Deep learning-based method and system for identifying infiltrative images of lung tumors

Abstract

The present invention discloses a method and system for identifying infiltrative images of lung tumors based on deep learning, which relates to the technical field of identifying infiltrative lung tumors and includes the following steps: obtaining medical images of lung tumors and simultaneously collecting well-annotated sample images; performing regional segmentation on the lung tumors based on the annotation results to obtain tumor regions; extracting edge features, morphological features, and density features of the tumor regions; performing deep learning on the identification of the infiltrative nature of lung tumors based on the edge features, morphological features, and density features of the sample images; The present invention is used to solve the problem that the existing techniques for identifying the infiltrative nature of lung tumors still rely too much on manual identification and the extracted tumor features lack specificity, resulting in inaccurate identification results for the infiltrative nature of lung tumors.

Description

Technical Field

[0001] The present invention relates to the technical field of lung tumor invasiveness recognition, and specifically to a method and system for lung tumor invasiveness image recognition based on deep learning. Background Art

[0002] The lung tumor invasiveness recognition technology refers to a method that uses computer vision and medical image analysis to help doctors and researchers identify and evaluate the invasiveness degree of lung tumors. The main goal of this technology is to accurately determine the invasion range and degree of lung tumors in lung tissue through an automated or semi-automated method.

[0003] The invasiveness of lung tumors is an important indicator for evaluating their malignancy, formulating treatment plans, and predicting the prognosis of patients. Traditional image-based invasiveness assessments rely on the experience of radiologists, and have problems such as strong subjectivity, low efficiency, and high misdiagnosis rates. Existing lung tumor invasiveness recognition technologies are not intelligent enough in recognizing the invasiveness of lung tumors, and the specificity of the extracted tumor features is insufficient, resulting in misjudgments when classifying invasiveness. For example, in the patent application with the publication number CN110533667A, a 3D segmentation method for CT images of lung tumors based on image pyramid fusion is disclosed. This solution is used to segment images within different scale ranges and finally fuses the segmentation results of different scale ranges. However, the feature fusion of the tumor features themselves is insufficient, and the accuracy of the recognition area is insufficient. Existing lung tumor invasiveness recognition technologies also rely too much on manual recognition and the specificity of the extracted tumor features is insufficient, resulting in inaccurate recognition results for the invasiveness of lung tumors. Summary of the Invention

[0004] The present invention aims to at least solve one of the technical problems in the existing technology to a certain extent. By obtaining the medical images of lung tumors and collecting well-annotated sample images at the same time, then performing regional segmentation on the lung tumors based on the annotation results to obtain the tumor regions, then extracting edge features based on the tumor regions of the sample images, extracting morphological features based on the tumor regions of the sample images, extracting density features based on the tumor regions of the sample images, and finally performing deep learning on the invasiveness recognition of lung tumors based on the edge features, morphological features, and density features of the sample images, to solve the problem that existing lung tumor invasiveness recognition technologies rely too much on manual recognition and the specificity of the extracted tumor features is insufficient, resulting in inaccurate recognition results for the invasiveness of lung tumors.

[0005] To achieve the above object, in the first aspect, the present application provides a method for lung tumor invasiveness image recognition based on deep learning, including the following steps:

[0006] Obtain the medical images of lung tumors and collect well-annotated sample images at the same time;

[0007] Perform regional segmentation on the lung tumor based on the annotation results to obtain the tumor region;

[0008] Extract the edge features, morphological features, and density features of the tumor region;

[0009] Perform deep learning on the invasive recognition of lung tumors based on the edge features, morphological features, and density features of the sample images.

[0010] Furthermore, obtain the medical images of the lung tumor, and collect well-annotated sample images, including the following sub-steps:

[0011] The medical image is a CT image;

[0012] Obtain a first quantity of medical images, and have a doctor perform regional annotation on the lung tumors in the medical images, and at the same time perform invasive annotation on the lung tumors to obtain sample images;

[0013] The regional annotation in the sample image is to circle the region where a lung tumor is located, and the invasive annotation includes non-invasive, micro-invasive, and invasive.

[0014] Furthermore, perform regional segmentation on the lung tumor based on the annotation results to obtain the tumor region, including the following sub-steps:

[0015] Mark the region circled by the regional annotation in the sample image as the roughly segmented region;

[0016] Perform edge extraction on the roughly segmented region to extract the contour of the lung tumor and obtain the tumor contour;

[0017] Segment the region enclosed by the tumor contour and mark it as the tumor region.

[0018] Furthermore, extract the edge features, morphological features, and density features of the tumor region, including the following sub-steps:

[0019] Extract edge features based on the tumor region of the sample image;

[0020] Extract morphological features based on the tumor region of the sample image;

[0021] Extract density features based on the tumor region of the sample image.

[0022] Furthermore, extract edge features based on the tumor region of the sample image, including the following sub-steps:

[0023] Mark the region of the non-tumor region in the sample image as the normal region;

[0024] Mark the pixel points on the tumor contour as contour pixel points, and mark the gray value of the contour pixel points as the contour gray value;

[0025] Mark the pixel points in the normal area as normal pixel points, and mark the gray value of the normal pixel points as the normal gray value;

[0026] Obtain the normal pixel points adjacent to the contour pixel points, mark them as adjacent pixel points, and mark the normal gray value of the adjacent pixel points as the adjacent gray value;

[0027] For any contour pixel point, calculate the difference between the contour gray value and the corresponding adjacent gray value, mark it as the edge gray difference, calculate the edge gray difference of each contour pixel point, count the number of the same edge gray differences, and mark it as the gray difference number;

[0028] Construct a histogram with the edge gray difference as the X-axis and the gray difference number as the Y-axis, mark it as the edge feature, and record the edge gray difference and the corresponding gray difference number into the edge feature.

[0029] Furthermore, extracting morphological features based on the tumor region of the sample image includes the following sub-steps:

[0030] Perform a rectangular selection on the tumor region to obtain a tumor rectangular region;

[0031] Mark the pixel points within the tumor region in the tumor rectangular region as tumor pixel points;

[0032] The tumor rectangular region is the morphological feature.

