A lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning

Abstract

The application provides a lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning, comprising: a collection module that acquires a pathological section image of a patient; a processing module that segments the pathological section image into a plurality of sub-images, analyzes each sub-image respectively to obtain corresponding depth information and semantic information, and clusters the sub-images with the same depth information to obtain a plurality of image classes; a comparison module that compares the semantic information of the sub-images in the same image class with different preset information samples respectively to obtain the pathological features corresponding to each sub-image; and an execution module that marks the pathological features on the pathological section image to generate an analysis report and transmit the report to a designated terminal for display. The pathological section image of the patient is segmented through semantic segmentation, and then the segmented image is compared using a weakly supervised learning technique, so that the lung pathological features of the patient can be obtained to generate an analysis report for reference by medical staff.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent diagnosis, in particular to a lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning. BACKGROUND

[0002] Lung cancer is the first malignant tumor in terms of mortality in China, and the incidence continues to rise. In 2020, there were 815,600 new lung cancer patients (accounting for 17.9% of new cancer patients), and 710,000 lung cancer patients died (accounting for 23.8% of the total number of cancer deaths) in that year. Surgery and chemotherapy are important treatment methods for lung cancer patients, but postoperative pathological adverse prognostic factors are crucial for the selection of chemotherapy programs, which can directly determine the patient's chemotherapy cycle and whether to perform postoperative radiotherapy. The prognosis of lung cancer does not strictly follow the tumor stage, and progression occurs in the short term. If the early intervention-related treatment is performed, the progression will be delayed, achieving the purpose of prevention.

[0003] Therefore, the present application provides a lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning. SUMMARY

[0004] The lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning of the present application segments the pathological section image of the patient through semantic segmentation, and then compares the segmented image using weakly supervised learning technology, so as to obtain the lung pathological characteristics of the patient, thereby generating an analysis report for reference by medical staff.

[0005] The present application provides a lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning, comprising:

[0006] The acquisition module is configured to obtain the pathological section image of the patient.

[0007] The processing module is configured to segment the pathological section image into a plurality of sub-images, analyze each sub-image respectively to obtain corresponding depth information and semantic information, cluster the sub-images with the same depth information to obtain a plurality of image classes.

[0008] The comparison module is configured to compare the semantic information of the sub-images in the same image class with different preset information samples respectively to obtain the pathological characteristics corresponding to each sub-image.

[0009] The execution module is configured to mark the pathological characteristics on the pathological section image to generate an analysis report and transmit the analysis report to a designated terminal for display.

[0010] In an implementable manner,

[0011] The acquisition module comprises:

[0012] An information checking unit is configured to acquire basic information of the patient, find historical information of the patient in a historical record, and determine whether the basic information is consistent with the historical information.

[0013] An information processing unit is configured to, when the basic information of the patient is consistent with the historical information, call a disease history of the patient, and determine a diseased organ of the patient.

[0014] A scanning execution unit is configured to perform a computer tomography scan on the patient, and perform a secondary scan on the diseased organ to obtain a pathological slice image.

[0015] In an implementable manner,

[0016] The processing module comprises:

[0017] A segmentation unit is configured to acquire a plurality of image images contained in the pathological slice image, acquire an image specification corresponding to each image image respectively, match a corresponding segmentation rule according to the image specification, segment the corresponding image image based on the segmentation rule, and obtain a plurality of sub-images.

[0018] A processing unit is configured to perform down-sampling on each sub-image respectively to obtain a corresponding sub-image, perform edge detection on the sub-image, mark a corresponding image edge on each sub-image to generate a trajectory image, and obtain depth information of each sub-image according to the trajectory image.

[0019] A classification unit is configured to mark sub-images with the same depth information as a class, and obtain a plurality of image classes.

[0020] In an implementable manner,

[0021] The processing unit is further configured to acquire a pixel distribution corresponding to each sub-image, establish a pixel matrix based on the pixel distribution, input all elements in the pixel matrix into a preset semantic layering model to perform pixel layering, obtain a semantic layer corresponding to each pixel, and establish semantic information.

[0022] In an implementable manner,

[0023] The comparison module comprises:

[0024] A preprocessing unit is configured to establish a depth label for each sub-image in a corresponding image class according to the depth information corresponding to each image class, and establish a semantic label according to the semantic information corresponding to each sub-image in the same image class.

