A method for extracting lesion regions from lung cancer images based on region segmentation

Abstract

The present invention relates to the field of image processing technology, and in particular to a method for extracting lung cancer image lesion regions based on regional segmentation, comprising: obtaining a grayscale image of a patient's lungs; obtaining a morphological lesion coefficient of each tissue connected domain; obtaining atomization features of each local extreme point; obtaining atomization isolation coefficient of the grayscale image; filtering the grayscale image according to the atomization isolation coefficient to obtain a restored image; obtaining an interference factor of each tissue connected domain; obtaining the lesion probability of each tissue connected domain according to the morphological lesion coefficient and the interference factor of each tissue connected domain; and extracting lung cancer lesion regions in the grayscale image according to the lesion probability. The present invention improves the accuracy of extracting lung cancer lesion regions.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a method for extracting lung cancer image lesion regions based on region segmentation. Background Art

[0002] When the patient's lung image is segmented and the lung cancer lesion area is extracted by using the existing technology, there is a large amount of mist secretions in the patient's lung image, which are widely distributed and attached in the lung tissue. Since the mist secretions are similar in basic morphology to the lung cancer lesion area, it is difficult to distinguish between the mist secretions and the lung cancer lesion area, which interferes with the extraction of the lung cancer lesion area and makes it impossible to accurately extract the lung cancer lesion area. Summary of the invention

[0003] In order to solve the technical problem in the prior art that the mist secretions in the non-lesion area and the lung cancer lesion area are difficult to distinguish, which leads to interference in the extraction of the lung cancer lesion area, the purpose of the present invention is to provide a lung cancer image lesion area extraction method based on region segmentation, and the technical scheme adopted is as follows:

[0004] The present invention provides a method for extracting lung cancer image lesion regions based on region segmentation, the method comprising:

[0005] Acquire a grayscale image of the patient's lungs; the grayscale image contains a plurality of tissue connected domains;

[0006] According to the position distribution of edge pixels in each tissue connected domain and the grayscale difference between the edge pixels and the surrounding area, the morphological pathology coefficient of each tissue connected domain is obtained;

[0007] According to the gray value distribution of the local position in the gray image, all local extreme points are obtained; according to the gray distribution and position distribution between each local extreme point and the adjacent local extreme points, the fog feature of each local extreme point is obtained; according to the occurrence probability of different types of fog features in the gray image, the fog isolation coefficient of the gray image is obtained; according to the fog isolation coefficient, the gray image is filtered to obtain the restored image; according to the difference in gray distribution of each tissue connected domain in the gray image and the restored image, the interference factor of each tissue connected domain is obtained; according to the morphological pathology coefficient and interference factor of each tissue connected domain, the pathology probability of each tissue connected domain is obtained;

[0008] The lung cancer lesion area in the grayscale image is extracted according to the lesion probability.

[0009] Preferably, the morphological pathology coefficient of each tissue connected domain is obtained according to the position distribution of edge pixels in each tissue connected domain and the grayscale difference between the edge pixels and the surrounding area, including the specific method of:

[0010] According to the position distribution of edge pixel points in each tissue connected domain, the unsmooth burr coefficient of each tissue connected domain is obtained;

[0011] According to the grayscale difference between each tissue connected domain and the surrounding area, the ground glass coefficient of each tissue connected domain is obtained.

[0012] The product of the unsmooth burr coefficient of each tissue connected domain and the ground glass coefficient of each tissue connected domain is used as the morphological lesion factor of each tissue connected domain;

[0013] The morphological lesion factors of all tissue connected domains in the grayscale image are obtained; all morphological lesion factors are linearly normalized, and each normalized morphological lesion factor is recorded as a morphological lesion coefficient.

[0014] Preferably, the method of obtaining the unsmooth burr coefficient of each tissue connected domain according to the position distribution of edge pixels in each tissue connected domain includes the following specific methods:

[0015] The Euclidean distance between each edge pixel point in the tissue connected domain and the centroid of the tissue connected domain is recorded as the first distance of each edge pixel point; the variance of the first distance of all edge pixels in the tissue connected domain is used as the unsmooth burr coefficient of the tissue connected domain.

[0016] Preferably, the method of obtaining the ground glass coefficient of each tissue connected domain according to the grayscale difference between each tissue connected domain and the surrounding area includes:

[0017] The area formed by the maximum circumscribed circle of the tissue connected domain is recorded as the morphological range of the tissue connected domain; the area within the morphological range of the tissue connected domain except the tissue connected domain is recorded as the external area of ​​the tissue connected domain; the area in the grayscale image except all tissue connected domains is recorded as the lung background area of ​​the grayscale image; the mean of the grayscale values ​​of all pixels in the external area of ​​the tissue connected domain is recorded as the first mean; the mean of the grayscale values ​​of all pixels in the lung background area of ​​the grayscale image is recorded as the second mean; the difference between the first mean and the second mean is taken as the ground glass coefficient of the tissue connected domain.

