Classification detection system for lung cancer combined embolism

Abstract

The invention relates to the technical field of medical treatment, and particularly discloses a classification detection system for lung cancer and pulmonary embolism, which is used for solving the problem that in the prior art, texture features related to lung cancer and pulmonary embolism are not extracted, and a model does not divide a focus area, so that an embolism position cannot be quickly found out when a film is read by using the model. According to the method, a healthy lung X-ray image set is acquired and summarized into a first lung image set, a lung cancer combined embolism lung X-ray image set is acquired and summarized into a second lung image set, embolism features are extracted according to the first lung image set and the second lung image set, and an embolism area is positioned and marked according to the embolism features; according to the method, extraction and analysis of the vascular sparseness and right cardiac shadow increase features are integrated, the embolism area is positioned in combination with machine learning, and the detection efficiency and accuracy of lung cancer combined pulmonary embolism are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and more particularly to a classification detection system for lung cancer combined with embolism. Background Art

[0002] Pulmonary thromboembolism is a group of diseases caused by blood clots blocking the pulmonary arteries and their branches. Pulmonary thromboembolism presents with diverse clinical manifestations and a high mortality rate, so early detection plays a crucial role in reducing the mortality rate. Malignant tumors can easily lead to a hypercoagulable state, so patients with a history of cancer are at increased risk for pulmonary embolism. Lung cancer has a high incidence rate and is the leading cause of cancer-related mortality. It is also a common malignant tumor associated with pulmonary embolism. Chest X-rays are a commonly used lung examination method. Traditionally, doctors review each image individually and make judgments based on experience. This process takes a long time, and due to significant individual variability, doctors' interpretation skills depend on their personal experience. This long period of time can lead to fatigue, which can affect the accuracy of diagnostic results. In chest X-rays, existing literature ([1] Wang Xuejuan. Research on classification algorithm of lung cancer combined with pulmonary embolism based on machine learning [D]. Yunnan Normal University, 2022. DOI: 10.27459 / d.cnki.gynfc.2022.001019.) proposed a clinical text data classifier based on machine learning and an X-ray image classifier based on a deep learning model to detect whether lung cancer patients have pulmonary embolism, so as to achieve early treatment, better prognosis, and prolong survival time. However, the existing technology does not extract texture features related to lung cancer combined with pulmonary embolism, and the model does not segment the lesion area, resulting in the inability to quickly find the embolism location when reading the film using the model. In order to solve the above problems, a technical solution is now provided. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides a classification detection system for lung cancer combined with pulmonary embolism, which is used to solve the problem that the prior art does not extract texture features related to lung cancer combined with pulmonary embolism, and the model does not segment the lesion area, resulting in the inability to quickly find the embolism location when reading the film using the model, so as to solve the problems raised in the above-mentioned background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A classification detection system for lung cancer combined with embolism includes an image processing module, an embolism feature extraction module, and a feature fusion detection module; the image processing module is used to obtain a healthy lung X-ray image set and summarize it into a first lung image set, and obtain a lung X-ray image set of lung cancer combined with embolism and summarize it into a second lung image set; the embolism feature extraction module is used to extract embolism features based on the first lung image set and the second lung image set; the embolism feature extraction module includes an image processing unit, a first feature recognition unit, and a first feature analysis unit; the image processing unit is used to grayscale, normalize, and enhance the first lung image set and the second lung image set; the first feature recognition unit Used to enhance the pulmonary vascular structure map of the first lung image set into a standard pulmonary vascular image through a vascular enhancement filter, and extract the pulmonary vascular structure map of the second lung image set into a second pulmonary vascular image; the first feature analysis unit is used to perform vascular sparsity analysis based on the standard lung image and the second pulmonary vascular image: extract the standard lung image and the second pulmonary vascular image, obtain the first eigenvalue and the first vascular distribution coefficient of the standard lung image, obtain the second eigenvalue and the second vascular distribution coefficient of the second pulmonary vascular image, and construct a first feature analysis model based on the first eigenvalue, the second eigenvalue, the first vascular distribution coefficient and the second vascular distribution coefficient to obtain the first feature coefficient.

