A cell morphology classification method based on morphology and Shannon entropy

Abstract

The invention relates to the field of image processing technology, and in particular to a cell morphology classification method based on morphology and Shannon entropy, comprising acquiring a cell image and preprocessing the cell image, determining the center of gravity and edge of the cell morphology after the preprocessing, setting a cell edge arc length microelement, and extracting a discrete polar coordinate set describing the cell edge; calculating an angle increment and its probability distribution, and calculating an index characterizing a circumferential angle feature of the cell morphology; calculating a radius increment and its probability distribution, and calculating an index characterizing a radial radius feature of the cell morphology; drawing a scatter plot of each morphology with an index of the circumferential angle feature of the cell morphology as a horizontal coordinate and an index of the radial radius feature of the cell morphology as a vertical coordinate, obtaining a corresponding data center under each morphology, and calculating the distance of the cell image data center to confirm its morphology label; the method of the invention has the characteristics of high efficiency and rapidity, and is of great significance in assisting red blood cell classification and thalassemia diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and also to the technical fields of biology, physics, statistics, information theory, etc., and in particular to a cell morphology classification method based on morphology and Shannon entropy. Background Art

[0002] Thalassemia, also known as thalassemia, is a hemoglobin disease caused by gene deletion or mutation. Clinically, it often manifests as hereditary hemolytic anemia, and is often accompanied by characteristics such as uneven red blood cell size and abnormal morphology, such as spherical, target-shaped, teardrop-shaped, etc. The more severe the anemia symptoms, the more obvious the heteromorphism. Therefore, it is extremely important to develop a fast and effective method to analyze the morphological characteristics of red blood cells, and then evaluate the functional status of red blood cells and assist in the diagnosis and treatment of thalassemia.

[0003] Inspired by the ideas of cell morphological dynamics and Shannon entropy, the present invention studies the physical morphological changes of red blood cells in thalassemia cases and proposes a thalassemia red blood cell classification method combining morphology and Shannon entropy, which achieves the purpose of rapid classification of red blood cells from two-dimensional morphology and can provide important guiding significance for the clinical diagnosis of thalassemia. Summary of the invention

[0004] In order to quickly and efficiently analyze the morphological change characteristics of cells, the present invention provides a cell morphology classification method based on morphology and Shannon entropy, which specifically includes the following steps:

[0005] Acquire cell images and preprocess the cell images, determine the center of gravity and edge of the cell morphology after preprocessing, set the arc length microelement of the cell edge, and extract a discrete polar coordinate set describing the cell edge;

[0006] According to the obtained angular coordinates of the cell edge, the angle increment and its probability distribution are calculated, and the index characterizing the circumferential angle characteristics of the cell morphology is calculated;

[0007] According to the obtained radius coordinates of the cell edge, the radius increment and its probability distribution are calculated, and the index characterizing the radial radius characteristics of the cell morphology is calculated;

[0008] Based on the cell images containing morphological classification labels, a scatter plot is drawn with the index of the cell morphological circumferential angle feature as the horizontal axis and the index of the cell morphological radial radius feature as the vertical axis to obtain the corresponding data center under each morphology;

[0009] The distance between the cell image without morphological labels and the data center corresponding to each morphology is calculated, and the morphology corresponding to the closest distance is used as the morphological label corresponding to the cell image.

[0010] Furthermore, the process of setting the cell edge arc length microelement and extracting a discrete polar coordinate set describing the cell edge includes:

[0011] Set the value of the cell edge arc length element dN;

[0012] The center of gravity of the cell image is taken as the origin of the polar coordinate system, and the edge point with angle θ = 0 in the polar coordinate system is taken as the starting point, and its coordinates (θ1, r1) are recorded;

[0013] Starting from the starting point, rotate counterclockwise by a cell edge arc length microelement dN to obtain the coordinates of a new edge point (θ2, r2);

[0014] And so on, until you get coordinates and treats them as a set of discrete polar coordinates describing the cell edge.

[0015] Furthermore, the index characterizing the circumferential angle characteristics of cell morphology is expressed as:

[0016]

[0017] Among them, CME[p(Δθ i )] is an index to characterize the angular characteristics of cell morphology; p(Δθ i ) represents the probability distribution of the i-th angle increment; p j (Δθ i ) represents the probability distribution of the i-th angle increment corresponding to the j-th event; n represents the number of events in the probability distribution; log2() represents the calculation of the logarithm with base 2.

[0018] Furthermore, the i-th angle increment is the difference between the angle value of the edge point in the i-th polar coordinate system and the angle value of the edge point in the i-1-th polar coordinate system.

[0019] Furthermore, the index characterizing the radial radius characteristics of cell morphology is expressed as:

[0020]

[0021] Among them, CME[p(Δr)] is an index to characterize the radial radius characteristics of cell morphology; p(Δr i ) represents the probability distribution of the i-th radius increment; p j (Δr i ) represents the probability distribution of the i-th radius increment corresponding to the j-th event.