[0025] The comparison unit is used to establish supervision information based on the semantic and depth labels corresponding to the same sub-image, and to traverse the supervision information using preset information samples to obtain the similarity between each sub-image and different sample images.

[0026] The parsing unit is used to generate pathological features corresponding to each sub-image based on the similarity between each sub-image and different sample images, combined with the sample features corresponding to each sample image.

[0027] In one feasible approach

[0028] The execution module includes:

[0029] The marking unit is used to obtain the marking position of each pathological feature based on the image position of each sub-image on the pathological slide image, and to mark the pathological feature on the pathological slide image to obtain a marked pathological slide image.

[0030] The analysis unit is used to generate tissue structure data based on the labeled pathological slide image, filter out a number of abnormal data from the tissue structure data, search for the patient's lung area in historical information based on the ratio of the number of abnormal data to the number of tissue structure data, and obtain the abnormal lung area of ​​the patient by combining the ratio.

[0031] The execution unit is configured to acquire the feature set within the nine-square grid range of each pathological feature on the labeled pathological slide image, determine whether the feature set belongs to the abnormal feature set, enhance the contrast of the abnormal feature set on the labeled pathological slide image if it is, acquire the area of ​​the region corresponding to all enhanced regions on the labeled pathological slide image, and determine that all abnormal feature sets have been extracted when the area of ​​the region is consistent with the abnormal area of ​​the lung; otherwise, continue to extract the abnormal feature sets on the labeled pathological slide image.

[0032] The generation unit is used to acquire the tissue slice corresponding to each enhanced region on the labeled pathological slide image, generate a tissue abnormality analysis report for the patient, and transmit it to a designated terminal for display.

[0033] In one feasible approach

[0034] The analysis unit includes:

[0035] The first analysis subunit is used to filter out several abnormal data from the organizational structure data;

[0036] The second analysis subunit is used to determine that there are no abnormalities in the patient's lungs when the number of abnormal data is 0, generate a tissue normal analysis report, and transmit it to a designated terminal for display.

[0037] The third analysis subunit is configured to, when the number of the abnormal data is not 0, acquire a first data amount corresponding to the abnormal data and a second data amount corresponding to the organizational structure data, and obtain the lung abnormal area of the patient according to a ratio of the first data amount and the second data amount.

[0038] In an implementable manner,

[0039] The analysis unit comprises:

[0040] The first analysis subunit is configured to, based on the similarity between the same sub-image and different sample images, establish a fusion weight for each sample image respectively, and generate a plurality of sets of sample images to be fused.

[0041] The second analysis subunit is configured to, according to the fusion weight between the same sub-image and each sample image, establish a fusion model, select one sample image to be fused in each set of sample images to be fused respectively, and input the sample image to be fused into the fusion model for image fusion to obtain a plurality of fused images.

[0042] The third analysis subunit is configured to acquire an image feature corresponding to each fused image respectively, and extract a target image feature with the highest occurrence frequency.

[0043] The fourth analysis subunit is configured to, according to the target image feature, establish a pathological feature corresponding to each sub-image.

[0044] In an implementable manner,

[0045] The analysis module further comprises:

[0046] The analysis module is configured to acquire a historical report of the patient, combine the analysis report to obtain a body change trend range of the patient, and transmit the body change trend range to a designated terminal for display.

[0047] In an implementable manner,

[0048] The analysis module comprises:

[0049] The first analysis unit is configured to acquire and analyze the historical report of the patient to obtain a disease symptom corresponding to each period of the patient.

[0050] The second analysis unit is configured to establish a body constitution change graph of the patient according to the disease symptom corresponding to each period of the patient, and predict a body change trend range of the patient according to the body constitution change graph.

[0051] The third analysis unit is configured to analyze the analysis report to obtain a current body constitution of the patient, determine whether the current body constitution of the patient is within the body change trend range, and if not, determine that the patient has an abnormal body condition.

[0052] The fourth analysis unit is configured to generate a prompt message when the determined physical condition of the patient is abnormal, and transmit the prompt message to the designated terminal for display.