[0033] Furthermore, extracting density features based on the tumor region of the sample image includes the following sub-steps:

[0034] Mark the gray value of the tumor pixel points as the tumor gray value, and mark the number of the same tumor gray values as the tumor gray density;

[0035] Sort and number the tumor gray density in ascending order, and represent it by the symbol H n where n is a positive integer and n is the serial number of H;

[0036] Establish a plane rectangular coordinate system with n as the horizontal axis and the tumor gray density as the vertical axis, name it the density feature map, and record the tumor gray density according to the corresponding H n into the density feature map;

[0037] Perform linear regression on the density feature map, and mark the slope of the regression function as the density feature.

[0038] Furthermore, based on the edge feature, morphological feature, and density feature of the sample image, performing deep learning on the invasive recognition of lung tumors includes the following sub-steps:

[0039] Based on the invasive annotation of the sample images, the sample images are divided into different invasive groups, where the invasive groups include a non-invasive group, a micro-invasive group, and a strong-invasive group, and the strong-invasive group is the set of sample images with an invasive annotation of invasive;

[0040] For any invasive group, number the sample images therein, represented by the symbol P m where m is a positive integer and m is the serial number of P. Set natural numbers i and j, where i is any digit in m, and the value range of j is the same as that of m but j≠i;

[0041] For any i, obtain the edge features of P i and P j . Mark the histogram in the edge features of P i as the first histogram, and mark the histogram in the edge features of P j as the second histogram. Combine the edge features of P i and P j . Obtain the area of the first histogram, the area of the second histogram, and the area of the overlapping part between the first histogram and the second histogram, and mark them as S1, S2, and S3 respectively;

[0042] Calculate (S3 / S1 + S3 / S2) / 2, mark the calculation result as the edge similarity, and analyze P i with each P j . Calculate max(m)-1 edge similarities, find the maximum value among them, and mark it as the edge maximum similarity, where max() is the maximum value operator. Analyze each value of i to obtain max(m) edge maximum similarities, and find the minimum value among them, and mark it as the edge similarity limit;

[0043] For any i, obtain the morphological features of P i and P j . Enlarge the morphological feature with a smaller resolution until it has the same resolution as the morphological feature with a larger resolution. Mark the tumor pixel points in the morphological features of P i and P j as the first pixel point and the second pixel point respectively. Count the number of the first pixel point, the second pixel point, and the overlapping pixel points between the first pixel point and the second pixel point, and mark them as N1, N2, and N3 in sequence;

[0044] Calculate (N3 / N1 + N3 / N2) / 2, mark the calculation result as the morphological similarity, and analyze P i with each P jPerform an analysis, calculate max(m) - 1 morphological similarity degrees, find the maximum value among them, mark it as the maximum morphological similarity degree. Analyze for each value of i to obtain max(m) maximum morphological similarity degrees, and find the minimum value among them, mark it as the morphological similarity limit value;

[0045] For any i, obtain P i and P j 's density features, mark them as K1 and K2 respectively. If K1 ≥ K2, calculate K2 / K1; if K1 < K2, calculate K1 / K2. Mark the calculation result as the density similarity degree. For P i and each P j perform an analysis, calculate max(m) - 1 density similarity degrees, find the maximum value among them, mark it as the maximum density similarity degree. Analyze for each value of i to obtain max(m) maximum density similarity degrees, and find the minimum value among them, mark it as the density similarity limit value;

[0046] Based on the edge similarity limit value, morphological similarity limit value, and density similarity limit value, perform deep learning for the invasive identification of lung tumors.

[0047] Furthermore, based on the edge similarity limit value, morphological similarity limit value, and density similarity limit value, performing deep learning for the invasive identification of lung tumors includes the following sub - steps:

[0048] The deep learning is to perform deep learning on the analysis process of the edge similarity limit value, morphological similarity limit value, and density similarity limit value, and is used to continuously update the sample images and update the edge similarity limit value, morphological similarity limit value, and density similarity limit value;

[0049] The edge similarity limit value includes a non - invasive edge similarity limit value, a micro - invasive edge similarity limit value, and a strong - invasive edge similarity limit value, which are represented by the symbols EA, EB, and EC in sequence; the morphological similarity limit value includes a non - invasive morphological similarity limit value, a micro - invasive morphological similarity limit value, and a strong - invasive morphological similarity limit value, which are represented by the symbols FA, FB, and FC in sequence; the density similarity limit value includes a non - invasive density similarity limit value, a micro - invasive density similarity limit value, and a strong - invasive density similarity limit value, which are represented by the symbols DA, DB, and DC in sequence;

[0050] Obtain the medical images of the patient, mark them as images to be recognized, extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the images to be recognized in the non-invasive group, and mark them as GAE, GAF, and GAD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the images to be recognized in the micro-invasive group, and mark them as GBE, GBF, and GBD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the images to be recognized in the strongly invasive group, and mark them as GCE, GCF, and GCD respectively;

[0051] Calculate |GAE - EA|, |GAF - FA|, and |GAD - DA| respectively, and mark the calculation results as A1, A2, and A3 respectively; calculate |GBE - EB|, |GBF - FB|, and |GBD - DB| respectively, and mark the calculation results as B1, B2, and B3 respectively; calculate |GCE - EC|, |GCF - FC|, and |GCD - DC| respectively, and mark the calculation results as C1, C2, and C3 respectively;

[0052] Calculate A1 + A2 + A3 = QA, calculate B1 + B2 + B3 = QB, calculate C1 + C2 + C3 = QC, where QA is the non-invasive deviation degree, QB is the micro-invasive deviation degree, and QC is the strongly invasive deviation degree;

[0053] Compare the sizes of QA, QB, and QC, find the minimum value among them, mark it as the minimum deviation degree, and output the corresponding invasive annotation of the minimum deviation degree as the invasive annotation of the image to be recognized.

