An intelligent classification method for ENT lesion images
By acquiring the performance parameters and pathological similarity coefficients of the throat grayscale image, preliminary classification and adaptive grayscale enhancement are solved, and the problem of inaccurate classification of throat lesions in the existing technology is improved, and the accuracy of intelligent classification of ENT ERs and nose lesions is improved and refinement effects.
Patent Information
- Application Number
- CN202411252213.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-09
AI Technical Summary
The prior art in the intelligent classification process of throat lesion images is not accurate enough due to blurred image features, and direct local enhancement may introduce noise or lead to feature distortion, affecting the classification effect.
By obtaining the performance parameters and pathological similarity coefficients in the throat grayscale image, preliminary classification is carried out and grayscale enhancement is adaptive to highlight the characteristics of the disease, avoid distortion, and then refine classification is carried out.
It improves the intelligent classification accuracy of ENT lesions images, reduces the complexity of analysis, and ensures the accuracy of disease characteristics and the refinement of classification results.
Smart Images

Figure CN118762246B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image analysis and processing, and in particular to an intelligent classification method for ENT lesion images. Background Art
[0002] With the intensification of environmental pollution and the changes in modern lifestyles, ear, nose and throat diseases have become a disease with a high incidence rate in today's society, causing serious impacts on patients' work and life. Pharyngitis is a frequent disease among ear, nose and throat diseases. Throat lesions not only affect patients' breathing and swallowing functions, but in severe cases may also affect voice and cause various complications. At present, the images of throat lesions of different severity can be intelligently classified to assist relevant medical personnel in visual diagnosis, and then quickly evaluate the cause of the disease by observing the lesion site for diagnosis and treatment, thereby improving the efficiency of diagnosis and treatment.
[0003] However, in the intelligent classification process of throat lesion images, some images are incorrectly classified into the same category due to the fuzzy features of the images, resulting in inaccurate classification of ENT lesion images, which in turn affects the subsequent auxiliary diagnosis and treatment effects; therefore, it is crucial to perform local enhancement of the lesion site in the throat lesion image. However, in the prior art, direct local enhancement may introduce additional noise, and excessive enhancement may also cause defects such as distortion of the features of the lesion site, which in turn leads to poor subsequent intelligent classification of throat lesion images. Summary of the invention
[0004] In order to solve the technical problem that the prior art has poor intelligent classification effect on throat lesion images, the purpose of the present invention is to provide an intelligent classification method for ENT lesion images, and the technical solution adopted is as follows:
[0005] Obtain a grayscale image of the throat of each pharyngitis patient;
[0006] All pharyngeal grayscale images are classified using a preset classification method to obtain initial categories of all images; the preset classification method includes: in each pharyngeal grayscale image, all suspected lesion areas are obtained, and the performance parameters of each pixel are obtained according to the grayscale information of each pixel and the position distribution relative to the suspected lesion area; all lesion areas in each pharyngeal grayscale image are obtained according to the performance parameters; according to the grayscale differences and morphological feature differences between the lesion areas in different pharyngeal grayscale images, combined with the differences in the performance parameters of the pixels between the lesion areas, the pathological similarity coefficients between different pharyngeal grayscale images are obtained; according to the pathological similarity coefficients, all pharyngeal grayscale images are classified;
[0007] A reference throat grayscale image is selected from each of the initial categories of the image according to the image texture information; an enhancement factor of each of the lesion areas in each throat grayscale image is obtained according to the performance parameters of all pixels in each lesion area in each throat grayscale image and the grayscale difference inside and outside the area, combined with the pathological similarity coefficient between the throat grayscale image and the corresponding reference throat grayscale image; and the grayscale of the pixels in the lesion area in the corresponding throat grayscale image is enhanced according to the enhancement factor;
[0008] All enhanced throat grayscale images are classified using a preset classification method to obtain the final categories of all images.
[0009] Furthermore, the method for obtaining the performance parameters includes:
[0010] In each pharyngeal grayscale image, the grayscale coefficient of each pixel is obtained according to the deviation of the grayscale value of each pixel relative to the grayscale values of all pixels;
[0011] Taking any pixel point as a target pixel point, obtaining a position coefficient of the target pixel point according to the minimum Euclidean distance between the target pixel point and all the suspected lesion areas; the minimum Euclidean distance is negatively correlated with the position coefficient;
[0012] The performance parameter of the corresponding pixel point is obtained according to the grayscale coefficient and the position coefficient, and both the grayscale coefficient and the position coefficient are positively correlated with the performance parameter.
[0013] Furthermore, the method for obtaining the lesion area includes:
[0014] The pixel points whose performance parameters are greater than the preset parameter threshold are regarded as lesion pixels, and all lesion areas in each throat grayscale image are obtained according to the aggregation of all the lesion pixels.
[0015] Furthermore, the method for obtaining the pathological similarity coefficient includes:
[0016] In each pharyngeal grayscale image, according to the grayscale concentration features of all the pixels in the lesion area, the pathological grayscale parameters of each pharyngeal grayscale image are obtained;
[0017] Obtaining pathological morphological parameters of each pharyngeal grayscale image according to the boundary rules and area of all the lesion areas in each pharyngeal grayscale image;
[0018] Obtaining pathological performance parameters of each pharynx grayscale image according to the concentrated features of the performance parameters of all pixel points in the lesion area in each pharynx grayscale image;
[0019] According to the differences in the pathological grayscale parameters, the differences in the pathological morphological parameters and the differences in the pathological manifestation parameters between different throat grayscale images, the pathological similarity coefficients between different throat grayscale images are obtained; the differences in the pathological grayscale parameters, the differences in the pathological morphological parameters and the differences in the pathological manifestation parameters are all negatively correlated with the pathological similarity coefficient.
