Esophageal cancer and lung cancer diagnosis CT image lymph node drainage area automatic sketching and lymph node diagnosis model based on deep learning

Abstract

The invention relates to prediction equipment, a prediction method and a computer readable storage medium, relates to the technical field of medical image processing, and can reduce the treatment cost. The prediction device is configured to execute the following steps: determining a lymph node sketching area based on a preoperative medical image containing a lymph node corresponding to a target object; performing feature extraction on the lymph node delineation area to obtain first feature information corresponding to radiomics features, and obtaining second feature information corresponding to clinical features of the target object; and inputting the first feature information and the second feature information into a diagnosis model to obtain postoperative lymph node diagnosis information.

Description

Technical Field

[0001] The present application relates to the technical field of medical image processing, and in particular to a prediction device, a prediction method, an apparatus, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the development of medical imaging technology, medical imaging can provide detailed images of the internal structure of the target object and assist in the processing of various medical tasks.

[0003] In related technologies, after obtaining the medical image of the target object before surgery, the relevant operators mainly use the medical image to analyze the current condition of the target object, and then promptly formulate or adjust a treatment plan that matches the current condition of the target object to improve treatment efficiency.

[0004] However, although the treatment plan formulated by the above-mentioned treatment method matches the current condition of the target object, the effect obtained after the operation may not meet the expectations, resulting in the need to perform surgery on the target object again or multiple times, which significantly increases the manpower and material costs. Summary of the invention

[0005] Based on this, it is necessary to provide a prediction device, prediction method, apparatus, computer equipment, computer-readable storage medium and computer program product to address the above technical problems.

[0006] In a first aspect, the present application provides a prediction device,

[0007] The prediction device is configured to perform the following steps:

[0008] Determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object;

[0009] Extracting features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtaining second feature information corresponding to the clinical feature of the target object;

[0010] The first characteristic information and the second characteristic information are input into a diagnosis model to obtain postoperative lymph node diagnosis information.

[0011] In one embodiment, the prediction device is trained to obtain the diagnosis model through the following steps:

[0012] Obtaining a lymph node delineation result of a preoperative medical image sample corresponding to the sample object, extracting third feature information of the lymph node delineation result under multiple candidate imaging genomics features, and obtaining fourth feature information of the sample object under multiple candidate clinical features;

[0013] According to the combination of multiple feature information of each of the third feature information and each of the fourth feature information, multiple predictions are made on the postoperative lymph node abnormality information, and according to the multiple prediction results, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features;

[0014] The diagnostic model is trained using the third feature information of the sample object under the key imaging features and the fourth feature information under the key clinical features.

[0015] In one embodiment, the postoperative lymph node abnormality information is predicted multiple times based on the combination of multiple feature information of each of the third feature information and each of the fourth feature information, and key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features based on the multiple prediction results, including:

[0016] Determine current feature weight coefficients of various candidate imaging features and various candidate clinical features, and predict the probability of postoperative lymph node abnormality based on each of the feature weight coefficients, each of the third feature information, and each of the fourth feature information to obtain a prediction result;

[0017] Determine a loss value according to the difference between the prediction result and the postoperative lymph node metastasis information of the sample object, and the sum of the coefficients of each of the feature weight coefficients;

[0018] According to the loss value, at least part of each of the feature weight coefficients is adjusted downward, and the step of determining current feature weight coefficients of various candidate radiomics features and various candidate clinical features is returned to be executed until an iteration end condition is met;

[0019] According to the current feature weight coefficients, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features.

[0020] In one embodiment, determining the key radiomics features and key clinical features whose correlations satisfy the correlation conditions from various candidate radiomics features and various candidate clinical features according to the current feature weight coefficients includes:

[0021] Comparing the coefficient values ​​of the current feature weight coefficients, and determining, according to the comparison results, from various candidate imaging features and various candidate clinical features, a plurality of key features whose coefficient values ​​are greater than any other features; the plurality of key features include imaging features and clinical features, and the other features are other features of various candidate imaging features and various candidate clinical features except the key features;

[0022] Based on the multiple key features, key radiomics features and key clinical features whose correlations meet correlation conditions are obtained.

