Cervical cancer cytology screening system and method based on image analysis

Abstract

The present invention discloses a cervical cancer cytological screening system and method based on image analysis, which relates to the technical field of image processing. The screening system includes an image acquisition module, an image processing module, and also includes a cell marking module, a feature extraction module, a comparative analysis module, a central control module, an identification and inspection module, and a result output module. Compared with the prior art, the screening method proposed by the technical solution of the present invention adds a self-inspection process for the identification results of diseased cells. Using the marked area in the cytopathological smear scan image of the detected object individual as a reference, the cell types in other areas are compared and identified, which can improve the accuracy of identifying diseased cells when there is an overlap phenomenon of multiple types of cells and other substances.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly relates to a cervical cancer cytology screening system and method based on image analysis. Background Art

[0002] TCT cervical cancer screening refers to collecting exfoliated cells at the cervical orifice through a liquid-based thin-layer cell kit, using a fully automatic thin-layer cell preparation machine to prepare slides, and performing cytological classification diagnosis based on the nuclear morphology to determine whether there are canceration symptoms. The initial cervical cancer cytology screening was manually observed by detecting physicians, which had a large workload and was prone to missed detection and false negative results. With the continuous development of image processing technology, artificial intelligence technology has been applied to cervical cancer cytology screening. By scanning cervical epithelial cells with a photoelectric sensor, optical characteristic data and current attenuation curves are obtained, and then through an artificial intelligence expert evaluation system, the acquired data is automatically identified and verified, and automatically analyzed and compared with the database of millions of samples that have been diagnosed in the database through algorithms to achieve rapid screening.

[0003] However, in the screening method steps of the existing screening system, there is a lack of a self-check process for identifying and recognizing diseased cells. Due to the complexity of the scanned images of cytopathological smears obtained and the differences between individual detection objects, when there are overlaps of multiple types of cells and other substances, diseased cells cannot be recognized or there are deviations between the cell recognition results and the actual situation. For this reason, we propose a cervical cancer cytology screening system and method based on image analysis. Summary of the Invention

[0004] The main object of the present invention is to provide a cervical cancer cytology screening system and method based on image analysis, which can effectively solve the problems in the background art.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A cervical cancer cytology screening system based on image analysis includes an image acquisition module, an image processing module, and also includes a cell marking module, a feature extraction module, a comparative analysis module, a central control module, an identification and inspection module, and a result output module, where;

[0007] The image acquisition module is used to scan a cervical liquid-based smear to obtain a smear scan image of the cervical liquid-based smear;

[0008] The image processing module is connected to the image acquisition module and is used to preprocess and segment the obtained smear scan image to obtain all regions of interest of the segmented smear scan image.

[0009] The cell marking module is connected to the image segmentation module, and is used to classify some of the obtained regions of interest as different marked samples according to the types of substances contained in the regions of interest in the segmented image;

[0010] The feature extraction module is connected to the cell marking module, and is used to extract the feature parameters of the marked samples;

[0011] The comparison and analysis module is connected to the feature extraction module, and is used to perform one-by-one comparison and recognition on the regions of interest not marked by the cell marking module according to the feature parameters of the marked samples, and classify the unmarked regions of interest according to the recognition results;

[0012] The recognition and verification module is connected to the comparison and analysis module, and is used to verify the classification results of the unmarked regions of interest and calculate the classification accuracy rate of the unmarked regions of interest;

[0013] The central control module is connected to the recognition and verification module, and is used to distinguish the levels of the classification accuracy rate of the unmarked regions of interest and determine whether to output the classification results by setting the output threshold of the classification accuracy rate;

[0014] The result output module is connected to the central control module, and is used to count the number of regions of interest that need to output the classification results and output the quantity statistical results and the corresponding images of the regions of interest when the central control module determines to output the classification results.

[0015] The screening system further includes a storage module, a processor, and a computer program stored on the storage module and operable on the processor. The storage module is connected to the image acquisition module and the result output module, and is used to store the smear scan image information of the cervical liquid-based smear and the quantity and image information of the regions of interest that need to output the classification results.