[0020] Furthermore, the method for obtaining the pathological morphological parameters includes:
[0021] According to the change of the slope of the line between adjacent pixel points on the region boundary of each lesion region, the contour expression parameters of each lesion region are obtained; according to the concentrated features of the contour expression parameters of all lesion regions in each pharyngeal grayscale image, the boundary feature parameters of each pharyngeal grayscale image are obtained;
[0022] In each pharyngeal grayscale image, the severity of the disease in each pharyngeal grayscale image is obtained according to the area of all the lesion areas; the area of the area is positively correlated with the severity of the disease;
[0023] According to the boundary feature parameters and the severity of the disease, the pathological morphological parameters of the corresponding throat grayscale image are obtained; the boundary feature parameters and the severity of the disease are both positively correlated with the pathological morphological parameters.
[0024] Furthermore, the method of classifying all throat grayscale images according to the pathological similarity coefficient includes:
[0025] The pathological similarity coefficient is used as a metric distance, and all throat grayscale images are clustered based on a distance clustering algorithm, and the clustering result is used as a classification result of the throat grayscale image.
[0026] Furthermore, the method for obtaining the reference throat grayscale image includes:
[0027] In each of the initial image categories, the number of gray levels of each throat gray image is obtained, and the throat gray image with the largest number of gray levels is used as a reference throat gray image in the initial image category.
[0028] Furthermore, the method for obtaining the enhancement factor includes:
[0029] The grayscale mean of all pixels in each lesion area is used as the first grayscale; each lesion area is expanded outward by a preset number of circles, and the grayscale mean of all pixels in the expanded area is used as the second grayscale; the difference between the first grayscale and the second grayscale is used as the first enhancement parameter;
[0030] Acquire a second enhancement parameter of each lesion area according to the concentrated features of the performance parameters of all pixels in each lesion area;
[0031] According to the first enhancement parameter and the second enhancement parameter of each lesion area in each throat grayscale image, combined with the pathological similarity coefficient between each throat grayscale image and the reference throat grayscale image in the initial image category to which it belongs, an enhancement factor for each lesion area in each throat grayscale image is obtained; the first enhancement parameter and the second enhancement parameter are both positively correlated with the enhancement factor, and the pathological similarity coefficient is negatively correlated with the enhancement factor.
[0032] Furthermore, the method for performing grayscale enhancement on the pixel points in the lesion area in the corresponding throat grayscale image according to the enhancement factor includes:
[0033] The sum of the normalized enhancement factor and the constant 1 is used as the grayscale enhancement weight, and the grayscale value of each pixel in each lesion area in each throat grayscale image is weighted using the grayscale enhancement weight, and the weighted result is used as the enhanced grayscale value of the corresponding pixel.
[0034] Furthermore, the method for obtaining the suspected lesion area includes:
[0035] Threshold segmentation is performed on each pharyngeal grayscale image to obtain all foreground areas; the foreground areas with a total number of pixels greater than a preset threshold are regarded as suspected lesion areas.
[0036] The present invention has the following beneficial effects:
[0037] The embodiment of the present invention obtains the pharynx grayscale image of each pharyngitis patient; uses a preset classification method to obtain the initial category of all images; the preset classification method includes: in each pharynx grayscale image, all suspected lesion areas are obtained, and the performance parameters of each pixel are obtained according to the grayscale information of each pixel point and the position distribution relative to the suspected lesion area, the performance parameters quantify the unique performance characteristics of each pixel point in the entire pharynx grayscale image, and the larger the performance parameter, the greater the possibility that it is a pixel point in the lesion area; all lesion areas in each pharynx grayscale image are obtained according to the performance parameters, and the situation that the regional boundary pixel point division effect is poor due to the regional division based on the grayscale value alone is compensated, and a more accurate lesion area is obtained; according to the grayscale difference and morphological feature difference between the lesion areas in different pharynx grayscale images, combined with the difference in the performance parameters of the pixels between the lesion areas, the pathological relationship between different pharynx grayscale images is obtained. Similarity coefficient, case similarity coefficient obtains the similarity of symptoms based on the color characteristics and morphological characteristics of the lesion area, so that throat grayscale images of similar severity can be subsequently divided into one category, thereby helping to quickly and centrally analyze the characteristics of the patient's symptoms; according to the pathological similarity coefficient, all throat grayscale images are classified; according to the image texture information, reference throat grayscale images are screened out from the initial category of each image; according to the performance parameters of all pixels in each lesion area in each throat grayscale image, and the grayscale difference inside and outside the area, combined with the pathological similarity coefficient between the throat grayscale image and the corresponding reference throat grayscale image, the enhancement factor of each lesion area in each throat grayscale image is obtained; according to the enhancement factor, the grayscale of the pixels in the lesion area in the corresponding throat grayscale image is enhanced to highlight the characteristics of the disease while avoiding distortion; all enhanced throat grayscale images are classified using a preset classification method to obtain the final category of all images. The present invention firstly performs a preliminary classification of images based on the pathological similarities of different throat grayscale images to reduce the complexity of subsequent analysis, then measures the degree of blur according to the texture information of the throat grayscale images for adaptive enhancement, highlights the disease characteristics while avoiding distortion, and finally refines the classification of the enhanced throat grayscale images, thereby improving the accuracy of intelligent classification of ENT lesion images. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0039] Figure 1 A flowchart of an intelligent classification method for ENT lesion images provided by an embodiment of the present invention;
[0040] Figure 2 A comparison diagram of a pharyngeal grayscale image and a corresponding threshold segmentation image provided by an embodiment of the present invention;
[0041] Figure 3 A schematic diagram of comparison of grayscale images of the throat of a mild pharyngitis symptom and a severe pharyngitis symptom provided by an embodiment of the present invention;
[0042] Figure 4 A flow chart of a method for obtaining a pathological similarity coefficient provided by an embodiment of the present invention;
[0043] Figure 5 The present invention is a flowchart of a method for obtaining an enhancement factor provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0044] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, features and effects of an intelligent classification method for ENT lesion images proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0045] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0046] The specific scheme of the intelligent classification method for ENT lesion images provided by the present invention is described in detail below with reference to the accompanying drawings.
