A method and system for segmenting ENT examination images

Abstract

The present invention relates to the field of image processing technology, and in particular to a method and system for segmenting ENT examination images. The method comprises the following steps: acquiring an ENT examination image and performing image quality assessment to obtain image quality assessment data, performing image clarity assessment based on the data to obtain image clarity data, performing cutting accuracy requirement analysis on the image to obtain image cutting accuracy data, and constructing a graph cut bias matrix in combination with the image clarity data, performing image complexity analysis to obtain image complexity data, normalizing the image and injecting matrix data to generate a level set bias matrix, performing threshold preset on the graph cut bias matrix and the level set bias matrix to obtain a cutting bias threshold, and generating an image cutting scheme for an ENT examination image based on the threshold. The present invention is conducive to adaptive image segmentation and improved image accuracy by analyzing the image.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an ENT examination image segmentation method and system. Background Art

[0002] In recent years, with the development of electronic endoscope technology, the resolution of ENT examination images has been continuously improved, providing more information for the early diagnosis and treatment of diseases. Manual analysis of these images is time-consuming and laborious, easily affected by subjective factors, and is not suitable for large-scale research or scenarios that require processing of large amounts of data. Therefore, there is an urgent need to develop automated and accurate ENT examination image segmentation technology to assist doctors in diagnosis and improve diagnostic efficiency and accuracy. Traditional ENT examination image segmentation methods mainly rely on edge detection, threshold segmentation, and region growing technology, which have the advantages of simple principles, easy implementation, small amount of calculation, and fast processing speed. However, they are sensitive to noise. ENT examination images are often accompanied by noise and uneven illumination. Traditional methods are difficult to accurately identify the edge of the target, resulting in inaccurate segmentation results. For images with complex structures, such as adhesions and blurred boundaries of lesion areas, traditional methods are difficult to accurately segment. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide an ENT examination image segmentation method and system to solve at least one of the above technical problems.

[0004] A method for segmenting an ear, nose and throat examination image comprises the following steps:

[0005] Step S1: acquiring an ENT examination image; performing image quality assessment on the ENT examination image to obtain image quality assessment data; performing clarity assessment on the ENT examination image based on the image quality assessment data to obtain image clarity data;

[0006] Step S2: Analyze the cutting accuracy requirements of the ENT examination image to obtain image cutting accuracy data; construct a graph cutting bias matrix for the image cutting accuracy data and the image clarity data to obtain a graph cutting bias matrix;

[0007] Step S3: performing image complexity analysis on the ENT examination image to obtain image complexity data; performing complexity normalization on the image complexity data to obtain image normalized complexity data; injecting the image normalized complexity data into matrix data to generate a level set bias matrix;

[0008] Step S4: Preset thresholds for the graph cut bias matrix and the level set bias matrix to obtain a cutting bias threshold; generate an image cutting scheme for the ENT examination image based on the cutting bias threshold to obtain an image cutting scheme.

[0009] The present invention can effectively identify low-quality images by evaluating the image quality of the acquired ENT examination images and evaluating the clarity based on the evaluation data. This provides a basis for subsequent image processing, ensures the quality of the input image, and thus improves the accuracy of subsequent segmentation; when performing image cutting, not only the clarity of the image itself is considered, but also the specific cutting accuracy requirements are analyzed. By integrating the image clarity data and the cutting accuracy data to construct a graph cut bias matrix, the segmentation algorithm can be guided to pay more attention to the area that needs fine segmentation, thereby improving the segmentation accuracy of the key area while ensuring the overall segmentation effect; By analyzing the complexity of the image and performing normalization processing, the complexity of different images can be mapped to the same scale for comparison. Injecting the normalized data into the matrix to generate the level set bias matrix can make the segmentation algorithm better adapt to images of different complexities. For example, for images with simple structures and clear boundaries, the algorithm can quickly and accurately complete the segmentation; while for images with complex structures and blurred boundaries, the algorithm will be more cautious in processing, thereby improving the robustness of the overall segmentation; by presetting the thresholds of the graph cut bias matrix and the level set bias matrix, the segmentation scheme can be flexibly adjusted according to actual needs and image features. This means that different segmentation strategies can be formulated for different types of ENT examination images, such as the nasal cavity, throat, etc., to achieve more accurate and efficient image segmentation, and ultimately improve the accuracy and efficiency of clinical diagnosis.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: Acquire ENT examination images;

[0012] Step S12: performing image edge density analysis on the ENT examination image to obtain image edge density data;

[0013] Step S13: acquiring an image frequency domain data set for the ENT examination image to obtain an image frequency domain data set;

[0014] Step S14: performing image quality assessment on the ENT examination image based on the image edge density data and the image frequency domain data set to obtain image quality assessment data;

[0015] Step S15: evaluating the clarity of the ENT examination image based on the image quality assessment data to obtain image clarity data;

[0016] The present invention provides raw data for subsequent image analysis and processing by acquiring an ENT examination image to be processed; the acquired ENT examination image is subjected to image edge density analysis to extract the quantity and distribution characteristics of edge information in the image. Areas with high edge density usually correspond to important anatomical structures or lesion areas in the image, while areas with low edge density represent relatively flat or simple textured tissues. This information can provide an important reference basis for subsequent image quality assessment and segmentation; the acquired ENT examination image is subjected to image edge density analysis to extract the quantity and distribution characteristics of edge information in the image, which can provide an important reference basis for subsequent image quality assessment and segmentation; based on the image edge density data and image frequency domain data set acquired in the previous steps, a more comprehensive and objective image quality assessment can be performed on the ENT examination image; and the clarity of the ENT examination image can be evaluated. Clarity is an important factor affecting subsequent image segmentation, and images with high clarity can improve the accuracy and reliability of segmentation. For images with low clarity, image enhancement technology can be considered for preprocessing to improve image quality and provide more favorable conditions for subsequent image segmentation.

[0017] Preferably, step S12 comprises the following steps:

[0018] Step S121: acquiring edge pixel features of the ENT examination image to obtain edge pixel feature data;

[0019] Step S122: marking edge pixels of the ENT examination image based on the edge pixel feature data to obtain edge pixel marking data;

[0020] Step S123: dividing the ENT examination image into unit areas to obtain a unit area examination image;

[0021] Step S124: calculating the number of edge pixels per unit area of ​​the inspection image based on the edge pixel identification data to obtain data on the number of edge pixels per unit area;

[0022] Step S125: Calculate the image edge density of the ENT examination image based on the edge pixel number per unit area data to obtain image edge density data.

[0023] The present invention helps to highlight the key areas and details in the ENT examination image by acquiring edge pixel features. By identifying and extracting edge pixels, information on visual features such as the shape, texture and contour of the image object can be obtained, providing basic data support for subsequent analysis and processing; edge pixel identification can clearly mark the area of ​​interest in the ENT examination image. Through identification, the image can be divided into meaningful parts, which is convenient for subsequent quantitative analysis and feature extraction of the image, and effectively improves the accuracy of abnormal structure or diseased tissue detection in the image; unit area division can standardize and quantify the ENT examination image. By dividing the image into grids or blocks of uniform size, the image can be evaluated more carefully and accurately, which is convenient for statistics and comparison of the number, size and distribution of abnormal areas in the image, thereby improving the accuracy and consistency of detection and diagnosis; the calculation of the number of edge pixels per unit area of ​​the image can provide local feature information of the ENT examination image. Through calculation, the edge complexity and detail richness of different areas in the image can be analyzed and compared, which helps to identify and locate diseased tissue or abnormal structure, and provides quantitative indicators for disease diagnosis and classification; image edge density calculation can quantitatively describe the overall complexity and detail level of the ENT examination image. Through this step, the differences between different ENT examination images can be evaluated and compared, providing objective quantitative indicators for disease diagnosis, grading and prognosis, and assisting clinicians in making more accurate and reliable decisions.

[0024] Preferably, step S13 comprises the following steps:

[0025] Step S131: performing frequency domain conversion on the ENT examination image to obtain an image frequency domain;

[0026] Step S132: acquiring the frequency spectrum of the image in the frequency domain to obtain the image frequency spectrum;

[0027] Step S133: performing high-frequency energy recognition on the image frequency spectrum to obtain high-frequency energy data of the image frequency spectrum;

[0028] Step S134: performing data mapping on the ENT examination image based on the high-frequency energy data of the image frequency spectrum to obtain a high-frequency energy image;

[0029] Step S135: performing high-frequency distribution analysis on the high-frequency energy image to obtain high-frequency energy distribution data;

[0030] Step S136: performing spectrum flatness analysis on the image frequency spectrum based on the high-frequency energy distribution data to obtain spectrum flatness data;

[0031] Step S137: performing data aggregation on the high-frequency energy distribution data and the spectrum flatness data to obtain an image frequency domain data set.

[0032] The present invention can reveal the characteristics of ENT examination images in frequency space through frequency domain conversion. By converting the image from the spatial domain to the frequency domain, information of different frequency components in the image can be obtained, which is convenient for subsequent analysis and processing. This step helps to highlight the subtle details and key features of the image; obtaining the frequency spectrum of the image can intuitively show the energy distribution of the image at different frequencies. Frequency spectrum analysis can help understand the frequency composition of the image and identify the high-frequency and low-frequency components in the image, so that the image features at different frequencies can be extracted and analyzed respectively; frequency energy recognition can highlight the edges, details and other high-frequency features in the ENT examination image. By identifying and extracting high-frequency energy, the visibility of small abnormal structures or diseased tissues in the image can be enhanced, which is convenient for subsequent detection and analysis; data mapping can visually present the distribution of high-frequency energy in the ENT examination image. Through mapping, the concentrated area of ​​high-frequency components in the image can be intuitively displayed, providing visual guidance for further analysis and diagnosis; high-frequency energy distribution analysis can quantitatively describe the concentration and dispersion of high-frequency energy in the ENT examination image. Through analysis, the local detail richness and overall complexity of the image can be evaluated, providing quantitative indicators for disease diagnosis and grading, and improving the accuracy and consistency of clinical decision-making; spectrum flatness analysis can evaluate the flatness of the frequency spectrum of ENT examination images. A higher flatness indicates that the image frequency spectrum energy distribution is relatively even. A lower flatness indicates that the image frequency spectrum energy distribution is uneven, and there are edges, details or other high-frequency features, which require further analysis and diagnosis; the data set forms an image frequency domain data set, which integrates high-frequency energy distribution data and spectrum flatness data, laying the foundation for subsequent further analysis and modeling based on the frequency domain. Through the data set, ENT examination images can be explored and studied more comprehensively and in-depth.