[0053] The present application can achieve the beneficial effects that: in order to analyze the prognosis of the patient's lung condition, the pathological section image of the patient is first collected, and then the pathological section image is segmented, so as to enhance the analysis accuracy of the pathological section image, then each sub-image is analyzed, and the sub-image is classified according to the depth information, then the semantic information of the sub-image in the same category is compared with the preset information sample, so as to obtain the pathological characteristics of each sub-image, and then the pathological characteristics can be marked on the pathological section image, thereby generating an analysis report for reference by the doctor, so that the prognosis of the patient can be determined through the analysis report, and the doctor can give corresponding medical guidance.

[0054] Additional features and advantages of the application will be set forth in the description that follows, and in part will be apparent from the description, or can be learned by practice of the application. The objectives and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

[0055] The technical solutions of the present application will be further described in detail below with the help of the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0056] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and serve to explain the application without limiting the application. In the drawings:

[0057] Figure 1 It is a schematic diagram of the composition of a lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning in an embodiment of the present application.

[0058] Figure 2 It is a schematic diagram of the operation that a doctor can perform after the execution of embodiment 1 of a lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning in an embodiment of the present application.

[0059] Figure 3 It is a schematic diagram of the composition of a processing module of a lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning in an embodiment of the present application. DETAILED DESCRIPTION

[0060] The preferred embodiments of the present application will be described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0061] Embodiment 1

[0062] This embodiment provides a lung cancer prognostic analysis system based on semantic segmentation and weakly supervised learning, such as Figure 1 As shown, it includes:

[0063] The acquisition module is used to acquire images of the patient's pathological slides;

[0064] The processing module is used to segment the pathological slide image into several sub-images, parse each sub-image to obtain the corresponding depth information and semantic information, and cluster the sub-images with the same depth information to obtain several image classes;

[0065] The comparison module is used to compare the semantic information corresponding to sub-images in the same image class with different preset information samples to obtain the pathological features corresponding to each sub-image.

[0066] The execution module is used to mark the pathological features on the pathological slide image, generate an analysis report, and transmit it to a designated terminal for display.

[0067] In this example, depth information represents the number of bits used to store each pixel in a sub-image;

[0068] In this example, semantic information represents the image content reflected by a sub-image in the visual layer, object layer, and conceptual layer, respectively;

[0069] In this example, the image class represents a collection of images consisting of sub-images with the same depth information;

[0070] In this example, the pathological feature represents the shadows in the lung image on the sub-image;

[0071] In this case, adverse prognoses for lung cancer patients include: micropapillary tumors, solid tumors, airway dissemination, and vascular tumor emboli.

[0072] In this example, the purpose of clustering sub-images with the same depth information is to group images of the same lung region of the patient on the pathological slide image into one category. For example, if there are 3 images on the pathological slide image that are about the second-level bronchus of the patient's left lung, then the images of the second-level bronchus of the patient's left lung on the pathological slide image have the same depth, so the 3 images on the pathological slide image can be grouped into one category.

[0073] In this example, based on the full-scan pathological images of postoperative specimens from stage I lung cancer patients, the system can automatically analyze their key clinical features, predict and quantify key gene mutations such as EGFR, ALK, KRAS, BRAF, and TP53, and simultaneously collect information such as clinical treatment and prognostic results and pTNM staging. The system can also describe the impact of different factors on patient prognosis using a proportional hazards regression model.

[0074] As Figure 2 shown, after completing the example, (a) four prognostic high-risk morphological factors (micro-papilla, solid, airway dissemination and vascular embolus) can be automatically identified by a supervised learning model DeepLab v3, improving efficiency and assisting clinicians in developing better treatment strategies; (b) through the weakly supervised learning technology CAMEL developed by the applicant, the internal relationship between lung tissue morphological changes and lung cancer gene mutations can be found, and only cases diagnosed by doctors are subjected to molecular detection to reduce medical expenses.

[0075] The working principle and beneficial effects of the above technical solutions are as follows: In order to analyze the prognosis of the patient's lung, the pathological section image of the patient is first collected, and then the pathological section image is segmented, so that the analysis accuracy of the pathological section image can be improved. Then analyze each sub-image, classify the sub-image according to its depth information, then compare the semantic information of the sub-image in the same class with the preset information sample, so as to obtain the pathological characteristics of each sub-image, and then mark the pathological characteristics on the pathological section image to generate an analysis report for the doctor to refer to. In this way, the prognosis of the patient can be determined through the analysis report, and the corresponding medical guidance can be given by the doctor.