[0006] As a further solution of the present invention, the specific steps of constructing the first characteristic analysis model to obtain the first characteristic coefficient include:

[0007] Extract the absolute difference between the average value of the vascularity distribution coefficients of all samples in the standard lung image set and the vascularity distribution coefficient of the current b-th second lung image;

[0008] Divide the current b-th second lung image into n image blocks, and perform the following processing on each image block: obtain a first deviation feature by taking the difference between the second eigenvalue of the image block in the b-th second lung image and the mean of the first eigenvalue of the corresponding block in the standard lung image, and obtain a second deviation feature by taking the difference between the mean of the maximum eigenvalue of the standard image and the maximum eigenvalue of the b-th second lung image. Obtain the proportion of the first deviation feature in the second deviation feature to obtain a normalized deviation ratio;

[0009] The first characteristic coefficient is obtained by averaging the deviation ratio of the sum of the absolute difference and each image block.

[0010] As a further solution of the present invention, a standard lung image and a second lung vascular image are obtained, the standard lung image is divided into several image blocks, and the vascular network density in each image block is obtained as a first eigenvalue, where the first eigenvalue is the ratio of the number of pixels in the image block belonging to blood vessels to the total number of pixels in the image block. The first eigenvalues ​​of all image blocks are calculated by variance to obtain a first vascular distribution coefficient of the standard lung image; the second lung vascular image is divided into several image blocks, and the vascular network density in each image block is obtained as a second eigenvalue, where the second eigenvalue is the ratio of the number of pixels in the image block belonging to blood vessels to the total number of pixels in the image block. The second eigenvalues ​​of all image blocks are calculated by variance to obtain a second vascular distribution coefficient of the second lung vascular image.

[0011] As a further embodiment of the present invention, a first standard eigenvalue is obtained based on the first characteristic coefficient and the first vascularity distribution coefficient. Specifically, for all standard lung images, a normalized block feature of each image block in each image is obtained. The proportion of the first eigenvalue of each image block in the difference between the maximum and minimum eigenvalues ​​of the standard lung image is obtained, and the normalized values ​​are summed and averaged. Next, the absolute value of the difference between the overall vascularity distribution coefficient of the image and the average normalized block feature is used to represent the degree of deviation of the image. The first standard eigenvalue is obtained based on the degree of deviation of the standard lung image.

[0012] The first characteristic standard value and the first characteristic coefficient of the second lung image are extracted, and the first characteristic coefficient of the second lung image is compared with the first characteristic standard value. If the first characteristic coefficient of the second lung image is greater than or equal to the first characteristic standard value, the vascular sparsity of the second lung image is normal; if the first characteristic coefficient of the second lung image is less than the first characteristic standard value, the vascular sparsity of the second lung image is abnormal.

[0013] As a further solution of the present invention, the embolism feature extraction module also includes a second feature recognition unit and a second feature analysis unit; the second feature analysis unit is used to analyze the right heart shadow enlargement feature based on the standard area image and the first area image: obtain the standard area image and the first area image, divide the standard area image into several area blocks, calculate the area ratio of the right heart shadow in each area block in the standard area image as the first area feature value, and perform variance calculation on the first area feature values ​​of all area blocks to obtain the first area distribution coefficient of the standard area image; divide the first area image into several area blocks, calculate the area ratio of the right heart shadow in each area block in the first area image as the second area feature value, and perform variance calculation on the second area feature values ​​of all area blocks to obtain the second area distribution coefficient of the first area image; construct a second feature analysis model based on the first area feature value, the second area feature value, the first area distribution coefficient and the second area distribution coefficient to analyze and obtain the second feature coefficient.

[0014] As a further solution of the present invention, the step of constructing a second characteristic analysis model to analyze and obtain a second characteristic coefficient includes:

[0015] Extracting the absolute difference between the mean of the first region distribution coefficient and the second region distribution coefficient;

[0016] The current first region image is divided into m image blocks, and each image block is processed as follows: a third deviation feature is obtained by taking the difference between the second eigenvalue of the image block in the first region image and the mean value of the first region feature of the region block in the standard region image, and a fourth deviation feature is obtained by taking the difference between the mean value of the maximum first region feature in the standard region image and the maximum second region feature value in the first region image, and a ratio of the third deviation feature to the fourth deviation feature is obtained to obtain a normalized deviation ratio;

[0017] The second characteristic coefficient is obtained by averaging the deviation ratio of the sum of the absolute difference and each image block.

[0018] As a further solution of the present invention, a second standard characteristic value is obtained based on the first regional characteristic value and the first regional distribution coefficient analysis, specifically:

[0019] The standard region image is divided into m region blocks, the first region feature value of each region block is taken, and the difference between the maximum region block feature value and the minimum region block feature value in the standard region image is used (each first region feature value is normalized, and then the normalized values ​​are summed and averaged to obtain the average normalized block feature of the standard region image;

[0020] Taking the difference between the first regional distribution coefficient of the image and the average normalized block feature and taking the absolute value to obtain the degree of deviation between the overall distribution of the standard regional image and the regional block feature;

[0021] Repeat the above steps for all standard region images, obtain the absolute deviation value of each standard region image, perform mean processing on it, and obtain the second standard feature value.