[0033] Preferably, step S2 comprises the following steps:

[0034] Step S21: Preset the cutting result of the ENT examination image to obtain a preset cutting result;

[0035] Step S22: obtaining a cutting result of the ENT examination image to obtain an actual cutting result;

[0036] Step S23: analyzing the overlap degree of the preset cutting results based on the actual cutting results to obtain the overlapping degree of the cutting data;

[0037] Step S24: performing cutting accuracy requirement analysis on the ENT examination image based on the degree of overlapping of the cutting data to obtain image cutting accuracy data;

[0038] Step S25: construct a graph cut bias matrix for the image cutting accuracy data and the image clarity data to obtain a graph cut bias matrix.

[0039] The present invention can provide an initial guide and benchmark for the cutting of ENT examination images by setting a preset cutting result. By preset cutting, the preliminary results of image cutting can be quickly obtained, which helps to guide the subsequent cutting direction and strategy and ensure cutting efficiency and accuracy; the actual cutting result acquisition is a key step in the ENT examination image cutting process. Through special cutting algorithms and techniques, objects or regions of interest can be extracted from the image to obtain image fragments that are more suitable for subsequent analysis and diagnosis, thereby improving the accuracy of disease detection and diagnosis; the overlap degree analysis can quantitatively evaluate the consistency and accuracy of the preset cutting results and the actual cutting results. By analyzing the overlap degree of the two results, the differences and deviations of the cutting can be identified, which helps to optimize and improve the cutting algorithm and improve the accuracy and reliability of the cutting; the cutting accuracy requirement analysis can dynamically adjust the cutting accuracy according to the actual medical scene and image features. By analyzing the image clarity and overlap degree, the required accuracy level of cutting can be determined, and a balance can be found between speed and accuracy to meet the needs and requirements of clinical diagnosis; the construction of the graph cut bias matrix integrates key information such as image clarity and cutting accuracy. This matrix can more comprehensively and systematically describe and analyze the cutting requirements and characteristics of ENT examination images, provide important parameters and references for subsequent disease detection, diagnosis and analysis based on image cutting, and improve the effectiveness and accuracy of the overall process.

[0040] Preferably, step S3 comprises the following steps:

[0041] Step S31: performing image complexity analysis on the ENT examination image to obtain image complexity data;

[0042] Step S32: weighting the image complexity data to obtain image complexity weight data;

[0043] Step S33: weightedly merging the image complexity data based on the image complexity weighted data to obtain an image complexity data set;

[0044] Step S34: normalizing the complexity of the image complex data set to obtain image normalized complexity data;

[0045] Step S35: injecting the normalized complexity data of the image into the matrix data to generate a level set bias matrix.

[0046] The present invention helps to evaluate the feature richness and processing difficulty of the image by analyzing the complexity of the ENT examination image. Through the quantitative analysis of the image complexity, the information such as the number of objects in the image, the richness of details, the variability of texture, etc. can be obtained, which provides important reference data for subsequent processing and analysis; setting the image complexity weight can highlight the key features and areas of interest in the ENT examination image. By assigning weights to the image complexity data, the relative importance of different image parts in terms of complexity can be expressed, thereby providing a basis for subsequent weighted processing and decision-making; weighted merging can integrate image complexity data and complex weight data. Through merging, a data set that describes the complexity of the image more comprehensively and accurately can be obtained, providing basic input data for complexity normalization and other subsequent processing steps; complexity normalization can standardize and compare the complexity of different ENT examination images. Through normalization processing, the complexity differences caused by factors such as image acquisition, lighting, and noise can be eliminated, helping subsequent analysis and processing steps to more effectively process different image data; the generation of the level set bias matrix integrates the image complexity normalization data and other related data. This matrix can more comprehensively and systematically describe and analyze the complexity, importance, and feature distribution of ENT examination images, provide valuable information and reference for disease detection, diagnosis, and analysis tasks, and improve the accuracy and effectiveness of overall decision-making.

[0047] Preferably, step S31 includes the following steps:

[0048] Step S311: performing boundary contour feature analysis on the ENT examination image to obtain boundary contour feature data;

[0049] Step S312: performing boundary contour recognition on the ENT examination image based on the boundary contour feature data to obtain image boundary contour data;

[0050] Step S313: extracting geometric data from the image boundary contour data to obtain image geometric data;

[0051] Step S314: performing geometric feature analysis on the image geometric data to obtain image geometric feature data;

[0052] Step S315: merging feature vectors of the image boundary contour data and the image geometric feature data to obtain an image feature vector;

[0053] Step S316: performing image complexity analysis on the ENT examination image based on the image feature vector to obtain image complexity data.

[0054] The present invention can extract and describe the shape and contour information of objects in ENT examination images through boundary contour feature analysis. By identifying and analyzing edges, inflection points, curvature and other contour features, the structure and layout information of objects in the image can be obtained, providing important features for subsequent identification, classification and diagnosis; boundary contour recognition can clearly distinguish and mark the objects of interest and background areas in ENT examination images. Through recognition, the precise contour of the object can be obtained, which helps to focus and separate the target object, facilitates subsequent quantitative analysis and feature extraction, and improves the accuracy of disease detection and diagnosis; geometric data extraction can quantify the shape, size and position of objects in ENT examination images. By extracting geometric data such as length, width, angle, area, etc., the object can be accurately described and analyzed, laying the foundation for subsequent shape analysis, size comparison and spatial relationship research; geometric feature analysis can reveal the symmetry, regularity and other geometric characteristics of objects in ENT examination images. Through analysis, abnormal or unique features of objects in shape and structure can be discovered and quantified, providing important indicators for disease diagnosis, grading and classification, and helping to improve the accuracy of clinical decision-making; feature vectors merge and integrate image boundary contour data and geometric feature data. This merging step can more comprehensively and concisely represent the features of image objects, retain key information in the original data, and reduce the dimension and noise of the data; image complexity analysis can quantitatively describe the edges, textures, details and other complex features of ENT examination images. By analyzing the image feature vector, the processing difficulty, feature richness and information content of the image can be evaluated, providing important reference for subsequent image classification, segmentation, detection and other tasks, and helping algorithms and models to better process and understand image data.

[0055] Preferably, step S4 comprises the following steps:

[0056] Step S41: aligning the dimensions of the graph cut bias matrix and the level set bias matrix to obtain a graph cut bias matrix with the same dimension and a level set bias matrix with the same dimension;

[0057] Step S42: Preset thresholds for the graph cut bias isodimensional matrix and the level set bias isodimensional matrix to obtain a cut bias threshold;

[0058] Step S43: performing image cutting operation analysis on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold to obtain image cutting operation data;

[0059] Step S44: generating an image cutting scheme for the ENT examination image based on the image cutting operation data to obtain an image cutting scheme.

[0060] The present invention can make the graph cut bias matrix and the level set bias matrix consistent in dimension through dimension alignment, ensuring that the two can be added or combined. Through dimension alignment, the dimension mismatch problem caused by different original data or processing methods can be eliminated, ensuring that the subsequent processing and analysis steps can be consistently and accurately applied to the two matrices; the threshold preset can provide quantitative standards and decision-making basis for image cutting operation analysis. By setting the cutting bias threshold, the sensitivity and specificity requirements of image cutting can be defined, helping to find a balance between speed and accuracy, and ensuring that the cutting results meet the actual diagnostic needs and clinical requirements; the image cutting operation analysis can guide and optimize the image cutting scheme according to the preset cutting bias threshold. By analyzing the graph cut bias same-dimensional matrix and the level set bias same-dimensional matrix, the strategies, parameters and techniques required for image cutting can be determined, including selecting a suitable cutting algorithm, setting cutting parameters and optimizing the cutting process to achieve the expected results; the generation of the image cutting scheme is based on the graph cutting bias data and image feature information. By integrating the results of the image cutting operation analysis, a set of cutting schemes can be customized for ENT examination images, including cutting steps, cutting sequence, cutting algorithm selection, cutting parameter setting, etc. Customized image cutting solutions can improve cutting efficiency, ensure cutting accuracy, and meet the needs of clinical diagnosis.

[0061] Preferably, step S43 includes the following steps:

[0062] Step S431: performing bias data identification on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold, and obtaining matrix bias data, which includes graph cut bias data and level set bias data;

[0063] Step S432: Compare the graph cut bias data with the level set bias data. If the graph cut bias data is greater than the level set bias data, mark the matrix bias data as graph cut bias data. If the graph cut bias data is less than the level set bias data, mark the matrix bias data as level set bias data. If the graph cut bias data is equal to the level set bias data, mark the matrix bias data as unbiased data.

[0064] Step S433: integrating the image cutting high bias data, the image cutting low bias data and the unbiased data to obtain the image cutting bias data;

[0065] Step S434: performing image cutting processing on the ENT examination image based on the image cutting bias data. If the image cutting bias data is graph cutting bias data, image graph cutting processing is performed to produce a graph cutting cutting image; if the image cutting bias data is level set bias data, image level set processing is performed to generate a level set cutting image; if the image cutting bias data is unbiased data, image level set and graph cutting dual processing is performed to generate a level set graph cutting cutting image;

[0066] Step S435: integrating the graph cut bias data, the level set cut image and the level set graph cut cut image to obtain image cut operation data.