[0018] Preferably, the method of obtaining the atomization feature of each local extreme point according to the grayscale distribution and position distribution between each local extreme point and adjacent local extreme points includes the following specific methods:

[0019] According to the Euclidean distance and grayscale distribution between each local extreme point and its adjacent local extreme point, the target local extreme point and the atomization characteristic grayscale line of each local extreme point are obtained;

[0020] The mean of the grayscale values ​​of all pixels on the grayscale line of the fog feature of each local extreme point is recorded as the third mean; the mean of the grayscale values ​​of each local extreme point and its target local extreme point is recorded as the fourth mean; the product of the ratio of the third mean to the fourth mean and the Euclidean distance between each local extreme point and its target local extreme point is taken as the fog feature of each local extreme point.

[0021] Preferably, the specific method of obtaining the target local extreme point and the atomization characteristic grayscale line of each local extreme point includes:

[0022] For any local extreme point in the grayscale image of the patient's lungs, the Euclidean distance between the local extreme point and all other local extreme points is obtained, and the local extreme point with the smallest Euclidean distance is used as the target local extreme point of the local extreme point; the straight line formed between the local extreme point and its target local extreme point is recorded as the atomization characteristic grayscale line of the local extreme point.

[0023] Preferably, the method of obtaining the fog isolation coefficient of the grayscale image according to the occurrence probabilities of different types of fog features in the grayscale image includes the following specific methods:

[0024] All the fog-like features in the grayscale image are sorted in ascending order according to the probability of occurrence, and all the sorted fog-like features are used as the fog feature sequence of the grayscale image;

[0025] For any type of fog feature in the fog feature sequence of the grayscale image;

[0026] Obtaining a relative non-lesion atomization feature and a relative lesion atomization feature of the atomization feature;

[0027] The ratio of the variance of the occurrence probabilities of all relative non-lesion atomization features of this type of atomization features to the mean of the occurrence probabilities of all relative non-lesion atomization features of this type of atomization features is recorded as the first ratio; the inverse proportional value of the information entropy of the occurrence probabilities of all relative lesion atomization features of this type of atomization features in the occurrence probabilities of all atomization features in the atomization feature sequence of the grayscale image is recorded as the first inverse proportional value; the sum of the first ratio and the first inverse proportional value is taken as the objective function value of this type of atomization features;

[0028] The atomization feature with the largest objective function value is recorded as the target atomization feature;

[0029] The ratio of the cumulative sum of the occurrence probabilities of all relative non-lesion atomized features of the target atomized feature to the cumulative sum of the occurrence probabilities of all relative lesion atomized features of the target atomized feature is used as the atomization isolation coefficient of the grayscale image.

[0030] Preferably, the specific method of obtaining the relative non-lesion atomization feature and the relative lesion atomization feature of the atomization feature includes:

[0031] In the fog feature sequence of the grayscale image, all fog-like features before the fog-like feature are recorded as relative non-lesion fog features of the fog-like feature; all fog-like features after the fog-like feature are recorded as relative pathological fog features of the fog-like feature.

[0032] Preferably, the method of obtaining the interference factor of each tissue connected domain according to the grayscale distribution difference between the grayscale image and the restored image of each tissue connected domain includes the following specific methods:

[0033] For any tissue connected domain in the grayscale image, in the restored image, the area formed by the maximum circumscribed circle of the tissue connected domain is used as the interference layer area of ​​the tissue connected domain;

[0034] The ratio of the mean grayscale value of the pixels in the interference layer area of ​​the tissue connected domain to the mean grayscale value of all pixels in the lung background area of ​​the grayscale image is used as the interference factor of the tissue connected domain.

[0035] Preferably, the specific method of obtaining the lesion probability of each tissue connected domain according to the morphological lesion coefficient and the interference factor of each tissue connected domain is as follows:

[0036] The product of the morphological lesion coefficient of the tissue connectivity domain and the interference factor is used as the lesion factor of the tissue connectivity domain;

[0037] Obtain the lesion factors of all tissue connected domains in the grayscale image; perform linear normalization on all lesion factors, and record each normalized lesion factor as the lesion probability.

[0038] The beneficial effects of the technical solution of the present invention are as follows: the present invention obtains the morphological lesion coefficient of each tissue connected domain according to the position distribution of edge pixel points in each tissue connected domain and the grayscale difference between the pixel points and the surrounding area; thereby obtaining the preliminary probability that the tissue connected domain is a lung cancer lesion area through the basic morphological information of the tissue connected domain; obtaining the atomization isolation coefficient of the grayscale image according to the occurrence probability of different types of atomization features in the grayscale image; filtering the grayscale image according to the atomization isolation coefficient to obtain a restored image; thereby distinguishing foggy secretions from lung cancer lesion areas; obtaining the interference factor of each tissue connected domain according to the grayscale distribution difference between the grayscale image and the restored image of each tissue connected domain; obtaining the lesion probability of each tissue connected domain according to the morphological lesion coefficient and the interference factor of each tissue connected domain; thereby correcting the morphological lesion coefficient of the tissue connected domain through the interference factor of the tissue connected domain to obtain the lesion probability of the tissue connected domain; thereby avoiding the interference of foggy secretions in non-lesion areas on lung cancer lesion areas, thereby improving the accuracy of extracting lung cancer lesion areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be 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.

