image, the brightness features BF c of the fluorescent region of interest in the CD45 feature image, the brightness features BF v of the fluorescent region of interest in the Vimentin feature image, and the brightness features BF e of the fluorescent region of interest in the EpCAM feature image are obtained.

[0036] The cell classification unit obtains four kinds of brightness features BF d , BFc , BF v , BF e Classify the cell region image, and refer to the cell classification process Figure 3

[0037] Step a, if BF d > Δ1, classify the cell in the cell region image as a live cell, and execute step b; otherwise, classify the cell in the cell region image as a dead cell, and end the classification;

[0038] Step b, if BF c < Δ2, classify the cell in the cell region image as an abnormal cell, and execute step d; otherwise, execute step c; wherein Δ1 and Δ2 represent two luminance thresholds, in this embodiment, Δ1 = 50 DPI and Δ2 = 20 DPI;

[0039] Step c, determine whether BF v / BF c ≤ δ and BF e / BF c ≤ δ are true, if true, classify the cell in the cell region image as a normal cell, and end the classification; otherwise, execute step d; wherein δ represents a proportional luminance threshold, in this embodiment, δ = 1.7;

[0040] Step d, if BF v / BF c ≤ δ and BF e / BF c > δ, classify the cell in the cell region image as an epithelial abnormal cell;

[0041] If BF v / BF c > δ and BF e / BF c ≤ δ, classify the cell in the cell region image as a mesenchymal abnormal cell;

[0042] If BF v / BF c > δ and BF e / BF c > δ, classify the cell in the cell region image as an epithelial-mesenchymal mixed abnormal cell.

[0043] The image type labeling module labels the type of the cell region image classified by the image processing module, and obtains the cell region image with the label.

[0044] ​The cell image classification detection of the application is applied to medical detection, can pre-screen the cell detection, automatically classifies and labels the fluorescent cell type, and can batch process the original fluorescent image, achieves the same classification effect, greatly improves the detection efficiency. Compared with the manual detection method, not only the detection time is shortened, but also the accuracy of cell recognition of the system is improved.

Claims

1. A classification detection system based on cell fluorescence images, characterized by, The application relates to a type labeling method for a fluorescence image, and belongs to the field of image processing. The method comprises the following steps: An image acquisition module, an image processing module and a type labeling module are included; The image acquisition module generates four fluorescence characteristic images, including a DAPI characteristic image, a CD45 characteristic image, a Vimentin characteristic image and an EpCAM characteristic image, through an upright microscope, and the four fluorescence characteristic images are fused through the upright microscope to obtain a Merge characteristic image with four fluorescence characteristics; The image processing module comprises a fluorescence image preprocessing unit, a fluorescence cell cluster segmentation unit, a fluorescence cell positioning unit, a fluorescence cell feature extraction unit and a fluorescence cell classification unit; The fluorescence image preprocessing unit performs a preprocessing operation on the Merge characteristic image, including converting a three-channel Merge image into a single-channel gray image, then performing noise filtering on the gray image through a Gaussian filtering method to obtain a filtered single-channel gray image, and then performing thresholding processing on the filtered single-channel gray image through an OTSU adaptive threshold segmentation method to obtain a binary image of the fluorescence image, wherein white areas in the binary image are cell areas, and black areas are background areas; The fluorescence cell cluster segmentation unit performs segmentation processing on cell clusters in the binary image to obtain a plurality of cell area images; The fluorescence cell feature extraction unit respectively changes four fluorescence feature images into four HSV images, sets a cell region of interest of each of the four HSV images according to a center point of the cell, and statistically processes a brightness feature of the cell region of interest on each of the HSV images, so as to obtain a brightness feature BF of the fluorescence region of interest in the DAPI feature image d , a brightness feature BF of the fluorescence region of interest in the CD45 feature image c , a brightness feature BF of the fluorescence region of interest in the Vimentin feature image v , and a brightness feature BF of the fluorescence region of interest in the EpCAM feature image e ; The cell classification unit classifies the cell region image according to four brightness features BF d , BF c , BF v , BF e ​ Step a, if BF d > Δ1, then classify the cell in the cell region image as a live cell and perform step b; otherwise, classify the cell in the cell region image as a dead cell and end the classification. Step b, if BF c <Δ2, the cell in the cell region image is classified as an abnormal cell, and step d is performed; otherwise, step c is performed; wherein, Δ1, Δ2 represent two brightness thresholds, and Δ2<Δ1; Δ1=50 DPI, Δ2=20 DPI. Step c, judging and is established, the cells in the cell region image are classified as normal cells, and the classification ends; otherwise, step d is performed; wherein, represents the proportional brightness threshold value; Step d, if and then classifying the cell in the cell region image as an epithelial-type abnormal cell; If and then classifying the cell in the cell region image as a stromal abnormal cell; If and then classifying the cell in the cell region image as an epithelial-mesenchymal mixed abnormal cell; The fluorescence cell positioning unit extracts cell area edges in the cell area image to obtain a cell contour edge point set, and then calculates the center point of the cell through a statistical method; The type labeling module labels the type of the cell area image classified by the image processing module to obtain a cell area image with labels.

Images