[0054] In a second aspect, the present application provides a lung tumor invasiveness image recognition system based on deep learning, including an image collection module, a region segmentation module, a feature extraction module, and a deep learning module; the image collection module, the region segmentation module, and the deep learning module are respectively connected to the feature extraction module for data connection;

[0055] The image collection module is used to obtain the medical images of lung tumors and collect well-annotated sample images at the same time;

[0056] The region segmentation module is used to perform region segmentation on the lung tumor based on the annotation results to obtain the tumor region;

[0057] The feature extraction module is used to extract the edge features, morphological features, and density features of the tumor region;

[0058] The deep learning module is used to perform deep learning on the invasiveness recognition of lung tumors based on the edge features, morphological features, and density features of the sample images.

[0059] Advantages of the present invention: By obtaining medical images of lung tumors and collecting well-annotated sample images, the present invention then performs regional segmentation on lung tumors based on the annotation results to obtain the tumor region. Subsequently, edge features are extracted based on the tumor region of the sample image, morphological features are extracted based on the tumor region of the sample image, and density features are extracted based on the tumor region of the sample image. The advantage lies in that there are usually certain differences between the tumor region and the normal region. Feature extraction of the tumor region through the three features of edge, morphology, and density ensures sufficient specificity and improves the accuracy and effectiveness of the recognition of lung tumor invasiveness.

[0060] The present invention performs deep learning on the recognition of lung tumor invasiveness based on the edge features, morphological features, and density features of the sample image, calculates the edge similarity limit value, morphological similarity limit value, and density similarity limit value of the edge features, morphological features, and density features, and then matches the image to be recognized with different infiltration groups to obtain the final result of the recognition of invasiveness. The advantage is that the matching of a single feature is not sufficient for the classification and recognition of invasiveness. Matching and analyzing the features in multiple dimensions can avoid the phenomenon of misjudgment easily occurring during the classification of invasiveness. At the same time, through the deep learning method, it can perform intelligent recognition, removing the subjectivity of manual recognition and improving the efficiency and accuracy of the recognition of lung tumor invasiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is the principle block diagram of the system of the present invention;

[0062] Figure 2 is the schematic diagram of the sample image of the present invention;

[0063] Figure 3 is the schematic diagram of the rough segmentation region of the present invention;

[0064] Figure 4 is the schematic diagram of the tumor region of the present invention;

[0065] Figure 5 is the schematic diagram of rendering the tumor contour in gray of the present invention;

[0066] Figure 6 is the edge feature of the present invention;

[0067] Figure 7 is the schematic diagram of the tumor rectangular region of the present invention;

[0068] Figure 8 is the density feature map of the present invention;

[0069] Figure 9 is the edge feature of P2 of the present invention;

[0070] Figure 10 This is the histogram obtained after merging the edge features of P1 and P2 of the present invention;

[0071] Figure 11 This is a schematic diagram of the morphological features of P1 and P2 of the present invention;

[0072] Figure 12 This is the flowchart of the steps of the method of the present invention. Detailed implementation manners

[0073] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0074] Example 1, please refer to Figure 1 As shown, the present application provides a pulmonary tumor infiltrative image recognition system based on deep learning, including an image collection module, a region segmentation module, a feature extraction module, and a deep learning module; the image collection module, the region segmentation module, and the deep learning module are respectively connected to the feature extraction module for data connection;

[0075] The image collection module is used to obtain medical images of pulmonary tumors and collect well-annotated sample images at the same time;

[0076] The image collection module is configured with an image collection strategy, and the image collection strategy includes:

[0077] The medical image is a CT image;

[0078] Please refer to Figure 2 As shown, obtain a first number of medical images. The first number is a specific value in the process of selecting medical images. Specifically, the first number is set to 50 in implementation. Experienced radiologists use segmentation tools to perform region annotation on the pulmonary tumors in the medical images and perform infiltrative annotation on the pulmonary tumors at the same time to obtain sample images;

[0079] The region annotation in the sample image is to circle the region where a pulmonary tumor is located. The infiltrative annotation includes non-infiltrative, micro-infiltrative, and infiltrative;

[0080] In practical applications, the region annotation is as Figure 2 shown. The infiltrative annotation is to mark a label for the sample image. Figure 2 The sample image shown is micro-infiltrative.

[0081] The region segmentation module is used to perform region segmentation on the pulmonary tumor based on the annotation result to obtain the tumor region;

[0082] The region segmentation module is configured with a region segmentation strategy, and the region segmentation strategy includes:

[0083] Please refer to Figures 3 to 4 As shown, mark the region outlined by the region annotation in the sample image as the roughly segmented region;

[0084] Extract the edges of the roughly segmented region to obtain the contour of the lung tumor, and get the tumor contour;

[0085] Segment the region enclosed by the tumor contour and mark it as the tumor region;

[0086] In practical applications, for the roughly segmented region as Figure 3 shown, extract the edges of the roughly segmented region, and get the tumor region as Figure 4 shown, Figure 4 The white part in

[0087] is the tumor region, and the junction between the white part and the black part is the tumor contour.

[0088] The feature extraction module is used to extract the edge features, morphological features, and density features of the tumor region; the feature extraction module includes an edge feature extraction unit, a morphological feature extraction unit, and a density feature extraction unit;

[0089] The edge feature extraction unit is used to extract edge features based on the tumor region of the sample image;

[0090] Mark the region of the non-tumor region in the sample image as the normal region;

[0091] Mark the pixel points on the tumor contour as contour pixel points, and mark the gray value of the contour pixel points as the contour gray value;

[0092] Mark the pixel points of the normal region as normal pixel points, and mark the gray value of the normal pixel points as the normal gray value;

[0093] Please refer to Figure 5 shown, obtain the normal pixel points adjacent to the contour pixel points, mark them as adjacent pixel points, and mark the normal gray value of the adjacent pixel points as the adjacent gray value;

[0094] For any contour pixel point, calculate the difference between the contour gray value and the corresponding adjacent gray value, mark it as the edge gray difference, calculate the edge gray difference of each contour pixel point, and count the number of the same edge gray differences, mark it as the gray difference number;