[0040] Figure 1 It is a flowchart of the steps of the method for extracting lung cancer image lesion area based on region segmentation of the present invention;

[0041] Figure 2 The feature relationship flow chart of the lung cancer image lesion region extraction method based on region segmentation of the present invention. DETAILED DESCRIPTION

[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of the lung cancer image lesion area extraction method based on regional segmentation proposed by the present invention 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.

[0043] 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.

[0044] The specific scheme of the method for extracting lung cancer lesion regions based on region segmentation provided by the present invention is described in detail below with reference to the accompanying drawings.

[0045] See also Figure 1 , which shows a flowchart of a method for extracting lung cancer image lesion regions based on region segmentation provided by an embodiment of the present invention, the method comprising the following steps:

[0046] Step S001: Acquire a grayscale image of the patient's lungs.

[0047] It should be noted that this embodiment uses an edge detection algorithm to extract the lung cancer lesion area in the patient's lung image, but the misty secretions in the non-lesion area interfere with the extraction of the lung cancer lesion area, making it impossible to accurately extract the lung cancer lesion area in the patient's lung image.

[0048] Specifically, we first need to collect a grayscale image of the patient's lungs. The specific process is as follows:

[0049] The patient's lung image is acquired through an X-ray detection device, and the patient's lung image is subjected to median filtering, denoising and grayscale operations to obtain a grayscale image of the patient's lung. The median filtering and grayscale operations are prior arts, and are not described in detail in this embodiment.

[0050] So far, a grayscale image of the patient's lungs is obtained through the above method.

[0051] Step S002: Obtain the morphological pathology coefficient of each tissue connected domain according to the position distribution of edge pixels in each tissue connected domain and the grayscale difference between the edge pixels and the surrounding area.

[0052] It should be noted that in order to extract the lung cancer lesion area in the patient's lung image, edge detection can be performed on the grayscale image of the patient's lung to extract several closed areas inside the grayscale image of the patient's lung, wherein some closed areas are formed by normal lung nodules and blood vessels, and some are formed by lung cancer lesion areas; since the edge features of normal lung tissue are strip-shaped blood vessels or circular-shaped normal nodules, and the lung cancer lesion area is an irregular pattern with burrs on the edge, it is necessary to analyze the edge fluctuation degree of each closed area, and since the lung cancer lesion area will form a surrounding ground-glass shadow on the lung image; therefore, in order to extract the lung cancer lesion area in the patient's lung image, it is necessary to use the edge features of the lung cancer lesion area for analysis and screening.

[0053] Specifically, the Canny edge detection algorithm is used to perform edge detection on the grayscale image of the patient's lungs to obtain an edge detection result map of the patient's lungs; each closed area formed by edge pixels in the edge detection result map corresponding to the grayscale image of the patient's lungs is used as a tissue connected domain in the grayscale image of the patient's lungs.

[0054] It should be noted that due to the continuous increase in the tissue density of lung cancer cells, lymphocyte infiltration, lung tissue infection, inflammation, etc. are more likely to occur. Therefore, the lung cancer lesion area is irregular and foggy and has burrs on the edge. There are also ground-glass shadows around the lung cancer lesion area. Therefore, for the tissue connected domain belonging to the lung cancer lesion area, there are more external foggy white spots; then, according to the position distribution of the edge pixel points in each tissue connected domain and the grayscale difference between the surrounding area, the morphological lesion coefficient of each tissue connected domain is obtained; since the larger the unsmooth burr coefficient and ground-glass coefficient of the tissue connected domain, the larger the morphological lesion coefficient of the tissue connected domain, which indicates that the edge of the tissue connected domain has irregular foggy characteristics, then the larger the morphological lesion coefficient of the tissue connected domain.

[0055] Preferably, in some implementations of the embodiments of the present invention, for any tissue connected domain in a grayscale image, a calculation method for obtaining the morphological pathology coefficient of the tissue connected domain is:

[0056] According to the position distribution of edge pixel points in each tissue connected domain, the unsmooth burr coefficient of each tissue connected domain is obtained;

[0057] According to the grayscale difference between each tissue connected domain and the surrounding area, the ground glass coefficient of each tissue connected domain is obtained.

[0058] The product of the unsmooth burr coefficient of each tissue connected domain and the ground glass coefficient of each tissue connected domain is used as the morphological lesion factor of each tissue connected domain;

[0059] The morphological lesion factors of all tissue connected domains in the grayscale image are obtained; all morphological lesion factors are linearly normalized, and each normalized morphological lesion factor is recorded as a morphological lesion coefficient.