[0047] See also Figure 1 , which shows a flow chart of an intelligent classification method for ENT lesion images provided by an embodiment of the present invention, specifically comprising:
[0048] Step S1, obtaining a grayscale image of the throat of each pharyngitis patient.
[0049] It should be noted that the embodiments of the present invention are mainly aimed at intelligent classification of throat images of pharyngitis patients.
[0050] In one embodiment of the present invention, first, under the guidance of relevant medical personnel, a high-definition camera is used to collect throat images of the oropharynx of each pharyngitis patient, and then the throat images are preprocessed, including at least grayscale and noise reduction, to obtain all throat grayscale images for subsequent analysis. Grayscale and noise reduction are both well-known technologies and will not be described in detail here.
[0051] Step S2, classify all throat grayscale images using a preset classification method to obtain initial categories of all images; the preset classification method includes: in each throat grayscale image, obtain all suspected lesion areas, and obtain the performance parameters of each pixel based on the grayscale information of each pixel and the position distribution relative to the suspected lesion area; obtain all lesion areas in each throat grayscale image based on the performance parameters; obtain the pathological similarity coefficient between different throat grayscale images based on the grayscale differences and morphological feature differences between the lesion areas in different throat grayscale images, combined with the differences in the performance parameters of the pixels between the lesion areas; and classify all throat grayscale images according to the pathological similarity coefficient.
[0052] Since the acquisition process of the pharyngeal grayscale image may be affected by factors such as the device itself or the acquisition operation, such as jitter and inaccurate focus, the image will be blurred and difficult to analyze the lesion features, which will lead to poor subsequent intelligent classification results. Therefore, the embodiment of the present invention first performs a preliminary classification on all pharyngeal grayscale images, and based on the preliminary classification results, the local features of the pharyngeal grayscale images are enhanced to highlight the image features, and then the classification is refined to improve the accuracy of the intelligent classification results of ENT lesion images.
[0053] Considering that the color of normal tissue in the oropharynx is usually pink and presents relatively uniform texture features in the grayscale image of the throat; and lesions may cause changes in the texture features of the tissue, usually appearing as milky white or light yellow granular areas and concentrated distribution; therefore, the embodiment of the present invention obtains the suspected lesion area based on the distinguishing texture features of the diseased tissue and the normal tissue area.
[0054] Preferably, in one embodiment of the present invention, considering that the threshold segmentation can perform image segmentation according to the distinguishing grayscale features of the pixel points, thereby distinguishing the suspected lesion area from the normal tissue area; and considering that the grayscale image of the throat may reflect light due to the presence of saliva attached, it is easy to mistakenly determine the saliva reflection area as the suspected lesion area after the threshold segmentation; please refer to Figure 2 , which shows a comparison diagram of a pharyngeal grayscale image and a corresponding threshold segmentation image provided by an embodiment of the present invention; Figure 2 The left image in the middle is a grayscale image of the throat, and the right image is a binary image after threshold segmentation. The white area is the foreground area, which may contain saliva reflection areas. However, the saliva reflection area is smaller than the suspected lesion area, so the possibility of misidentification of the suspected lesion area can be reduced. The methods for obtaining the suspected lesion area include:
[0055] Threshold segmentation is performed on each pharyngeal grayscale image to obtain all foreground areas; the foreground areas with a total number of pixels greater than a preset threshold are regarded as suspected lesion areas.
[0056] As an example, each pharyngeal grayscale image is adaptively binarized to obtain a foreground area and a background area, wherein the foreground area is an area with a higher grayscale value, which is more likely to be a suspected lesion area and a saliva reflection area, while the background area is an area with a lower grayscale value, which is more likely to be a normal pharyngeal tissue area; then, among all foreground areas, the foreground area with a total number of pixels greater than a preset threshold is taken as a suspected lesion area, wherein the preset threshold is 20, and interference from the saliva reflection area is excluded.
[0057] It should be noted that adaptive binarization to obtain foreground and background areas is a well-known technology; implementers can also set the segmentation threshold for threshold segmentation according to actual needs, and can also set a preset threshold for the total number of pixels.
[0058] In another embodiment of the present invention, in each pharyngeal grayscale image, the grayscale value and position coordinates of each pixel are used as vector elements to construct a pixel vector, and then the similarity between the vectors is used as a metric distance, and the elbow method is used to obtain a preset K value, and all pixel clusters are obtained based on the K-means clustering algorithm, and the pixels in the same cluster are assigned the same cluster label, and the cluster labels of different clusters are inconsistent, and then the cluster labels are mapped to the pharyngeal grayscale image instead of the pixel grayscale values, and a regional connectivity test is performed on all labeled pixels to obtain all regions, and the area of each region is sorted in descending order; considering that the area of the suspected lesion area is larger than that of the normal tissue The area of the region is small and larger than the area of the saliva reflection area. Therefore, the first-order difference is further performed after descending sorting, and the suspected lesion area is obtained according to the first-order difference result; for example, the descending sorting sequence is {278, 259, 89, 86, 76, 15, 13}, and the corresponding first-order difference result is {19, 170, 3, 10, 61, 2}, among which 170 and 61 are the regional area mutation points. Then, in the regional area descending sorting sequence, 278 and 259 are the corresponding regional areas of the normal tissue area, 89, 86, and 76 are the corresponding regional areas of the suspected lesion area, and 15 and 13 are the corresponding regional areas of the saliva reflection area, thereby screening out all suspected lesion areas.