[0023] In one embodiment, the using the third feature information of the sample object under the key radiomics feature and the fourth feature information under the key clinical feature to train the diagnostic model includes:

[0024] Identify multiple machine learning models;

[0025] For each of the machine learning models, the third feature information of the sample object under the key imaging genomics feature and the fourth feature information under the key clinical feature are used to train the machine learning model, and a model training effect corresponding to the machine learning model is obtained; the model training effect represents the accuracy of the prediction result of the machine learning model predicting the probability of postoperative lymph node abnormality using the third feature information and the fourth feature information;

[0026] From the multiple machine learning models, a machine learning model whose model training effect meets the preset training effect screening conditions is screened out as the trained diagnostic model.

[0027] In one embodiment, determining the lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object includes:

[0028] The preoperative medical image is input into the lymph node delineation model, and the lymph node delineation area is output.

[0029] In one embodiment, the prediction device is trained to obtain the lymph node delineation model through the following steps:

[0030] Obtaining a preoperative medical image sample corresponding to the sample object, lymph node annotation information corresponding to the preoperative medical image sample, and a lymph node pathology detection result;

[0031] Acquire prompt information for excluding the correlation between the candidate delineated areas and the lymph node delineation regions that does not meet the preset conditions, and input the prompt information and the preoperative medical image sample into the lymph node delineation model to be trained, and the lymph node delineation model extracts the image features corresponding to the candidate delineated areas in the preoperative medical image sample according to the prompt information, and outputs the lymph node delineation results and the predicted pathological detection results according to the image features;

[0032] According to the difference between the lymph node delineation result and the lymph node annotation information and the difference between the predicted pathological result and the lymph node pathological detection result, the model parameters of the lymph node delineation model are adjusted to obtain a trained lymph node delineation model.

[0033] In one embodiment, the preoperative medical image sample includes the target structure tissue, lymph nodes within a preset range of the target structure tissue, and adjacent structure tissues outside the target structure tissue;

[0034] The prediction device obtains the prompt information through the following steps:

[0035] Acquiring clinical prior information indicating abnormal conditions of target structural tissue of the sample object, and acquiring anatomical structural restriction information of the adjacent structural tissue;

[0036] The prompt information is obtained according to the clinical prior information and the anatomical structure restriction information.

[0037] In a second aspect, the present application also provides a prediction method, the method comprising:

[0038] Determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object;

[0039] Extracting features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtaining second feature information corresponding to the clinical feature of the target object;

[0040] The first characteristic information and the second characteristic information are input into a diagnosis model to obtain postoperative lymph node diagnosis information.

[0041] In a third aspect, the present application also provides a prediction device, comprising:

[0042] A lymph node delineation module, for determining a lymph node delineation region based on a preoperative medical image containing lymph nodes corresponding to a target object;

[0043] A feature information extraction module, used to extract features of the lymph node delineation area, obtain first feature information corresponding to the radiomics feature, and obtain second feature information corresponding to the clinical feature of the target object;

[0044] The prediction module is used to input the first characteristic information and the second characteristic information into the diagnosis model to obtain postoperative lymph node diagnosis information.

[0045] In a fourth aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0046] Determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object;

[0047] Extracting features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtaining second feature information corresponding to the clinical feature of the target object;

[0048] The first characteristic information and the second characteristic information are input into a diagnosis model to obtain postoperative lymph node diagnosis information.

[0049] In a fifth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0050] Inputting the preoperative medical images containing lymph nodes corresponding to the target object into the trained lymph node delineation model, and determining each lymph node delineation area according to the model output result;

[0051] Performing feature extraction on each of the lymph node delineation regions to obtain first feature information corresponding to key imaging genomics features, and obtaining second feature information of the target object under key clinical features;

[0052] The first feature information and the second feature information are input into a trained diagnostic model to obtain postoperative lymph node abnormality information predicted by the diagnostic model.

[0053] In a sixth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0054] Determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object;

[0055] Extracting features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtaining second feature information corresponding to the clinical feature of the target object;

[0056] The first characteristic information and the second characteristic information are input into a diagnosis model to obtain postoperative lymph node diagnosis information.