[0060] The specific formula for obtaining the morphological pathological factor of the tissue connectivity domain is:

[0061] ;

[0062] In the formula, Morphopathic factors representing domains of tissue connectivity; The unsmooth burr coefficient represents the connected domain of the tissue; The ground glass coefficient represents the connected area of ​​the tissue.

[0063] It should be noted that the larger the rough burr coefficient and ground glass coefficient of the tissue connected domain are, the larger the morphological lesion coefficient of the tissue connected domain is, indicating that the edge of the tissue connected domain has irregular fog-like features.

[0064] Preferably, in some implementations of the embodiments of the present invention, since the lung cancer lesion area is irregularly lumpy and has burrs on the edge, the distance between its edge pixel point and its centroid is quite different; therefore, the method for obtaining the unsmooth burr coefficient of each tissue connected domain includes:

[0065] The Euclidean distance between each edge pixel point in the tissue connected domain and the centroid of the tissue connected domain is recorded as the first distance of each edge pixel point; the variance of the first distance of all edge pixels in the tissue connected domain is used as the unsmooth burr coefficient of the tissue connected domain.

[0066] Preferably, in some implementations of the embodiments of the present invention, since there is a ground-glass opacity in the outer area of ​​the lung cancer lesion area, there will be more white pixels with higher grayscale values ​​in the outer area, while the grayscale value of the outer area of ​​the normal lung tissue area is lower; therefore, the method for obtaining the ground-glass coefficient of each tissue connected domain includes:

[0067] The area formed by the maximum circumscribed circle of the tissue connected domain is recorded as the morphological range of the tissue connected domain; the area within the morphological range of the tissue connected domain except the tissue connected domain is recorded as the external area of ​​the tissue connected domain; the area in the grayscale image except all tissue connected domains is recorded as the lung background area of ​​the grayscale image; the mean of the grayscale values ​​of all pixels in the external area of ​​the tissue connected domain is recorded as the first mean; the mean of the grayscale values ​​of all pixels in the lung background area of ​​the grayscale image is recorded as the second mean; the difference between the first mean and the second mean is taken as the ground glass coefficient of the tissue connected domain.

[0068] At this point, the morphological pathological factor of each tissue connectivity domain is obtained through the above method.

[0069] Step S003: Obtain all local extreme points in the grayscale image; obtain the fog feature of each local extreme point according to the grayscale distribution and position distribution between each local extreme point and the adjacent local extreme points; obtain the fog isolation coefficient of the grayscale image according to the occurrence probability of different types of fog features in the grayscale image; filter the grayscale image according to the fog isolation coefficient to obtain a restored image; obtain the interference factor of each tissue connected domain according to the difference in grayscale distribution of each tissue connected domain in the grayscale image and the restored image; obtain the lesion probability of each tissue connected domain according to the morphological lesion coefficient and the interference factor of each tissue connected domain.

[0070] It should be noted that foggy secretions often appear in the grayscale images of the patient's lungs. Therefore, it is too ideal to extract the lung cancer lesion area based only on the basic morphological information of the tissue connectivity domain. It is not very practical and requires further analysis of the foggy secretions. The foggy secretions that are not generated by the lung cancer lesion area are randomly distributed, while the lung cancer lesion area will invade the surrounding tissues, so the distribution of the foggy secretions generated by it has diffusion characteristics. The biggest difference between the two is that the foggy secretions of the secretions are superimposed on the lung tissue in pieces, while the foggy area generated by the lung cancer lesions invading the lung tissue directly changes the image information of the lung tissue. The components are cross-distributed with high brightness and low brightness, so the local extreme point is used to represent the position of the fog secretion in the grayscale image; by analyzing the fog characteristics of the local extreme point, the fog isolation coefficient used to obtain the fog component of the lung cancer lesion area and the fog secretion in the non-lesion area is obtained; the grayscale image is filtered according to the fog isolation coefficient to obtain the restored image; according to the difference in grayscale distribution of each tissue connected domain in the grayscale image and the restored image, the interference factor representing the interference information within the scope of each tissue connected domain is obtained; the larger the morphological lesion coefficient of each tissue connected domain, the more interference information is received within the scope of the tissue connected domain, and the greater the lesion probability of the tissue connected domain.

[0071] 1. Obtain the fog features of all local extreme points in the grayscale image.

[0072] It should be noted that since the atomized components are cross-distributed with high brightness and low brightness, the atomized components are all high-frequency components. However, the reason why the atomized secretions of normal lung tissue secretions appear atomized is simply because the density is different, and the density distribution is random and extremely uneven, so its brightness and darkness changes are relatively more high-frequency; and the reason why the atomized areas produced by lung cancer lesions invading lung tissue appear atomized is simply because they have not been completely eroded, and the distribution is concentrated and diffuse, with low randomness, and the lung cancer lesion area is more likely to produce abnormal secretions and be covered by other atomized secretions, so its brightness and darkness changes are relatively low-frequency; therefore, the local extreme points are used to represent the highlight positions in the grayscale image; and then the atomization characteristics of each local extreme point are obtained according to the grayscale distribution and position distribution between each local extreme point and the adjacent local extreme point.