[0067] The present invention can extract and identify their bias from the graph cut bias isodimensional matrix and the level set bias isodimensional matrix through bias data identification. Through this step, the graph cut bias and level set bias in the two matrices can be clarified, providing key input data for subsequent decision-making and image cutting operations, ensuring that the cutting scheme meets expectations and requirements; comparing the graph cut bias data and the level set bias data can clarify the size relationship of the two biases, and mark the matrix bias data according to the results. This step helps to determine whether the graph cut method or the level set method is more inclined in the image cutting operation, providing guidance for selecting appropriate cutting algorithms and parameters, and improving cutting efficiency and accuracy; data integration can summarize the graph cut bias data and the level set bias data to form comprehensive image cutting bias data. Through integration, the graph cut and the high-bias level set method can be considered at the same time to ensure the comprehensiveness and flexibility of the cutting scheme, so as to process different types of ENT examination images and improve the cutting effect; the image cutting process adopts different cutting strategies according to the image cutting bias data. For biased graph cuts, the graph cut algorithm is used to focus on analyzing and processing the graphics and topological structure of the image; for level set bias, the level set method is used to accurately capture and process the shape and edges of the image. For unbiased images, a dual processing method is used to combine the advantages of the two methods to ensure the cutting effect; data integration summarizes the graph cut bias data and cutting results, including graph cut cutting images, level set cutting images, and level set graph cut cutting images. This step can comprehensively present the results of the image cutting operation, providing a complete set of cutting images for subsequent image analysis, object recognition, disease diagnosis and other tasks to meet different analysis needs and application scenarios.

[0068] Preferably, the present invention further provides an ENT examination image segmentation system for executing the ENT examination image segmentation method as described above, the ENT examination image segmentation system comprising:

[0069] An image clarity analysis module is used to obtain an ENT examination image; perform image quality assessment on the ENT examination image to obtain image quality assessment data; perform clarity assessment on the ENT examination image based on the image quality assessment data to obtain image clarity data;

[0070] A graph cut matrix generation module is used to analyze the cutting accuracy requirements of ENT examination images to obtain image cutting accuracy data; a graph cut bias matrix is ​​constructed based on the image cutting accuracy data and image clarity data to obtain a graph cut bias matrix;

[0071] The level set matrix generation module is used to analyze the image complexity of the ENT examination image to obtain image complexity data; perform complexity normalization on the image complexity data to obtain image normalized complexity data; inject the image normalized complexity data into the matrix data to generate a level set bias matrix;

[0072] The image cutting scheme generation module is used to preset the thresholds of the graph cutting bias matrix and the level set bias matrix to obtain the cutting bias threshold; based on the cutting bias threshold, the image cutting scheme is generated for the ENT examination image to obtain the image cutting scheme.

[0073] In summary, the present invention provides an ENT examination image segmentation method and system. The ENT examination image segmentation system consists of an image clarity analysis module, a graph cut matrix generation module, a level set matrix generation module, and an image cutting scheme generation module. It can implement any ENT examination image segmentation method described in the present invention, and is used to combine the operations between computer programs running on each module to implement any ENT examination image segmentation method. The internal structures of the system cooperate with each other, which can greatly reduce repetitive work and manpower investment, and can quickly and effectively provide a more accurate and efficient ENT examination image segmentation process, thereby simplifying the operation process of the ENT examination image segmentation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0075] Figure 1 A schematic diagram of the steps of a method for segmenting an ear, nose and throat examination image according to the present invention;

[0076] Figure 2 for Figure 1 Detailed step flow diagram of step S4;

[0077] Figure 3 for Figure 2 Detailed step flow chart of step S43 in FIG. DETAILED DESCRIPTION

[0078] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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 technicians in this field without creative work are within the scope of protection of the present invention.

[0079] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0080] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0081] To achieve this, please refer to Figures 1 to 3 The present invention provides an ENT examination image segmentation method, comprising the following steps:

[0082] Step S1: acquiring an ENT examination image; performing image quality assessment on the ENT examination image to obtain image quality assessment data; performing clarity assessment on the ENT examination image based on the image quality assessment data to obtain image clarity data;

[0083] Step S2: Analyze the cutting accuracy requirements of the ENT examination image to obtain image cutting accuracy data; construct a graph cutting bias matrix for the image cutting accuracy data and the image clarity data to obtain a graph cutting bias matrix;

[0084] Step S3: performing image complexity analysis on the ENT examination image to obtain image complexity data; performing complexity normalization on the image complexity data to obtain image normalized complexity data; injecting the image normalized complexity data into matrix data to generate a level set bias matrix;

[0085] Step S4: Preset thresholds for the graph cut bias matrix and the level set bias matrix to obtain a cutting bias threshold; generate an image cutting scheme for the ENT examination image based on the cutting bias threshold to obtain an image cutting scheme.

[0086] In the embodiment of the present invention, please refer to Figure 1 FIG. 1 is a schematic diagram of the steps of the ENT examination image segmentation method of the present invention. In this example, the ENT examination image segmentation method includes the following steps:

[0087] Step S1: acquiring an ENT examination image; performing image quality assessment on the ENT examination image to obtain image quality assessment data; performing clarity assessment on the ENT examination image based on the image quality assessment data to obtain image clarity data;

[0088] In an embodiment of the present invention, by connecting an ENT endoscope device to an image acquisition system, a clear ENT examination image is obtained and saved in a digital format; then, the obtained image is preprocessed, including graying, denoising and other operations, and an edge detection algorithm, such as the Canny operator, is used to calculate the image edge density value as image edge density data; then, the image is Fourier transformed, and frequency domain feature parameters such as high-frequency energy and spectrum flatness are extracted to construct an image frequency domain data set; then, the image edge density data is fused with the image frequency domain data set to construct a feature vector, and an image quality assessment model such as a support vector machine or a neural network is used to predict the image quality score to obtain image quality assessment data; finally, indicators related to clarity are extracted from the image quality assessment data, mapped to corresponding clarity levels according to a predefined clarity level classification standard, and the image clarity data is output.

[0089] Step S2: Analyze the cutting accuracy requirements of the ENT examination image to obtain image cutting accuracy data; construct a graph cutting bias matrix for the image cutting accuracy data and the image clarity data to obtain a graph cutting bias matrix;

[0090] In an embodiment of the present invention, a medical image annotation tool is used to manually annotate the boundary of a target area of ​​an ENT examination image, and the annotation result is saved as a preset cutting result; then, a suitable image segmentation algorithm, such as U-Net or Graph Cut, is selected, and the annotated data set is used for training or parameter adjustment to segment the ENT examination image to obtain an actual cutting result; then, the actual cutting result is compared with the preset cutting result at the pixel level, and the Dice coefficient is calculated to measure the degree of overlap between the two results; then, a Dice coefficient threshold is set according to clinical diagnostic requirements, and the calculated Dice coefficient is compared with the threshold to generate image cutting accuracy data; finally, an image quality assessment algorithm is used to obtain image clarity data, and the image cutting accuracy data is fused with the image clarity data to generate a graph cut bias matrix, each element of which represents the degree of bias of the corresponding pixel in the graph cut algorithm, guiding the graph cut algorithm to pay more attention to the segmentation effect of a specific area.

[0091] Step S3: performing image complexity analysis on the ENT examination image to obtain image complexity data; performing complexity normalization on the image complexity data to obtain image normalized complexity data; injecting the image normalized complexity data into matrix data to generate a level set bias matrix;

[0092] In an embodiment of the present invention, by extracting features such as texture, edge, color, etc. of an ENT examination image, and performing feature fusion to obtain a numerical value reflecting the complexity of the image, image complexity data is generated; then, different weight coefficients are set for different areas of the image according to task requirements to obtain image complexity weighted data; then, the image complexity weighted data are weightedly summed to obtain an image complex data set; subsequently, the image complex data set is normalized to obtain image normalized complexity data; finally, the image normalized complexity data is arranged into a matrix according to the spatial position information of the image to generate a level set bias matrix, in which each element of the matrix represents the image normalized complexity value of the corresponding pixel point, guiding the level set curve to move to the area with high image complexity, thereby improving the segmentation accuracy.

[0093] Step S4: Preset thresholds for the graph cut bias matrix and the level set bias matrix to obtain a cutting bias threshold; generate an image cutting scheme for the ENT examination image based on the cutting bias threshold to obtain an image cutting scheme.

[0094] In an embodiment of the present invention, by aligning the dimensions of a graph cut bias matrix and a level set bias matrix of the same dimension, a graph cut bias matrix and a level set bias matrix of the same dimension are obtained; then, thresholds are preset for the two matrices to obtain a cutting bias threshold; then, image cutting operations are performed on the two matrices based on the cutting bias threshold to obtain image cutting operation data; finally, a cutting scheme for an ENT examination image is generated based on the image cutting operation data, and the scheme includes information such as the coordinates of the cutting area, a cutting path, and a cutting order.

[0095] The present invention can effectively identify low-quality images by evaluating the image quality of the acquired ENT examination images and evaluating the clarity based on the evaluation data. This provides a basis for subsequent image processing, ensures the quality of the input image, and thus improves the accuracy of subsequent segmentation; when performing image cutting, not only the clarity of the image itself is considered, but also the specific cutting accuracy requirements are analyzed. By integrating the image clarity data and the cutting accuracy data to construct a graph cut bias matrix, the segmentation algorithm can be guided to pay more attention to the area that needs fine segmentation, thereby improving the segmentation accuracy of the key area while ensuring the overall segmentation effect; By analyzing the complexity of the image and performing normalization processing, the complexity of different images can be mapped to the same scale for comparison. Injecting the normalized data into the matrix to generate the level set bias matrix can make the segmentation algorithm better adapt to images of different complexities. For example, for images with simple structures and clear boundaries, the algorithm can quickly and accurately complete the segmentation; while for images with complex structures and blurred boundaries, the algorithm will be more cautious in processing, thereby improving the robustness of the overall segmentation; by presetting the thresholds of the graph cut bias matrix and the level set bias matrix, the segmentation scheme can be flexibly adjusted according to actual needs and image features. This means that different segmentation strategies can be formulated for different types of ENT examination images, such as the nasal cavity, throat, etc., to achieve more accurate and efficient image segmentation, and ultimately improve the accuracy and efficiency of clinical diagnosis.