[0057] The above prediction device, prediction method, apparatus, computer device, computer-readable storage medium and computer program product can determine the lymph node delineation area based on the preoperative medical image containing lymph nodes corresponding to the target object; then the lymph node delineation area can be feature extracted to obtain the first feature information corresponding to the imaging genomics feature, and the second feature information corresponding to the clinical characteristics of the target object can be obtained; then, the first feature information and the second feature information can be input into the diagnosis model to obtain the postoperative lymph node diagnosis information. In this embodiment, on the one hand, by inputting the first feature information of the imaging genomics feature and the second feature information corresponding to the clinical characteristics of the target object as the input parameters of the two modalities into the diagnosis model, the postoperative lymph node prognosis can be predicted by combining the lymph node image information and the clinical condition of the target object, and the prediction accuracy of the diagnosis model can be effectively improved; on the other hand, by obtaining the postoperative lymph node diagnosis information based on the imaging genomics features of the lymph nodes in the preoperative medical images and the clinical characteristics of the target object, the treatment strategy can be adaptively adjusted in advance according to the abnormal lymph nodes that may occur after surgery in the target object, reducing the possibility of performing surgery on the target object again, and effectively reducing the treatment cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 A schematic flow chart of a prediction method performed by a prediction device in one embodiment;

[0060] Figure 2 A flowchart of a step of training a diagnostic model in one embodiment;

[0061] Figure 3 A schematic diagram of a flow chart of another prediction method in one embodiment;

[0062] Figure 4 A schematic flow chart of a step of obtaining a lymph node delineation model in one embodiment;

[0063] Figure 5 is a structural block diagram of a prediction device in an embodiment;

[0064] Figure 6 is an internal structure diagram of a computer device in one embodiment;

[0065] Figure 7 The figure is a diagram of the internal structure of another computer device in one embodiment. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0067] In one embodiment, a prediction device is provided. In some exemplary embodiments, the prediction device may be a terminal, a server, or a system including a terminal and a server, and information prediction is achieved through interaction between the terminal and the server. In some embodiments, the system may also include a medical imaging device, and exemplarily, may include one or more of the following medical imaging devices: Computed Tomography (CT) equipment, Magnetic Resonance Imaging (MRI) equipment, and Nuclear Medicine Imaging (NM) imaging equipment.

[0068] In this embodiment, if Figure 1 As shown, the prediction device is configured to perform the following steps:

[0069] S101, determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to a target object.

[0070] The target object may be any object to be operated on.

[0071] In a specific implementation, the preoperative medical image corresponding to the target object can be obtained before the operation. In some optional embodiments, one or more medical imaging devices can be used to scan and photograph the preset area of ​​the target object to obtain the preoperative medical image. Among them, the preoperative medical image of the target object includes image information of one or more lymph nodes, and the preoperative medical image can provide reference information for the formulation of the treatment plan. In some examples, the preoperative medical image includes a CT image. Taking the CT image obtained in the treatment of esophageal cancer and lung cancer as an example, the preoperative medical image may contain the lymph node drainage area, and may also include one or more image information such as the target area, surrounding organs at risk, and adjacent organ tissues.

[0072] Preoperative medical images can provide detailed images of the internal structure of the target object and can provide a variety of decision-making support information. Taking the radiotherapy of esophageal cancer and lung cancer as an example, with the clinical popularization of image-guided radiotherapy (IGRT), preoperative medical images can provide anatomical structure information of esophageal cancer and lung cancer lesions and surrounding involved organs and parts. By performing image scanning before each radiotherapy fraction of the target object and obtaining the preoperative medical images of the target object, it can help doctors to timely understand the changes in the position of the target object, the volume, shape, and displacement of the target area at that fraction relative to the time when the treatment plan was formulated, and assist doctors to modify the outline of the target area and organs at risk and adjust the radiotherapy plan in time based on the images at that fraction, so as to achieve the purpose of accurate radiotherapy and improve the efficacy.

[0073] In this step, a preoperative medical image corresponding to the target object can be obtained, and the preoperative medical image contains lymph nodes. The prediction device can obtain the lymph node delineation area in the preoperative medical image based on the preoperative medical image; wherein the lymph node delineation area can be the area containing lymph nodes in the preoperative medical image, for example, each lymph node delineation area can include one or more lymph nodes.

[0074] In some exemplary embodiments, the lymph node delineation area may be obtained by automatic delineation by a prediction device, or by manual delineation, or by a combination of automatic delineation by a prediction device and manual delineation.

[0075] S102, extracting features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtaining second feature information corresponding to the clinical feature of the target object.