[0073] Specifically, for any pixel point in the grayscale image of the patient's lungs, if the pixel point is the maximum grayscale value of the pixels within its eight-domain range, the pixel point is recorded as a local extreme point.

[0074] Specifically, as an example, for the grayscale image of the patient's lungs local extreme points, and obtain the The calculation method of the atomization characteristics of local extreme points is:

[0075] According to The Euclidean distance and grayscale distribution between the local extreme point and the adjacent local extreme point are obtained. The target local extreme point of the local extreme point and the atomization characteristic grayscale line;

[0076] The first The mean of the grayscale values ​​of all pixels on the fog feature grayscale line of the local extreme point is recorded as the third mean; The mean of the grayscale values ​​of the local extreme point and its target local extreme point is recorded as the fourth mean; the ratio of the third mean to the fourth mean is calculated by The product of the Euclidean distances between the local extreme point and its target local extreme point is taken as the The atomization characteristics of the local extreme points.

[0077] The specific formula is:

[0078] ;

[0079] In the formula, Indicates Atomization characteristics of local extreme points; Indicates that the The mean gray value of all pixels on the fog feature gray line of the local extreme point; Indicates The mean of the grayscale values ​​of the local extreme points and their target local extreme points; Indicates The Euclidean distance between a local extreme point and its target local extreme point.

[0080] It should be noted that the smaller the ratio of the third mean to the fourth mean, the The smaller the Euclidean distance between the local extreme point and its target local extreme point, the The faster the brightness change between the local extreme point and its target local extreme point, the The smaller the atomization characteristics of the local extreme points are.

[0081] Preferably, in some implementations of the embodiments of the present invention, a method for acquiring a target local extreme point and a fogging characteristic grayscale line of each local extreme point includes:

[0082] For any local extreme point in the grayscale image of the patient's lungs, the Euclidean distance between the local extreme point and all other local extreme points is obtained, and the local extreme point with the smallest Euclidean distance is used as the target local extreme point of the local extreme point; the straight line formed between the local extreme point and its target local extreme point is recorded as the atomization characteristic grayscale line of the local extreme point.

[0083] At this point, the fogging features of all local extreme points in the grayscale image are obtained.

[0084] 2. Get the fog isolation coefficient of the grayscale image.

[0085] It should be noted that by analyzing the atomization characteristics of local extreme points, the atomization isolation coefficient for obtaining the atomization components that divide the lung cancer lesion area and the atomization secretions in the non-lesion area is obtained; since the smaller the atomization characteristics, the faster the change in light and darkness, and the closer the atomization secretions are to the non-lesion area, the objective function value of each type of atomization feature is obtained one by one in the order of the probability of occurrence of the atomization characteristics from small to large, which is used to estimate the isolation node between the atomization components of the lung cancer lesion area and the atomization secretions in the non-lesion area.

[0086] Specifically, all the fog-like features in the grayscale image are sorted from small to large according to the probability of occurrence, and all the sorted fog-like features are used as the fog feature sequence of the grayscale image.

[0087] It should be noted that, according to the occurrence probability of fog features in the grayscale image, the objective function value of each type of fog feature is obtained; according to the objective function value, all fog features are screened to obtain the target fog feature; according to the target fog feature, the fog isolation coefficient of the grayscale image is obtained.

[0088] Obtaining a relative non-lesion atomization feature and a relative lesion atomization feature of the atomization feature;

[0089] The ratio of the variance of the occurrence probabilities of all relatively non-lesion atomization features of this type of atomization features to the mean of the occurrence probabilities of all relatively non-lesion atomization features of this type of atomization features is recorded as the first ratio; the inverse proportional value of the information entropy of the occurrence probabilities of all relatively lesion atomization features of this type of atomization features in the occurrence probabilities of all atomization features in the atomization feature sequence of the grayscale image is recorded as the first inverse proportional value; the sum of the first ratio and the first inverse proportional value is taken as the objective function value of this type of atomization features.

[0090] The specific formula is:

[0091] ;

[0092] In the formula, The objective function value representing the atomization characteristics; represents the variance of the occurrence probability of all atomized features relative to non-lesion atomized features; represents the mean of the occurrence probabilities of all atomization features relative to non-lesion atomization features; The information entropy of the occurrence probabilities of all relative lesion atomization features representing atomization features in the atomization feature sequence of the grayscale image is the information entropy of the occurrence probabilities of all atomization features in the atomization feature sequence of the grayscale image; represents an exponential function with a natural constant as the base, and the embodiment adopts The model shows an inverse proportional relationship. As the input of the model, implementers can choose the inverse proportional function according to actual conditions.