[0096] Preferably, step S1 comprises the following steps:

[0097] Step S11: Acquire ENT examination images;

[0098] In an embodiment of the present invention, an ENT endoscope device is connected to an image acquisition system to ensure that the device is working normally and is in image acquisition mode, the endoscope is gently inserted into the patient's ENT area, the angle and focal length of the endoscope are adjusted to obtain clear images of the examination area, the acquired images are transmitted to the image acquisition system in real time, and saved in a digital format, such as JPEG, PNG or DICOM format.

[0099] Step S12: performing image edge density analysis on the ENT examination image to obtain image edge density data;

[0100] In an embodiment of the present invention, the acquired ENT examination image is preprocessed, including grayscale, denoising and other operations to reduce noise, and a suitable edge detection algorithm, such as the Canny operator, the Sobel operator, etc., is used to perform edge detection on the preprocessed image, and the number of edge pixels in the image is counted and divided by the total number of pixels in the image to obtain the edge density value of the image, and the calculated image edge density value is output as image edge density data.

[0101] Step S13: acquiring an image frequency domain data set for the ENT examination image to obtain an image frequency domain data set;

[0102] In an embodiment of the present invention, the acquired ENT examination image is subjected to Fourier transform to convert it from the spatial domain to the frequency domain to obtain the spectral information of the image, and the spectral information of the image is analyzed to extract characteristic parameters such as high-frequency energy and spectral flatness to characterize the details, texture and other information of the image, and the extracted frequency domain characteristic parameters are integrated together to construct an image frequency domain data set.

[0103] Step S14: performing image quality assessment on the ENT examination image based on the image edge density data and the image frequency domain data set to obtain image quality assessment data;

[0104] In an embodiment of the present invention, by fusing the obtained image edge density data with the obtained image frequency domain data set, a feature vector that comprehensively reflects the image quality is constructed, and a suitable image quality assessment model, such as a support vector machine, a neural network, etc., is selected or trained to evaluate image clarity, contrast, noise and other indicators. The trained image quality assessment model is used to predict the fused feature vector to obtain the image quality score, which is output as image quality assessment data.

[0105] Step S15: evaluating the clarity of the ENT examination image based on the image quality assessment data to obtain image clarity data;

[0106] In an embodiment of the present invention, by extracting clarity-related indicators from image quality assessment data, such as edge-based clarity indicators and frequency-based clarity indicators, the extracted clarity indicators are mapped to corresponding levels according to predefined clarity level classification standards, such as clear, general, blurred, etc., and the assessed clarity level is output as image clarity data.

[0107] The present invention provides raw data for subsequent image analysis and processing by acquiring an ENT examination image to be processed; the acquired ENT examination image is subjected to image edge density analysis to extract the quantity and distribution characteristics of edge information in the image. Areas with high edge density usually correspond to important anatomical structures or lesion areas in the image, while areas with low edge density represent relatively flat or simple textured tissues. This information can provide an important reference basis for subsequent image quality assessment and segmentation; the acquired ENT examination image is subjected to image edge density analysis to extract the quantity and distribution characteristics of edge information in the image, which can provide an important reference basis for subsequent image quality assessment and segmentation; based on the image edge density data and image frequency domain data set acquired in the previous steps, a more comprehensive and objective image quality assessment can be performed on the ENT examination image; and the clarity of the ENT examination image can be evaluated. Clarity is an important factor affecting subsequent image segmentation, and images with high clarity can improve the accuracy and reliability of segmentation. For images with low clarity, image enhancement technology can be considered for preprocessing to improve image quality and provide more favorable conditions for subsequent image segmentation.

[0108] Preferably, step S12 comprises the following steps:

[0109] Step S121: acquiring edge pixel features of the ENT examination image to obtain edge pixel feature data;

[0110] In an embodiment of the present invention, the acquired ENT examination image is grayed to convert the color image into a grayscale image, and the grayscale image is convolved using a Sobel operator or other gradient operator to calculate the gradient value and gradient direction of each pixel point. The gradient value represents the degree of change in the brightness of the pixel point. According to the gradient value and gradient direction, the feature information of the edge pixel is extracted, such as the gradient amplitude, gradient direction, edge strength, edge direction, etc., and the feature information is stored as edge pixel feature data.

[0111] Step S122: marking edge pixels of the ENT examination image based on the edge pixel feature data to obtain edge pixel marking data;

[0112] In an embodiment of the present invention, a suitable gradient threshold is set through statistical analysis, and each pixel in the image is traversed. If the gradient amplitude of the pixel is greater than the preset threshold, it is marked as an edge pixel, otherwise it is marked as a non-edge pixel, and the edge identification information of each pixel is stored as edge pixel identification data, which can be stored using a binary image, where 1 represents an edge pixel and 0 represents a non-edge pixel.

[0113] Step S123: dividing the ENT examination image into unit areas to obtain a unit area examination image;

[0114] In an embodiment of the present invention, a suitable unit area size is determined according to the size of the image and the analysis requirements. For example, the image can be divided into a number of square areas of equal size. According to the set unit area size, the original image is divided into a number of non-overlapping unit area areas, and each area represents a part of the image content.

[0115] Step S124: calculating the number of edge pixels per unit area of ​​the inspection image based on the edge pixel identification data to obtain data on the number of edge pixels per unit area;

[0116] In an embodiment of the present invention, by reading the generated edge pixel identification data and the divided unit area region information, for each unit area region, the edge identification information of all pixels in the region is traversed, the number of edge pixels is counted, and the number of edge pixels in each unit area region is stored as unit area edge pixel number data, which can be stored using an array or other data structure.

[0117] Step S125: Calculate the image edge density of the ENT examination image based on the edge pixel number per unit area data to obtain image edge density data.

[0118] In an embodiment of the present invention, for each unit area, the number of edge pixels in the area is divided by the total number of pixels in the area to obtain the edge density value of the area, and the edge density values ​​of all unit area areas are integrated together to form complete image edge density data, which can be stored using a matrix or other data structure.

[0119] The present invention helps to highlight the key areas and details in the ENT examination image by acquiring edge pixel features. By identifying and extracting edge pixels, information on visual features such as the shape, texture and contour of the image object can be obtained, providing basic data support for subsequent analysis and processing; edge pixel identification can clearly mark the area of ​​interest in the ENT examination image. Through identification, the image can be divided into meaningful parts, which is convenient for subsequent quantitative analysis and feature extraction of the image, and effectively improves the accuracy of abnormal structure or diseased tissue detection in the image; unit area division can standardize and quantify the ENT examination image. By dividing the image into grids or blocks of uniform size, the image can be evaluated more carefully and accurately, which is convenient for statistics and comparison of the number, size and distribution of abnormal areas in the image, thereby improving the accuracy and consistency of detection and diagnosis; the calculation of the number of edge pixels per unit area of ​​the image can provide local feature information of the ENT examination image. Through calculation, the edge complexity and detail richness of different areas in the image can be analyzed and compared, which helps to identify and locate diseased tissue or abnormal structure, and provides quantitative indicators for disease diagnosis and classification; image edge density calculation can quantitatively describe the overall complexity and detail level of the ENT examination image. Through this step, the differences between different ENT examination images can be evaluated and compared, providing objective quantitative indicators for disease diagnosis, grading and prognosis, and assisting clinicians in making more accurate and reliable decisions.

[0120] Preferably, step S13 comprises the following steps:

[0121] Step S131: performing frequency domain conversion on the ENT examination image to obtain an image frequency domain;

[0122] In an embodiment of the present invention, the acquired ENT examination image is preprocessed, such as grayscale, image normalization and other operations, and a two-dimensional fast Fourier transform algorithm is applied to the preprocessed image to convert the image from the spatial domain to the frequency domain. The transformed frequency information is represented in complex form, including two parts of information: amplitude spectrum and phase spectrum.

[0123] Step S132: acquiring the frequency spectrum of the image in the frequency domain to obtain the image frequency spectrum;

[0124] In an embodiment of the present invention, amplitude spectrum information is extracted from the obtained frequency domain data, and the square of the amplitude spectrum is usually used to represent the energy distribution of the image. The amplitude spectrum is centered, that is, low-frequency information is moved to the center of the image, and high-frequency information is moved to the edge of the image. The centered amplitude spectrum is logarithmically transformed to compress the dynamic range and enhance the display effect.

[0125] Step S133: performing high-frequency energy recognition on the image frequency spectrum to obtain high-frequency energy data of the image frequency spectrum;

[0126] In an embodiment of the present invention, a suitable frequency threshold or region is defined according to the characteristics of the image and the analysis requirements to identify high-frequency information, and the sum of the energy values ​​of all frequency points in the defined high-frequency region is calculated as the high-frequency energy data of the image frequency spectrum.

[0127] Step S134: performing data mapping on the ENT examination image based on the high-frequency energy data of the image frequency spectrum to obtain a high-frequency energy image;

[0128] In an embodiment of the present invention, the obtained high-frequency energy data is normalized and mapped to a grayscale range of 0 to 255, and the normalized high-frequency energy data is mapped to the corresponding pixel position of the original image to generate a high-frequency energy image, in which the pixel value represents the high-frequency energy intensity at that position.

[0129] Step S135: performing high-frequency distribution analysis on the high-frequency energy image to obtain high-frequency energy distribution data;

[0130] In an embodiment of the present invention, a region of interest, such as a lesion area or a tissue structure that needs to be analyzed in detail, is selected on a high-frequency energy image according to medical knowledge and analysis objectives, and image segmentation techniques, such as threshold segmentation and region growing, are used to segment the ROI from the high-frequency energy image. A high-frequency energy distribution analysis is performed on the segmented ROI, and multiple statistical feature parameters are extracted, such as the mean, the average high-frequency energy value of all pixels in the ROI, and the standard deviation, which reflects the degree of discreteness of the high-frequency energy distribution in the ROI.

[0131] Step S136: performing spectrum flatness analysis on the image frequency spectrum based on the high-frequency energy distribution data to obtain spectrum flatness data;

[0132] In the embodiment of the present invention, the uniformity of the image frequency distribution is quantified by selecting a suitable spectrum flatness index. Commonly used indexes include Wiener entropy, the larger the value, the flatter the spectrum and the more uniform the image information; spectral entropy, the closer the value is to 1, the flatter the spectrum; Flatness Measure, the closer the value is to 1, the flatter the spectrum. According to the selected index formula, the global spectrum flatness value is calculated using the image frequency spectrum, the image frequency spectrum is divided into multiple sub-regions, the spectrum flatness of each sub-region is calculated respectively, and the results are integrated as spectrum flatness data.

