Cell sample picture extraction analysis comparison detection method

Abstract

The application provides a cell sample picture extraction analysis comparison detection method, obtains multiple cell sample pictures, performs gray scale processing on the cell sample pictures, sets the pixels with the largest area as white or black, and synchronously increases the color brightness of other positions; performs entropy processing on other colors of the cell sample pictures, and then obtains a cell sample rough image by using a maximum class difference method binarization segmentation to determine a rough position of a target; the rough position of the target is subjected to a segmentation value clustering algorithm, the picture is subjected to a sharpening processing, and a required cell sample picture is obtained; basic cases are obtained by analyzing the cell sample picture, uncertain decision adjustment is performed on the obtained basic cases, CBR / RBR retrieval analysis is performed, and similarity evaluation analysis is performed on the basic cases and existing cases in a case library. The method has the advantages of high recognition rate and fast recognition speed, no subsequent processing is required, and processing time can be greatly saved.

Description

Technical Field

[0001] This invention relates to the field of cell sample image extraction and analysis, and in particular to a method for comparative detection of cell sample image extraction and analysis. Background Technology

[0002] With the rapid development of artificial intelligence, the intelligentization of various industries is becoming increasingly apparent. However, the preparation work before observing cell sample images is insufficient in terms of its level of intelligence. In places such as hospitals, biological or medical research institutes, where a large number of cell sample images need to be observed and analyzed, a single cervical cell sample contains many images, of which one-third of the scanned images are useless, meaning they contain little information or are blank images. Automatic scaling by the computer cannot accurately reveal the specific image of the cervical cell sample, and some cervical cell cases are difficult to identify. This makes it impossible to simply determine the case based on the cervical cell sample images later. When making a decision, it is necessary to compare with existing cases in the case database to determine the closest match before making a better decision. However, current retrieval methods cannot obtain the most accurate data, leading to difficulties in making case decisions. Summary of the Invention

[0003] The main objective of this invention is to provide a method for extracting, analyzing, comparing, and detecting cell sample images. This method addresses the difficulty in identifying cervical cell cases, which makes it impossible to easily determine the case based solely on cervical cell sample images. Decision-making requires comparison with existing cases in a database to determine the closest match before making a better decision. However, current retrieval methods cannot obtain the most accurate data, leading to difficulties in making case decisions.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for extracting, analyzing, comparing, and detecting cell sample images, the detection method comprising:

[0005] S1. Obtain multiple cell sample images, perform grayscale processing on the cell sample images, set the largest pixel in the area to white or black, and increase the color brightness of other areas simultaneously.

[0006] S2. Perform entropy processing on other colors in the cell sample image, and then use the maximum class difference method for binarization segmentation to obtain a rough image of the cell sample and determine the rough location of the target.

[0007] S3. The target's approximate location is segmented using a clustering algorithm, and the image is sharpened to obtain the required cell sample image.

[0008] S4. Analyze cell sample images to obtain basic cases, and then perform CBR / RBR retrieval analysis on the obtained basic cases under uncertain decision adjustment, and conduct similarity evaluation analysis between the basic cases and existing cases in the case database.

[0009] In the preferred embodiment, the specific implementation method of step S2 is as follows:

[0010] The cell sample image size is x×y. The entropy value of the cell sample image is obtained according to the entropy processing formula. The grayscale image is obtained according to the entropy processing formula. The entropy processing formula is:

[0011]

[0012] In the formula, f(i,j) represents the gray value of the point at coordinates (i,j);

[0013] Then, the Otsu's method is used to perform binary segmentation on the image;

[0014] The segmentation method is as follows: Let the cell sample image be D. The number of pixels with a grayscale value less than a threshold E is denoted as Q0, and the number of pixels with a grayscale value greater than a threshold T is denoted as Q1. The proportion of the target pixel count to the entire image is denoted as... but:

[0015] Let its average grayscale value be γ0, and the proportion of background pixels to the entire image be... for:

[0016] The overall average gray level of the image is γ, and the inter-class variance is denoted as O. s ,but:

[0017]

[0018] Let E traverse within the grayscale levels, such that O s When E is at its maximum, then the optimal segmentation threshold T is obtained. s ,Right now:

[0019]

[0020] This yields a binary image, indicating the approximate location of the target.