[0076] Embodiment 2

[0077] Based on embodiment 1, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning, the collection module comprises:

[0078] An information checking unit is configured to obtain the basic information of the patient, find the historical information of the patient in the historical record, and determine whether the basic information and the historical information are consistent.

[0079] An information processing unit is configured to, when the basic information and the historical information of the patient are consistent, retrieve the medical history of the patient and determine the diseased organ of the patient.

[0080] A scanning execution unit is configured to perform computed tomography on the patient and perform secondary scanning on the diseased organ to obtain a pathological section image.

[0081] In this example, the basic information includes the patient's name and patient number.

[0082] In this example, the historical record contains the basic information of each postoperative patient.

[0083] In this example, the diseased organ represents the device where the patient's lung is diseased.

[0084] The working principle and beneficial effects of the technical solution are as follows: in order to further analyze the lung condition of the patient and avoid information disorder, the basic information of the patient is collected first, then information checking is performed, the patient's medical history is extracted after the checking is successful, so that the diseased organ can be obtained, and then the diseased organ is scanned again during scanning, the information of the diseased organ can be collected from multiple directions, and accurate scanning is realized.

[0085] Embodiment 3

[0086] Based on the embodiment 1, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning comprises a processing module, a classification module and a display module. Figure 3 The processing module comprises a segmentation unit and a processing unit.

[0087] The segmentation unit is configured to acquire a plurality of image images contained in the pathological section image, acquire an image specification corresponding to each image image respectively, match a corresponding segmentation rule according to the image specification, segment the corresponding image image based on the segmentation rule, and obtain a plurality of sub-images.

[0088] The processing unit is configured to perform down-sampling on each sub-image to obtain a corresponding sub-image, perform edge detection on the sub-image, mark a corresponding image edge on each sub-image to generate a track image, establish a pixel distribution map of each sub-image, obtain depth information of each sub-image according to the track image, and establish a pixel matrix according to the pixel distribution map, input all elements in the pixel matrix into a preset semantic layering model to perform pixel layering, obtain a semantic layer corresponding to each pixel, and establish semantic information.

[0089] The classification unit is configured to mark sub-images with the same depth information as a class, and obtain a plurality of image classes.

[0090] In this example, the image image represents an image of the pathological section image, which is a part of the pathological section image.

[0091] In this example, the image specification represents the length and width specification of the image image.

[0092] In this example, the segmentation rule represents a rule for dividing an image image into a plurality of sub-images, and a general rule is to divide the image image into equal-area rectangles according to the image specification.

[0093] In this example, down-sampling represents a compression operation on the sub-image, and a sub-image, i.e. a thumbnail of the image, can be generated through down-sampling, so that the image conforms to the size of the display area, the dimension of the features of the sub-image can be reduced, and the edges in the image can be determined.

[0094] In this example, the edge detection represents a process of marking pixels in the sub-sampled image whose intensity changes in a preset range;

[0095] In this example, the image edge represents a place where the region attribute in the sub-sampled image changes abruptly;

[0096] In this example, the trajectory map represents a trajectory where the image edge changes, that is, a texture trajectory of different regions of the lung;

[0097] The working principle and beneficial effects of the above technical solutions are as follows: in order to realize image classification, first, the pathological section image is segmented to obtain a plurality of sub-images, then the sub-images are down-sampled to obtain thumbnail sub-images, the sub-images are edge detected, and the detected edges are marked on the sub-images, so that a trajectory map of the sub-images can be generated, the depth information of the sub-images can be obtained through the trajectory map, and finally the sub-images can be classified according to the depth information, thereby realizing image classification and facilitating subsequent image comparison.

[0098] Embodiment 4

[0099] On the basis of embodiment 3, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning comprises:

[0100] The processing unit is further configured to obtain a pixel distribution corresponding to each sub-image, establish a pixel matrix based on the pixel distribution, input all elements in the pixel matrix into a preset semantic layering model to perform pixel layering, obtain a semantic layer corresponding to each pixel, and establish semantic information.

[0101] In this example, the pixel matrix represents a matrix established according to pixel values of different pixels as elements;

[0102] In this example, the semantic layer comprises a visual layer, an object layer, and a concept layer.