[0133] Step S137: performing data aggregation on the high-frequency energy distribution data and the spectrum flatness data to obtain an image frequency domain data set.

[0134] In an embodiment of the present invention, a plurality of extracted high-frequency energy distribution characteristic parameters and the spectrum flatness values ​​calculated in the steps are organized into a feature vector, the feature vector of each ENT examination image is taken as a sample, and the feature vectors and corresponding label information of all samples, such as image quality level, disease category, etc., are integrated to construct a complete image frequency domain data set, and the constructed image frequency domain data set is stored.

[0135] The present invention can reveal the characteristics of ENT examination images in frequency space through frequency domain conversion. By converting the image from the spatial domain to the frequency domain, information of different frequency components in the image can be obtained, which is convenient for subsequent analysis and processing. This step helps to highlight the subtle details and key features of the image; obtaining the frequency spectrum of the image can intuitively show the energy distribution of the image at different frequencies. Frequency spectrum analysis can help understand the frequency composition of the image and identify the high-frequency and low-frequency components in the image, so that the image features at different frequencies can be extracted and analyzed respectively; frequency energy recognition can highlight the edges, details and other high-frequency features in the ENT examination image. By identifying and extracting high-frequency energy, the visibility of small abnormal structures or diseased tissues in the image can be enhanced, which is convenient for subsequent detection and analysis; data mapping can visually present the distribution of high-frequency energy in the ENT examination image. Through mapping, the concentrated area of ​​high-frequency components in the image can be intuitively displayed, providing visual guidance for further analysis and diagnosis; high-frequency energy distribution analysis can quantitatively describe the concentration and dispersion of high-frequency energy in the ENT examination image. Through analysis, the local detail richness and overall complexity of the image can be evaluated, providing quantitative indicators for disease diagnosis and grading, and improving the accuracy and consistency of clinical decision-making; spectrum flatness analysis can evaluate the flatness of the frequency spectrum of ENT examination images. A higher flatness indicates that the image frequency spectrum energy distribution is relatively even. A lower flatness indicates that the image frequency spectrum energy distribution is uneven, and there are edges, details or other high-frequency features, which require further analysis and diagnosis; the data set forms an image frequency domain data set, which integrates high-frequency energy distribution data and spectrum flatness data, laying the foundation for subsequent further analysis and modeling based on the frequency domain. Through the data set, ENT examination images can be explored and studied more comprehensively and in-depth.

[0136] Preferably, step S2 comprises the following steps:

[0137] Step S21: Preset the cutting result of the ENT examination image to obtain a preset cutting result;

[0138] In an embodiment of the present invention, by using a professional medical image annotation tool, the boundaries of the target area, such as the lesion area and the organ contour, are manually annotated on the ENT examination image, and the annotation result is saved as a binary image, in which the pixel value of the target area is 1 and the pixel value of the background area is 0, as the preset cutting result.

[0139] Step S22: obtaining a cutting result of the ENT examination image to obtain an actual cutting result;

[0140] In an embodiment of the present invention, a suitable image segmentation algorithm is selected according to image characteristics and segmentation targets, such as deep learning models such as U-Net and SegNet, or traditional segmentation algorithms such as Graph Cut and Level Set, and trained using a labeled image data set; when using a traditional algorithm, it is necessary to adjust the algorithm parameters according to the image characteristics, and the ENT examination image is segmented using the trained model or the algorithm with adjusted parameters to obtain a segmentation result, which is also represented in the form of a binary image.

[0141] Step S23: analyzing the overlap degree of the preset cutting results based on the actual cutting results to obtain the overlapping degree of the cutting data;

[0142] In the embodiment of the present invention, the actual cutting result and the preset cutting result are compared at the pixel level, the number of pixels overlapping the two results is counted, and the Dice coefficient is used to measure the degree of overlap between the two results. The value range of the Dice coefficient is 0 to 1. The larger the value, the higher the degree of overlap. The calculation formula is: Dice = (2 * number of pixels in the overlapping area) / (number of pixels in the actual segmentation area + number of pixels in the preset segmentation area).

[0143] Step S24: performing cutting accuracy requirement analysis on the ENT examination image based on the degree of overlapping of the cutting data to obtain image cutting accuracy data;

[0144] In an embodiment of the present invention, a reasonable Dice coefficient threshold, such as 0.8, is set according to clinical diagnosis requirements, and the calculated Dice coefficient is compared with the preset threshold. If the Dice coefficient is higher than the threshold, it is considered that the segmentation result of the region meets the accuracy requirement; otherwise, it is considered that the region requires a higher segmentation accuracy. According to the accuracy judgment result of each pixel, image segmentation accuracy data is generated, which can be represented by a binary image, in which the pixel value of the area requiring higher accuracy is 1, and the pixel value of other areas is 0.

[0145] Step S25: construct a graph cut bias matrix for the image cutting accuracy data and the image clarity data to obtain a graph cut bias matrix.

[0146] In an embodiment of the present invention, the clarity of ENT examination images is evaluated by using an image quality assessment algorithm to obtain image clarity data, which can be represented by a grayscale image. The higher the pixel value, the higher the clarity. The image cutting accuracy data and the image clarity data are fused to generate a graph cut bias matrix. Each element of the graph cut bias matrix represents the degree of bias of the corresponding pixel in the graph cut algorithm. For example, for areas that require high-precision segmentation and have higher clarity, a higher bias value is assigned to guide the graph cut algorithm to pay more attention to the segmentation effect of these areas; for areas with low accuracy requirements or lower clarity, a lower bias value is assigned to allow the graph cut algorithm to perform a coarser segmentation in these areas.

[0147] The present invention can provide an initial guide and benchmark for the cutting of ENT examination images by setting a preset cutting result. By preset cutting, the preliminary results of image cutting can be quickly obtained, which helps to guide the subsequent cutting direction and strategy and ensure cutting efficiency and accuracy; the actual cutting result acquisition is a key step in the ENT examination image cutting process. Through special cutting algorithms and techniques, objects or regions of interest can be extracted from the image to obtain image fragments that are more suitable for subsequent analysis and diagnosis, thereby improving the accuracy of disease detection and diagnosis; the overlap degree analysis can quantitatively evaluate the consistency and accuracy of the preset cutting results and the actual cutting results. By analyzing the overlap degree of the two results, the differences and deviations of the cutting can be identified, which helps to optimize and improve the cutting algorithm and improve the accuracy and reliability of the cutting; the cutting accuracy requirement analysis can dynamically adjust the cutting accuracy according to the actual medical scene and image features. By analyzing the image clarity and overlap degree, the required accuracy level of cutting can be determined, and a balance can be found between speed and accuracy to meet the needs and requirements of clinical diagnosis; the construction of the graph cut bias matrix integrates key information such as image clarity and cutting accuracy. This matrix can more comprehensively and systematically describe and analyze the cutting requirements and characteristics of ENT examination images, provide important parameters and references for subsequent disease detection, diagnosis and analysis based on image cutting, and improve the effectiveness and accuracy of the overall process.

[0148] Preferably, step S3 comprises the following steps:

[0149] Step S31: performing image complexity analysis on the ENT examination image to obtain image complexity data;

[0150] In an embodiment of the present invention, a plurality of features reflecting the complexity of the image are extracted from the ENT examination image, for example: texture features, such as energy, contrast, correlation, entropy, etc., are extracted using a gray-level co-occurrence matrix. Edge features, edge information is extracted using an edge detection operator, and features such as edge density and edge strength are calculated. Color features: the image is in color, and features such as color histogram and color moment can be extracted. Feature fusion: the extracted multiple features are fused, and principal component analysis or other dimensionality reduction methods can be used to reduce the multidimensional features to one dimension to obtain a value that comprehensively reflects the complexity of the image, and the complexity value of each pixel is mapped to the image to generate image complexity data.

[0151] Step S32: weighting the image complexity data to obtain image complexity weight data;

[0152] In the embodiment of the present invention, the segmentation accuracy of the determined region is more important. For example, if the goal is to segment the lesion region, the segmentation accuracy of the lesion edge is more important; if the goal is to segment the organ contour, the segmentation accuracy of the contour edge is more important. Different weight coefficients are set for different regions of the image according to the task requirements. For example, a higher weight coefficient, such as 1.2, is assigned to important regions. A normal weight coefficient, such as 1.0, is assigned to general regions. A lower weight coefficient, such as 0.8, is assigned to unimportant regions.

[0153] Step S33: weightedly merging the image complexity data based on the image complexity weighted data to obtain an image complexity data set;

[0154] In an embodiment of the present invention, the obtained image complexity weighted data is weighted and summed to obtain a comprehensive complexity value for each image. The weighted summation process can adopt a linear weighted or nonlinear weighted method, and the specific method can be selected according to the actual situation. For example, the weighted complexity values ​​of all features can be summed to obtain a comprehensive complexity value, or the comprehensive complexity value can be calculated by weighted average. Finally, a data set containing the comprehensive complexity values ​​of all images is obtained, that is, an image complex data set. For example, the weighted complexity of the texture feature of an image is 3, the weighted complexity of the edge feature is 2, and the weighted complexity of the frequency feature is 1, then its comprehensive complexity value is 3+2+1=6, and the comprehensive complexity values ​​of all images are combined into a data set, that is, an image complex data set.

[0155] Step S34: normalizing the complexity of the image complex data set to obtain image normalized complexity data;

[0156] In an embodiment of the present invention, in order to eliminate the influence of the numerical differences between different images in the image complex data set, it is necessary to normalize the data. The normalization method may adopt the maximum and minimum value normalization, Z-score normalization and other methods. For example, the maximum and minimum value normalization method may be used to map all the numerical values ​​in the image complex data set to between 0 and 1. The normalized data is called image normalized complexity data.

[0157] Step S35: injecting the normalized complexity data of the image into the matrix data to generate a level set bias matrix.