[0016] Further, the screening method of the screening system includes the following steps:

[0017] Step 1, prepare a cervical liquid-based smear by using the thin-layer liquid-based technology, and scan the cervical liquid-based smear through the image acquisition module to obtain at least one group of smear scan images of the cervical liquid-based smear;

[0018] Step 2, perform enhancement processing on the obtained smear scan images through the image processing module, and perform segmentation on the obtained smear scan images by using the threshold segmentation method. The algorithm model is:

[0019]

[0020] Among them, i and j respectively represent the horizontal and vertical coordinates of the pixel; g(i, j) represents the local characteristics of the pixel; f(i, j) represents the pixel gray value; T is the segmentation threshold. Among them, the value of the segmentation threshold T is determined by the empirical method according to the quality of the smear scan image of the cervical liquid-based smear obtained;

[0021] After segmenting the image, all regions of interest of the segmented smear scan image are extracted according to the types of substances contained, including seven types: normal cell region, diseased cell region, impurity region, overlapping region of normal cells and diseased cells, overlapping region of normal cells and impurities, overlapping region of diseased cells and impurities, and overlapping region of normal cells, diseased cells and impurities;

[0022] Step three, sequentially number the regions of interest as A1, A2,..., A7 according to the type. Respectively select several regions of each type of region of interest in the smear scan image as samples, mark the selected samples to obtain marked samples, and number the marked samples as A ij , where i is the region of interest type number, i = 1, 2,..., 7; j is the marked sample number, j = 1, 2,..., n, and n is a positive integer;

[0023] Step four, according to the feature set for automatically identifying the pathological changes of cervical cancer cells, use an image recognition algorithm to extract the feature parameters of the density feature, size feature, edge feature, shape feature, and texture feature of the marked samples. Among them, the density feature includes average gray value, standard deviation, entropy, target-background contrast; the size feature includes area, perimeter, major axis, and minor axis; the edge feature includes edge strength, edge variance, and edge blur coefficient; the shape feature includes convexity, circularity, shape factor, eccentricity, Fourier descriptor, average normalized radius, normalized radius variance, normalized radius entropy, area ratio, and roughness; the texture feature includes fractal dimension, gray level co-occurrence matrix, local binary pattern mean, and LBP variance. The feature parameters of the marked samples extracted include at least one feature in the density feature, at least one feature in the size feature, at least one feature in the edge feature, at least one feature in the shape feature, and at least one feature in the texture feature;

[0024] Step five, respectively obtain the feature parameters of the unmarked regions of interest, and construct an evaluation data group of the unmarked regions of interest, denoted as Q t =(q 1t ,q 2t ,...,q lt), where t is the number of unlabeled regions of interest, t = 1, 2,..., n, n is a positive integer, l is the number of items of the characteristic parameters of the unlabeled regions of interest, l = 1, 2,..., u, u is a positive integer, and the mean value of the characteristic parameters of the labeled samples is used as the comparison data group, denoted as P = (p1, p2,..., p l ), calculate the Euclidean distance between the evaluation data group and the comparison data as the similarity Si of the data group, and the calculation formula is:

[0025]

[0026] Classify all unlabeled regions of interest according to the similarity;

[0027] Step Six, use the random sampling method to extract W test samples after classifying the unlabeled regions of interest. The number of test samples W is determined by the statistical formula: Determine, where N is the number of the number of test samples W; Z w is the sampling survey confidence level, usually taken as 95%; P w is the dispersion degree of the sampling sample; E w is the sampling error range, usually taken as ±3%, test and verify the extracted test samples, and calculate the classification accuracy rate Cr of each type of unlabeled region of interest according to the verification result v , and the calculation formula is: where v represents the type of unlabeled region of interest, v = 1, 2,..., 7; W v represents the total number of test samples of the v-th unlabeled region of interest; w v represents the number of correct items of the test samples of the v-th unlabeled region of interest;

[0028] Step Seven, divide the level of the classification accuracy rate Cr v of the unlabeled regions of interest. The method for dividing the level of the classification accuracy rate Cr v is as follows:

[0029] After obtaining the classification accuracy rate Cr v , create a sample set using the value of the classification accuracy rate Cr v , and obtain the mean value and standard deviation in the sample set. Standardize the data using the mean value and standard deviation. The standardization formula is In this formula, z is the standard parameter, σ is the variance of the sample data, μ is the mean value of the sample data. After completing the standardization, adjust the numerical interval of the standard parameter to between [0, 1] using , and classify the coincidence rate using the function value of f(k). The classification mechanism is:

[0030] When , the classification accuracy rate Crv Classified as first level;

[0031] When the classification accuracy rate Cr v is classified as second level;