[0059] When obtaining suspected lesion areas, relying solely on local grayscale information is susceptible to interference, resulting in poor segmentation of pixels at the boundaries of adjacent areas of the suspected lesion area, thereby affecting the subsequent lesion feature analysis effect; at the same time, the grayscale performance and position distribution of each pixel in the throat grayscale image have certain uniqueness, and its performance characteristics in the entire throat grayscale image are unique; therefore, after obtaining all suspected lesion areas in each throat grayscale image, the embodiment of the present invention further obtains the performance parameters of each pixel based on the grayscale information of each pixel and the position distribution relative to the suspected lesion area; the performance parameters quantify the unique performance characteristics of each pixel in the entire throat grayscale image, and the larger the performance parameter, the greater the possibility that it is a pixel in the lesion area.
[0060] Preferably, in one embodiment of the present invention, considering that the deviation of the grayscale of a pixel point can indirectly reflect its unique grayscale characteristics, the position distribution of the pixel point relative to the adjacent suspected lesion area can indirectly reflect the unique position characteristics of the pixel point; therefore, the method for obtaining the performance parameter includes:
[0061] In each pharyngeal grayscale image, the grayscale coefficient of each pixel is obtained according to the deviation of the grayscale value of each pixel relative to the grayscale values of all pixels;
[0062] Taking any pixel as the target pixel, the position coefficient of the target pixel is obtained according to the minimum Euclidean distance between the target pixel and all suspected lesion areas; the minimum Euclidean distance is negatively correlated with the position coefficient;
[0063] The performance parameters of the corresponding pixel points are obtained according to the grayscale coefficient and the position coefficient, and both the grayscale coefficient and the position coefficient are positively correlated with the performance parameters.
[0064] As an example, the performance parameter is calculated as: ;in, For each throat grayscale image The performance coefficient of each pixel; is the standard normalization function; For each throat grayscale image The gray value of each pixel; is the mode of the grayscale values of all pixels in each pharyngeal grayscale image; is the absolute value symbol; For each throat grayscale image The grayscale coefficient of each pixel; For each throat grayscale image The minimum Euclidean distance between a pixel and the pixel in the nearest suspected lesion area; is a natural constant.
[0065] It should be noted that the minimum Euclidean distance is the minimum Euclidean distance between a pixel point and a pixel point in a suspected lesion area. The specific method for obtaining the minimum Euclidean distance is as follows: the Euclidean distance between the target pixel point and each pixel point in each suspected lesion area is calculated, and then the minimum Euclidean distance is selected from all pixel points in all suspected lesion areas.
[0066] It should be noted that the normalization functions used in the embodiments of the present invention are all linear normalizations, and the same applies to the subsequent steps. The implementer may also select other normalization functions according to actual needs, which will not be elaborated here.
[0067] In the performance parameter calculation formula, the deviation of the gray value is evaluated according to the absolute value of the difference between the gray value and the gray value mode to obtain the gray coefficient; the minimum Euclidean distance is used as an exponential function with the natural constant e as the base. In Negative correlation mapping is performed. The larger the distance, the closer it is to the suspected lesion area. The larger the position coefficient, the greater the possibility that it is the corresponding pixel point in the lesion area, and the greater the distinguishing feature from the normal tissue area. Then, after multiplying the grayscale coefficient and the position coefficient, the product is linearly normalized to obtain the performance parameters of the pixel point.
[0068] In other examples, the minimum Euclidean distance may be negatively mapped using other negative correlation mapping methods such as inverse operation, and then the grayscale coefficient and the position parameter may be further combined by addition or weighted summation. The implementer may also directly use the Euclidean distance of each pixel point to the geometric center of the nearest suspected lesion area to approximate the position coefficient, which will not be elaborated here.
[0069] Considering that the larger the performance coefficient is, the greater the possibility that the pixel point corresponds to the lesion area, the embodiment of the present invention further obtains all lesion areas in each throat grayscale image according to the performance parameter. Determining the final lesion area by the performance parameter can make up for the poor effect of dividing the area boundary pixels caused by dividing the area only according to the grayscale value, and obtain a more accurate lesion area.
[0070] Preferably, in one embodiment of the present invention, the method for acquiring the lesion area includes:
[0071] The pixels whose performance parameters are greater than the preset parameter threshold are regarded as lesion pixels, and all lesion areas in each throat grayscale image are obtained according to the aggregation of all lesion pixels.
[0072] The preset parameter threshold is 0.6, which can be adjusted by the implementer; all lesion pixels are clustered based on their position coordinates in the grayscale image of the throat, and can be clustered based on the K-means algorithm, wherein the number of clusters, i.e., the preset K value, is obtained based on the elbow method, or the preset K value is directly set to the total number of suspected lesion areas; then all clusters are subjected to regional connectivity detection to obtain all lesion areas; in other embodiments, the implementer can also directly perform regional connectivity detection on all lesion pixels to obtain all lesion areas. It should be noted that the elbow method, K-means algorithm and regional connectivity detection algorithm are all well-known technologies and will not be described in detail.
[0073] Considering that the severity of pharyngitis symptoms in different patients is not consistent, classifying the throat grayscale images of patients with similar severity into one category is helpful for quickly and centrally analyzing the characteristics of the patient's symptoms, thereby facilitating the diagnosis and treatment of relevant medical staff; considering that the difference in symptom characteristics is mainly reflected in the color and morphology of the lesions, and the difference in the performance parameters of the pixel points can further help evaluate the difference between the lesion characteristics, thereby better measuring the similarity of symptoms between different throat grayscale images; therefore, the embodiment of the present invention obtains the pathological similarity coefficient between different throat grayscale images based on the grayscale difference and morphological feature difference between the lesion areas in different throat grayscale images, combined with the difference in pixel performance parameters between the lesion areas.
[0074] For easier understanding, see Figure 3 , which shows a schematic diagram of comparison of pharynx grayscale images of mild pharyngitis and severe pharyngitis provided by an embodiment of the present invention; Figure 3 The left picture in the figure is a grayscale image of the throat with mild pharyngitis, and the right picture is a grayscale image of the throat with severe pharyngitis.