[0022] The second characteristic standard value and the second characteristic coefficient of the second area image are extracted, and the second characteristic coefficient of the second area image is compared with the second characteristic standard value. If the second characteristic coefficient of the second area image is less than the second characteristic standard value, there is no risk of enlargement of the right heart shadow in the second area image; if the second characteristic coefficient of the second area image is greater than or equal to the second characteristic standard value, there is a risk of enlargement of the right heart shadow in the second area image.

[0023] As a further solution of the present invention, the feature fusion detection module is used to locate and mark the embolism area according to the embolism characteristics: obtain the first characteristic coefficient of the second lung image and the second characteristic coefficient of the second area image, and when the first characteristic coefficient of the second lung image is less than the first characteristic standard value, mark the pulmonary vascular structure diagram of the second lung image; when the second characteristic coefficient of the second area image is less than the second characteristic standard value, mark the right heart shadow area in the second area image.

[0024] The technical effects and advantages of the classification detection system for lung cancer combined with embolism of the present invention are as follows: the present invention obtains a set of healthy lung X-ray images and summarizes it into a first lung image set, and obtains a set of lung X-ray images of lung cancer combined with embolism and summarizes it into a second lung image set, extracts embolism features based on the first lung image set and the second lung image set, and locates and marks the embolism area based on the embolism features; the present invention integrates the extraction and analysis of vascular sparsity and right heart shadow enlargement features, combines machine learning to locate the embolism area, and greatly improves the detection efficiency and accuracy of lung cancer combined with pulmonary embolism; directly locates and marks the pulmonary embolism area through the feature fusion detection module, providing doctors with an intuitive visual reference and reducing the workload of reading the film; combines the feature extraction of pulmonary vascular sparsity and right heart shadow enlargement to ensure the rapid discovery of the lesion area, which is particularly suitable for cases with complex distribution of embolism areas; analyzes the changes in pulmonary vascular sparsity and the characteristics of right heart shadow enlargement to make the diagnosis more comprehensive and reduce the possibility of missed diagnosis, and significantly improves the detection sensitivity and accuracy of abnormal features by performing standardized comparison between lesion images and healthy images. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic structural diagram of a classification detection system for lung cancer combined with embolism provided by the present invention;

[0026] Figure 2 Comparison of lung imaging features and risk index analysis of healthy and pathological samples provided by the present invention;

[0027] Figure 3 This is a distribution analysis diagram of pixel intensity, edge and texture features in various areas of the diseased lung provided by the present invention. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the accompanying drawings. Obviously, the technical solutions described are only part of the present invention, not the entire invention. Based on the technical solutions of the present invention, all other technical solutions obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0029] A classification detection system for lung cancer combined with embolism includes an image processing module, an embolism feature extraction module, and a feature fusion detection module. The image processing module is connected to the embolism feature extraction module, and the embolism feature extraction module is connected to the feature fusion detection module.

[0030] The image processing module is used to obtain a healthy lung X-ray image set and summarize it into a first lung image set, and obtain a lung X-ray image set of lung cancer combined with embolism and summarize it into a second lung image set;

[0031] The embolism feature extraction module is used to extract embolism features based on the first lung image set and the second lung image set;

[0032] The feature fusion detection module is used to locate and mark the embolic area according to the embolic features.

[0033] Specifically, the embolism feature extraction module includes an image processing unit, a first feature recognition unit, a first feature analysis unit, a second feature recognition unit, and a second feature analysis unit; the image processing unit is connected to the first feature recognition unit and the second feature recognition unit respectively, the first feature recognition unit is connected to the first feature analysis unit, and the second feature recognition unit is connected to the second feature analysis unit;

[0034] The image processing unit is used to perform grayscale conversion, normalization, and image enhancement processing on the first lung image set and the second lung image set;

[0035] The first feature recognition unit is used to enhance the pulmonary vascular structure map of the first lung image set into a standard pulmonary vascular image through a vascular enhancement filter, and extract the pulmonary vascular structure map of the second lung image set into a second pulmonary vascular image;

[0036] The first feature analysis unit is used to perform vascular sparsity analysis based on the standard lung image and the second lung vessel image;

[0037] The second feature recognition unit is used to mark the right heart shadow area contour of the first lung image set as the standard area image and mark the right heart shadow area contour of the second lung image set as the first area image by using a grayscale gradient and edge detection algorithm;

[0038] The second feature analysis unit is used to analyze the feature of right heart enlargement based on the standard region image and the first region image.