[0032] where f(k)min and f(k)max are respectively the minimum and maximum values of the function values of f(k), and it is determined whether to output the classification result according to the output threshold and the discrimination result. The determination principle is:

[0033] When the classification accuracy rate Cr v is classified as first level, it is determined not to output the classification result;

[0034] When the classification accuracy rate Cr v is classified as second level, it is determined to output the classification result;

[0035] Step 8, when the central control module determines to output the classification result, count the number of regions of interest for which the classification result needs to be output, and output the quantity statistical result and the corresponding region of interest image. When the central control module determines not to output the classification result, return to Step 6 to re-extract the test samples of the unlabeled regions of interest that are not output, and perform the classification accuracy verification again. If the verification result still determines not to output the classification result, return to Step 5.

[0036] Furthermore, the labeled samples are classified into normal cell labeled samples, diseased cell labeled samples, impurity labeled samples, labeled samples where normal cells and diseased cells overlap, labeled samples where normal cells and impurities overlap, labeled samples where diseased cells and impurities overlap, and labeled samples where normal cells, diseased cells, and impurities overlap according to the types of substances contained in the regions of interest of the segmented image.

[0037] The present invention has the following beneficial effects:

[0038] (1) The technical solution of the present invention proposes a screening system and a screening method. By preprocessing and segmenting the scanned image of the cervical liquid-based smear, all regions of interest in the segmented smear scanned image are obtained. According to the types of substances contained in the regions of interest in the segmented image, some of the obtained regions of interest are classified into different labeled samples. The characteristic parameters of the labeled samples are extracted, and the regions of interest not labeled by the cell labeling module are compared and identified one by one according to the characteristic parameters of the labeled samples. According to the identification results, the unlabeled regions of interest are classified, and the classification results of the unlabeled regions of interest are tested. The classification accuracy rate of the unlabeled regions of interest is calculated, and a classification accuracy rate output threshold is set to distinguish the levels of the classification accuracy rate of the unlabeled regions of interest. It is determined whether to output the classification results. When it is determined to output the classification results, the number of regions of interest and the region of interest images that need to be output with classification results are statistically calculated and output. Compared with the prior art, the self-checking process for identifying and recognizing diseased cells is increased. Taking the labeled regions in the cytopathological smear scanned image of the detected object individual as a reference, the cell types in other regions are compared and recognized, which can improve the recognition accuracy of diseased cells when there are overlapping phenomena of multiple types of cells and other substances. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a schematic diagram of the overall structure of the cervical cancer cytology screening system based on image analysis of the present invention;

[0040] Figure 2 is a flowchart of the cervical cancer cytology screening method based on image analysis of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following further describes the present invention in conjunction with the specific embodiments. Among them, the drawings are only for illustrative purposes, showing only schematic diagrams, not physical diagrams, and should not be construed as limiting the present invention. In order to better illustrate the specific embodiments of the present invention, some components in the drawings will be omitted, enlarged or reduced, and do not represent the dimensions of the actual product.

[0042] Embodiment 1

[0043] As Figure 1 shown in the structure of the cervical cancer cytology screening system based on image analysis and Figure 2 shown in the working process steps of the screening system,

[0044] The screening method of the screening system includes the following steps:

[0045] Step 1: Prepare a cervical liquid-based smear using the thin-layer liquid-based technology and scan the cervical liquid-based smear through the image acquisition module. During the automatic smear scanning process, the focusing strategy adopts the three-point progressive uniform sampling method that can simultaneously process the unimodal or bimodal criterion function waveform to obtain at least one set of smear scanning images of the cervical liquid-based smear. When there is sufficient cervical fluid, multiple sets of cervical liquid-based smears can be prepared to obtain a larger number of smear scanning images.

[0046] Step 2: Enhance the obtained smear scanning images through the image processing module and segment the obtained smear scanning images using the threshold segmentation method. The algorithm model is:

[0047]

[0048] where i and j respectively represent the horizontal and vertical coordinates of the pixel; g(i, j) represents the local characteristics of the pixel; f(i, j) represents the pixel gray value; T is the segmentation threshold. The value of the segmentation threshold T is determined by the empirical method according to the quality of the smear scanning images of the obtained cervical liquid-based smear. Generally speaking, in the specific implementation process, the Otsu algorithm can be used to determine the threshold, which will not be elaborated here.