[0075] Preferably, in one embodiment of the present invention, the method for obtaining the pathological similarity coefficient includes:
[0076] See also Figure 4 , which shows a flow chart of a method for obtaining a pathological similarity coefficient provided by an embodiment of the present invention, specifically comprising:
[0077] Step S401, in each pharynx grayscale image, the pathological grayscale parameters of each pharynx grayscale image are obtained according to the grayscale concentration features of the pixels in all the lesion areas.
[0078] Considering that the grayscale of the corresponding lesion area of patients with pharyngitis of different severity is quite different, for example, it usually appears milky white when the inflammation is mild, and usually light yellow when the inflammation is severe; therefore, the pathological grayscale parameters of each throat grayscale image can be obtained based on the grayscale information; the pathological grayscale parameters quantify the severity of the disease corresponding to the throat grayscale image based on the color information, so as to subsequently evaluate the pathological similarity of different throat grayscale images.
[0079] As an example, in each throat grayscale image, the mean grayscale value of all pixels corresponding to all lesion areas is used as the pathological grayscale parameter of each throat grayscale image; in other examples, implementers may also use concentrated feature measurement parameters such as the mode of grayscale values instead of the mean to obtain the pathological grayscale parameter.
[0080] Step S402, obtaining the pathological morphological parameters of each pharynx grayscale image according to the boundary rules and area of all lesion regions in each pharynx grayscale image.
[0081] Considering that the morphological characteristics of the corresponding lesion areas of patients with pharyngitis of different severity are quite different, for example, when the inflammation is mild, it is usually an area with fuzzy and irregular boundaries and a small area, while when the inflammation is severe, it is usually manifested as a clear and rounded boundary and a large area; therefore, the pathological morphological parameters of each throat grayscale image can be obtained according to the boundary regularity and regional area of all lesion areas; the pathological morphological parameters quantify the severity of the disease corresponding to the throat grayscale image based on the regional morphological characteristics, so as to subsequently evaluate the pathological similarity of different throat grayscale images.
[0082] In a preferred embodiment of the present invention, considering the change in the slope of the line between adjacent pixel points on the regional boundary of the lesion area, the change in the contour of the region is reflected to a certain extent, and the regional area of the lesion area reflects the severity of the disease, and then the two can be combined to evaluate the pathological morphological parameters; the method for obtaining the pathological morphological parameters includes:
[0083] According to the change of the slope of the line between adjacent pixel points on the regional boundary of each lesion area, the contour expression parameters of each lesion area are obtained; according to the concentrated features of the contour expression parameters of all lesion areas in each pharyngeal grayscale image, the boundary feature parameters of each pharyngeal grayscale image are obtained;
[0084] In each pharyngeal grayscale image, the severity of the disease in each pharyngeal grayscale image is obtained according to the area of all lesion areas; the area is positively correlated with the severity of the disease;
[0085] According to the boundary feature parameters and the severity of the disease, the pathological morphological parameters of the corresponding throat grayscale image are obtained; the boundary feature parameters and the severity of the disease are positively correlated with the pathological morphological parameters.
[0086] As an example, the boundary characteristic parameter is calculated as: ;in, For each throat grayscale image The contour performance parameters of the lesion area; is the serial number of the lesion area in each pharyngeal grayscale image; is the total number of lesion areas in each pharyngeal grayscale image; For each throat grayscale image The serial number of the pixel points on the boundary of the lesion area; For each throat grayscale image The total number of pixels on the boundary of the lesion area; For each throat grayscale image On the regional boundary of the lesion area, Pixels and The slope of the line connecting the pixels; For each throat grayscale image On the regional boundary of the lesion area, Pixels and The slope of the line connecting the pixels; For each throat grayscale image The contour expression parameters of the lesion area.
[0087] It should be noted that the pixel points on the regional boundary of the lesion area are the outermost pixel points of the lesion area; the slope is specifically calculated using a two-point formula, which is a well-known technology and will not be described in detail here.
[0088] This example further uses the cumulative sum of the area of all lesion regions in each throat grayscale image as the disease severity of each throat grayscale image, and further multiplies the disease severity with the boundary feature parameters to obtain the pathological morphological parameters of each throat grayscale image.
[0089] In other examples, implementers may also combine the severity of the disease and the boundary feature parameters through other basic mathematical operations such as addition or weighted summation and related mapping means, which will not be described in detail here.
[0090] Step S403, obtaining the pathological performance parameters of each pharynx grayscale image according to the concentrated features of the performance parameters of all pixels in the lesion area in each pharynx grayscale image.
[0091] As an example, the mean of the performance parameters of all pixels in the lesion area in each pharyngeal grayscale image is used as the pathological performance parameter of each pharyngeal grayscale image. In other examples, the implementer may also use a concentrated feature measurement parameter such as mode instead of the mean to obtain the pathological performance parameter, which will not be described in detail here.
[0092] Step S404, obtaining the pathological similarity coefficient between different throat grayscale images according to the differences in pathological grayscale parameters, pathological morphological parameters and pathological manifestation parameters between different throat grayscale images; the differences in pathological grayscale parameters, pathological morphological parameters and pathological manifestation parameters are all negatively correlated with the pathological similarity coefficient.
[0093] As an example, the calculation formula of the pathological similarity coefficient is:
[0094] ;in, For the The grayscale image of the throat and the Case similarity coefficient between pharyngeal grayscale images; is the standard normalization function; For the Pathological grayscale parameters of the pharyngeal grayscale image; For the Pathological grayscale parameters of the pharyngeal grayscale image; For the Pathological morphological parameters of the pharyngeal grayscale image; For the Pathological morphological parameters of the pharyngeal grayscale image; It is a very small positive parameter, which is set to 0.01 in this example to prevent the denominator from being 0; is an exponential function with the natural constant e as base; For the Pathological parameters of the pharyngeal grayscale image; For the Pathological parameters of the pharyngeal grayscale image; is the absolute value symbol.