[0022] Furthermore, the i-th radius increment is the difference between the radius value of the edge point in the i-th polar coordinate system and the radius value of the edge point in the i-1-th polar coordinate system.

[0023] Furthermore, if the cell image is a thalassemia red blood cell, the thalassemia red blood cells are divided into three categories according to round, spiny and teardrop morphologies, and the data center coordinates of the three categories are calculated with the index of the circumferential angle characteristics of the cell morphology as the horizontal coordinate and the index of the radial radius characteristics of the cell morphology as the vertical coordinate. The data center corresponding to the circular morphology is at (0.23, 0.54), the data center corresponding to the spiny morphology is at (0.36, 0.76), and the data center corresponding to the teardrop morphology is at (0.68, 0.85).

[0024] Compared with the prior art, the cell morphology classification method based on morphology and Shannon entropy of the present invention has the following beneficial effects:

[0025] (1) This invention develops a morphological entropy method based on the Shannon entropy concept. It studies the circumferential angle characteristics and radial radius characteristics of cell morphology from the perspective of entropy for the first time, and can comprehensively characterize the changing characteristics of the physical morphology of red blood cells in thalassemia.

[0026] (2) The invention can classify red blood cells in thalassemia based on cell morphological entropy, and the operation is simple, which greatly reduces the complexity of calculation and can provide important value for the clinical diagnosis and treatment of thalassemia quickly and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a process for classifying red blood cells in an embodiment of a cell morphology classification method based on morphology and Shannon entropy of the present invention;

[0028] Figure 2 A schematic diagram and a related probability diagram of extracting two-dimensional polar coordinate values ​​of red blood cell edges in a preferred embodiment of the present invention;

[0029] Figure 3 A statistical histogram of morphological entropy components and average values ​​in a preferred embodiment of the present invention;

[0030] Figure 4 is a scatter plot of the morphological entropy component in a preferred embodiment of the present invention;

[0031] Figure 5 It is a morphological entropy scatter density map and classification results of the example in a preferred embodiment of the present invention;

[0032] Figure 6 The present invention is a flow chart of a cell morphology classification method based on morphology and Shannon entropy. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] The present invention provides a cell morphology classification method based on morphology and Shannon entropy, such as Figure 6 , specifically including the following steps:

[0035] Acquire cell images and preprocess the cell images, determine the center of gravity and edge of the cell morphology after preprocessing, set the arc length microelement of the cell edge, and extract a discrete polar coordinate set describing the cell edge;

[0036] According to the obtained angular coordinates of the cell edge, the angle increment and its probability distribution are calculated, and the index characterizing the circumferential angle characteristics of the cell morphology is calculated;

[0037] According to the obtained radius coordinates of the cell edge, the radius increment and its probability distribution are calculated, and the index characterizing the radial radius characteristics of the cell morphology is calculated;

[0038] Based on the cell images containing morphological classification labels, a scatter plot is drawn with the index of the cell morphological circumferential angle feature as the horizontal axis and the index of the cell morphological radial radius feature as the vertical axis to obtain the corresponding data center under each morphology;

[0039] The distance between the cell image without morphological labels and the data center corresponding to each morphology is calculated, and the morphology corresponding to the closest distance is used as the morphological label corresponding to the cell image.

[0040] This embodiment uses a thalassemia red blood cell dataset published in a journal article as an example of an original two-dimensional red blood cell image. In this embodiment, image preprocessing includes the following steps:

[0041] (1.1) A total of 3856 red blood cell images were obtained, which can be divided into three categories. The first category has teardrop-shaped features (teardrop), totaling 2076 images; the second category has features that are approximately normal round (normal), totaling 1426 images; the third category has spiny features (acantocyte), totaling 354 images.

[0042] (1.2) The acquired red blood cell image is converted into a grayscale image, and a Gaussian filter is applied to it to obtain a grayscale image with reduced noise. On this basis, the threshold of the grayscale image is determined and converted into a binary image.

[0043] (1.3) Fill the “holes” in the binary image to obtain multiple connected domains of different sizes, and determine that the connected domain with the largest area is the cell morphology.

[0044] The filtering, grayscale conversion, filling and other processing of the above red blood cell images can be adjusted by those skilled in the art according to actual needs. The present invention does not limit the image processing method, that is, those skilled in the art can use any existing technology to obtain the center of gravity and edge of the cell.