[0103] In this example, the pixel distribution map represents a respective case of different pixel values in the sub-sampled image;

[0104] The working principle and beneficial effects of the above technical solutions are as follows: according to the pixel distribution of the sub-sampled image, a pixel matrix is established, and finally a preset semantic layering model is used for semantic layering to obtain semantic information of the image, thereby facilitating subsequent image comparison.

[0105] Embodiment 5

[0106] On the basis of embodiment 1, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning comprises:

[0107] a preprocessing unit configured to establish a depth label for each sub-image in a corresponding image class according to depth information corresponding to the image class, and establish a semantic label according to semantic information corresponding to each sub-image in the same image class;

[0108] a comparison unit configured to establish supervision information according to the semantic label and the depth label corresponding to the same sub-image, traverse the supervision information using a preset information sample, and obtain a similarity between each sub-image and a different sample image;

[0109] an analysis unit configured to generate a pathological feature corresponding to each sub-image according to the similarity between each sub-image and a different sample image and a sample feature corresponding to each sample image.

[0110] In this example, the depth label represents a label used to distinguish different image classes, that is, the depth label of an image class is the same;

[0111] In this example, the semantic label represents a label used to distinguish different sub-images in an image class;

[0112] In this example, the supervision information represents information that trains the depth label and the semantic label of a sub-image into a format that can be used for supervised learning;

[0113] In this example, the sample image corresponds to the preset information sample one by one, and each sample image has a sample feature.

[0114] The working principle and beneficial effects of the above technical solution are as follows: the supervision information of the sub-image is determined by establishing the label format for each sub-image, then the preset information sample is used for comparison to obtain the similarity between the sub-image and the sample image, and then the sample feature of each sample image is combined to generate the pathological feature of each sub-image, which is the basis for subsequent feature labeling.

[0115] Embodiment 6

[0116] Based on the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning in Embodiment 1, the execution module comprises:

[0117] a labeling unit configured to obtain a labeling position of each pathological feature according to an image position of each sub-image on the pathological section image, label the pathological feature on the pathological section image to obtain a labeled pathological section image;

[0118] an analysis unit configured to generate tissue structure data according to the labeled pathological section image, filter a plurality of abnormal data from the tissue structure data, find a lung area of the patient in historical information according to a quantity ratio of the abnormal data to the tissue structure data, and obtain an abnormal lung area of the patient in combination with the quantity ratio.

[0119] The execution unit is configured to acquire a feature set in a nine-square grid range where each pathological feature is located on the labeled pathological section image, determine whether the feature set belongs to an abnormal feature set, if yes, enhance the contrast of the abnormal feature set on the labeled pathological section image, acquire a region area corresponding to all enhanced regions on the labeled pathological section image, and determine that extraction of all abnormal feature sets is completed when the region area is consistent with the lung abnormal area, otherwise, continue to extract the abnormal feature set on the labeled pathological section image.

[0120] The generation unit is configured to acquire a section tissue corresponding to each enhanced region on the labeled pathological section image, establish a tissue abnormality analysis report for the patient, and transmit the report to a designated terminal for display.

[0121] In this example, the image position and the marker position are one-to-one corresponding;

[0122] In this example, the tissue structure data includes tissue area data, tissue texture data, tissue shadow data, and tissue smoothness data.

[0123] In this example, the lung abnormal area represents the area of the patient's lung lesion.

[0124] In this example, when a pathological feature is at the edge of the labeled pathological section image, a preset standard feature is used to fill in the blank squares of the nine-square grid when acquiring the feature set.

[0125] In this example, the purpose of enhancing the contrast of the abnormal feature on the labeled pathological section image is to highlight the abnormal feature on the labeled pathological section image.

[0126] In this example, when the region area is consistent with the lung abnormal area, it means that all abnormalities are marked on the labeled pathological section image.

[0127] The working principle and beneficial effects of the above technical solution are as follows: In order to further determine the patient's disease condition, first mark the pathological features on the pathological section image to generate tissue structure data, then filter out abnormal data from the tissue structure data, determine the lung abnormal area of the patient according to the quantity ratio of the abnormal data and the tissue structure data, further analyze using the feature set, highlight the abnormal feature set on the labeled pathological section image, determine the lung abnormal position, use the quantity information of the lung abnormal area and the region area to determine whether all abnormal regions have been found, and finally generate a lung abnormality analysis report and transmit it to the terminal for reference by the doctor.