[0039] The feature fusion detection module directly locates and marks the pulmonary embolism area, providing doctors with an intuitive visual reference and reducing the workload of reading films; combining the feature extraction of pulmonary vascular sparsity and right heart shadow enlargement to ensure the rapid discovery of the lesion area, especially suitable for cases with complex distribution of embolism areas; analyzing the changes in pulmonary vascular sparsity and the characteristics of right heart shadow enlargement to make the diagnosis more comprehensive and reduce the possibility of missed diagnosis; by comparing the lesion image with the healthy image in a standardized manner, the detection sensitivity and accuracy of abnormal features are significantly improved; including the fully automatic execution of grayscale, normalization and enhancement processing to ensure the consistency of input image quality and improve the efficiency of subsequent analysis Reliability, vascular enhancement and right heart shadow area contour marking are all automatically completed by the system, avoiding errors that may be caused by manual intervention; by calculating the characteristic value and distribution coefficient of the sparseness of pulmonary blood vessels, quantitative indicators are provided to facilitate doctors to judge the severity of embolism, and quantitative feature calculation of the right heart shadow area (area ratio, distribution coefficient, etc.) helps doctors identify possible cor pulmonale, pulmonary hypertension and other related complications; through the feature fusion detection module, the system can mark the embolic area on the image, making it clear to doctors at a glance, significantly improving the diagnosis speed, extracting texture features related to lung cancer combined with pulmonary embolism, and filling the gaps in such information missing in existing models.

[0040] Specifically, the first feature analysis unit is used to perform vascular sparsity analysis based on the standard lung image and the second lung vascular image: obtain a standard lung image and a second lung vascular image, divide the standard lung image into several image blocks, and obtain the vascular network density in each image block as a first eigenvalue, where the first eigenvalue is the ratio of the number of pixels belonging to blood vessels in the image block to the total number of pixels in the image block, and perform variance calculation on the first eigenvalues ​​of all image blocks to obtain the first vascular distribution coefficient of the standard lung image; divide the second lung vascular image into several image blocks, and obtain the vascular network density in each image block as a second eigenvalue, where the second eigenvalue is the ratio of the number of pixels belonging to blood vessels in the image block to the total number of pixels in the image block, and perform variance calculation on the second eigenvalues ​​of all image blocks to obtain the second vascular distribution coefficient of the second lung vascular image; construct a first feature analysis model based on the first eigenvalue, the second eigenvalue, the first vascular distribution coefficient, and the second vascular distribution coefficient to obtain the first feature coefficient.

[0041] The specific steps of constructing the first characteristic analysis model to obtain the first characteristic coefficient include:

[0042] Extract the absolute difference between the average value of the vascularity distribution coefficients of all samples in the standard lung image set and the vascularity distribution coefficient of the current b-th second lung image;

[0043] Divide the current b-th second lung image into n image blocks, and perform the following processing on each image block: obtain a first deviation feature by taking the difference between the second eigenvalue of the image block in the b-th second lung image and the mean of the first eigenvalue of the corresponding block in the standard lung image, and obtain a second deviation feature by taking the difference between the mean of the maximum eigenvalue of the standard image and the maximum eigenvalue of the b-th second lung image. Obtain the proportion of the first deviation feature in the second deviation feature to obtain a normalized deviation ratio;

[0044] The first characteristic coefficient is obtained by averaging the deviation ratio of the sum of the absolute difference and each image block.

[0045] A standard lung image and a second lung vascular image are obtained, the standard lung image is divided into several image blocks, and the vascular network density in each image block is obtained as a first eigenvalue, where the first eigenvalue is the ratio of the number of pixels belonging to blood vessels in the image block to the total number of pixels in the image block. The first eigenvalues ​​of all image blocks are calculated with variance to obtain a first vascular distribution coefficient of the standard lung image; the second lung vascular image is divided into several image blocks, and the vascular network density in each image block is obtained as a second eigenvalue, where the second eigenvalue is the ratio of the number of pixels belonging to blood vessels in the image block to the total number of pixels in the image block. The second eigenvalues ​​of all image blocks are calculated with variance to obtain a second vascular distribution coefficient of the second lung vascular image.