[0045] After obtaining the center of gravity and the edge of the cell, the center of gravity of the cell is used as the origin of polar coordinates to extract a discrete polar coordinate set describing the edge of the cell, which specifically includes the following steps:

[0046] Setting the value of the cell edge arc length microelement dN, which determines the accuracy of the present invention. In the present invention, the value is set as small as possible. Preferably, the value of the cell edge arc length microelement dN is less than or equal to 2;

[0047] The center of gravity of the cell image is taken as the origin of the polar coordinate system, and the edge point with angle θ = 0 in the polar coordinate system is taken as the starting point, and its coordinates (θ1, r1) are recorded;

[0048] Starting from the starting point, rotate counterclockwise by a cell edge arc length microelement dN to obtain the coordinates of a new edge point (θ2, r2);

[0049] And so on, rotate counterclockwise until you get coordinates and treats them as a set of discrete polar coordinates describing the cell edge.

[0050] Based on the red blood cell image after the above preprocessing, the polar coordinates and their increments describing the cell boundary are extracted, and the corresponding probability distribution (such as Figure 2 As shown). Based on this, the Shannon entropy formula is introduced and the morphological entropy is calculated to quantitatively characterize the characteristics of cell morphology. The specific steps include:

[0051] (2.1) Calculate the coordinates of the pixel points on the cell morphology boundary (x i ,y i ), and the coordinates of the center of gravity (x0, y0), and then transform them into the polar coordinate system to obtain the corresponding pixel coordinates (θ i , r i ) and the centroid coordinates (θ0, r0). Then calculate the increments between adjacent coordinates, including the angle increment Δθ i and radius increment Δr i , the calculation formula is as follows:

[0052] Δθ i =θ i+1 -θ i ,

[0053] Δr i =r i+1 -r i .

[0054] Among them, θ is the angle value of the edge point in the polar coordinate system, Δθ is the increment of the angle, r is the radius value of the edge point in the polar coordinate system, Δr is the increment of the radius, and the subscript i represents the serial number of the discrete polar coordinates.

[0055] (2.2) Plot the angle increment Δθ i and radius increment Δr i The statistical histogram is obtained, and the frequency is converted into probability by dividing it by the total number of coordinates, and the probability density distributions p(Δθ) and p(Δr) of the angle increment and radius increment are obtained.

[0056] (2.3) Substituting the above probability density distribution p(Δθ) into the Shannon entropy formula, we can obtain the index CME[p(Δθ)] that characterizes the circumferential angle characteristics of cell morphology, as shown below:

[0057]

[0058] Among them, j is the ordinal number of the corresponding event in the probability distribution, n is the total number of events in the probability distribution, and log2 represents the logarithm with base 2.

[0059] (2.4) The index CME[p(Δr)] that characterizes the radial radius characteristics of cell morphology is further calculated as follows:

[0060]

[0061] (2.5) After calculating the index CMEa and the index CMEr, the distance between the cell to be detected and the center coordinates of each morphological type is calculated, and the morphological label corresponding to the center with the closest distance is used as the classification result of the cell to be detected.

[0062] (2.6) In this embodiment, another classifier is proposed. In this classifier, based on the above two indicators, the average value of cell morphological entropy is calculated by the following formula:

[0063] CME=(CMEa+CMEr) / 2.

[0064] Combining the cell morphological characteristics and the physical meaning of Shannon entropy, it can be seen that the morphological entropy is between [0, 1]. When the cell morphology is absolutely circular, it takes 0, and when it is far from a circle, it tends to 1. Therefore, the morphological entropy can be used to characterize the morphological dynamics of the cell. Therefore, cells are classified according to the average value CME of the cell morphological entropy. That is, if the average value CME of the morphological entropy of a cell is closer to 0, the cell morphology is closer to a circle, and an indicator is obtained to characterize whether the cell morphology is close to a circle.

[0065] Based on the morphological entropy calculated above, this example further draws the statistical distribution diagram of three types of morphological entropy, such as Figure 3 The results show that these three types of entropy also show bimodal characteristics. For CMEa, the peak positions are approximately 0.25 and 0.65; for CMEr, the peak positions are approximately 0.55 and 0.85; for CME, the peak positions are approximately 0.40 and 0.75.

[0066] Different from the above explanation of the statistical characteristics hidden behind the morphological entropy ( Figure 3 ), this embodiment also focuses on revealing the quantitative relationship between the morphological entropy components (CMEa and CMEr), such as Figure 4 As shown. Figure 4 In the figure, the horizontal axis represents CMEa and the vertical axis represents CMEr. The scatter plot mainly shows three categories, with their centers at approximately (0.23, 0.54), (0.36, 0.76) and (0.68, 0.85). Figure 2 The two categories presented are different. Figure 4 The scatter plot further subdivides the above two categories into three morphological categories. The specific differences can be compared and judged based on the scatter plot and the statistical histogram on the side.