[0095] Please refer to Figure 6As shown in the figure, a histogram is constructed with the edge gray - level difference as the X - axis and the number of gray - level differences as the Y - axis, marked as edge features. The edge gray - level differences and the corresponding number of gray - level differences are recorded into the edge features;

[0096] In practical applications, Figure 4 the black area in [[ ]] is the normal area, and the white area is the tumor area. Figure 5 In [[ ]], the tumor contour is rendered in gray for distinction. Taking Figure 5 the contour pixel point in the upper - left corner of [[ ]] as an example, the adjacent pixel points adjacent to the contour pixel point are three dark - gray normal pixel points. Calculate their edge gray - level differences respectively, and the obtained edge gray - level differences are 12, 9, and 17. The constructed edge features are as Figure 6 shown;

[0097] The morphological feature extraction unit is used to extract morphological features based on the tumor area of the sample image;

[0098] The morphological feature extraction unit is configured with a morphological feature extraction strategy, and the morphological feature extraction strategy includes:

[0099] Please refer to Figure 7 shown in the figure, the tumor area is framed with a rectangle to obtain the tumor rectangle area;

[0100] The pixel points within the tumor area in the tumor rectangle area are marked as tumor pixel points;

[0101] The tumor rectangle area is the morphological feature;

[0102] In practical applications, the obtained tumor rectangle area is as Figure 7 shown;

[0103] The density feature extraction unit is used to extract density features based on the tumor area of the sample image;

[0104] The density feature extraction unit is configured with a density feature extraction strategy, and the density feature extraction strategy includes:

[0105] The gray - level value of the tumor pixel points is marked as the tumor gray - level value, and the number of the same tumor gray - level values is marked as the tumor gray - level density;

[0106] Sort and number the tumor gray - level densities in ascending order, and represent them by the symbol H n where n is a positive integer and n is the serial number of H;

[0107] Please refer to Figure 8 shown in the figure, establish a plane rectangular coordinate system with n as the horizontal axis and the tumor gray - level density as the vertical axis, named the density feature map, and record the tumor gray - level density into the density feature map according to the corresponding H n ;

[0108] Perform linear regression on the density feature map, and label the slope of the regression function as the density feature;

[0109] In practical applications, the magnitude of the tumor gray value is not important. The density feature only uses the change trend of the density of the tumor gray value as a feature. Therefore, in this embodiment, the n corresponding tumor gray value is not specifically displayed, and the density feature map is constructed as Figure 8 shown. Through linear regression analysis, the density feature is obtained as 1.0345.

[0110] The deep learning module is used to perform deep learning on the invasive recognition of lung tumors based on the edge features, morphological features, and density features of the sample images; the deep learning module includes a similarity limit analysis unit and an invasive recognition unit;

[0111] The similarity limit analysis unit is configured with a similarity limit analysis strategy, and the similarity limit analysis strategy includes:

[0112] Based on the invasive annotation of the sample images, the sample images are divided into different invasive groups, and the invasive groups include a non-invasive group, a micro-invasive group, and a strong-invasive group. The strong-invasive group is the set of sample images with an invasive annotation of invasive;

[0113] For any invasive group, number the sample images therein, and represent them by the symbol P m where m is a positive integer and m is the serial number of P. Set natural numbers i and j, where i is any digit in m, and the value range of j is the same as that of m but j ≠ i;

[0114] Please refer to Figures 9 to 10 shown. For any i, obtain the edge features of P i and P j . Mark the histogram in the edge features of P i as the first histogram, mark the histogram in the edge features of P j as the second histogram, merge the edge features of P i and P j . Obtain the area of the first histogram, the area of the second histogram, and the area of the overlapping part between the first histogram and the second histogram, and mark them as S1, S2, and S3 respectively;

[0115] Calculate (S3 / S1 + S3 / S2) / 2, mark the calculation result as the edge similarity, and compare P i with each P jPerform an analysis to calculate max(m)-1 edge similarities, find the maximum value among them, and mark it as the maximum edge similarity. Here, max() is the maximum value operator. Analyze each value of i to obtain max(m) maximum edge similarities, and find the minimum value among them, which is marked as the edge similarity limit value;

[0116] In practical applications, this embodiment takes the analysis and calculation process of the edge similarity limit value, morphological similarity limit value, and density similarity limit value of the minimally invasive group as an example to illustrate the identification of lung tumor invasiveness. For the minimally invasive group, taking i = 1 and j = 2 as an example, obtain the edge features of P1 and P2, where Figure 6 is the edge feature of P1, Figure 9 is the edge feature of P2, Figure 10 is the histogram obtained after merging the edge features of P1 and P2. Obtain that S1, S2, and S3 are 145, 124, and 116 in sequence, calculate the edge similarity to be 0.87, and retain two decimal places for the calculation result. Analyze P1 with all P j to obtain max(m)-1 edge similarities. Through searching, the maximum edge similarity of P1 is found to be 0.89. Each P m has a maximum edge similarity. Through searching, the edge similarity limit value is found to be 0.82;

[0117] Please refer to Figure 11 As shown, for any i, obtain the morphological features of P i and P j . Enlarge the morphological features with a smaller resolution until they have the same resolution as the morphological features with a larger resolution. Mark the tumor pixel points in the morphological features of P i and P j as the first pixel point and the second pixel point respectively. Count the number of the first pixel point, the second pixel point, and the overlapping pixel points of the first pixel point and the second pixel point, and mark them as N1, N2, and N3 in sequence;

[0118] Calculate (N3 / N1 + N3 / N2) / 2, and mark the calculation result as the morphological similarity. Analyze P i with each P j to calculate max(m)-1 morphological similarities, find the maximum value among them, and mark it as the maximum morphological similarity. Analyze each value of i to obtain max(m) maximum morphological similarities, and find the minimum value among them, which is marked as the morphological similarity limit value;