[0095] In other examples, implementers may also use other basic mathematical operations or related mapping methods to combine the differences in the three pathological parameters between different throat grayscale images, which will not be described in detail here.
[0096] After obtaining the case similarity parameters, all throat grayscale images can be preliminarily classified according to the pathological similarity coefficient to obtain the initial categories of all images; during the preliminary classification process, the image classification effect is not good due to the fuzzy local features of some throat grayscale images, but the throat grayscale images are roughly divided into multiple initial categories, thereby reducing the complexity of subsequent analysis and calculation, and facilitating more detailed enhancement of the throat grayscale images in each initial category, helping to improve classification efficiency and accuracy.
[0097] Preferably, in one embodiment of the present invention, according to the pathological similarity coefficient, the method for classifying all the pharyngeal grayscale images includes: using the pathological similarity coefficient as a metric distance, clustering all the pharyngeal grayscale images based on a distance clustering algorithm, and using the clustering result as the classification result of the pharyngeal grayscale image. Specifically, clustering can be performed based on a K-means algorithm, and the preset K value is obtained based on the elbow method, so as to perform clustering to obtain the classification results of all the pharyngeal grayscale images.
[0098] It should be noted that the classification result obtained based on the above preset classification method is the preliminary classification result of all throat grayscale images, thereby obtaining the initial categories of all images.
[0099] Step S3, select reference throat grayscale images from each image initial category according to the image texture information; obtain the enhancement factor of each lesion area in each throat grayscale image according to the performance parameters of all pixels in each lesion area in each throat grayscale image, and the grayscale difference inside and outside the area, combined with the pathological similarity coefficient between the throat grayscale image and the corresponding reference throat grayscale image; perform grayscale enhancement on the pixels in the lesion area in the corresponding throat grayscale image according to the enhancement factor.
[0100] Considering that there is at least one throat grayscale image in each image initial category that can relatively clearly characterize the characteristics of throat diseases; and considering that the image texture information can reflect the clarity of the image; therefore, the embodiment of the present invention filters out the reference throat grayscale image in each image initial category according to the image texture information.
[0101] Preferably, in one embodiment of the present invention, considering that more gray levels mean richer details and layers of the image, the method for obtaining the reference pharynx grayscale image includes: in each initial image category, obtaining the number of gray levels of each pharynx grayscale image, and using the pharynx grayscale image with the largest number of gray levels as the reference pharynx grayscale image in the initial image category. The acquisition of the number of gray levels is a well-known technique and will not be described in detail here.
[0102] In other embodiments of the present invention, the implementer may also use other means to select the clearest and most texture-rich throat grayscale image as a reference throat grayscale image; for example, the gradient variance of all pixels of each throat grayscale image may be used to evaluate the texture richness, or the image may be converted to the frequency domain for analysis and evaluation of the texture richness, thereby selecting an initial reference throat grayscale image. These are all existing technologies and will not be described in detail here.
[0103] Considering that the lower the pathological similarity between the throat grayscale image and the reference throat grayscale image in the initial category of the corresponding image, the more blurred the corresponding throat grayscale image is, and the greater the possibility that its disease characteristics are blurred, appropriate enhancement should be performed to highlight the disease characteristics; considering that the throat grayscale image has different blur levels, the enhancement levels of the lesion area under different blur conditions should be different to ensure that the disease characteristics of the lesion area in the throat grayscale image are not distorted; the smaller the grayscale difference inside and outside the lesion area, the smaller the expression coefficient of the pixel points in the lesion area, the more blurred the disease characteristics, and the enhancement level should be relatively small, otherwise it should be appropriately increased to avoid feature distortion while enhancing.
[0104] Therefore, the embodiment of the present invention obtains the enhancement factor of each lesion area in each throat grayscale image based on the performance parameters of all pixels in each lesion area in each throat grayscale image, and the grayscale difference inside and outside the area, combined with the pathological similarity coefficient between the throat grayscale image and the corresponding reference throat grayscale image.
[0105] Preferably, in one embodiment of the present invention, the method for obtaining the enhancement factor includes:
[0106] See also Figure 5 , which shows a flow chart of a method for obtaining an enhancement factor provided by an embodiment of the present invention, specifically comprising:
[0107] Step S501, taking the grayscale mean of all pixels in each lesion area as the first grayscale; expanding each lesion area outward by a preset number of circles, taking the grayscale mean of all pixels in the expanded area as the second grayscale; taking the difference between the first grayscale and the second grayscale as the first enhancement parameter.
[0108] As an example, the preset number of circles is 1, that is, a circle of pixels outside the lesion area is used as the expansion area, and the difference between the first grayscale and the second grayscale is used as the first enhancement parameter. The smaller the first enhancement parameter is, the smaller the grayscale difference between the pixels inside and outside the lesion area is, and the greater the degree of blurring of the disease characteristics of the lesion area. On the basis of the original lesion area, the enhancement degree should be appropriately smaller to avoid distortion after enhancement; conversely, the larger the grayscale difference is, the greater the enhancement degree can be appropriately increased.
[0109] In other examples, implementers may also set the preset number of circles to expand outward, but it should not be too large to avoid reducing the approximate evaluation effect of the blurriness of the lesion area, thereby affecting the enhancement effect.
[0110] Step S502: acquiring a second enhancement parameter of each lesion region according to the concentrated features of the performance parameters of all pixels in each lesion region.
[0111] As an example, the mean of the performance parameters of all pixels in each lesion area is used as the second enhancement parameter of each lesion area. The larger the second enhancement parameter is, the lower the blurring degree of the pixels in the lesion area is, and the corresponding enhancement degree is appropriately larger. In other examples, implementers may also use other methods such as concentrated feature measurement parameters to obtain the second enhancement parameter.