[0076] Specifically, on the one hand, during the development of cancer, tumor cells may enter the lymph nodes through the lymphatic vessels and form lymph node metastasis. On the other hand, the postoperative prognosis is also related to the state of the target object itself. In this regard, in this step, after obtaining the lymph node delineation area, the lymph node delineation area can be feature extracted to determine the characteristic information of the lymph node delineation area under the imaging genomics feature. For the convenience of distinction, this characteristic information is called the first characteristic information. In some examples, the first characteristic information can be understood as the specific characteristic value under the corresponding imaging genomics feature. In addition, the characteristic information of the target object under the clinical characteristics can also be obtained. For the convenience of distinction, this characteristic information is called the second characteristic information. In some examples, the second characteristic information can be understood as the specific characteristic value under the corresponding clinical characteristics. For example, if the imaging genomics feature is a wavelet filter feature, the lymph node delineation result can be processed by wavelet filtering, and the processing result can be used as the feature information under the wavelet filter feature, that is, the first feature information. For example, if the clinical feature is the diagnostic stage, the specific diagnostic stage of the target object (such as stage X) can be used as the feature information under the diagnostic stage, that is, the second feature information.

[0077] In some exemplary embodiments, the radiomics features and / or clinical features may be unscreened, or may be obtained after screening multiple candidate radiomics features and / or clinical features according to a preset screening rule or screening rules. For example, the radiomics features may be screened according to the preset screening rules to obtain some of the radiomics features, and then the first feature information of the target object under the partial radiomics features may be obtained. On the other hand, the second feature information of the target object under multiple unscreened clinical features may be obtained. The specific radiomics features and clinical features used may be selected according to actual conditions.

[0078] S103, inputting the first characteristic information and the second characteristic information into the diagnosis model to obtain postoperative lymph node diagnosis information.

[0079] In a specific implementation, a diagnostic model can be pre-trained, which can predict the postoperative lymph node abnormality information of the target object based on the input information of the two modalities; wherein the input of the two modalities includes imaging genomics features obtained based on images, and clinical features determined by collecting physiological parameters or physiological information of the target object.

[0080] After obtaining the first characteristic information and the second characteristic information, the first characteristic information and the second characteristic information can be used together as input parameters for postoperative lymph node abnormality prediction and input into the diagnostic model. The diagnostic model predicts the possibility of postoperative lymph node abnormality of the target object based on the input first characteristic information and second characteristic information, and obtains postoperative lymph node diagnostic information; wherein, the postoperative lymph node diagnostic information can be a prediction result obtained by predicting the prognosis of the target object's lymph nodes after surgery, which can characterize whether the target object has abnormalities after treatment, such as whether lymph node metastasis occurs and the number of lymph nodes with lymph node metastasis.

[0081] The prognosis of postoperative lymph nodes (such as postoperative lymph node metastasis) is an important reference factor, which often directly affects the treatment plan and follow-up plan of the target object. In this embodiment, the metastasis of the lymph nodes after treatment can be predicted prospectively based on the preoperative medical images before treatment, that is, before the treatment intervention (such as surgical resection), the possibility of abnormalities in the postoperative lymph nodes can be predicted prospectively in advance, so that the formulation of individualized follow-up strategies and individualized treatment strategies can be carried out in advance in a targeted manner, which can greatly reduce the scope of surgical resection, reduce the invasive damage to the target object, reduce the cost of human and material resources, and even achieve surgical exemption for some target objects, which is of great significance for reducing the workload of doctors and realizing the integration of personalized diagnosis and treatment.

[0082] The prediction device provided in the above embodiment can determine the lymph node delineation area based on the preoperative medical image containing lymph nodes corresponding to the target object; then, the lymph node delineation area can be feature extracted to obtain the first feature information corresponding to the imaging genomics feature, and the second feature information corresponding to the clinical feature of the target object can be obtained; then, the first feature information and the second feature information can be input into the diagnosis model to obtain the postoperative lymph node diagnosis information. In this embodiment, on the one hand, by inputting the first feature information of the imaging genomics feature and the second feature information corresponding to the clinical feature of the target object as the input parameters of the two modalities into the diagnosis model, the postoperative lymph node prognosis can be predicted by combining the lymph node image information and the clinical condition of the target object, and the prediction accuracy of the diagnosis model can be effectively improved; on the other hand, by obtaining the postoperative lymph node diagnosis information based on the imaging genomics features of the lymph nodes in the preoperative medical images and the clinical characteristics of the target object, the treatment strategy can be adaptively adjusted in advance according to the abnormal lymph nodes that may occur after surgery in the target object, reducing the possibility of performing surgery on the target object again, and effectively reducing the treatment cost.

[0083] In an exemplary embodiment, the prediction device trains the diagnosis model through the following steps:

[0084] S201, obtaining the lymph node delineation result of the preoperative medical image sample corresponding to the sample object, extracting the third feature information of the lymph node delineation result under multiple candidate imaging genomics features, and obtaining the fourth feature information of the sample object under multiple candidate clinical features.