[0021] In the preferred embodiment, the segmentation value clustering algorithm method in step S3 is as follows: the position of the target is determined by detecting the segmented binary image based on the local entropy, and a binary image D2 is obtained. The segmented binary image D2 is processed by the segmentation value clustering algorithm to obtain image D3, which is divided into k clusters. The final image result is denoted as D4.

[0022] A1. Search D2 and store the coordinates of the pixels with a value of 1 into an array, denoted as Tarray(i);

[0023] A2. Traverse the array Tarray(i) and find the points in D3 whose coordinates are the data in the array. The cluster they belong to is k1.

[0024] A3. Set the gray value of the pixels clustered into k1 to 255, and set the gray value of the remaining pixels to 0 to obtain the desired cell sample image result D4.

[0025] In the preferred scheme, the method for similarity evaluation analysis of cases in step S4 is as follows: Basic pathology is derived from cell sample images, and then CBR / RBR retrieval analysis is performed on the basic pathology. CBR / RBR retrieval analysis is achieved by comparing the attribute differences between cases in the case library and the target case. The similarity function used is to analyze the similarity between the two cases. Euclidean distance between T attributes;

[0026] The CBR / RBR similarity evaluation function is defined as follows:

[0027]

[0028]

[0029]

[0030] In the formula, The attribute 'a' represents the case in the case database and the target case. j The eigenvalues ​​are all determined; Attribute a of the target case j The eigenvalues ​​are all uncertain; The attribute 'a' represents a case in the case database. j The eigenvalues ​​are all uncertain; and For conditional probabilities in Bayesian networks used in inference models;

[0031] In the formula, (U, T, A, V) is an information matrix of corresponding dimensions, and θ is a weakly balanced random variable with a mathematical expectation of 0, which represents the model error, parameter measurement error, and external disturbance uncertainty.

[0032] In the preferred scheme, U is the cell type instance set, A is the cell type case attribute set, V is the attribute feature set, and T is the target case for emergency decision-making, where T can play a decision-making role.

[0033] The case set U = {u i |i=1,2,3...n},u i For the i-th cell type pathology;

[0034] Where case set A = {a j |j=1,2,3...m},a j For the j-th cell type, the pathological attribute is...

[0035] Case Studies For u i The attribute feature vector,

[0036] V ij ,u i The j-th attribute feature value.

[0037] This invention provides a method for extracting, analyzing, comparing, and detecting cell sample images. The method first performs local grayscale processing on the cell sample image, then binarizes and segments the useful areas to obtain a rough image of the cell sample, thus determining the approximate location of the target. This method reduces areas with limited information or blank images, offering advantages such as high recognition rate and fast recognition speed. No further processing is required, significantly saving processing time. The obtained cell sample image is analyzed and written into a basic case study. Data from a case database is then retrieved, and the closest data case is compared with the basic case study to provide a decision-making reference, improving the decision-making effectiveness of the case. Attached Figure Description

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0039] Figure 1 The image contains too little information about this invention.

[0040] Figure 2 This invention relates to entropy processing and grayscale processing of photographs.

[0041] Figure 3 The present invention uses binarization segmentation to obtain a coarse image;

[0042] Figure 4 The cell sample images obtained after the segmentation value clustering algorithm of this invention are shown. Detailed Implementation

[0043] like Figures 1-4 As shown, a method for extracting, analyzing, comparing, and detecting cell sample images is described. The detection method includes:

[0044] S1. Obtain multiple cell sample images, perform grayscale processing on the cell sample images, set the largest pixel in the area to white or black, and increase the color brightness of other areas simultaneously.