[0128] Embodiment 7

[0129] On the basis of embodiment 1, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning, the analysis unit comprises:

[0130] The first analysis subunit is used for screening a plurality of abnormal data from the tissue structure data;

[0131] The second analysis subunit is used for determining that the patient's lung does not have abnormalities when the number of abnormal data is 0, generating a normal tissue analysis report, and transmitting the report to a designated terminal for display;

[0132] The third analysis subunit is used for obtaining a first data amount corresponding to the abnormal data and a second data amount corresponding to the tissue structure data when the number of abnormal data is not 0, and obtaining the patient's lung abnormal area by combining the first data amount and the second data amount.

[0133] The working principle and beneficial effects of the above technical solution are as follows: the abnormal data is screened to preliminarily determine whether the patient's lung has abnormalities, and when the number of abnormal data is 0, it is determined that the patient's lung is healthy, otherwise it indicates that the patient's lung has abnormalities, and then the ratio of the first data amount of the abnormal data to the second data amount of the tissue structure data is used to determine the patient's lung abnormal area, thereby laying a foundation for subsequent judgment of the integrity of the area.

[0134] Embodiment 8

[0135] On the basis of embodiment 5, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning, the analysis unit comprises:

[0136] The first analysis subunit is used for establishing a fusion weight for each sample image based on the similarity between the same sub-image and different sample images, and generating a plurality of to-be-fused sample image sets;

[0137] The second analysis subunit is used for establishing a fusion model according to the fusion weight between the same sub-image and each sample image, selecting a to-be-fused sample image in each to-be-fused sample image set, and inputting the to-be-fused sample image into the fusion model for image fusion to obtain a plurality of fused images;

[0138] The third analysis subunit is used for obtaining an image feature corresponding to each fused image, and extracting a target image feature with the highest frequency of occurrence;

[0139] The fourth analysis subunit is used for establishing a pathological feature corresponding to each sub-image according to the target image feature.

[0140] In this example, the fusion weight represents a parameter for fusing sample images, and the higher the fusion weight, the higher the fusion degree.

[0141] In this example, the set of sample images to be fused represents similar images generated according to the similarity between the sub-image and different sample images, and the set of sample images to be fused is composed of similar images.

[0142] In this example, the weight fusion model represents a model for mutual fusion of images established according to the fusion weight between the sub-image and different sample images, wherein the fusion ratio is consistent with the fusion weight.

[0143] The working principle and beneficial effects of the above technical solution are as follows: In order to obtain the pathological characteristics of each sub-image, the fusion weight of the sample image is determined according to the similarity between the sub-image and the sample image, and the corresponding set of sample images to be fused is generated, then the fusion model is established according to the fusion weight, and the images in the set of sample images are fused, a plurality of fusion images can be obtained, and finally the image features of the fusion images are used to reflect the pathological characteristics of the corresponding sub-image from the side, which provides a basis for subsequent feature labeling and generation of analysis reports.

[0144] Embodiment 9

[0145] Based on the embodiment 1, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning further comprises:

[0146] The analysis module is configured to obtain the historical report of the patient, combine the analysis report, obtain the body change trend range of the patient, and transmit the body change trend range to a designated terminal for display.

[0147] The working principle and beneficial effects of the above technical solution are as follows: In order to achieve the purpose of prognosis analysis, the patient's historical report and analysis report are used for overall analysis, so as to judge the change trend of the patient's body, and the doctor can make a more effective treatment plan.

[0148] Embodiment 10

[0149] Based on the embodiment 9, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning, the analysis module comprises:

[0150] The first analysis unit is configured to obtain and parse the historical report of the patient to obtain the corresponding disease symptoms of the patient at different periods.

[0151] The second analysis unit is configured to establish a patient body quality change graph according to the corresponding disease symptoms of the patient at different periods, and predict the body change trend range of the patient according to the body quality change graph.

[0152] A third analysis unit is configured to analyze the analysis report to obtain the current physical condition of the patient, determine whether the current physical condition of the patient is within the physical change trend range, and determine that the physical condition of the patient is abnormal if the current physical condition of the patient is not within the physical change trend range.