[0158] In an embodiment of the present invention, the image normalized complexity data is arranged according to the spatial position information of the image to form a matrix, which is called a level set bias matrix. Each element in the matrix represents the image normalized complexity value of the corresponding pixel. The level set bias matrix can be used as prior information for the level set segmentation algorithm to guide the level set curve to move to the area with high image complexity, thereby improving the segmentation accuracy. For example, if there are multiple lesions in the image, the element value corresponding to the lesion area in the level set bias matrix will be relatively large, guiding the level set curve to preferentially segment these areas, and finally generating a matrix containing all the image normalized complexity values, namely the level set bias matrix.

[0159] The present invention helps to evaluate the feature richness and processing difficulty of the image by analyzing the complexity of the ENT examination image. Through the quantitative analysis of the image complexity, the information such as the number of objects in the image, the richness of details, the variability of texture, etc. can be obtained, which provides important reference data for subsequent processing and analysis; setting the image complexity weight can highlight the key features and areas of interest in the ENT examination image. By assigning weights to the image complexity data, the relative importance of different image parts in terms of complexity can be expressed, thereby providing a basis for subsequent weighted processing and decision-making; weighted merging can integrate image complexity data and complex weight data. Through merging, a data set that describes the complexity of the image more comprehensively and accurately can be obtained, providing basic input data for complexity normalization and other subsequent processing steps; complexity normalization can standardize and compare the complexity of different ENT examination images. Through normalization processing, the complexity differences caused by factors such as image acquisition, lighting, and noise can be eliminated, helping subsequent analysis and processing steps to more effectively process different image data; the generation of the level set bias matrix integrates the image complexity normalization data and other related data. This matrix can more comprehensively and systematically describe and analyze the complexity, importance, and feature distribution of ENT examination images, provide valuable information and reference for disease detection, diagnosis, and analysis tasks, and improve the accuracy and effectiveness of overall decision-making.

[0160] Preferably, step S31 includes the following steps:

[0161] Step S311: performing boundary contour feature analysis on the ENT examination image to obtain boundary contour feature data;

[0162] In an embodiment of the present invention, the collected ENT examination image is preprocessed, including operations such as image enhancement, denoising and segmentation, to remove irrelevant information and highlight the target area, such as the nasal cavity, pharynx or larynx. Then, edge detection algorithms such as the Canny operator or the Sobel operator are used to extract edge information in the image to obtain the boundary contour of the image. Next, the extracted boundary contour is analyzed to extract its features, such as boundary length, curvature, number of inflection points, shape factor, etc. The boundary length can reflect the size of the target area, the curvature can reflect the degree of curvature of the target area, the number of inflection points can reflect the complexity of the target area, and the shape factor can reflect the shape characteristics of the target area. Finally, the extracted boundary contour feature data is stored in a data structure. For example, the features such as the boundary length, curvature, number of inflection points and shape factor of each image can be stored in a table or matrix to form boundary contour feature data.

[0163] Step S312: performing boundary contour recognition on the ENT examination image based on the boundary contour feature data to obtain image boundary contour data;

[0164] In an embodiment of the present invention, a machine learning model, such as a support vector machine or a random forest, is trained by using the extracted boundary contour feature data to identify boundary contours in an image. The training data set may include a variety of different types of ENT examination images, and corresponding boundary contour annotation information. After the training is completed, the trained model is used to perform boundary contour recognition on new ENT examination images, that is, the boundary contour in the image is automatically recognized and extracted based on the edge information and boundary contour features of the image. For example, the model can determine whether there is a boundary of the nasal cavity, pharynx or larynx in the image based on the edge information and boundary contour features of the image, and extract it to form image boundary contour data, such as a series of coordinate points.

[0165] Step S313: extracting geometric data from the image boundary contour data to obtain image geometric data;

[0166] In an embodiment of the present invention, geometric features are extracted from the obtained image boundary contour data, such as calculating the area, perimeter, centroid, major axis and minor axis of the boundary contour. The area can reflect the size of the target area, the perimeter can reflect the length of the target area, the centroid can reflect the center position of the target area, and the major axis and minor axis can reflect the shape of the target area. For example, the geometric features such as the area, perimeter, centroid, major axis and minor axis of the nasal cavity, pharynx or larynx can be calculated to form image geometric data, such as a table or structure containing these geometric feature values.

[0167] Step S314: performing geometric feature analysis on the image geometric data to obtain image geometric feature data;

[0168] In an embodiment of the present invention, the extracted image geometric data is analyzed to extract its features, such as shape factor, circularity, rectangularity, etc. The shape factor can reflect the shape feature of the target area, the circularity can reflect the circularity of the target area, and the rectangularity can reflect the rectangularity of the target area. For example, the shape factor, circularity, rectangularity, and other geometric features of the nasal cavity, pharynx, or larynx area can be calculated to form image geometric feature data, such as a vector containing these geometric feature values.

[0169] Step S315: merging feature vectors of the image boundary contour data and the image geometric feature data to obtain an image feature vector;

[0170] In an embodiment of the present invention, a feature vector containing multiple features is formed by merging the obtained image boundary contour data and the obtained image geometric feature data. For example, features such as boundary length, curvature, number of inflection points, shape factor, area, perimeter, centroid, major axis, minor axis, shape factor, circularity and rectangularity can be merged into a feature vector, and each feature vector corresponds to an ENT examination image. The image feature vector finally obtained can be a vector containing all these feature values.

[0171] Step S316: performing image complexity analysis on the ENT examination image based on the image feature vector to obtain image complexity data.

[0172] In an embodiment of the present invention, a machine learning model, such as a support vector machine or a random forest, is trained by using the obtained image feature vector to analyze the complexity of the image. The training data set may include a variety of different types of ENT examination images and corresponding complexity annotation information. After the training is completed, the trained model is used to perform complexity analysis on new ENT examination images, that is, the complexity of the image is automatically determined based on the feature vector of the image, and it is quantified into a numerical value, such as 0-10 points, where the higher the value, the more complex the image.

[0173] The present invention can extract and describe the shape and contour information of objects in ENT examination images through boundary contour feature analysis. By identifying and analyzing edges, inflection points, curvature and other contour features, the structure and layout information of objects in the image can be obtained, providing important features for subsequent identification, classification and diagnosis; boundary contour recognition can clearly distinguish and mark the objects of interest and background areas in ENT examination images. Through recognition, the precise contour of the object can be obtained, which helps to focus and separate the target object, facilitates subsequent quantitative analysis and feature extraction, and improves the accuracy of disease detection and diagnosis; geometric data extraction can quantify the shape, size and position of objects in ENT examination images. By extracting geometric data such as length, width, angle, area, etc., the object can be accurately described and analyzed, laying the foundation for subsequent shape analysis, size comparison and spatial relationship research; geometric feature analysis can reveal the symmetry, regularity and other geometric characteristics of objects in ENT examination images. Through analysis, abnormal or unique features of objects in shape and structure can be discovered and quantified, providing important indicators for disease diagnosis, grading and classification, and helping to improve the accuracy of clinical decision-making; feature vectors merge and integrate image boundary contour data and geometric feature data. This merging step can more comprehensively and concisely represent the features of image objects, retain key information in the original data, and reduce the dimension and noise of the data; image complexity analysis can quantitatively describe the edges, textures, details and other complex features of ENT examination images. By analyzing the image feature vector, the processing difficulty, feature richness and information content of the image can be evaluated, providing important reference for subsequent image classification, segmentation, detection and other tasks, and helping algorithms and models to better process and understand image data.

[0174] Preferably, step S4 comprises the following steps:

[0175] Step S41: aligning the dimensions of the graph cut bias matrix and the level set bias matrix to obtain a graph cut bias matrix with the same dimension and a level set bias matrix with the same dimension;

[0176] Step S42: Preset thresholds for the graph cut bias isodimensional matrix and the level set bias isodimensional matrix to obtain a cut bias threshold;

[0177] Step S43: performing image cutting operation analysis on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold to obtain image cutting operation data;

[0178] Step S44: generating an image cutting scheme for the ENT examination image based on the image cutting operation data to obtain an image cutting scheme.

[0179] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1Detailed step flow diagram of step S4 in FIG. 1 , in this embodiment, step S4 includes the following steps:

[0180] Step S41: aligning the dimensions of the graph cut bias matrix and the level set bias matrix to obtain a graph cut bias matrix with the same dimension and a level set bias matrix with the same dimension;

[0181] In an embodiment of the present invention, since the dimensions of the graph cut bias matrix and the level set bias matrix are inconsistent due to different generation methods or data sources, dimensional alignment can be achieved through interpolation, cropping or other image processing methods. For example, the dimension of the graph cut bias matrix is ​​512x512, while the dimension of the level set bias matrix is ​​256x256. The level set bias matrix can be enlarged to 512x512 by bilinear interpolation so that it has the same dimension as the graph cut bias matrix. Alternatively, the graph cut bias matrix can be cropped so that it has the same dimension as the level set bias matrix. Finally, a graph cut bias matrix with the same dimension as the level set bias matrix and a level set bias matrix with the same dimension as the graph cut bias matrix are obtained.

[0182] Step S42: Preset thresholds for the graph cut bias isodimensional matrix and the level set bias isodimensional matrix to obtain a cut bias threshold;

[0183] In an embodiment of the present invention, in order to determine the area for image cutting, it is necessary to set a threshold for the graph cut biased isodimensional matrix and the level set biased isodimensional matrix. The threshold can be set according to the characteristics of the image, clinical experience or other prior knowledge. For example, a threshold can be set according to the differences in tissue structure in the image, the morphological characteristics of the lesion or other information, and the elements in the matrix greater than the threshold are regarded as areas that need to be cut, and the elements less than the threshold are regarded as areas that do not need to be cut. The threshold can be a single value or a range. For example, a threshold range can be set, and the elements in the matrix greater than the upper threshold are regarded as areas that need to be cut, and the elements less than the lower threshold are regarded as areas that do not need to be cut. The elements between the two are judged whether they need to be cut according to other conditions.