[0045] S2. Perform entropy processing on other colors in the cell sample image, and then use the maximum class difference method for binarization segmentation to obtain a rough image of the cell sample and determine the rough location of the target.

[0046] S3. The target's approximate location is segmented using a clustering algorithm, and the image is sharpened to obtain the required cell sample image.

[0047] S4. Analyze cell sample images to obtain basic cases, and then perform CBR / RBR retrieval analysis on the obtained basic cases under uncertain decision adjustment, and conduct similarity evaluation analysis between the basic cases and existing cases in the case database.

[0048] The cell sample image is first processed locally to achieve grayscale. The resulting grayscale image is then binarized and segmented into useful locations to obtain a rough image of the cell sample, thus determining the approximate location of the target. This method reduces areas with limited information or blank images, offering advantages such as high recognition rate and fast recognition speed. No further processing is required, which can significantly save processing time. The obtained cell sample image is then analyzed to create a basic case study. Data from the case database is then retrieved, and the closest data case is compared with the basic case study to provide a decision-making reference and improve the decision-making effectiveness of the case.

[0049] In the preferred embodiment, the specific implementation method of step S2 is as follows:

[0050] The cell sample image size is x×y. The entropy value of the cell sample image is obtained according to the entropy processing formula. The grayscale image is obtained according to the entropy processing formula. The entropy processing formula is:

[0051]

[0052] In the formula, f(i,j) represents the gray value of the point at coordinates (i,j);

[0053] Next, the Otsu's method is used for binary segmentation of the image. For the grayscale image after the aforementioned local entropy processing, to facilitate target location determination, this study employs Otsu's method for binary segmentation. Otsu's method is an adaptive threshold determination method that divides the image into background and target parts based on its grayscale characteristics. A larger inter-class variance between the background and target indicates a greater difference between the two parts of the image. Misclassifying a target as background or vice versa will decrease the inter-class variance. Therefore, segmentation that maximizes the inter-class variance implies a minimum misclassification probability.

[0054] The segmentation method is as follows: Let the cell sample image be D. The number of pixels with a grayscale value less than a threshold E is denoted as Q0, and the number of pixels with a grayscale value greater than a threshold T is denoted as Q1. The proportion of the target pixel count to the entire image is denoted as... but:

[0055] Let its average grayscale value be γ0, and the proportion of background pixels to the entire image be... for:

[0056] The overall average gray level of the image is γ, and the inter-class variance is denoted as O. s ,but:

[0057]

[0058] Let E traverse within the grayscale levels, such that O s When E is at its maximum, then the optimal segmentation threshold T is obtained. s ,Right now:

[0059]

[0060] This yields a binary image, indicating the approximate location of the target.

[0061] In the preferred scheme, the K-means clustering algorithm is used for segmentation value clustering. K-means is a relatively basic clustering algorithm. This study first arbitrarily selects k objects from n data objects as initial cluster centers. For the remaining objects, they are assigned to the clusters most similar to these centers based on their similarity (distance). Then, the cluster centers of each newly obtained cluster are calculated, and this process is repeated until the standard measure function begins to converge. The clustering measure function is generally the sum of squared errors. The number of iterations is generally determined by the dataset size, the number of clusters, and the data distribution. Since thermal infrared images are generally single-band, the data dimension in the K-means algorithm is 1, resulting in a time complexity of O(nk). To save computation time and improve efficiency, this study will improve upon this approach in the experiments.

[0062] The segmentation value clustering algorithm method in step S3 is as follows: the position of the target is determined by detecting the segmented binary image based on the local entropy, and the binary image D2 is obtained. The segmented binary image D2 is processed by the segmentation value clustering algorithm to obtain image D3, which is divided into k clusters. The final image result is denoted as D4.

[0063] A1. Search D2 and store the coordinates of the pixels with a value of 1 into an array, denoted as Tarray(i);

[0064] A2. Traverse the array Tarray(i) and find the points in D3 whose coordinates are the data in the array. The cluster they belong to is k1.