[0119] In practical applications, the morphological features of P1 and P2 are as Figure 11As shown, among them, the morphological feature of P1 is the image on the left, with a resolution of 26×25, while the morphological feature of P2 is the image on the right, with a resolution of 24×23. The morphological feature of P2 is smaller. Change its resolution to 26×25, then overlap them. By statistics, N1, N2, and N3 are 534, 516, and 474 in sequence. By calculation, the morphological similarity between P1 and P2 is 0.90, and the calculation result is reserved to two decimal places. Compare P1 with each P j for analysis, find the maximum value among them, and the maximum morphological similarity is 0.91. Each P m corresponds to a maximum morphological similarity, and the morphological similarity limit value is found to be 0.81;

[0120] For any i, obtain the density features of P i and P j , which are marked as K1 and K2 respectively. If K1≥K2, then calculate K2 / K1. If K1<K2, then calculate K1 / K2. Mark the calculation result as the density similarity. Compare P i with each P j for analysis, calculate max(m)-1 density similarities, find the maximum value among them, and mark it as the maximum density similarity. Analyze each value of i to obtain max(m) maximum density similarities, and find the minimum value among them, which is marked as the density similarity limit value;

[0121] In practical applications, the density features K1 and K2 of P1 and P2 are obtained as 1.0345 and 1.1798 respectively. Since K1 is less than K2, calculate K1 / K2, and the density similarity is 0.88. The calculation result is reserved to two decimal places. Compare P1 with each P j for analysis, obtain max(m)-1 density similarities, and find that the maximum density similarity of P1 is 0.95. Each P m corresponds to a maximum density similarity, and the density similarity limit value is found to be 0.78;

[0122] The invasiveness recognition unit is used for deep learning of the invasiveness recognition of lung tumors based on the edge similarity limit value, morphological similarity limit value, and density similarity limit value;

[0123] The invasiveness recognition unit is configured with an invasiveness recognition strategy, and the invasiveness recognition strategy includes:

[0124] The deep learning is a process of deep learning for the analysis of the edge similarity limit value, morphological similarity limit value, and density similarity limit value, and is used to continuously update the sample images and update the edge similarity limit value, morphological similarity limit value, and density similarity limit value;

[0125] The edge similarity limits include non-invasive edge similarity limits, micro-invasive edge similarity limits, and strong-invasive edge similarity limits, which are represented by the symbols EA, EB, and EC in sequence; the morphological similarity limits include non-invasive morphological similarity limits, micro-invasive morphological similarity limits, and strong-invasive morphological similarity limits, which are represented by FA, FB, and FC in sequence; the density similarity limits include non-invasive density similarity limits, micro-invasive density similarity limits, and strong-invasive density similarity limits, which are represented by DA, DB, and DC in sequence.

[0126] Obtain the medical image of the patient, mark it as the image to be recognized, and extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be recognized in the non-invasive group, which are marked as GAE, GAF, and GAD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be recognized in the micro-invasive group, which are marked as GBE, GBF, and GBD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be recognized in the strong-invasive group, which are marked as GCE, GCF, and GCD respectively.

[0127] Calculate |GAE - EA|, |GAF - FA|, and |GAD - DA| respectively, and mark the calculation results as A1, A2, and A3 respectively; calculate |GBE - EB|, |GBF - FB|, and |GBD - DB| respectively, and mark the calculation results as B1, B2, and B3 respectively; calculate |GCE - EC|, |GCF - FC|, and |GCD - DC| respectively, and mark the calculation results as C1, C2, and C3 respectively.

[0128] Calculate A1 + A2 + A3 = QA, calculate B1 + B2 + B3 = QB, calculate C1 + C2 + C3 = QC, where QA is the non-invasive deviation degree, QB is the micro-invasive deviation degree, and QC is the strong-invasive deviation degree.

[0129] Compare the magnitudes of QA, QB, and QC, find the minimum value among them, mark it as the minimum deviation degree, and output the infiltration annotation corresponding to the minimum deviation degree as the infiltration annotation of the image to be recognized.

[0130] In practical applications, the recognition of the invasiveness of lung tumors has been described in detail in the invasive recognition strategy. The relationships between different data and the calculation process have been given. In this embodiment, no further explanation will be provided, and only the final numerical values and results will be given, and deep learning will be further explained. Through the calculations in the invasive recognition strategy, QA, QB, and QC are 0.18, 0.33, and 0.25 respectively. By comparison, the minimum deviation is 0.18, corresponding to the non-invasive group. That is, the medical image of the patient is marked as non-invasive. At the same time, the medical image of the patient is entered into the non-invasive group, and the deep learning model re-analyzes and updates the edge similarity limit value, shape similarity limit value, and density similarity limit value of the non-invasive group.

[0131] Example 2, please refer to Figure 12 As shown, the present application provides a method for recognizing the invasiveness of lung tumor images based on deep learning, including the following steps:

[0132] Step S1, obtain the medical image of the lung tumor and collect well-annotated sample images at the same time. Step S1 includes the following sub-steps:

[0133] Step S101, the medical image is a CT image;

[0134] Step S102, obtain a first quantity of medical images, and use a segmentation tool by an experienced radiologist to perform regional annotation on the lung tumor in the medical image, and at the same time perform invasive annotation on the lung tumor to obtain sample images;

[0135] Step S103, the regional annotation in the sample image is to circle the area where a lung tumor is located, and the invasive annotation includes non-invasive, micro-invasive, and invasive;

[0136] Step S2, perform regional segmentation on the lung tumor based on the annotation results to obtain the tumor region. Step S2 includes the following sub-steps:

[0137] Step S201, mark the area circled by the regional annotation in the sample image as the rough segmentation area;

[0138] Step S202, extract the edge of the rough segmentation area to extract the contour of the lung tumor to obtain the tumor contour;

[0139] Step S203, segment the area enclosed by the tumor contour and mark it as the tumor region;

[0140] Step S3, extract the edge feature, shape feature, and density feature of the tumor region. Step S3 includes the following sub-steps:

[0141] Step S301, extract the edge feature based on the tumor region of the sample image;

[0142] Step S301 includes the following sub-steps:

[0143] Step S301.1, mark the non-tumor regions in the sample image as normal regions;

[0144] Step S301.2, mark the pixel points on the tumor contour as contour pixel points, and mark the gray values of the contour pixel points as contour gray values;