[0093] It should be noted that when all the relative non-lesion atomization features of this type of atomization features are atomization secretions in non-lesion areas, due to the random distribution of atomization secretions in non-lesion areas, the number of atomization feature classes is large and scattered, but the probability distribution of each class is small, so The larger the numerator and the smaller the denominator, the more chaotic the distribution of the probability of occurrence of all relatively non-lesion atomization features of this type of atomization feature is, and the more dispersed the probability is, the more this feature conforms to the mist secretion in the non-lesion area; the smaller the information entropy of the occurrence probability of all relatively lesion atomization features of this type of atomization feature in the atomization feature sequence of the grayscale image is, the The larger it is, the simpler and more regular the atomization feature information of all relative lesions of this type of atomization feature is, that is, the more consistent it is with the atomization components of the lesion area; the larger the objective function value of this type of atomization feature is.

[0094] Specifically, as an example, the calculation method for obtaining the fog isolation coefficient of the grayscale image is:

[0095] The atomization feature with the largest objective function value is recorded as the target atomization feature;

[0096] The ratio of the cumulative sum of the occurrence probabilities of all relative non-lesion atomized features of the target atomized feature to the cumulative sum of the occurrence probabilities of all relative lesion atomized features of the target atomized feature is used as the atomization isolation coefficient of the grayscale image.

[0097] The specific formula is:

[0098] ;

[0099] In the formula, Represents the fog isolation coefficient of the grayscale image; represents the cumulative sum of the occurrence probabilities of all relative non-lesion atomization features of the target atomization feature; Represents the cumulative sum of the occurrence probabilities of all relative lesion atomization features of the target atomization feature.

[0100] Preferably, in some implementations of the embodiments of the present invention, the method for acquiring the relative non-lesion atomization feature and the relative lesion atomization feature of each type of atomization feature includes:

[0101] For any type of haze feature in the haze feature sequence of the grayscale image, in the haze feature sequence of the grayscale image, all the haze-like features before the type of haze feature are recorded as the relative non-lesion haze features of the type of haze feature; and all the haze-like features after the type of haze feature are recorded as the relative pathological haze features of the type of haze feature.

[0102] It should be noted that, because the greater the occurrence probability, the more obvious the atomization degree is, the relative non-lesion atomization characteristics and the relative pathological atomization characteristics of each type of atomization characteristics can be obtained according to the occurrence probability of the atomization characteristics.

[0103] At this point, the fog isolation coefficient of the grayscale image is obtained.

[0104] 3. Obtain the lesion probability of each tissue connectivity domain.

[0105] It should be noted that the grayscale image is filtered according to the fog isolation coefficient of the grayscale image, and the fog components of the lung tissue information and the lung cancer lesion area in the grayscale image are masked to obtain the fog secretion distribution image of the non-lesion area, that is, the restored image; according to the difference in grayscale distribution of each tissue connected domain in the grayscale image and the restored image, the interference factor of each tissue connected domain is obtained; according to the morphological lesion coefficient and the interference factor of each tissue connected domain, the lesion probability of each tissue connected domain is obtained.

[0106] The grayscale image is filtered according to the fog isolation coefficient to obtain a restored image; specifically, the frequency spectrum of the grayscale image is obtained by using the Fourier transform algorithm; the fog isolation coefficient of the grayscale image is used as a cutoff filter of a high-pass filter, and the frequency spectrum of the grayscale image is filtered by using the high-pass filter to obtain a filtered image, which is recorded as the restored image.

[0107] Among them, the Fourier transform algorithm and high-pass filtering are existing technologies, and will not be described in detail in this embodiment.

[0108] According to the difference in grayscale distribution of each tissue connectivity domain in the grayscale image and the restored image, the interference factor of each tissue connectivity domain is obtained; specifically, as an example, the calculation method for obtaining the interference factor of each tissue connectivity domain is:

[0109] For any tissue connected domain in the grayscale image, in the restored image, the area formed by the maximum circumscribed circle of the tissue connected domain is used as the interference layer area of ​​the tissue connected domain;

[0110] The ratio of the mean grayscale value of the pixels in the interference layer area of ​​the tissue connected domain to the mean grayscale value of all pixels in the lung background area of ​​the grayscale image is used as the interference factor of the tissue connected domain.

[0111] The specific formula is:

[0112] ;

[0113] In the formula, The disturbed factor representing the tissue connectivity domain; The mean gray value of the pixels in the interference layer area representing the tissue connectivity domain; Represents the mean grayscale value of all pixels in the lung background area of ​​the grayscale image.

[0114] It should be noted that The smaller the ratio is, the greater the difference between the mean grayscale value of the pixels in the interference layer area of ​​the tissue connectivity domain and the mean grayscale value of all pixels in the lung background area, and the more interference information is received within the scope of the tissue connectivity domain, that is, the greater the interference factor of the tissue connectivity domain.