[0184] Step S43: performing image cutting operation analysis on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold to obtain image cutting operation data;

[0185] In an embodiment of the present invention, the image cutting operation analysis is performed on the graph cut biased isodimensional matrix and the level set biased isodimensional matrix by using the set cutting bias threshold. The image cutting operation can adopt different methods, such as threshold-based binarization, region growing, watershed algorithm, etc. For example, the elements greater than the threshold in the graph cut biased isodimensional matrix and the level set biased isodimensional matrix can be set to 1, and the elements less than the threshold can be set to 0 to obtain two binary images. Then, logical operations can be performed on the two binary images, such as taking the intersection or union, to obtain the final image cutting area. Alternatively, a region growing algorithm can be used, starting from the pixel points greater than the threshold, and expanding to the surrounding pixel points until a certain condition is met to stop and obtain the image cutting area.

[0186] Step S44: generating an image cutting scheme for the ENT examination image based on the image cutting operation data to obtain an image cutting scheme.

[0187] In an embodiment of the present invention, a cutting scheme for an ENT examination image is generated by using the obtained image cutting operation data. The cutting scheme may include information such as the coordinates of the cutting area, the cutting path, and the cutting sequence. For example, a graph including the coordinates of the cutting area may be generated based on the image cutting operation data, or an instruction sequence including the cutting path and the cutting sequence may be generated.

[0188] The present invention can make the graph cut bias matrix and the level set bias matrix consistent in dimension through dimension alignment, ensuring that the two can be added or combined. Through dimension alignment, the dimension mismatch problem caused by different original data or processing methods can be eliminated, ensuring that the subsequent processing and analysis steps can be consistently and accurately applied to the two matrices; the threshold preset can provide quantitative standards and decision-making basis for image cutting operation analysis. By setting the cutting bias threshold, the sensitivity and specificity requirements of image cutting can be defined, helping to find a balance between speed and accuracy, and ensuring that the cutting results meet the actual diagnostic needs and clinical requirements; the image cutting operation analysis can guide and optimize the image cutting scheme according to the preset cutting bias threshold. By analyzing the graph cut bias same-dimensional matrix and the level set bias same-dimensional matrix, the strategies, parameters and techniques required for image cutting can be determined, including selecting a suitable cutting algorithm, setting cutting parameters and optimizing the cutting process to achieve the expected results; the generation of the image cutting scheme is based on the graph cutting bias data and image feature information. By integrating the results of the image cutting operation analysis, a set of cutting schemes can be customized for ENT examination images, including cutting steps, cutting sequence, cutting algorithm selection, cutting parameter setting, etc. Customized image cutting solutions can improve cutting efficiency, ensure cutting accuracy, and meet the needs of clinical diagnosis.

[0189] Preferably, step S43 includes the following steps:

[0190] Step S431: performing bias data identification on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold, and obtaining matrix bias data, which includes graph cut bias data and level set bias data;

[0191] Step S432: Compare the graph cut bias data with the level set bias data. If the graph cut bias data is greater than the level set bias data, mark the matrix bias data as graph cut bias data. If the graph cut bias data is less than the level set bias data, mark the matrix bias data as level set bias data. If the graph cut bias data is equal to the level set bias data, mark the matrix bias data as unbiased data.

[0192] Step S433: integrating the image cutting high bias data, the image cutting low bias data and the unbiased data to obtain the image cutting bias data;

[0193] Step S434: performing image cutting processing on the ENT examination image based on the image cutting bias data. If the image cutting bias data is graph cutting bias data, image graph cutting processing is performed to produce a graph cutting cutting image; if the image cutting bias data is level set bias data, image level set processing is performed to generate a level set cutting image; if the image cutting bias data is unbiased data, image level set and graph cutting dual processing is performed to generate a level set graph cutting cutting image;

[0194] Step S435: integrating the graph cut bias data, the level set cut image and the level set graph cut cut image to obtain image cut operation data.

[0195] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 2 FIG. 4 is a schematic diagram of a detailed step flow chart of step S43 in FIG. 4 . In this embodiment, step S43 includes the following steps:

[0196] Step S431: performing bias data identification on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold, and obtaining matrix bias data, which includes graph cut bias data and level set bias data;

[0197] In an embodiment of the present invention, each pixel point in the graph cut bias matrix of the same dimension and the level set bias matrix of the same dimension is judged according to the set cutting bias threshold value to determine whether it belongs to the area that needs to be cut. If the matrix value corresponding to the pixel point is greater than the threshold value, it is considered that the pixel point belongs to the area that needs to be cut, and its corresponding matrix value is stored as graph cut bias data and level set bias data respectively. For example, the value of a pixel point in the graph cut bias matrix of the same dimension is 0.8, the value of the corresponding pixel point in the level set bias matrix of the same dimension is 0.6, and the cutting bias threshold is 0.5, then these two pixels are both identified as areas that need to be cut, and their corresponding matrix values ​​0.8 and 0.6 are stored as graph cut bias data and level set bias data respectively, and two matrices with the same dimensions as the original matrices are obtained, which respectively store graph cut bias data and level set bias data, and these data together constitute the matrix bias data.

[0198] Step S432: Compare the graph cut bias data with the level set bias data. If the graph cut bias data is greater than the level set bias data, mark the matrix bias data as graph cut bias data. If the graph cut bias data is less than the level set bias data, mark the matrix bias data as level set bias data. If the graph cut bias data is equal to the level set bias data, mark the matrix bias data as unbiased data.

[0199] In an embodiment of the present invention, the graph cut bias data and the level set bias data are compared pixel by pixel. For each pixel, the size of the corresponding graph cut bias data and the level set bias data is compared. If the graph cut bias data is larger than the level set bias data, it is considered that the pixel is more inclined to be cut using the graph cut algorithm, and the matrix bias data corresponding to the pixel is marked as graph cut bias data. If the graph cut bias data is smaller than the level set bias data, it is considered that the pixel is more inclined to be cut using the level set algorithm, and the matrix bias data corresponding to the pixel is marked as level set bias data. If the graph cut bias data is equal to the level set bias data, it is considered that the bias degrees of the two algorithms for the pixel are the same, and the corresponding matrix bias data is marked as unbiased data. Finally, a matrix with the same dimension as the graph cut bias matrix and the level set bias matrix is ​​obtained, in which each pixel is marked as graph cut bias, level set bias or unbiased.

[0200] Step S433: integrating the image cutting high bias data, the image cutting low bias data and the unbiased data to obtain the image cutting bias data;

[0201] In an embodiment of the present invention, the image cutting bias data is obtained by integrating the marked matrix bias data. The matrix values ​​corresponding to the pixel points marked as the image cutting bias data are stored in a data structure, such as a matrix or an array, representing the image cutting bias data. The matrix values ​​corresponding to the pixel points marked as the level set bias data are stored in another data structure, representing the level set bias data. The matrix values ​​corresponding to the pixel points marked as the unbiased data are stored in a third data structure, representing the unbiased data. Finally, three data structures are obtained, which respectively contain the image cutting bias data, the level set bias data and the unbiased data.

[0202] Step S434: performing image cutting processing on the ENT examination image based on the image cutting bias data. If the image cutting bias data is graph cutting bias data, image graph cutting processing is performed to produce a graph cutting cutting image; if the image cutting bias data is level set bias data, image level set processing is performed to generate a level set cutting image; if the image cutting bias data is unbiased data, image level set and graph cutting dual processing is performed to generate a level set graph cutting cutting image;

[0203] In an embodiment of the present invention, the image cutting bias data is used to perform image cutting processing on the ENT examination image, and each pixel of the image is traversed to read the corresponding image cutting bias data. The image cutting bias data corresponding to the pixel is the graph cutting bias data, which is used as the input of the graph cutting algorithm, and the graph cutting algorithm is executed to perform image segmentation. The graph cutting algorithm usually needs to set the source point and the sink point, as well as the weight map of the image, and find the best path for segmenting the image through the minimum cut algorithm, and finally generate the graph cutting image. The image cutting bias data corresponding to the pixel is the level set bias data, which is used as the input of the level set algorithm, and the level set algorithm is executed to perform image segmentation. The level set algorithm usually needs to define an initial contour, and iterates to update the contour so that it gradually approaches the boundary of the target area, and finally generates the level set cutting image. The image cutting bias data corresponding to the pixel is unbiased data, so the level set algorithm is first executed to obtain a preliminary segmentation result, and then the result is used as the input of the graph cutting algorithm to further optimize the segmentation result, and finally generate the level set graph cutting image.

[0204] Step S435: integrating the graph cut bias data, the level set cut image and the level set graph cut cut image to obtain image cut operation data.

[0205] In an embodiment of the present invention, the generated graph cut image, level set cut image, level set graph cut image and graph cut bias data are integrated to obtain image cut operation data. First, the graph cut image, level set cut image and level set graph cut image are spliced ​​according to the pixels to form a complete image cut result, the graph cut bias data is matched with the image cut result, and the cutting method corresponding to each pixel is recorded, such as graph cut cutting, level set cutting or level set graph cut cutting. In addition, the cutting result of each pixel can also be recorded, such as the segmentation area to which the pixel belongs, the label of the area, etc. Finally, this information is stored in a data structure, such as a multidimensional array or dictionary, to form image cut operation data.

[0206] The present invention can extract and identify their bias from the graph cut bias isodimensional matrix and the level set bias isodimensional matrix through bias data identification. Through this step, the graph cut bias and level set bias in the two matrices can be clarified, providing key input data for subsequent decision-making and image cutting operations, ensuring that the cutting scheme meets expectations and requirements; comparing the graph cut bias data and the level set bias data can clarify the size relationship of the two biases, and mark the matrix bias data according to the results. This step helps to determine whether the graph cut method or the level set method is more inclined in the image cutting operation, providing guidance for selecting appropriate cutting algorithms and parameters, and improving cutting efficiency and accuracy; data integration can summarize the graph cut bias data and the level set bias data to form comprehensive image cutting bias data. Through integration, the graph cut and the high-bias level set method can be considered at the same time to ensure the comprehensiveness and flexibility of the cutting scheme, so as to process different types of ENT examination images and improve the cutting effect; the image cutting process adopts different cutting strategies according to the image cutting bias data. If the bias is towards graph cut, the graph cut algorithm is used to focus on analyzing and processing the graphics and topological structure of the image; if the bias is towards level set, the level set method is used to accurately capture and process the shape and edges of the image. A dual processing method is used for unbiased images, combining the advantages of the two methods to ensure the cutting effect; data integration summarizes the graph cut bias data and cutting results, including graph cut cutting images, level set cutting images, and level set graph cut cutting images. This step can comprehensively present the results of the image cutting operation, providing a complete set of cutting images for subsequent image analysis, object recognition, disease diagnosis and other tasks to meet different analysis needs and application scenarios.