[0049] After segmenting the image, extract all the regions of interest of the segmented smear scanning image according to the types of substances contained, including seven types: normal cell region, diseased cell region, impurity region, overlapping region of normal cells and diseased cells, overlapping region of normal cells and impurities, overlapping region of diseased cells and impurities, and overlapping region of normal cells, diseased cells and impurities.

[0050] Step 3: Number the regions of interest in order of type as A1, A2,..., A7. Select several regions of each type of region of interest in the smear scanning image as samples, mark the selected samples to obtain marked samples. Similarly, the marked samples are classified into normal cell marked samples, diseased cell marked samples, impurity marked samples, marked samples of overlapping normal cells and diseased cells, marked samples of overlapping normal cells and impurities, marked samples of overlapping diseased cells and impurities, and marked samples of overlapping normal cells, diseased cells and impurities according to the types of substances contained in the regions of interest of the segmented image. Number the marked samples as Aij, where i is the region of interest type number, i = 1, 2,..., 7; j is the marked sample number, j = 1, 2,..., n, and n is a positive integer.

[0051] Step 4: According to the feature set for automatically identifying the pathological changes of cervical cancer cells, use an image recognition algorithm to extract the feature parameters of the density features, size features, edge features, shape features, and texture features of the labeled samples. Among them, the density features include average gray level, standard deviation, entropy, and object-background contrast; the size features include area, perimeter, major axis length, and minor axis length; the edge features include edge intensity, edge variance, and edge blur coefficient; the shape features include convexity, circularity, shape factor, eccentricity, Fourier descriptor, average normalized radius, normalized radius variance, normalized radius entropy, area ratio, and roughness; the texture features include fractal dimension, gray-level co-occurrence matrix, local binary pattern mean, and LBP variance. The extracted feature parameters of the labeled samples include at least one feature from the density features, at least one feature from the size features, at least one feature from the edge features, at least one feature from the shape features, and at least one feature from the texture features;

[0052] Step 5: Obtain the feature parameters of each unlabeled region of interest respectively, and construct an evaluation data group for the unlabeled region of interest, denoted as Q t =(q 1t ,q 2t ,...,q lt ), where t is the number of the unlabeled region of interest, t = 1, 2,..., n, n is a positive integer, l is the number of items of the feature parameters of the unlabeled region of interest, l = 1, 2,..., u, u is a positive integer. Use the mean value of the feature parameters of the labeled samples as the comparison data group, denoted as P=(p1, p2,..., p l ). Calculate the Euclidean distance between the evaluation data group and the comparison data as the similarity Si of the data group. The calculation formula is:

[0053]

[0054] Classify all unlabeled regions of interest into normal cell regions, diseased cell regions, impurity regions, regions where normal cells and diseased cells overlap, regions where normal cells and impurities overlap, regions where diseased cells and impurities overlap, and regions where normal cells, diseased cells, and impurities overlap according to the calculation results of the similarity;

[0055] Step 6: Use the random sampling method to extract W test samples after classifying the unlabeled regions of interest. The number of test samples W is determined by the statistical formula: where N is the number of the number of test samples W; Z w is the sampling survey confidence level, usually taking 95%; P w is the dispersion degree of the sampling sample; E wis the sampling error range, usually taken as ±3%, and the extracted test samples are tested and verified. According to the verification results, the classification accuracy rate Cr of various unlabeled regions of interest is calculated v , and the calculation formula is: where v represents the type of unlabeled region of interest, v = 1, 2,..., 7; W v represents the total number of test samples of the v-th unlabeled region of interest; w v represents the number of correct items in the test samples of the v-th unlabeled region of interest;

[0056] Step 7, classify the level of the classification accuracy rate Cr v of the unlabeled regions of interest. The method for classifying the level of the classification accuracy rate Cr v is as follows:

[0057] After obtaining the classification accuracy rate Cr v , a sample set is created using the value of the classification accuracy rate Cr v , and the mean and standard deviation in the sample set are obtained. The data is standardized using the mean and standard deviation, and the standardization formula is In this formula, z is the standard parameter, σ is the variance of the sample data, and μ is the mean of the sample data. After completing the standardization, the standard parameter is used to adjust the numerical interval to between [0, 1], and the coincidence rate is classified using the function value of f(k). The classification mechanism is:

[0058] When , the classification accuracy rate Cr v is classified as first level;

[0059] When , the classification accuracy rate Cr v is classified as second level;

[0060] where f(k)min and f(k)max are the minimum and maximum values of the function value of f(k) respectively. Whether to output the classification result is determined according to the output threshold and the discrimination result. The determination principle is:

[0061] When the classification accuracy rate Cr v is classified as first level, it is determined not to output the classification result;

[0062] When the classification accuracy rate Cr v is classified as second level, it is determined to output the classification result;

[0063] Step 8, when the central control module determines to output the classification result, count the number of regions of interest for which the classification result needs to be output, and output the quantity statistical result and the corresponding region-of-interest image. When the central control module determines not to output the classification result, go back to Step 6 to re-extract the test samples of the unlabeled regions of interest that are not to be output, and perform the classification accuracy verification again. If the verification result still determines not to output the classification result, go back to Step 5. It should be noted that due to the uncertainty of the sampling samples, when the sample size is large, the statistical result may be distorted. If both tests determine not to output, it may be due to an error or inaccuracy in obtaining the feature parameter of the unlabeled region of interest. It is necessary to re-identify and obtain the feature parameter to calculate the similarity Si, so as to perform the region classification again.

[0064] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A cervical cancer cytology screening system based on image analysis, comprising an image acquisition module and an image processing module, characterized in that, It further includes a cell marking module, a feature extraction module, a comparative analysis module, a central control module, an identification and verification module, and a result output module, where; The image acquisition module is used to scan the cervical liquid-based smear to obtain the smear scan image of the cervical liquid-based smear; The image processing module is connected to the image acquisition module and is used to preprocess and segment the obtained smear scan image to obtain all regions of interest in the segmented smear scan image. The cell marking module is connected to the image segmentation module and is used to classify some of the obtained regions of interest as different marked samples according to the types of substances contained in the regions of interest in the segmented image. The feature extraction module is connected to the cell marking module and is used to extract the feature parameters of the marked samples. The comparative analysis module is connected to the feature extraction module and is used to perform one-by-one comparison and identification on the regions of interest not marked by the cell marking module according to the feature parameters of the marked samples, and classify the unmarked regions of interest according to the identification results. The identification and verification module is connected to the comparative analysis module and is used to verify the classification results of the unmarked regions of interest and calculate the classification accuracy rate of the unmarked regions of interest. The central control module is connected to the identification and verification module and is used to distinguish the levels of the classification accuracy rate of the unmarked regions of interest by setting the output threshold of the classification accuracy rate and determine whether to output the classification results. The result output module is connected to the central control module and is used to count the number of regions of interest for which the classification results need to be output and output the quantity statistical results and the corresponding region of interest images when the central control module determines to output the classification results. The screening method of the screening system includes the following steps: Step 1: Prepare a cervical liquid-based smear using the thin-layer liquid-based technology, and scan the cervical liquid-based smear through the image acquisition module to obtain at least one group of smear scan images of the cervical liquid-based smear. Step 2: Perform enhancement processing on the obtained smear scan image through the image processing module, and segment the obtained smear scan image using the threshold segmentation method. The algorithm model is: where i and j respectively represent the horizontal and vertical coordinates of the pixel; g(i,j) represents the local characteristics of the pixel; f(i,j) represents the pixel gray value; T is the segmentation threshold. After segmenting the image, extract all regions of interest in the segmented smear scan image according to the types of substances contained, including seven types: normal cell region, diseased cell region, impurity region, overlapping region of normal cells and diseased cells, overlapping region of normal cells and impurities, overlapping region of diseased cells and impurities, and overlapping region of normal cells, diseased cells and impurities. Step 3: Sequentially number the regions of interest as A1, A2... A7 according to their types. Select several regions of each type of region of interest in the smear scan image as samples, mark the selected samples to obtain marked samples, and number the marked samples as A ij , where i is the region of interest type number, i = 1, 2... 7; j is the marked sample number, j = 1, 2... n, and n is a positive integer; Step 4: According to the feature set for automatically identifying the pathological changes of cervical cancer cells, use an image recognition algorithm to extract the feature parameters of the density feature, size feature, edge feature, shape feature, and texture feature of the marked samples. Step 5: Obtain the characteristic parameters of the unmarked regions of interest respectively, and construct an evaluation data set for the unmarked regions of interest, denoted as Q t =(q 1t , q 2t ... q lt ), where t is the number of the unmarked regions of interest, t = 1, 2... n, n is a positive integer, l is the number of items of the characteristic parameters of the unmarked regions of interest, l = 1, 2... u, u is a positive integer, and use the mean value of the characteristic parameters of the marked samples as the comparison data set, denoted as P = (p1, p2... p l ), calculate the Euclidean distance between the evaluation data set and the comparison data as the similarity Si of the data sets, and the calculation formula is: Classify all unmarked regions of interest according to the similarity. Step 6: Use the random sampling method to extract W test samples after classifying the unlabeled regions of interest, test and verify the extracted test samples, and calculate the classification accuracy rate Cr of each type of unlabeled region of interest according to the verification results v , and the calculation formula is as follows: where v represents the type of unlabeled region of interest, v = 1, 2... 7; W v represents the total number of test samples of the v-th unlabeled region of interest; w v represents the number of correct items of the test samples of the v-th unlabeled region of interest; Step 7, classify the classification accuracy rate Cr v of the unmarked region of interest, and determine whether to output the classification result according to the output threshold and the discrimination result; Step 8, when the central control module determines to output the classification result, count the number of regions of interest (ROIs) for which the classification result needs to be output, and output the quantity statistical result and the corresponding ROI images. When the central control module determines not to output the classification result, return to Step 6 to re-extract the test samples of the unlabeled ROIs that are not to be output, and perform the classification accuracy verification again. If the verification result still determines not to output the classification result, return to Step 5; Among the feature parameters, the density features include average gray level, standard deviation, entropy, and target-background contrast; the size features include area, perimeter, major axis length, and minor axis length; the edge features include edge strength, edge variance, and edge blur coefficient; the shape features include convexity, circularity, shape factor, eccentricity, Fourier descriptor, average normalized radius, normalized radius variance, normalized radius entropy, area ratio, and roughness; the texture features include fractal dimension, gray level co-occurrence matrix, local binary pattern mean, and LBP variance; In Step 4, the feature parameters of the extracted labeled samples include at least one feature from the density features, at least one feature from the size features, at least one feature from the edge features, at least one feature from the shape features, and at least one feature from the texture features; In step seven, the classification accuracy rate Cr v is classified as follows: Obtain the classification accuracy rate Cr v After that, use the classification accuracy rate Cr v to create a sample set, and obtain the mean and standard deviation in the sample set. Use the mean and standard deviation to standardize the data. The standardization formula is In this formula, z is the standard parameter, σ is the variance of the sample data, and μ is the mean of the sample data. After completing the standardization, use the standard parameter to adjust the numerical range to between [0, 1], and classify the coincidence rate using the function value of f(k). The classification mechanism is as follows: When the classification accuracy rate Cr v is classified as first level; When the classification accuracy rate Cr v is classified as secondary; where f(k)min and f(k)max are respectively the minimum and maximum values of the function values of f(k).