[0085] In a specific implementation, a preoperative medical image corresponding to the sample object may be obtained. For the sake of distinction, the preoperative medical image is referred to as a preoperative medical image sample, wherein the sample object is an object whose lymph nodes have been removed surgically. In some exemplary embodiments, the preoperative medical image sample may contain manual gold standard delineations of the lymph node drainage area, target area, and surrounding organs at risk. In addition, the pathological test results corresponding to the lymph nodes of the sample object may also be obtained.

[0086] In some embodiments, the preoperative medical image sample can be input into a trained lymph node delineation model, and the lymph node delineation model outputs the lymph node delineation result of the preoperative medical image sample. The lymph node delineation result can also be obtained by manually delineating on the preoperative medical image sample. The model delineation result can also be output by the trained lymph node delineation model, and the doctor can obtain the lymph node delineation result of the preoperative medical image sample after adjustment or confirmation.

[0087] Afterwards, multiple candidate radiomic features can be determined, and feature information of lymph node delineation results under various candidate radiomic features can be extracted. For the convenience of distinction, this feature information is called third feature information. On the other hand, feature information of sample objects under multiple candidate clinical features can also be obtained. For the convenience of distinction, this feature information is called fourth feature information.

[0088] In some exemplary embodiments, the feature information extraction under the candidate radiomics features may involve extracting nearly a thousand features including the following feature information: drainage area shape features, first-order features, second-order features, grayscale co-occurrence matrix features, grayscale size area matrix features, grayscale run matrix features, neighborhood grayscale difference matrix features, grayscale dependency matrix features, and wavelet filter features. The various third feature information extracted can be recorded in quantized numerical form for subsequent processing.

[0089] Candidate clinical features may include at least two of the following clinical indicator parameters at the diagnosis stage: diagnostic stage, gender, age, height, weight, smoking history, drinking history, KPS, history of hypertension, history of diabetes, history of heart disease, history of cerebral infarction, history of asthma, history of emphysema, history of pneumonia, and hematological indicators.

[0090] S202, based on the combination of multiple feature information of each third feature information and each fourth feature information, multiple predictions are made on the postoperative lymph node abnormality information, and based on the multiple prediction results, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features.

[0091] Among them, the postoperative lymph node abnormality information can be the prediction result obtained by predicting the prognosis of the sample object's lymph nodes after surgery, which can characterize whether the sample object has abnormalities after treatment, such as whether lymph node metastasis occurs and the number of lymph nodes with lymph node metastasis.

[0092] In a specific implementation, after obtaining each third feature information and each fourth feature information, the postoperative lymph node abnormality information can be predicted by combining part or all of the third feature information and each fourth feature information to obtain a prediction result. By combining each third feature information and each fourth feature information in a variety of ways, a variety of prediction results can be obtained, wherein the combination of each third feature information and each fourth feature information can be achieved by adjusting the feature weight coefficients of each candidate imaging genomics feature and various candidate clinical features. For example, by adjusting the feature weight coefficient of a certain imaging genomics feature to 0, the corresponding third feature information can be excluded from the prediction. Furthermore, the similarity between each prediction result and the actual postoperative condition of the sample object can be compared, and from various candidate imaging genomics features and various candidate clinical features, the imaging genomics features and clinical features related to the accuracy of the prediction results can be determined as the key imaging genomics features and key clinical features whose correlation meets the correlation condition.

[0093] S203, training a diagnostic model using the third feature information of the sample object under the key radiomics feature and the fourth feature information under the key clinical feature.

[0094] After obtaining the key radiomics features and key clinical features, the third feature information of the sample objects under the key radiomics features and the fourth feature information under the key clinical features can be used to train the diagnostic model.

[0095] In some exemplary embodiments, a data set containing multiple preoperative medical image samples can be obtained, wherein each preoperative medical image sample can be converted from an original format file to a target format file required for model input in the preprocessing stage, so as to perform subsequent radiomics feature extraction steps. At the same time, the above-collected and preprocessed data set can be divided into a training set and a test set, such as a 4:1 ratio, wherein the training set data will be used to train the diagnostic model, and the test set data will be used to test the prediction results of the pre-trained model.