[0065] A3. Set the gray value of the pixels clustered into k1 to 255, and set the gray value of the remaining pixels to 0 to obtain the desired cell sample image result D4.

[0066] In the preferred scheme, the method for similarity evaluation analysis of cases in step S4 is as follows: Basic pathology is derived from cell sample images, and then CBR / RBR retrieval analysis is performed on the basic pathology. CBR / RBR retrieval analysis is achieved by comparing the attribute differences between cases in the case library and the target case. The similarity function used is to analyze the similarity between the two cases. Euclidean distance between T attributes;

[0067] The CBR / RBR similarity evaluation function is defined as follows:

[0068]

[0069]

[0070]

[0071] In the formula, The attribute 'a' represents the case in the case database and the target case. j The eigenvalues ​​are all determined; Attribute a of the target case j The eigenvalues ​​are all uncertain; The attribute 'a' represents a case in the case database. j The eigenvalues ​​are all uncertain; and For conditional probabilities in Bayesian networks used in inference models;

[0072] In the formula, (U, T, A, V) is an information matrix of corresponding dimensions, and θ is a weakly balanced random variable with a mathematical expectation of 0, which represents the model error, parameter measurement error, and external disturbance uncertainty.

[0073] In the preferred scheme, U is the cell type instance set, A is the cell type case attribute set, V is the attribute feature set, and T is the target case for emergency decision-making, where T can play a decision-making role.

[0074] The case set U = {u i |i=1,2,3...n},u i For the i-th cell type pathology;

[0075] Where case set A = {a j |j=1,2,3...m},a j For the j-th cell type, the pathological attribute is...

[0076] Case Studies For u i The attribute feature vector,

[0077] V ij ,ui The j-th attribute feature value.

[0078] For knowledge representation of random terms, Bayesian networks demonstrate powerful processing capabilities. They can learn and reason under incomplete and uncertain information conditions, using conditional probabilities to represent the relationships between information elements for data prediction. Therefore, the Bayesian network method is chosen to represent uncertainty in knowledge reasoning. Constructing a robust Bayesian Network CBR / RBR (BN-CRBRBR) model combining steady-state and dynamic approaches effectively handles uncertain information in emergency decision-making, reducing the dependence of reasoning decisions on uncertain information. Based on the results of intelligent reasoning, increased uncertainty, changes in case attribute sets, and untimely updates to the intelligent reasoning information system limit the problem-solving ability of intelligent reasoning to a certain extent. Medical experts demonstrate superior capabilities in handling implicit and unstructured problems in reasoning, effectively improving decision-making efficiency. Therefore, this paper proposes using the subjective thinking of experts to supplement and correct intelligent reasoning. The results of intelligent decision-making reasoning, the medical experts' experience and knowledge in their field, and the actual situation of the patients are integrated through visual thinking analysis and decision-making, i.e., human-machine fusion decision-making, to optimize the decision results. The emergency decision-making process for cell sample cases, as a complex decision-making system, possesses inherent uncertainty, primarily due to model errors, parameter errors, and external uncertainties. To achieve reasonable and accurate diagnostic results, scientifically addressing uncertainty is a crucial primary issue. In uncertain situations, a widely adopted and effective method for handling the impact of external uncertainties and errors on system modeling is to treat these uncertainties as random variables. This method has undergone extensive and long-term testing in practical engineering. In emergency decision-making for cell sample cases, a deterministic robust inference model is established in the absence of uncertainties. However, actual emergency decision-making for cell sample cases may involve uncertainties, necessitating the use of robust stochastic hybrid methods to describe external disturbances or model errors through random terms, thus establishing a robust inference model. Therefore, a stable CBR / RBR fusion inference model is constructed for the emergency decision-making system for cell sample cases. Model errors, parameter errors, and external uncertainties in the inference decision-making process are described using random variables. Based on the steady-state study of CBR / RBR fusion inference, an uncertainty model for knowledge reasoning is established by adding random terms, and the impact of these uncertainties is analyzed.