[0145] Step S301.3, mark the pixel points in the normal regions as normal pixel points, and mark the gray values of the normal pixel points as normal gray values;

[0146] Step S301.4, obtain the normal pixel points adjacent to the contour pixel points, mark them as adjacent pixel points, and mark the normal gray values of the adjacent pixel points as adjacent gray values;

[0147] Step S301.5, for any contour pixel point, calculate the difference between the contour gray value and the corresponding adjacent gray value, mark it as the edge gray difference, calculate the edge gray difference of each contour pixel point, and count the number of the same edge gray differences, mark it as the gray difference quantity;

[0148] Step S301.6, construct a histogram with the edge gray difference as the X-axis and the gray difference quantity as the Y-axis, mark it as the edge feature, and enter the edge gray difference and the corresponding gray difference quantity into the edge feature;

[0149] Step S302, extract the morphological features based on the tumor region of the sample image;

[0150] Step S302 includes the following sub-steps:

[0151] Step S302.1, perform a rectangular selection on the tumor region to obtain a tumor rectangular region;

[0152] Step S302.2, mark the pixel points within the tumor region in the tumor rectangular region as tumor pixel points;

[0153] The tumor rectangular region is the morphological feature;

[0154] Step S303, extract the density features based on the tumor region of the sample image;

[0155] Step S303 includes the following sub-steps:

[0156] Step S303.1, mark the gray values of the tumor pixel points as tumor gray values, and mark the number of the same tumor gray values as the tumor gray density;

[0157] Step S303.2: Sort the tumor grayscale densities in ascending order and number them, denoted by the symbol H n where n is a positive integer and n is the serial number of H;

[0158] Step S303.3: Establish a plane rectangular coordinate system with n as the horizontal axis and the tumor grayscale density as the vertical axis, named the density feature map. Enter the tumor grayscale density into the density feature map according to the corresponding H n ;

[0159] Step S303.4: Perform linear regression on the density feature map and mark the slope of the regression function as the density feature;

[0160] Step S4: Perform deep learning on the invasion recognition of lung tumors based on the edge features, morphological features, and density features of the sample images. Step S4 includes the following sub-steps:

[0161] Step S401: Based on the invasion annotation of the sample images, divide the sample images into different invasion groups. The invasion groups include a non-invasion group, a micro-invasion group, and a strong-invasion group. The strong-invasion group is the set of sample images with an invasion annotation of invasive;

[0162] Step S402: For any invasion group, number the sample images in it, denoted by the symbol P m where m is a positive integer and m is the serial number of P. Set natural numbers i and j, where i is any digit in m, and the value range of j is the same as that of m but j≠i;

[0163] Step S403: For any i, obtain the edge features of P i and P j . Mark the histogram in the edge features of P i as the first histogram, mark the histogram in the edge features of P j as the second histogram, merge the edge features of P i and P j , and obtain the area of the first histogram, the area of the second histogram, and the area of the overlapping part between the first histogram and the second histogram, which are marked as S1, S2, and S3 respectively;

[0164] Step S404: Calculate (S3 / S1 + S3 / S2) / 2, mark the calculation result as the edge similarity, analyze P i with each P j , calculate to obtain max(m)-1 edge similarities, find the maximum value among them, mark it as the edge maximum similarity, where max() is the maximum value operator. Analyze each value of i to obtain max(m) edge maximum similarities, and find the minimum value among them, mark it as the edge similarity limit;

[0165] Step S405, for any i, obtain P i and P j 's morphological features, magnify the morphological features with a smaller resolution until the resolution is the same as that of the morphological features with a larger resolution, and for P i and P j 's morphological features, respectively mark the tumor pixel points as the first pixel points and the second pixel points, count the number of the first pixel points, the second pixel points, and the overlapping pixel points between the first pixel points and the second pixel points, and mark them as N1, N2, and N3 in sequence;

[0166] Step S406, calculate (N3 / N1 + N3 / N2) / 2, mark the calculation result as the morphological similarity, and analyze P i with each P j to calculate max(m) - 1 morphological similarities, find the maximum value among them, mark it as the maximum morphological similarity, analyze each value of i to obtain max(m) maximum morphological similarities, and find the minimum value among them, mark it as the morphological similarity limit;

[0167] Step S407, for any i, obtain P i and P j 's density features, mark them as K1 and K2 respectively. If K1 ≥ K2, calculate K2 / K1; if K1 < K2, calculate K1 / K2. Mark the calculation result as the density similarity, and analyze P i with each P j to calculate max(m) - 1 density similarities, find the maximum value among them, mark it as the maximum density similarity, analyze each value of i to obtain max(m) maximum density similarities, and find the minimum value among them, mark it as the density similarity limit;

[0168] Step S408, based on the edge similarity limit, the morphological similarity limit, and the density similarity limit, perform deep learning on the invasiveness recognition of lung tumors;

[0169] Step S408 includes the following sub - steps:

[0170] Step S408.1, the deep learning is to perform deep learning on the analysis process of the edge similarity limit, the morphological similarity limit, and the density similarity limit, which is used to continuously update the sample images and update the edge similarity limit, the morphological similarity limit, and the density similarity limit;

[0171] Step S408.2, the edge similarity limits include non-infiltrative edge similarity limit, micro-infiltrative edge similarity limit, and strong-infiltrative edge similarity limit, which are represented by symbols EA, EB, and EC in sequence; the morphological similarity limits include non-infiltrative morphological similarity limit, micro-infiltrative morphological similarity limit, and strong-infiltrative morphological similarity limit, which are represented by symbols FA, FB, and FC in sequence; the density similarity limits include non-infiltrative density similarity limit, micro-infiltrative density similarity limit, and strong-infiltrative density similarity limit, which are represented by symbols DA, DB, and DC in sequence;

[0172] Step S408.3, obtain the medical image of the patient, mark it as the image to be recognized, extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be recognized in the non-infiltrative group, and mark them as GAE, GAF, and GAD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be recognized in the micro-infiltrative group, and mark them as GBE, GBF, and GBD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be recognized in the strong-infiltrative group, and mark them as GCE, GCF, and GCD respectively;