[0112] Step S503, according to the first enhancement parameter and the second enhancement parameter of each lesion area in each throat grayscale image, combined with the pathological similarity coefficient between each throat grayscale image and the reference throat grayscale image in the initial category of the image to which it belongs, obtain the enhancement factor of each lesion area in each throat grayscale image; the first enhancement parameter and the second enhancement parameter are both positively correlated with the enhancement factor, and the pathological similarity coefficient is negatively correlated with the enhancement factor.
[0113] As an example, the enhancement factor is calculated as: ;in, For the The first image in the initial category In the grayscale image of the throat Enhancement factor of each lesion area; For the The first image in the initial category In the grayscale image of the throat The grayscale average of all pixels in the lesion area is the first grayscale; For the The first image in the initial category In the grayscale image of the throat The average grayscale value of all pixels in the expansion area outside the lesion area is the second grayscale; For the The first image in the initial category In the grayscale image of the throat The first enhancement parameter of the lesion area; For the The first image in the initial category In the grayscale image of the throat The second enhancement parameter of the lesion area; For the The first image in the initial category Pathological similarity coefficient between the pharyngeal grayscale image and the corresponding reference pharyngeal grayscale image; is the standard normalization function.
[0114] In this example, the first enhancement parameter and the second enhancement parameter are multiplied and combined, and then the pathological similarity coefficient is reciprocally operated for negative correlation mapping, and the product is multiplied and combined with the negative correlation mapping result to obtain the enhancement factor; when the pathological similarity coefficient is smaller, the image blur is greater and the enhancement is greater; on this basis, the enhancement factor is determined in combination with the first enhancement parameter and the second enhancement parameter. The larger the first enhancement parameter and the second enhancement parameter, the greater the enhancement can be. On the contrary, the smaller the first enhancement parameter and the second enhancement parameter, the smaller the enhancement is, so as to avoid distortion caused by excessive enhancement. In other examples, the implementer may also use other basic mathematical operations or related mapping methods to combine the three, or other normalization methods, which will not be repeated.
[0115] After obtaining the enhancement factor of each lesion area, the pixel points in the lesion area in the corresponding throat grayscale image can be adaptively enhanced in grayscale according to the enhancement factor, thereby improving the subsequent classification effect.
[0116] Preferably, in one embodiment of the present invention, the method for performing grayscale enhancement on the pixel points in the lesion area in the corresponding throat grayscale image according to the enhancement factor comprises:
[0117] The sum of the normalized enhancement factor and the constant 1 is used as the grayscale enhancement weight, and the grayscale enhancement weight is used to weight the grayscale value of each pixel in each lesion area in each throat grayscale image, and the weighted result is used as the enhanced grayscale value of the corresponding pixel. Specifically, the constant 1 is used as the adjustment basis, and the enhancement factor is used as the upward adjustment amplitude parameter to obtain the grayscale enhancement weight, and the grayscale enhancement weight is multiplied by the corresponding grayscale value to obtain the enhanced grayscale value, thereby realizing grayscale enhancement based on the grayscale value of the pixel in each lesion area, highlighting the symptom characteristics of the lesion area.
[0118] It should be noted that, for the reference throat grayscale image in each initial category of the image, the grayscale enhancement method is also used for enhancement, except that when calculating the enhancement factor, the pathological similarity coefficient is set to 1 to adaptively enhance all throat grayscale images.
[0119] Step S4, classify all enhanced throat grayscale images using a preset classification method to obtain the final categories of all images.
[0120] After all enhanced pharyngeal grayscale images are obtained, further refined classification can be performed to improve the accuracy of the classification results. In an embodiment of the present invention, the preset classification method described in step S2 is repeated to classify all enhanced pharyngeal grayscale images to obtain the final category of all images; the final category of the image is the final intelligent classification result of the image.
[0121] In summary, the present invention first obtains all pharyngeal grayscale images, and obtains the initial categories of all images using a preset classification method; wherein the preset classification method includes: evaluating the symptom differences between the lesion areas in different pharyngeal grayscale images, obtaining the pathological similarity coefficients between different pharyngeal grayscale images, and then classifying according to the pathological similarity coefficients to obtain the initial categories of all images; then in each image initial category, according to the texture information of the pharyngeal grayscale image, combined with the comparison of the reference pharyngeal grayscale image screened out, the lesion blur degree is measured, and the enhancement factor of each lesion area therein is correspondingly obtained, so as to grayscale enhance the lesion area; then all enhanced pharyngeal grayscale images are classified using a preset classification method to obtain the final categories of all images. The present invention first performs a preliminary classification of images based on the pathological similarity of the lesion areas in different pharyngeal grayscale images, reduces the complexity of subsequent analysis to perform more refined image enhancement, and then measures the blur degree according to the texture information of the pharyngeal grayscale image to adaptively enhance, highlight the symptom characteristics under the premise of avoiding distortion, and finally refines the classification of the enhanced pharyngeal grayscale image, thereby improving the accuracy of the intelligent classification results of the ENT lesion images.