[0067] Finally, this example also calculates Figure 4 The density distribution of the scatter plot, such as Figure 5 As shown in Figure 2, the results clearly show three dense concentration areas, whose characteristics are similar to human "fingerprints". Figure 4 , Figure 5 The figure shows more clearly that the analyzed red blood cells have three categories, namely, approximately circular, far from circular, and the shapes between the two. By combining the morphology of thalassemia red blood cells disclosed in the journal article, this embodiment can clearly indicate that the center at (0.23, 0.54) corresponds to the approximately circular feature (normal); the center at (0.36, 0.76) corresponds to the thorny feature (acantocyte); and the center at (0.68, 0.85) corresponds to the teardrop feature (teardrop).

[0068] As a specific implementation, the corresponding morphological entropy can be calculated, and the red blood cells in thalassemia can be classified according to the statistical graph, scatter plot, and density graph of the morphological entropy components and the average value. In the specific implementation, the present invention does not limit the edge arc length microelement dN, the coordinates of each center point, etc., and those skilled in the art can adjust them according to actual conditions to obtain the best results.

[0069] The experimental results show that the results obtained by the present invention are consistent with the existing results for the classification of red blood cells in thalassemia, and the accuracy rate reaches 100%. The present invention can be used to classify other cell structures that can obtain significant edges by classifying cell morphology. The cell classification method proposed by the present invention can be used to further study red blood cells with different characteristics, or other cell types, such as cancer cells, to improve the accuracy and robustness of classification.

[0070] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A cell morphology classification method based on morphology and Shannon entropy, characterized in that: The specific steps include: Acquire cell images and preprocess the cell images, determine the center of gravity and edge of the cell morphology after preprocessing, set the arc length microelement of the cell edge, and extract a discrete polar coordinate set describing the cell edge; According to the obtained angular coordinates of the cell edge, the angle increment and its probability distribution are calculated, and the index characterizing the circumferential angle characteristics of the cell morphology is calculated, namely: Among them, CME[p(Δθ i )] is an index to characterize the angular characteristics of cell morphology; p(Δθ i ) represents the probability distribution of the i-th angle increment; p j (Δθ i ) represents the probability distribution of the i-th angle increment corresponding to the j-th event; n represents the number of events in the probability distribution; log2() represents the calculation of the logarithm with base 2; According to the radius coordinates of the cell edge obtained, the radius increment and its probability distribution are calculated, and the index characterizing the radial radius characteristics of the cell morphology is calculated, namely: Among them, CME[p(Δr i )] is an index to characterize the radial radius characteristics of cell morphology; p(Δr i ) represents the probability distribution of the i-th radius increment; p j (Δr i ) represents the probability distribution of the i-th radius increment corresponding to the j-th event; Based on the cell images containing morphological classification labels, a scatter plot is drawn with the index of the cell morphological circumferential angle feature as the horizontal axis and the index of the cell morphological radial radius feature as the vertical axis to obtain the corresponding data center under each morphology; The distance between the cell image without morphological labels and the data center corresponding to each morphology is calculated, and the morphology corresponding to the closest distance is used as the morphological label corresponding to the cell image.

2. A cell morphology classification method based on morphology and Shannon entropy according to claim 1, characterized in that: The process of setting the cell edge arc length microelement and extracting the discrete polar coordinate set describing the cell edge includes: Set the value of the cell edge arc length element dN; The center of gravity of the cell image is taken as the origin of the polar coordinate system, and the edge point with angle θ = 0 in the polar coordinate system is taken as the starting point, and its coordinates (θ1, r1) are recorded; Starting from the starting point, rotate counterclockwise by a cell edge arc length microelement dN to obtain the coordinates of a new edge point (θ2, r2); And so on, until you get coordinates and treats them as a set of discrete polar coordinates describing the cell edge.

3. A cell morphology classification method based on morphology and Shannon entropy according to claim 2, characterized in that: The i-th angle increment is the difference between the angle value of the edge point in the i-th polar coordinate system and the angle value of the edge point in the i-1-th polar coordinate system.

4. A cell morphology classification method based on morphology and Shannon entropy according to claim 1, characterized in that: If the cell image is a thalassemia red blood cell, the thalassemia red blood cell is divided into three categories according to round, spinous and teardrop morphologies, and the data center coordinates of the three categories are calculated with the index of the circumferential angle feature of the cell morphology as the horizontal coordinate and the index of the radial radius feature of the cell morphology as the vertical coordinate. The data center corresponding to the circular morphology is at (0.23, 0.54), the data center corresponding to the spinous morphology is at (0.36, 0.76), and the data center corresponding to the teardrop morphology is at (0.68, 0.85).

5. The cell morphology classification method based on morphology and Shannon entropy according to claim 1, characterized in that: Preprocessing of cell images includes: The cell image is converted into a grayscale image and filtered by a Gaussian filter to obtain a grayscale image with reduced noise; Set a threshold of the grayscale image and convert the grayscale image into a binary image; Fill the gaps in the binary image to obtain multiple connected domains of different sizes, and determine the connected domain with the largest area as the cell morphology.

Images