[0173] Step S408.4, calculate |GAE - EA|, |GAF - FA|, and |GAD - DA| respectively, and mark the calculation results as A1, A2, and A3 respectively; calculate |GBE - EB|, |GBF - FB|, and |GBD - DB| respectively, and mark the calculation results as B1, B2, and B3 respectively; calculate |GCE - EC|, |GCF - FC|, and |GCD - DC| respectively, and mark the calculation results as C1, C2, and C3 respectively;

[0174] Step S408.5, calculate A1 + A2 + A3 = QA, calculate B1 + B2 + B3 = QB, calculate C1 + C2 + C3 = QC, where QA is the non-infiltrative deviation degree, QB is the micro-infiltrative deviation degree, and QC is the strong-infiltrative deviation degree;

[0175] Step S408.6, compare the magnitudes of QA, QB, and QC, find the minimum value among them, mark it as the minimum deviation degree, and output the infiltration annotation corresponding to the minimum deviation degree as the infiltration annotation of the image to be recognized.

[0176] Embodiment 3. The present application provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus. The memory stores computer-readable instructions, and the processor can call the instructions in the memory. When the computer-readable instructions are executed by the processor, the steps in the method for identifying infiltrative images of lung tumors based on deep learning are run to achieve the following functions: obtaining medical images of lung tumors and simultaneously collecting well-annotated sample images; performing regional segmentation on the lung tumors based on the annotation results to obtain tumor regions; extracting edge features, morphological features, and density features of the tumor regions; and performing deep learning on the infiltrative identification of lung tumors based on the edge features, morphological features, and density features of the sample images.

[0177] In addition, when the logical instructions in the above-mentioned memory are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0178] Embodiment 4. The present application also provides a computer-readable storage medium. The present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the method for identifying infiltrative images of lung tumors based on deep learning as described above are run to achieve the following functions: obtaining medical images of lung tumors and simultaneously collecting well-annotated sample images; performing regional segmentation on the lung tumors based on the annotation results to obtain tumor regions; extracting edge features, morphological features, and density features of the tumor regions; and performing deep learning on the infiltrative identification of lung tumors based on the edge features, morphological features, and density features of the sample images.

[0179] Through the description of the above embodiments, the embodiments of the present invention can be provided as a method, a system or a computer program product. Based on such understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0180] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The above-described embodiments are merely illustrative. For example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation. For another example, multiple modules or units can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some communication interfaces, and the indirect couplings or communication connections of systems, modules, and units can be electrical, mechanical, or other forms.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying infiltrative images of lung tumors based on deep learning, characterized in that, It includes the following steps: Obtain the medical images of lung tumors and collect well-annotated sample images at the same time; Perform regional segmentation on lung tumors based on the annotation results to obtain the tumor regions; Extract the edge features, morphological features, and density features of the tumor regions; extract edge features based on the tumor regions of the sample images; Extracting edge features based on the tumor regions of the sample images includes the following sub-steps: Mark the regions in the sample images that are not tumor regions as normal regions; Mark the pixel points on the tumor contour as contour pixel points and mark the gray values of the contour pixel points as contour gray values; Mark the pixel points in the normal regions as normal pixel points and mark the gray values of the normal pixel points as normal gray values; Obtain the normal pixel points adjacent to the contour pixel points, mark them as adjacent pixel points, and mark the normal gray values of the adjacent pixel points as adjacent gray values; For any contour pixel point, calculate the difference between the contour gray value and the corresponding adjacent gray value, mark it as the edge gray difference, calculate the edge gray difference of each contour pixel point, and count the number of the same edge gray differences, mark it as the gray difference quantity; Construct a histogram with the edge gray difference as the X-axis and the gray difference quantity as the Y-axis, mark it as the edge feature, and record the edge gray difference and the corresponding gray difference quantity into the edge feature; Based on the edge features, morphological features, and density features of the sample images, perform deep learning on the identification of the invasiveness of lung tumors.

2. The method for identifying infiltrative images of lung tumors based on deep learning according to claim 1, wherein Obtaining the medical images of lung tumors and collecting well-annotated sample images at the same time includes the following sub-steps: The medical images are CT images; Obtain a first quantity of medical images, and have doctors perform regional annotation on the lung tumors in the medical images and perform invasiveness annotation on the lung tumors at the same time to obtain sample images; The regional annotation in the sample images is to circle the region where a lung tumor is located, and the invasiveness annotation includes non-invasive, micro-invasive, and invasive.

3. The method for identifying infiltrative images of lung tumors based on deep learning according to claim 2, wherein Performing regional segmentation on lung tumors based on the annotation results to obtain the tumor regions includes the following sub-steps: Mark the regions circled by the regional annotation in the sample images as roughly segmented regions; Perform edge extraction on the roughly segmented regions to extract the contour of the lung tumor to obtain the tumor contour; Segment the region enclosed by the tumor contour and mark it as the tumor region.

4. The method for identifying infiltrative images of lung tumors based on deep learning according to claim 3, wherein, Extracting the edge features, morphological features, and density features of the tumor regions includes the following sub-steps: Extract morphological features based on the tumor regions of the sample images; Extract density features based on the tumor regions of the sample images.

5. The method for identifying infiltrative images of lung tumors based on deep learning according to claim 4, wherein Extracting morphological features based on the tumor regions of the sample images includes the following sub-steps: Perform rectangular selection on the tumor regions to obtain tumor rectangular regions; Mark the pixel points within the tumor regions in the tumor rectangular regions as tumor pixel points; The tumor rectangular region is the morphological feature.

6. The method for identifying infiltrative images of lung tumors based on deep learning according to claim 5, wherein Extracting density features based on the tumor regions of the sample images includes the following sub-steps: Mark the gray values of the tumor pixel points as tumor gray values, and mark the number of the same tumor gray values as tumor gray density; Sort and number the tumor gray density in ascending order, denoted by the symbol H n where n is a positive integer and n is the serial number of H; Taking n as the horizontal axis and the tumor gray density as the vertical axis, a plane rectangular coordinate system is established and named the density feature map. The tumor gray density is input into the density feature map according to the corresponding H n Enter it into the density feature map; Perform linear regression on the density feature map and mark the slope of the regression function as the density feature.