[0122] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0123] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
Claims
1. An intelligent classification method for ENT lesion images, characterized in that: The method comprises: Obtain a grayscale image of the throat of each pharyngitis patient; All pharyngeal grayscale images are classified using a preset classification method to obtain initial categories of all images; the preset classification method includes: in each pharyngeal grayscale image, all suspected lesion areas are obtained, and the performance parameters of each pixel are obtained according to the grayscale information of each pixel and the position distribution relative to the suspected lesion area; all lesion areas in each pharyngeal grayscale image are obtained according to the performance parameters; according to the grayscale differences and morphological feature differences between the lesion areas in different pharyngeal grayscale images, combined with the differences in the performance parameters of the pixels between the lesion areas, the pathological similarity coefficients between different pharyngeal grayscale images are obtained; according to the pathological similarity coefficients, all pharyngeal grayscale images are classified; A reference throat grayscale image is selected from each of the initial categories of the image according to the image texture information; an enhancement factor of each of the lesion areas in each throat grayscale image is obtained according to the performance parameters of all pixels in each lesion area in each throat grayscale image and the grayscale difference inside and outside the area, combined with the pathological similarity coefficient between the throat grayscale image and the corresponding reference throat grayscale image; and the grayscale of the pixels in the lesion area in the corresponding throat grayscale image is enhanced according to the enhancement factor; Classify all enhanced throat grayscale images using a preset classification method to obtain the final categories of all images; The method for obtaining the performance parameters includes: In each pharyngeal grayscale image, the grayscale coefficient of each pixel is obtained according to the deviation of the grayscale value of each pixel relative to the grayscale values of all pixels; Taking any pixel point as a target pixel point, obtaining a position coefficient of the target pixel point according to the minimum Euclidean distance between the target pixel point and all the suspected lesion areas; the minimum Euclidean distance is negatively correlated with the position coefficient; The performance parameter of the corresponding pixel point is obtained according to the grayscale coefficient and the position coefficient, and both the grayscale coefficient and the position coefficient are positively correlated with the performance parameter.
2. According to the method for intelligent classification of ENT lesion images described in claim 1, it is characterized in that: The method for obtaining the lesion area comprises: The pixel points whose performance parameters are greater than the preset parameter threshold are regarded as lesion pixels, and all lesion areas in each throat grayscale image are obtained according to the aggregation of all the lesion pixels.
3. According to the method for intelligent classification of ENT lesion images described in claim 1, it is characterized in that: The method for obtaining the pathological similarity coefficient includes: In each pharyngeal grayscale image, according to the grayscale concentration features of all the pixels in the lesion area, the pathological grayscale parameters of each pharyngeal grayscale image are obtained; Obtaining pathological morphological parameters of each pharyngeal grayscale image according to the boundary rules and area of all the lesion areas in each pharyngeal grayscale image; Obtaining pathological performance parameters of each pharynx grayscale image according to the concentrated features of the performance parameters of all pixel points in the lesion area in each pharynx grayscale image; According to the differences in the pathological grayscale parameters, the differences in the pathological morphological parameters and the differences in the pathological manifestation parameters between different throat grayscale images, the pathological similarity coefficients between different throat grayscale images are obtained; the differences in the pathological grayscale parameters, the differences in the pathological morphological parameters and the differences in the pathological manifestation parameters are all negatively correlated with the pathological similarity coefficient.
4. According to the method for intelligent classification of ENT lesion images described in claim 3, it is characterized in that: The method for obtaining the pathological morphological parameters includes: According to the change of the slope of the line between adjacent pixel points on the region boundary of each lesion region, the contour expression parameters of each lesion region are obtained; according to the concentrated features of the contour expression parameters of all lesion regions in each pharyngeal grayscale image, the boundary feature parameters of each pharyngeal grayscale image are obtained; In each pharyngeal grayscale image, the severity of the disease in each pharyngeal grayscale image is obtained according to the area of all the lesion areas; the area of the area is positively correlated with the severity of the disease; According to the boundary feature parameters and the severity of the disease, the pathological morphological parameters of the corresponding throat grayscale image are obtained; the boundary feature parameters and the severity of the disease are both positively correlated with the pathological morphological parameters.
5. According to the method for intelligent classification of ENT lesion images described in claim 1, it is characterized in that: The method for classifying all throat grayscale images according to the pathological similarity coefficient comprises: The pathological similarity coefficient is used as a metric distance, and all throat grayscale images are clustered based on a distance clustering algorithm, and the clustering result is used as a classification result of the throat grayscale image.
6. The method for intelligent classification of ENT lesion images according to claim 1, characterized in that: The method for obtaining the reference throat grayscale image comprises: In each of the initial image categories, the number of gray levels of each throat gray image is obtained, and the throat gray image with the largest number of gray levels is used as a reference throat gray image in the initial image category.
7. The method for intelligent classification of ENT lesion images according to claim 1, characterized in that: The method for obtaining the enhancement factor comprises: The grayscale mean of all pixels in each lesion area is used as the first grayscale; each lesion area is expanded outward by a preset number of circles, and the grayscale mean of all pixels in the expanded area is used as the second grayscale; the difference between the first grayscale and the second grayscale is used as the first enhancement parameter; Acquire a second enhancement parameter of each lesion area according to the concentrated features of the performance parameters of all pixels in each lesion area; According to the first enhancement parameter and the second enhancement parameter of each lesion area in each throat grayscale image, combined with the pathological similarity coefficient between each throat grayscale image and the reference throat grayscale image in the initial image category to which it belongs, an enhancement factor for each lesion area in each throat grayscale image is obtained; the first enhancement parameter and the second enhancement parameter are both positively correlated with the enhancement factor, and the pathological similarity coefficient is negatively correlated with the enhancement factor.
8. The method for intelligent classification of ENT lesion images according to claim 1, characterized in that: The method for performing grayscale enhancement on the pixel points in the lesion area in the corresponding throat grayscale image according to the enhancement factor comprises: The sum of the normalized enhancement factor and the constant 1 is used as the grayscale enhancement weight, and the grayscale value of each pixel in each lesion area in each throat grayscale image is weighted using the grayscale enhancement weight, and the weighted result is used as the enhanced grayscale value of the corresponding pixel.
9. The method for intelligent classification of ENT lesion images according to claim 1, characterized in that: The method for obtaining the suspected lesion area includes: Threshold segmentation is performed on each pharyngeal grayscale image to obtain all foreground areas; the foreground areas with a total number of pixels greater than a preset threshold are regarded as suspected lesion areas.