[0153] A fourth analysis unit is configured to generate a prompt message when it is determined that the physical condition of the patient is abnormal, and transmit the prompt message to the specified terminal for display.

[0154] In this example, the disease symptoms represent the corresponding diseases when the patient's physical condition is abnormal at different times;

[0155] In this example, when the current physical condition of the patient is not within the physical change trend range, the physician needs to determine whether the current physical condition of the patient belongs to recovery or deterioration.

[0156] The working principle and beneficial effects of the above technical solution are as follows: in order to further analyze the prognosis of the patient, the current physical condition of the patient is analyzed by judging the current physical condition of the patient and the predicted physical change trend range, and then a corresponding prompt is transmitted to the terminal for reference by the physician.

[0157] Embodiment 11

[0158] Based on the embodiment 8, the lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning comprises:

[0159] The first analysis subunit is further configured to obtain the pixel value corresponding to each pixel point in each of the sub-images, and calculate the signal-to-noise ratio of each sub-image according to formula (1).

[0160]

[0161] wherein B k represents the signal-to-noise ratio of the kth sub-image, f represents the dynamic range of the pixel value, generally takes a value of 255, n represents the number of pixel points in the kth sub-image, s ki represents the pixel value corresponding to the i-th pixel point in the kth sub-image.

[0162] According to the calculation result of formula (1), the sub-image to be adjusted whose signal-to-noise ratio is outside the preset signal-to-noise ratio range is extracted, the pixel of the sub-image to be adjusted is corrected to obtain an adjusted sub-image, and the adjusted sub-image is used to replace the corresponding sub-image.

[0163] The replaced adjusted sub-image is regarded as a sub-image, and the image mean and image variance corresponding to each sub-image are obtained.

[0164] The similarity between each sub-image and the sample image is calculated according to formula (2).

[0165]

[0166] wherein D p represents the similarity between the pth sub-image and the sample image, a p represents the image mean of the pth sub-image, a y represents the image mean of the yth sample image, x1 represents the mean correction parameter, d p represents the image variance of the pth sub-image, d y represents the image variance of the yth sample image, x2 represents the variance correction parameter;

[0167] According to the calculation result of formula (2), the similarity between each sub-image and different sample images is obtained.

[0168] The working principle and beneficial effects of the above technical solution are as follows: whether the image meets the comparison principle is determined by analyzing the signal-to-noise ratio of the sub-image, the sub-image with unqualified signal-to-noise ratio is corrected, and then the similarity between the sub-image and different sample images is determined. In this way, the interference of image noise points on the comparison result can be avoided, and the accuracy of subsequent generation of pathological features is improved.

[0169] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and variations.