[0046] The first standard eigenvalue is obtained based on the analysis of the first characteristic coefficient and the first vascular distribution coefficient. Specifically, for all standard lung images, the normalized block features of each image block in each image are obtained respectively: the proportion of the first eigenvalue of each image block in the difference between the maximum and minimum eigenvalues ​​of the standard lung image is obtained, and the normalized values ​​are summed and averaged; then, the absolute value of the difference between the overall vascular distribution coefficient of the image and the average normalized block feature is used to represent the degree of deviation of the image; and the first standard eigenvalue is obtained based on the deviation degree analysis of the standard lung image.

[0047] The first characteristic standard value and the first characteristic coefficient of the second lung image are extracted, and the first characteristic coefficient of the second lung image is compared with the first characteristic standard value. If the first characteristic coefficient of the second lung image is greater than or equal to the first characteristic standard value, the vascular sparsity of the second lung image is normal; if the first characteristic coefficient of the second lung image is less than the first characteristic standard value, the vascular sparsity of the second lung image is abnormal.

[0048] Based on multiple features, including the first characteristic coefficient, the first vascularity coefficient, and the second vascularity coefficient, the pulmonary vascular sparsity can be quantified from different perspectives. Comparing the standard value of the first characteristic with the characteristic value of the second lung image ensures objectivity and reduces the influence of subjective factors on the diagnostic results. Calculating characteristic values ​​by image block can not only capture the vascular sparsity distribution in local areas but also reflect the overall distribution trend. Variance analysis of the vascularity coefficient can measure the uniformity of vascular distribution and help identify local abnormal lung areas. A model is constructed based on standard lung images (healthy samples) and directly compared with abnormal lung images (pathological samples), adapting to differences between individuals and lesion types. Considering the range of characteristic values ​​(maximum and minimum values) can further correct for feature deviations caused by individual lung characteristics. The standard value of the first characteristic, used as a diagnostic benchmark, can quickly determine whether vascular sparsity is abnormal, improving diagnostic efficiency. Directly comparing the first characteristic value of the second lung image with the standard value can accurately identify areas with abnormal vascular sparsity. Variance calculation can effectively detect abnormal sparsity in the vascular network, especially local vascular sparsity caused by lung diseases such as pulmonary embolism.

[0049] Specifically, the second feature analysis unit is used to analyze the feature of right heart shadow enlargement based on the standard area image and the first area image: obtain the standard area image and the first area image, divide the standard area image into several area blocks, calculate the area ratio of the right heart shadow in each area block in the standard area image as the first area feature value, and perform variance calculation on the first area feature values ​​of all area blocks to obtain the first area distribution coefficient of the standard area image; divide the first area image into several area blocks, calculate the area ratio of the right heart shadow in each area block in the first area image as the second area feature value, and perform variance calculation on the second area feature values ​​of all area blocks to obtain the second area distribution coefficient of the first area image; construct a second feature analysis model based on the first area feature value, the second area feature value, the first area distribution coefficient and the second area distribution coefficient to analyze and obtain the second feature coefficient.

[0050] Extracting the absolute difference between the mean of the first region distribution coefficient and the second region distribution coefficient;

[0051] The current first region image is divided into m image blocks, and each image block is processed as follows: a third deviation feature is obtained by taking the difference between the second eigenvalue of the image block in the first region image and the mean value of the first region feature of the region block in the standard region image, and a fourth deviation feature is obtained by taking the difference between the mean value of the maximum first region feature in the standard region image and the maximum second region feature value in the first region image, and a ratio of the third deviation feature to the fourth deviation feature is obtained to obtain a normalized deviation ratio;

[0052] The second characteristic coefficient is obtained by averaging the deviation ratio of the sum of the absolute difference and each image block.

[0053] The second standard eigenvalue is obtained by analyzing the first regional eigenvalue and the first regional distribution coefficient, specifically:

[0054] The standard region image is divided into m region blocks, the first region feature value of each region block is taken, and the difference between the maximum region block feature value and the minimum region block feature value in the standard region image is used (each first region feature value is normalized, and then the normalized values ​​are summed and averaged to obtain the average normalized block feature of the standard region image;

[0055] Taking the difference between the first regional distribution coefficient of the image and the average normalized block feature and taking the absolute value to obtain the degree of deviation between the overall distribution of the standard regional image and the regional block feature;

[0056] Repeat the above steps for all standard region images, obtain the absolute deviation value of each standard region image, perform mean processing on it, and obtain the second standard feature value.