[0079] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for extracting, analyzing, comparing, and detecting cell sample images, characterized by: The detection methods include: S1. Obtain multiple cell sample images, perform grayscale processing on the cell sample images, set the largest pixel in the area to white or black, and increase the color brightness of other areas simultaneously. S2. Perform entropy processing on other colors of the cell sample image, and then use the maximum class difference method for binarization segmentation to obtain a rough image of the cell sample and determine the rough location of the target. S3. The target's approximate location is segmented using a clustering algorithm, and the image is sharpened to obtain the required cell sample image. S4. Analyze cell sample images to obtain basic cases, and then perform CBR / RBR retrieval analysis on the obtained basic cases under uncertain decision adjustment, and conduct similarity evaluation analysis between the basic cases and existing cases in the case database. The similarity evaluation analysis method for cases in step S4 is as follows: Basic pathology is derived from cell sample images, and then CBR / RBR search analysis is performed on the basic pathology. CBR / RBR search analysis is achieved by comparing the attribute differences between cases in the case library and the target case. The similarity function used is to analyze the similarity between the two cases. The Euclidean distance between T attributes; The CBR / RBR similarity evaluation function is defined as follows: In the formula, , This represents the attributes of cases in the case database and the target case. The eigenvalues ​​are all determined; Indicates the attributes of the target case The eigenvalues ​​are all uncertain; Indicates the attributes of cases in the case database The eigenvalues ​​are all uncertain; and For conditional probabilities in Bayesian networks used in inference models; In the formula, For information matrices of corresponding dimensions, Let be a weakly balanced random variable with a mathematical expectation of 0, representing the model error, parameter measurement error, and uncertainty of external disturbances.

2. The method for extracting, analyzing, comparing, and detecting cell sample images according to claim 1, characterized in that: The specific implementation method of step S2 is as follows: Cell sample image size is The entropy value of the cell sample image is obtained according to the entropy processing formula, and the grayscale image is obtained according to the entropy processing formula. The entropy processing formula is: In the formula, f(i,j) represents the gray value of the point at coordinates (i,j); Then, the Otsu's method is used to perform binary segmentation on the image; The segmentation method is as follows: Let the cell sample image be D. The number of pixels in the image whose grayscale value is less than a threshold E is denoted as... The number of pixels with a gray level greater than the threshold T is denoted as The proportion of the target pixels to the entire image is denoted as... ,but: ; Let its average grayscale value be . The proportion of background pixels to the entire image for: The total average gray level of the image is The variance between classes is denoted as ,but: Let E traverse within the grayscale levels, such that When E is at its maximum, then E achieves the optimal segmentation threshold. ,Right now: This yields a binary image, indicating the approximate location of the target.

3. The method for extracting, analyzing, comparing, and detecting cell sample images according to claim 1, characterized in that: The segmentation value clustering algorithm method in step S3 is as follows: Based on the location of the target determined by detecting the segmented binary image using local entropy, a binary image is obtained. Segmented binary image Image obtained through segmentation clustering algorithm. The image is divided into k clusters, and the final image result is denoted as . ; A1, Search Store the coordinates of the pixels with a value of 1 into an array, denoted as Tarray(i); A2. Traverse the array Tarray(i), from... Find the point whose coordinates are the data in the array, and determine its cluster. ; A3. Clustering into The grayscale value of the pixel is set to 255, and the grayscale value of the other pixels is set to 0 to obtain the desired cell sample image. .

4. The method for extracting, analyzing, comparing, and detecting cell sample images according to claim 1, characterized in that: U is the set of cell type instances, A is the set of cell type case attributes, V is the set of attribute features, and T is the target case for emergency decision-making, where T can play a decision-making role. Case Studies , The pathology of the i-th cell type; Case Studies , Pathological attributes of the j-th cell type; Case Studies for The attribute feature vector, , , The j-th attribute feature value.

Images