Claims

1. A lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning, characterized in that, include: The acquisition module is used to acquire images of the patient's pathological slides; The processing module is used to segment the pathological slide image into several sub-images, parse each sub-image to obtain the corresponding depth information and semantic information, and cluster the sub-images with the same depth information to obtain several image classes; The comparison module is used to compare the semantic information corresponding to sub-images in the same image class with different preset information samples to obtain the pathological features corresponding to each sub-image. The execution module is used to mark the pathological features on the pathological slide image, generate an analysis report, and transmit it to a designated terminal for display. The execution module includes: The marking unit is used to obtain the marking position of each pathological feature based on the image position of each sub-image on the pathological slide image, and to mark the pathological feature on the pathological slide image to obtain a marked pathological slide image. The analysis unit is used to generate tissue structure data based on the labeled pathological slide image, filter out a number of abnormal data from the tissue structure data, search for the patient's lung area in historical information based on the ratio of the number of abnormal data to the number of tissue structure data, and obtain the abnormal lung area of ​​the patient by combining the ratio. The execution unit is configured to acquire the feature set within the nine-square grid range of each pathological feature on the labeled pathological slide image, determine whether the feature set belongs to the abnormal feature set, enhance the contrast of the abnormal feature set on the labeled pathological slide image if it is, acquire the area of ​​the region corresponding to all enhanced regions on the labeled pathological slide image, and determine that all abnormal feature sets have been extracted when the area of ​​the region is consistent with the abnormal area of ​​the lung; otherwise, continue to extract the abnormal feature sets on the labeled pathological slide image. The generation unit is used to acquire the tissue slice corresponding to each enhanced region on the labeled pathological slide image, generate a tissue abnormality analysis report for the patient, and transmit it to a designated terminal for display. Also includes: The analysis module is used to obtain the patient's historical reports, combine the analysis reports to obtain the range of the patient's physical change trend, and transmit it to a designated terminal for display; The analysis module includes: The first analysis unit is used to acquire and parse the patient's historical reports to obtain the patient's symptoms at different times. The second analysis unit is used to establish a physical fitness change map of the patient based on the disease symptoms corresponding to different periods of the patient, and to predict the range of the patient's physical change trend based on the physical fitness change map. The third analysis unit is used to parse the analysis report, obtain the patient's current physical condition, determine whether the patient's current physical condition is within the range of physical change trends, and if not, determine that the patient's physical condition is abnormal. The fourth analysis unit is used to generate an alert message and transmit it to the designated terminal for display when it is determined that the patient's physical condition is abnormal. 2.The lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning of claim 1, wherein, The acquisition module includes: An information checking unit is configured to acquire basic information of the patient, find historical information of the patient in a historical record, and determine whether the basic information is consistent with the historical information; An information processing unit is configured to, when the basic information of the patient is consistent with the historical information, call a disease history of the patient, and determine a diseased organ of the patient; A collection execution unit is configured to collect a pathological section image of the patient. 3.The lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning of claim 1, wherein, The processing module comprises: A segmentation unit is configured to acquire a plurality of image images contained in the pathological section image, acquire an image specification corresponding to each image image, match a corresponding segmentation rule according to the image specification, segment the corresponding image image based on the segmentation rule, and obtain a plurality of sub-images; A processing unit is configured to perform down-sampling on each sub-image to obtain a corresponding sub-image, perform edge detection on the sub-image, mark a corresponding image edge on each sub-image to generate a trajectory image, and obtain depth information of each sub-image according to the trajectory image; A classification unit is configured to mark sub-images with the same depth information as a class, and obtain a plurality of image classes.

4. The lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning according to claim 3, wherein: The processing unit is further configured to acquire a pixel distribution corresponding to each sub-image, establish a pixel matrix based on the pixel distribution, input all elements in the pixel matrix into a preset semantic layering model to perform pixel layering, obtain a semantic layer corresponding to each pixel, and establish semantic information. 5.The lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning of claim 1, wherein, The comparison module comprises: A preprocessing unit is configured to establish a depth label for each sub-image in a corresponding image class according to depth information corresponding to the image class, and establish a semantic label according to semantic information corresponding to each sub-image in the same image class; A comparison unit is configured to establish supervision information according to the semantic label and the depth label corresponding to the same sub-image, traverse the supervision information by using a preset information sample, and obtain a similarity between each sub-image and a different sample image; An analysis unit is configured to generate a pathological feature corresponding to each sub-image according to the similarity between each sub-image and the different sample image, and combine a sample feature corresponding to each sample image.

6. The lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning of claim 1, wherein, The analysis unit comprises: A first analysis sub-unit is configured to filter a plurality of abnormal data from the tissue structure data; A second analysis sub-unit is configured to, when the number of abnormal data is 0, determine that the patient's lung does not have an abnormality, generate a normal tissue analysis report, and transmit the report to a designated terminal for display; A third analysis sub-unit is configured to, when the number of abnormal data is not 0, acquire a first data amount corresponding to the abnormal data and a second data amount corresponding to the tissue structure data, a ratio of the first data amount and the second data amount, and a lung area of the patient, and obtain an abnormal lung area of the patient.

7. The lung cancer prognosis analysis system based on semantic segmentation and weakly supervised learning of claim 5, wherein, The analysis unit comprises: A first analysis sub-unit is configured to, based on the similarity between the same sub-image and the different sample image, establish a fusion weight for each sample image, and generate a plurality of to-be-fused sample image sets; a second analysis subunit configured to establish a fusion model according to the fusion weight between the same sub-image and each sample image, select one sample image in each corresponding sample image set to be fused, input the sample image into the fusion model for image fusion, and obtain a plurality of fused images; a third analysis subunit configured to obtain the image features corresponding to each fused image, and extract a target image feature with the highest occurrence frequency of image features; a fourth analysis subunit configured to establish the pathological feature corresponding to each sub-image according to the target image feature.

Images