[0057] The second characteristic standard value and the second characteristic coefficient of the second area image are extracted, and the second characteristic coefficient of the second area image is compared with the second characteristic standard value. If the second characteristic coefficient of the second area image is less than the second characteristic standard value, there is no risk of enlargement of the right heart shadow in the second area image; if the second characteristic coefficient of the second area image is greater than or equal to the second characteristic standard value, there is a risk of enlargement of the right heart shadow in the second area image.

[0058] By calculating quantitative features such as the area ratio and distribution coefficient of the right heart shadow, the subjective judgment in traditional medical diagnosis is eliminated, and the accuracy and reliability of the results are ensured by comparing the characteristic values ​​and distribution coefficients with those of the standard area image; dividing the image into several area blocks and calculating the characteristic values ​​respectively can capture the local change characteristics of the right heart shadow, which is helpful for discovering abnormalities in a small range. The variance calculation provides a uniformity assessment of the right heart shadow distribution, which is helpful for analyzing the distribution pattern of the right heart shadow enlargement; the introduction of maximum and minimum values ​​ensures that the model can sensitively capture various types of lesions (such as local enlargement or overall enlargement); by comparing the characteristic values ​​with the standard values, it is clear whether there is a risk of enlargement of the right heart shadow, which is helpful for early screening and graded diagnosis. Based on comparison with standard area images, abnormal areas can be quickly identified, significantly improving diagnostic efficiency. Relying on multiple standard area images and target area images, a distribution law model is constructed to improve adaptability to complex cases. Through statistical analysis of the mean and variance, the model can suppress accidental errors and improve the robustness of feature extraction. It can generate visualization results such as heat maps and labeled areas to help doctors understand the location and severity of right heart enlargement. Combined with distribution coefficient analysis, it can clearly mark the area blocks with abnormal right heart shadows to facilitate further intervention. Through comparative analysis of standard area images and target images, a comprehensive assessment method for right heart enlargement is provided from a quantitative perspective, which can assist doctors in decision-making through automation, visualization and other technical means.

[0059] Specifically, the feature fusion detection module is used to locate and mark the embolism area according to the embolism characteristics: obtain the first characteristic coefficient of the second lung image and the second characteristic coefficient of the second area image, and when the first characteristic coefficient of the second lung image is less than the first characteristic standard coefficient, mark the pulmonary vascular structure diagram of the second lung image; when the second characteristic coefficient of the second area image is less than the second characteristic standard coefficient, mark the right heart shadow area in the second area image.

[0060] like Figure 2The present invention provides a comparison of lung imaging features and a risk index analysis chart for healthy samples and diseased samples; the chart shows the numerical differences in eight types of imaging features between healthy samples (green columns) and diseased (lung cancer combined with embolism) samples (red columns) through a columnar comparison, and the corresponding risk index (right vertical axis, %) is marked with a dotted line. Vascular density, number of branches, vascular diameter, and degree of torsion: healthy samples are significantly higher than diseased samples, suggesting that the vascular network structure is damaged or remodeled during lesions. Texture features, grayscale distribution, edge features, and morphological features: The differences between the two groups are also significant, among which the healthy samples maintain an advantage in texture features (390 vs 290) and edge features, while the risk index reaches a peak at the texture feature (about 72%), indicating that this item contributes the most to risk assessment. Risk index trend: From vascular structure to texture and morphological features, it shows a trend of first rising, then slightly falling, and then rising again, reflecting the dynamics of disease risk under the combined effect of multiple indicators.

[0061] like Figure 3 The present invention provides a distribution analysis diagram of pixel intensity, edge and texture features in each area of ​​the diseased lung; the diagram uses a combination of broken lines and columns to display the following in 8 predefined areas of the diseased lung: pixel intensity (blue broken line, left vertical axis), edge features (green broken line, left vertical axis), and texture features (yellow columns, right vertical axis). As can be seen from the figure: pixel intensity reaches a local peak (about 230-240) in areas 4 and 6, suggesting that these areas may have more obvious lesions or tissue density changes; edge features are also high in areas 2 and 4 (about 65-70), corresponding to clearer lesion boundaries; texture features are generally low and relatively flat (fluctuation between areas is 30-50), reflecting that fibrosis or nodule texture changes are relatively uniform. Taking "area 3" as an example, its pixel intensity is the highest (190), edge features are medium (55), and texture features are the lowest (42), suggesting that the image grayscale enhancement in this area is significant but the changes in edge and texture structure are relatively limited.