[0115] According to the morphological lesion coefficient and interference factor of each tissue connectivity domain, the lesion probability of each tissue connectivity domain is obtained.

[0116] Specifically, as an example, for any tissue connected domain in a grayscale image, the calculation method for obtaining the lesion probability of the tissue connected domain is:

[0117] The product of the morphological lesion coefficient of the tissue connectivity domain and the interference factor is used as the lesion factor of the tissue connectivity domain;

[0118] Obtain the lesion factors of all tissue connected domains in the grayscale image; perform linear normalization on all lesion factors, and record each normalized lesion factor as the lesion probability.

[0119] The specific formula for obtaining the pathological factor of the tissue connectivity domain is:

[0120] ;

[0121] In the formula, lesion factors representing domains of tissue connectivity; The disturbed factor representing the tissue connectivity domain; The morphological lesion coefficient represents the tissue connectivity domain.

[0122] It should be noted that the smaller the interference factor is, the greater the interference to the tissue connectivity domain is. Therefore, the lesion probability obtained by multiplying the interference factor by the morphological lesion coefficient should be smaller. Through the lesion probability, the interference of mist secretions in the non-lesion area on the lung cancer lesion area is avoided, making the feature extraction of the lung cancer lesion area more accurate.

[0123] At this point, the lesion probability of each tissue connected domain is obtained through the above method.

[0124] Step S004: extracting the lung cancer lesion area in the grayscale image according to the lesion probability.

[0125] Preset a threshold parameter , wherein this embodiment is based on This example is described as an example, and this embodiment is not specifically limited. It depends on the specific implementation situation.

[0126] Specifically, for any tissue connected domain in the grayscale image, if the lesion probability of the tissue connected domain is greater than or equal to the threshold parameter , the tissue connectivity domain is recorded as the lung cancer lesion area and marked.

[0127] It should be noted that by extracting the lung cancer lesion area in the grayscale image, doctors can be further assisted in making accurate judgments.

[0128] See also Figure 2 , which shows a feature relationship flow chart of a lung cancer image lesion region extraction method based on region segmentation.

[0129] At this point, this embodiment is completed.

[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting lung cancer lesion regions based on region segmentation, characterized in that: The method comprises the following steps: Acquire a grayscale image of the patient's lungs; the grayscale image contains a plurality of tissue connected domains; According to the position distribution of edge pixel points in each tissue connected domain and the grayscale difference between the edge pixel points and the surrounding area, the morphological pathology coefficient of each tissue connected domain is obtained; According to the gray value distribution of the local position in the gray image, all local extreme points are obtained; according to the gray distribution and position distribution between each local extreme point and the adjacent local extreme points, the fog feature of each local extreme point is obtained; according to the occurrence probability of different types of fog features in the gray image, the fog isolation coefficient of the gray image is obtained; according to the fog isolation coefficient, the gray image is filtered to obtain the restored image; according to the difference in gray distribution of each tissue connected domain in the gray image and the restored image, the interference factor of each tissue connected domain is obtained; according to the morphological pathology coefficient and interference factor of each tissue connected domain, the pathology probability of each tissue connected domain is obtained; Extract the lung cancer lesion area in the grayscale image according to the lesion probability; The specific method of obtaining the atomization feature of each local extreme point according to the grayscale distribution and position distribution between each local extreme point and adjacent local extreme points is as follows: According to the Euclidean distance and grayscale distribution between each local extreme point and its adjacent local extreme point, the target local extreme point and the atomization characteristic grayscale line of each local extreme point are obtained; The mean of the grayscale values ​​of all pixels on the grayscale line of the fog feature of each local extreme point is recorded as the third mean; the mean of the grayscale values ​​of each local extreme point and its target local extreme point is recorded as the fourth mean; the product of the ratio of the third mean to the fourth mean and the Euclidean distance between each local extreme point and its target local extreme point is taken as the fog feature of each local extreme point; The specific method of obtaining the target local extreme point and the atomization characteristic grayscale line of each local extreme point includes: For any local extreme point in the grayscale image of the patient's lungs, the Euclidean distance between the local extreme point and all other local extreme points is obtained, and the local extreme point with the smallest Euclidean distance is used as the target local extreme point of the local extreme point; the straight line formed between the local extreme point and its target local extreme point is recorded as the atomization characteristic grayscale line of the local extreme point.