[0207] Preferably, the present invention further provides an ENT examination image segmentation system for executing the ENT examination image segmentation method as described above, the ENT examination image segmentation system comprising:

[0208] An image clarity analysis module is used to obtain an ENT examination image; perform image quality assessment on the ENT examination image to obtain image quality assessment data; perform clarity assessment on the ENT examination image based on the image quality assessment data to obtain image clarity data;

[0209] A graph cut matrix generation module is used to analyze the cutting accuracy requirements of ENT examination images to obtain image cutting accuracy data; a graph cut bias matrix is ​​constructed based on the image cutting accuracy data and image clarity data to obtain a graph cut bias matrix;

[0210] The level set matrix generation module is used to analyze the image complexity of the ENT examination image to obtain image complexity data; perform complexity normalization on the image complexity data to obtain image normalized complexity data; inject the image normalized complexity data into the matrix data to generate a level set bias matrix;

[0211] The image cutting scheme generation module is used to preset the thresholds of the graph cutting bias matrix and the level set bias matrix to obtain the cutting bias threshold; based on the cutting bias threshold, the image cutting scheme is generated for the ENT examination image to obtain the image cutting scheme.

[0212] In summary, the present invention provides an ENT examination image segmentation method and system. The ENT examination image segmentation system consists of an image clarity analysis module, a graph cut matrix generation module, a level set matrix generation module, and an image cutting scheme generation module. It can implement any ENT examination image segmentation method described in the present invention, and is used to combine the operations between computer programs running on each module to implement any ENT examination image segmentation method. The internal structures of the system cooperate with each other, which can greatly reduce repetitive work and manpower investment, and can quickly and effectively provide a more accurate and efficient ENT examination image segmentation process, thereby simplifying the operation process of the ENT examination image segmentation system.

[0213] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for segmenting an ear, nose and throat examination image, characterized in that: The following steps are involved: Step S1: acquiring an ENT examination image; performing image quality assessment on the ENT examination image to obtain image quality assessment data; performing clarity assessment on the ENT examination image based on the image quality assessment data to obtain image clarity data; Step S2: Analyze the cutting accuracy requirements of the ENT examination image to obtain image cutting accuracy data; The image cutting accuracy data and the image clarity data are subjected to a graph cutting bias matrix construction to obtain a graph cutting bias matrix. Step S2 includes: Step S21: Preset the cutting result of the ENT examination image to obtain a preset cutting result; Step S22: obtaining a cutting result of the ENT examination image to obtain an actual cutting result; Step S23: analyzing the overlap degree of the preset cutting results based on the actual cutting results to obtain the overlapping degree of the cutting data; Step S24: performing cutting accuracy requirement analysis on the ENT examination image based on the degree of overlapping of the cutting data to obtain image cutting accuracy data; Step S25: constructing a graph cut bias matrix for the image cutting accuracy data and the image clarity data to obtain a graph cut bias matrix, wherein the elements of the graph cut bias matrix represent the degree of bias of the corresponding pixel in the graph cutting process; Step S3: performing image complexity analysis on the ENT examination image to obtain image complexity data; performing complexity normalization on the image complexity data to obtain image normalized complexity data; injecting the image normalized complexity data into matrix data to generate a level set bias matrix. Step S3 includes: Step S31: performing image complexity analysis on the ENT examination image to obtain image complexity data; Step S32: weighting the image complexity data to obtain image complexity weight data; Step S33: weightedly merging the image complexity data based on the image complexity weighted data to obtain an image complexity data set; Step S34: normalizing the complexity of the image complex data set to obtain image normalized complexity data; Step S35: injecting the normalized complexity data of the image into the matrix data to generate a level set bias matrix, wherein the elements in the level set bias matrix represent the normalized complexity values ​​of the image at the corresponding pixel points; Step S4: Preset thresholds for the graph cut bias matrix and the level set bias matrix to obtain a cutting bias threshold; generate an image cutting scheme for the ENT examination image based on the cutting bias threshold to obtain an image cutting scheme. Step S4 includes: Step S41: aligning the dimensions of the graph cut bias matrix and the level set bias matrix to obtain a graph cut bias matrix with the same dimension and a level set bias matrix with the same dimension; Step S42: Preset thresholds for the graph cut bias isodimensional matrix and the level set bias isodimensional matrix to obtain a cut bias threshold; Step S43: performing image cutting operation analysis on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold to obtain image cutting operation data; Step S44: generating an image cutting scheme for the ENT examination image based on the image cutting operation data to obtain an image cutting scheme.

2. The method for segmenting ENT examination images according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Acquire ENT examination images; Step S12: performing image edge density analysis on the ENT examination image to obtain image edge density data; Step S13: acquiring an image frequency domain data set for the ENT examination image to obtain an image frequency domain data set; Step S14: performing image quality assessment on the ENT examination image based on the image edge density data and the image frequency domain data set to obtain image quality assessment data; Step S15: Evaluate the clarity of the ENT examination image based on the image quality assessment data to obtain image clarity data.

3. The method for segmenting ENT examination images according to claim 2, characterized in that: Step S12 includes the following steps: Step S121: acquiring edge pixel features of the ENT examination image to obtain edge pixel feature data; Step S122: marking edge pixels of the ENT examination image based on the edge pixel feature data to obtain edge pixel marking data; Step S123: dividing the ENT examination image into unit areas to obtain a unit area examination image; Step S124: calculating the number of edge pixels per unit area of ​​the inspection image based on the edge pixel identification data to obtain data on the number of edge pixels per unit area; Step S125: Calculate the image edge density of the ENT examination image based on the edge pixel number per unit area data to obtain image edge density data.

4. The method for segmenting ENT examination images according to claim 2, characterized in that: Step S13 includes the following steps: Step S131: performing frequency domain conversion on the ENT examination image to obtain an image frequency domain; Step S132: acquiring the frequency spectrum of the image in the frequency domain to obtain the image frequency spectrum; Step S133: performing high-frequency energy recognition on the image frequency spectrum to obtain high-frequency energy data of the image frequency spectrum; Step S134: performing data mapping on the ENT examination image based on the high-frequency energy data of the image frequency spectrum to obtain a high-frequency energy image; Step S135: performing high-frequency distribution analysis on the high-frequency energy image to obtain high-frequency energy distribution data; Step S136: performing spectrum flatness analysis on the image frequency spectrum based on the high-frequency energy distribution data to obtain spectrum flatness data; Step S137: performing data aggregation on the high-frequency energy distribution data and the spectrum flatness data to obtain an image frequency domain data set.

5. The method for segmenting ENT examination images according to claim 1, characterized in that: Step S31 includes the following steps: Step S311: performing boundary contour feature analysis on the ENT examination image to obtain boundary contour feature data; Step S312: performing boundary contour recognition on the ENT examination image based on the boundary contour feature data to obtain image boundary contour data; Step S313: extracting geometric data from the image boundary contour data to obtain image geometric data; Step S314: performing geometric feature analysis on the image geometric data to obtain image geometric feature data; Step S315: merging feature vectors of the image boundary contour data and the image geometric feature data to obtain an image feature vector; Step S316: performing image complexity analysis on the ENT examination image based on the image feature vector to obtain image complexity data.

6. The method for segmenting ENT examination images according to claim 1, characterized in that: Step S43 includes the following steps: Step S431: performing bias data identification on the graph cut bias isodimensional matrix and the level set bias isodimensional matrix based on the cutting bias threshold, and obtaining matrix bias data, which includes graph cut bias data and level set bias data; Step S432: Compare the graph cut bias data with the level set bias data. If the graph cut bias data is greater than the level set bias data, mark the matrix bias data as graph cut bias data. If the graph cut bias data is less than the level set bias data, mark the matrix bias data as level set bias data. If the graph cut bias data is equal to the level set bias data, mark the matrix bias data as unbiased data. Step S433: integrating the image cutting high bias data, the image cutting low bias data and the unbiased data to obtain the image cutting bias data; Step S434: performing image cutting processing on the ENT examination image based on the image cutting bias data. If the image cutting bias data is graph cutting bias data, image graph cutting processing is performed to produce a graph cutting cutting image; if the image cutting bias data is level set bias data, image level set processing is performed to generate a level set cutting image; if the image cutting bias data is unbiased data, image level set and graph cutting dual processing is performed to generate a level set graph cutting cutting image; Step S435: integrating the graph cut bias data, the level set cut image and the level set graph cut cut image to obtain image cut operation data.

7. An ENT examination image segmentation system, characterized in that: Used to perform the ENT examination image segmentation method as claimed in claim 1, the ENT examination image segmentation system comprises: An image clarity analysis module is used to obtain an ENT examination image; perform image quality assessment on the ENT examination image to obtain image quality assessment data; perform clarity assessment on the ENT examination image based on the image quality assessment data to obtain image clarity data; A graph cut matrix generation module is used to analyze the cutting accuracy requirements of ENT examination images to obtain image cutting accuracy data; a graph cut bias matrix is ​​constructed based on the image cutting accuracy data and image clarity data to obtain a graph cut bias matrix; The level set matrix generation module is used to analyze the image complexity of the ENT examination image to obtain image complexity data; perform complexity normalization on the image complexity data to obtain image normalized complexity data; inject the image normalized complexity data into the matrix data to generate a level set bias matrix; The image cutting scheme generation module is used to preset the thresholds of the graph cutting bias matrix and the level set bias matrix to obtain the cutting bias threshold; based on the cutting bias threshold, the image cutting scheme is generated for the ENT examination image to obtain the image cutting scheme.

Images