[0062] The embodiment of the present invention obtains a set of healthy lung X-ray images to be summarized as a first lung image set, and obtains a set of lung X-ray images of lung cancer combined with embolism to be summarized as a second lung image set, extracts embolism features based on the first lung image set and the second lung image set, and locates and marks the embolism area based on the embolism features; the present invention integrates the extraction and analysis of vascular sparsity and right heart shadow enlargement features, combines machine learning to locate the embolism area, and greatly improves the detection efficiency and accuracy of lung cancer combined with pulmonary embolism; directly locates and marks the pulmonary embolism area through the feature fusion detection module, providing doctors with an intuitive visual reference and reducing the workload of reading the film; combines the feature extraction of pulmonary vascular sparsity and right heart shadow enlargement to ensure the rapid discovery of the lesion area, which is particularly suitable for cases with complex distribution of embolism areas; analyzes the changes in pulmonary vascular sparsity and the characteristics of right heart shadow enlargement to make the diagnosis more comprehensive and reduce the possibility of missed diagnosis, and significantly improves the detection sensitivity and accuracy of abnormal features by performing standardized comparison between lesion images and healthy images.

[0063] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0064] Finally: 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 spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A classification detection system for lung cancer combined with embolism, comprising an image processing module, an embolism feature extraction module, and a feature fusion detection module; characterized in that: The image processing module is used to obtain a healthy lung X-ray image set and summarize it into a first lung image set, and obtain a lung X-ray image set of lung cancer combined with embolism and summarize it into a second lung image set; the embolism feature extraction module is used to extract embolism features based on the first lung image set and the second lung image set; The embolism feature extraction module includes an image processing unit, a first feature recognition unit, and a first feature analysis unit; the image processing unit is used to perform grayscale, normalization, and image enhancement processing on the first lung image set and the second lung image set; The first feature recognition unit is used to enhance the pulmonary vascular structure map of the first lung image set into a standard pulmonary vascular image through a vascular enhancement filter, and extract the pulmonary vascular structure map of the second lung image set into a second pulmonary vascular image; the first feature analysis unit is used to perform vascular sparsity analysis based on the standard lung image and the second pulmonary vascular image: extract the standard lung image and the second pulmonary vascular image, obtain the first eigenvalue and the first vascular distribution coefficient of the standard lung image, obtain the second eigenvalue and the second vascular distribution coefficient of the second pulmonary vascular image, and construct a first feature analysis model based on the first eigenvalue, the second eigenvalue, the first vascular distribution coefficient and the second vascular distribution coefficient to obtain the first feature coefficient.

2. The classification and detection system for lung cancer combined with embolism according to claim 1, characterized in that: The specific steps of constructing the first characteristic analysis model to obtain the first characteristic coefficient include: Extract the absolute difference between the average value of the vascularity distribution coefficients of all samples in the standard lung image set and the vascularity distribution coefficient of the current b-th second lung image; Divide the current b-th second lung image into n image blocks, and perform the following processing on each image block: obtain a first deviation feature by taking the difference between the second eigenvalue of the image block in the b-th second lung image and the mean of the first eigenvalue of the corresponding block in the standard lung image, and obtain a second deviation feature by taking the difference between the mean of the maximum eigenvalue of the standard image and the maximum eigenvalue of the b-th second lung image. Obtain the proportion of the first deviation feature in the second deviation feature to obtain a normalized deviation ratio; The first characteristic coefficient is obtained by averaging the deviation ratio of the sum of the absolute difference and each image block.

3. The classification and detection system for lung cancer combined with embolism according to claim 1, characterized in that: A standard lung image and a second lung vascular image are obtained, the standard lung image is divided into several image blocks, and the vascular network density in each image block is obtained as a first eigenvalue, where the first eigenvalue is the ratio of the number of pixels belonging to blood vessels in the image block to the total number of pixels in the image block. The first eigenvalues ​​of all image blocks are calculated with variance to obtain a first vascular distribution coefficient of the standard lung image; the second lung vascular image is divided into several image blocks, and the vascular network density in each image block is obtained as a second eigenvalue, where the second eigenvalue is the ratio of the number of pixels belonging to blood vessels in the image block to the total number of pixels in the image block. The second eigenvalues ​​of all image blocks are calculated with variance to obtain a second vascular distribution coefficient of the second lung vascular image.