7. The method for identifying infiltrative images of lung tumors based on deep learning according to claim 6, wherein Deep learning for the invasive identification of lung tumors based on the edge features, morphological features, and density features of sample images includes the following sub-steps: Based on the invasive annotation of sample images, the sample images are divided into different invasive groups, including a non-invasive group, a micro-invasive group, and a strong-invasive group. The strong-invasive group is the set of sample images with an invasive annotation of invasive. For any infiltration group, number the sample images therein, and represent them by the symbol P m where m is a positive integer and m is the serial number of P. Set natural numbers i and j, where i is any one of the digits in m, and the value range of j is the same as that of m but j≠i; For any i, obtain P i and P j 's edge features. Mark the histogram in the edge features of P i as the first histogram, and mark the histogram in the edge features of P j as the second histogram. Merge the edge features of P i and P j to obtain the area of the first histogram, the area of the second histogram, and the area of the overlapping part between the first histogram and the second histogram, which are respectively marked as S1, S2, and S3; Calculate (S3 / S1 + S3 / S2) / 2, mark the calculation result as the edge similarity, and P i Analyze with each P j to calculate and obtain max(m)-1 edge similarities, find the maximum value among them, and mark it as the maximum edge similarity. Here, max() is the maximum value operator. Analyze each value of i to obtain max(m) maximum edge similarities, and find the minimum value among them, which is marked as the edge similarity limit value; For any i, obtain P i With P j For the morphological features, enlarge the morphological features with a smaller resolution until the resolution is the same as that of the morphological features with a larger resolution. For P i With P j In the morphological features, mark the tumor pixel points as the first pixel point and the second pixel point respectively, and count the number of the first pixel point, the second pixel point, and the overlapping pixel points of the first pixel point and the second pixel point, and mark them as N1, N2, and N3 in sequence; Calculate (N3 / N1 + N3 / N2) / 2, mark the calculation result as the morphological similarity, and mark P i Analyze with each P j to calculate and obtain max(m)-1 morphological similarities, find the maximum value among them, mark it as the maximum morphological similarity. Analyze for each value of i to obtain max(m) maximum morphological similarities, and find the minimum value among them, mark it as the morphological similarity limit value; For any i, obtain P i and P j 's density characteristics, which are respectively marked as K1 and K2. If K1 ≥ K2, then calculate K2 / K1. If K1 < K2, then calculate K1 / K2. Mark the calculation result as the density similarity. For P i and each P j conduct an analysis, calculate to obtain max(m) - 1 density similarities, find the maximum value among them, and mark it as the maximum density similarity. Conduct an analysis for each value of i to obtain max(m) maximum density similarities, and find the minimum value among them, which is marked as the density similarity limit value; Based on the edge similarity limit, morphological similarity limit, and density similarity limit, deep learning is performed for the invasive identification of lung tumors.

8. The method for identifying infiltrative images of lung tumors based on deep learning according to claim 7, wherein Based on the edge similarity limit, morphological similarity limit, and density similarity limit, deep learning for the invasive identification of lung tumors includes the following sub-steps: The deep learning is for the analysis process of the edge similarity limit, morphological similarity limit, and density similarity limit, and is used to continuously update the sample images and the edge similarity limit, morphological similarity limit, and density similarity limit. The edge similarity limit includes a non-invasive edge similarity limit, a micro-invasive edge similarity limit, and a strong-invasive edge similarity limit, which are represented by the symbols EA, EB, and EC in sequence; the morphological similarity limit includes a non-invasive morphological similarity limit, a micro-invasive morphological similarity limit, and a strong-invasive morphological similarity limit, which are represented by the symbols FA, FB, and FC in sequence; the density similarity limit includes a non-invasive density similarity limit, a micro-invasive density similarity limit, and a strong-invasive density similarity limit, which are represented by the symbols DA, DB, and DC in sequence. Obtain the medical image of the patient, mark it as the image to be identified, and extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be identified in the non-invasive group, which are marked as GAE, GAF, and GAD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be identified in the micro-invasive group, which are marked as GBE, GBF, and GBD respectively; extract the maximum edge similarity, maximum morphological similarity, and maximum density similarity of the image to be identified in the strong-invasive group, which are marked as GCE, GCF, and GCD respectively. Calculate |GAE - EA|, |GAF - FA|, and |GAD - DA| respectively, and mark the calculation results as A1, A2, and A3 respectively; calculate |GBE - EB|, |GBF - FB|, and |GBD - DB| respectively, and mark the calculation results as B1, B2, and B3 respectively; calculate |GCE - EC|, |GCF - FC|, and |GCD - DC| respectively, and mark the calculation results as C1, C2, and C3 respectively. Calculate A1 + A2 + A3 = QA, calculate B1 + B2 + B3 = QB, calculate C1 + C2 + C3 = QC, where QA is the non-invasive deviation, QB is the micro-invasive deviation, and QC is the strong-invasive deviation. Compare the magnitudes of QA, QB, and QC, find the minimum value among them, mark it as the minimum deviation, and output the invasive annotation corresponding to the minimum deviation as the invasive annotation of the image to be identified.

9. A deep learning-based pulmonary tumor invasive image recognition system for implementing the deep learning-based pulmonary tumor invasive image recognition method according to any one of claims 1-8, characterized in that, It includes an image collection module, a region segmentation module, a feature extraction module, and a deep learning module; the image collection module, the region segmentation module, and the deep learning module are respectively connected to the feature extraction module for data connection; The image collection module is used to obtain medical images of lung tumors and simultaneously collect well-annotated sample images; The region segmentation module is used to perform region segmentation on the lung tumors based on the annotation results to obtain tumor regions; The feature extraction module is used to extract the edge features, morphological features, and density features of the tumor regions; The deep learning module is used to perform deep learning on the identification of the invasiveness of lung tumors based on the edge features, morphological features, and density features of the sample images.

Images