2. The cervical cancer cytology screening system based on image analysis according to claim 1, characterized in that, The labeled samples are classified into normal cell labeled samples, diseased cell labeled samples, impurity labeled samples, labeled samples with normal cells and diseased cells overlapping, labeled samples with normal cells and impurities overlapping, labeled samples with diseased cells and impurities overlapping, and labeled samples with normal cells, diseased cells, and impurities overlapping according to the types of substances contained in the regions of interest in the segmented image.

3. The cervical cancer cytology screening system based on image analysis according to claim 1, characterized in that In Step 7, the principle for determining whether to output the classification result is: When the classification accuracy rate Cr v is classified as the first level, it is determined not to output the classification result; When the classification accuracy rate Cr v is classified as secondary level, it is determined as the output classification result.

4. The cervical cancer cytology screening system based on image analysis according to claim 1, wherein The screening system further includes a storage module, a processor, and a computer program stored on the storage module and executable on the processor. The storage module is connected to the image acquisition module and the result output module for storing the smear scan image information of the cervical liquid-based smear and the number and image information of the regions of interest for which the classification result needs to be output. Wherein, when the processor executes the program, it can implement the steps of the method described in Claim 1.

5. The cervical cancer cytology screening system based on image analysis according to claim 1, wherein In Step 2, the value of the segmentation threshold T is determined by the empirical method according to the quality of the smear scan image of the cervical liquid-based smear obtained.

6. The cervical cancer cytology screening system based on image analysis according to claim 1, characterized in that, In Step 6, the number of unlabeled regions of interest W in the test samples classified by the random sampling method is determined by the formula: where N is the number of test samples W; Z w is the sampling survey confidence level, taking 95%; P w is the dispersion degree of the sampling samples; E w is the sampling error range, taking ±3%.

Images