4. The classification and detection system for lung cancer combined with embolism according to claim 1, characterized in that: A first standard eigenvalue is obtained based on the first characteristic coefficient and the first vascularity coefficient analysis. Specifically, for all standard lung images, a normalized block feature is obtained for each image block in each image. The proportion of the first eigenvalue of each image block in the difference between the maximum and minimum eigenvalues ​​of the standard lung image is obtained, and the normalized values ​​are summed and averaged. Next, the absolute value of the difference between the overall vascularity coefficient of the image and the average normalized block feature is used to represent the degree of deviation of the image. The first standard eigenvalue is obtained based on the degree of deviation of the standard lung image. The first characteristic standard value and the first characteristic coefficient of the second lung image are extracted, and the first characteristic coefficient of the second lung image is compared with the first characteristic standard value. If the first characteristic coefficient of the second lung image is greater than or equal to the first characteristic standard value, the vascular sparsity of the second lung image is normal; if the first characteristic coefficient of the second lung image is less than the first characteristic standard value, the vascular sparsity of the second lung image is abnormal.

5. The classification and detection system for lung cancer combined with embolism according to claim 1, characterized in that: The embolism feature extraction module further includes a second feature recognition unit and a second feature analysis unit; The second feature analysis unit is used to analyze the feature of right heart shadow enlargement based on the standard area image and the first area image: obtain the standard area image and the first area image, divide the standard area image into several area blocks, calculate the area ratio of the right heart shadow in each area block in the standard area image as the first area feature value, and perform variance calculation on the first area feature values ​​of all area blocks to obtain the first area distribution coefficient of the standard area image; divide the first area image into several area blocks, calculate the area ratio of the right heart shadow in each area block in the first area image as the second area feature value, and perform variance calculation on the second area feature values ​​of all area blocks to obtain the second area distribution coefficient of the first area image; construct a second feature analysis model based on the first area feature value, the second area feature value, the first area distribution coefficient and the second area distribution coefficient to analyze and obtain the second feature coefficient.

6. The classification and detection system for lung cancer combined with embolism according to claim 5, characterized in that: The steps of constructing a second characteristic analysis model to analyze and obtain the second characteristic coefficient include: Extracting the absolute difference between the mean of the first region distribution coefficient and the second region distribution coefficient; The current first region image is divided into m image blocks, and each image block is processed as follows: a third deviation feature is obtained by taking the difference between the second eigenvalue of the image block in the first region image and the mean value of the first region feature of the region block in the standard region image, and a fourth deviation feature is obtained by taking the difference between the mean value of the maximum first region feature in the standard region image and the maximum second region feature value in the first region image, and a ratio of the third deviation feature to the fourth deviation feature is obtained to obtain a normalized deviation ratio; The second characteristic coefficient is obtained by averaging the deviation ratio of the sum of the absolute difference and each image block.

7. The classification and detection system for lung cancer combined with embolism according to claim 5, characterized in that: The second standard eigenvalue is obtained by analyzing the first regional eigenvalue and the first regional distribution coefficient, specifically: The standard region image is divided into m region blocks, the first region feature value of each region block is taken, and the difference between the maximum region block feature value and the minimum region block feature value in the standard region image is used (each first region feature value is normalized, and then the normalized values ​​are summed and averaged to obtain the average normalized block feature of the standard region image; Taking the difference between the first regional distribution coefficient of the image and the average normalized block feature and taking the absolute value to obtain the degree of deviation between the overall distribution of the standard regional image and the regional block feature; Repeat the above steps for all standard area images, obtain the absolute deviation value of each standard area image, perform mean processing on it, and obtain the second standard eigenvalue; The second characteristic standard value and the second characteristic coefficient of the second area image are extracted, and the second characteristic coefficient of the second area image is compared with the second characteristic standard value. If the second characteristic coefficient of the second area image is less than the second characteristic standard value, there is no risk of enlargement of the right heart shadow in the second area image; if the second characteristic coefficient of the second area image is greater than or equal to the second characteristic standard value, there is a risk of enlargement of the right heart shadow in the second area image.

8. The classification and detection system for lung cancer combined with embolism according to claim 1, characterized in that: The feature fusion detection module is used to locate and mark the embolism area according to the embolism characteristics: obtain the first feature coefficient of the second lung image and the second feature coefficient of the second area image; when the first feature coefficient of the second lung image is less than the first feature standard value, mark the pulmonary vascular structure diagram of the second lung image; when the second feature coefficient of the second area image is less than the second feature standard value, mark the right heart shadow area in the second area image.

Images