2. The method for extracting lung cancer lesion regions based on region segmentation according to claim 1, characterized in that: The method of obtaining the morphological pathology coefficient of each tissue connected domain according to the position distribution of edge pixels in each tissue connected domain and the grayscale difference between the edge pixels and the surrounding area includes: According to the position distribution of edge pixel points in each tissue connected domain, the unsmooth burr coefficient of each tissue connected domain is obtained; According to the grayscale difference between each tissue connected domain and the surrounding area, the ground glass coefficient of each tissue connected domain is obtained; The product of the unsmooth burr coefficient of each tissue connected domain and the ground glass coefficient of each tissue connected domain is used as the morphological lesion factor of each tissue connected domain; The morphological lesion factors of all tissue connected domains in the grayscale image are obtained; all morphological lesion factors are linearly normalized, and each normalized morphological lesion factor is recorded as a morphological lesion coefficient.

3. The method for extracting lung cancer lesion regions based on region segmentation according to claim 2, characterized in that: The specific method of obtaining the unsmooth burr coefficient of each tissue connected domain according to the position distribution of edge pixels in each tissue connected domain is as follows: The Euclidean distance between each edge pixel point in the tissue connected domain and the centroid of the tissue connected domain is recorded as the first distance of each edge pixel point; the variance of the first distance of all edge pixels in the tissue connected domain is used as the unsmooth burr coefficient of the tissue connected domain.

4. The method for extracting lung cancer lesion regions based on region segmentation according to claim 2, characterized in that: The method of obtaining the ground glass coefficient of each tissue connected domain according to the grayscale difference between each tissue connected domain and the surrounding area includes: The area formed by the maximum circumscribed circle of the tissue connected domain is recorded as the morphological range of the tissue connected domain; the area within the morphological range of the tissue connected domain except the tissue connected domain is recorded as the external area of ​​the tissue connected domain; the area in the grayscale image except all tissue connected domains is recorded as the lung background area of ​​the grayscale image; The mean of the grayscale values ​​of all pixels in the external area of ​​the tissue connected domain is recorded as the first mean; the mean of the grayscale values ​​of all pixels in the lung background area of ​​the grayscale image is recorded as the second mean; the difference between the first mean and the second mean is taken as the ground glass coefficient of the tissue connected domain.

5. The method for extracting lung cancer lesion regions based on region segmentation according to claim 1, characterized in that: The method of obtaining the fog isolation coefficient of the grayscale image according to the occurrence probability of different types of fog features in the grayscale image includes the following specific methods: All the fog-like features in the grayscale image are sorted in ascending order according to the probability of occurrence, and all the sorted fog-like features are used as the fog feature sequence of the grayscale image; For any type of fog feature in the fog feature sequence of the grayscale image; Obtaining a relative non-lesion atomization feature and a relative lesion atomization feature of the atomization feature; The ratio of the variance of the occurrence probabilities of all relatively non-lesion atomization features of this type of atomization features to the mean of the occurrence probabilities of all relatively non-lesion atomization features of this type of atomization features is recorded as the first ratio; the inverse proportional value of the occurrence probabilities of all relatively lesion atomization features of this type of atomization features to the information entropy of the occurrence probabilities of all atomization features in the atomization feature sequence of the grayscale image is recorded as the first inverse proportional value; The sum of the first ratio and the first inverse proportional value is used as the objective function value of the atomization feature; The atomization feature with the largest objective function value is recorded as the target atomization feature; The ratio of the cumulative sum of the occurrence probabilities of all relative non-lesion atomized features of the target atomized feature to the cumulative sum of the occurrence probabilities of all relative lesion atomized features of the target atomized feature is used as the atomization isolation coefficient of the grayscale image.

6. The method for extracting lung cancer lesion regions based on region segmentation according to claim 5, characterized in that: The specific method of obtaining the relative non-lesion atomization feature and the relative lesion atomization feature of this type of atomization feature is as follows: In the fog feature sequence of the grayscale image, all fog-like features before the fog-like feature are recorded as relative non-lesion fog features of the fog-like feature; all fog-like features after the fog-like feature are recorded as relative pathological fog features of the fog-like feature.

7. The method for extracting lung cancer lesion regions based on region segmentation according to claim 4, characterized in that: The method of obtaining the interference factor of each tissue connected domain according to the grayscale distribution difference between the grayscale image and the restored image of each tissue connected domain includes the following specific methods: For any tissue connected domain in the grayscale image, in the restored image, the area formed by the maximum circumscribed circle of the tissue connected domain is used as the interference layer area of ​​the tissue connected domain; The ratio of the mean grayscale value of the pixels in the interference layer area of ​​the tissue connected domain to the mean grayscale value of all pixels in the lung background area of ​​the grayscale image is used as the interference factor of the tissue connected domain.

8. The method for extracting lung cancer lesion regions based on region segmentation according to claim 1, characterized in that: The specific method of obtaining the lesion probability of each tissue connected domain according to the morphological lesion coefficient and the interference factor of each tissue connected domain is as follows: The product of the morphological lesion coefficient of the tissue connectivity domain and the interference factor is used as the lesion factor of the tissue connectivity domain; Obtain the lesion factors of all tissue connected domains in the grayscale image; perform linear normalization on all lesion factors, and record each normalized lesion factor as the lesion probability.

Images