[0096] In this embodiment, multiple predictions can be performed based on multiple feature information combinations of each third feature information and each fourth feature information, and the prediction results corresponding to each prediction can be obtained. By comparing the various prediction results, the key imaging features and key clinical features that are more relevant to the accuracy of lymph node abnormality information prediction can be determined from a variety of candidate imaging features and clinical features. The diagnostic model is subsequently trained using the feature information under the key imaging features and key clinical features, which can save computing resources while reducing the introduction of redundant information, thereby effectively improving the prediction accuracy of the diagnostic model.

[0097] In some exemplary embodiments, due to the complexity of factors such as the morphology and density of lymph nodes, the relevant imaging genomics feature extraction methods may not be able to fully capture all the characteristics of lymph nodes, thereby affecting the accuracy of postoperative abnormality assessment of lymph nodes.

[0098] In this regard, in other embodiments, multiple key features can be predetermined, including key imaging features and key clinical features, which are multiple key features that have a high degree of influence on the accuracy of postoperative lymph node abnormality information prediction. Compared with the evaluation of the postoperative condition of lymph nodes by extracting a large number of imaging features of medical images in some embodiments, by determining key imaging features from a large number of imaging features (such as nearly a thousand imaging features), specific features related to postoperative lymph node abnormality information can be refined, avoiding the uniform extraction of feature information of lymph node delineation areas under massive imaging features, effectively improving prediction accuracy while saving computing resources; on the other hand, by determining the key clinical features of the target object, it is possible to analyze the lymph node images while combining the clinical information reflecting the actual physiological state of the target object, and comprehensively analyze the postoperative lymph node abnormality of the target object, avoiding the association between the separate analysis of lymph node images and the separation of the target object's own state.

[0099] It can be understood that the above-mentioned use of the third feature information of the sample object under the key imaging omics feature and the fourth feature information under the key clinical feature to train the diagnostic model is only an illustrative example. In some other embodiments, the third feature information under multiple candidate imaging omics features and the fourth feature information under multiple candidate clinical features can also be used to train the diagnostic model. In other words, the model can be trained directly according to each candidate imaging omics feature and each candidate clinical feature without selecting the key feature. For example, a large number of training samples can be obtained, each training sample includes the third feature information of the sample object under multiple candidate imaging omics features, the fourth feature information under multiple candidate clinical features, and the postoperative lymph node prognosis of the sample object. Based on these training samples, multiple combinations of the third feature information and the fourth feature information and the postoperative lymph node prognosis of the sample object under each combination can be obtained. By using these training samples to supervise or semi-supervisedly learn the model, the model can learn the association between the postoperative lymph node prediction information and the third feature information and the fourth feature information, so that the model knows how the postoperative lymph node prediction information is affected by various candidate imaging omics features and clinical features. In addition, if new training samples are continuously input, online learning or incremental learning methods can also be used to enable the diagnostic model to continuously adapt to new data. For example, the training model can be re-predicted regularly based on new training samples to incorporate new data and features to maintain the timeliness and accuracy of the model. Accordingly, when using the diagnostic model, the input of the diagnostic model can include unscreened feature information under multiple imaging omics features and clinical features.

[0100] In one embodiment, the LASSO regression method can be used to select key features with a high correlation with the postoperative pathological diagnosis results from various candidate imaging features and various candidate clinical features by linear regression. In step S202, based on the combination of multiple feature information of each third feature information and each fourth feature information, multiple predictions are made on the postoperative lymph node abnormality information, and based on the multiple prediction results, key imaging features and key clinical features whose correlations meet the correlation conditions are determined from various candidate imaging features and various candidate clinical features, which can include the following steps:

[0101] Determine the current feature weight coefficients of various candidate imaging features and various candidate clinical features, and predict the probability of postoperative lymph node abnormality based on each feature weight coefficient, each third feature information and each fourth feature information to obtain a prediction result; determine the loss value based on the difference between the prediction result and the postoperative lymph node metastasis information of the sample object, and the sum of the coefficients of each feature weight coefficient; according to the loss value, at least part of each feature weight coefficient is lowered, and return to execute the step of determining the current feature weight coefficients of various candidate imaging features and various candidate clinical features until the iteration end condition is met; according to the current feature weight coefficients, determine the key imaging features and key clinical features whose correlations meet the correlation conditions from various candidate imaging features and various candidate clinical features.

[0102] In one example, the prediction result may be a pathological examination result (eg, 1 or 0) of the lymph node drainage area of ​​the sample object.

[0103] Specifically, the feature weight coefficients of various candidate radiomic features and various candidate clinical features can be initialized to obtain the current feature weight coefficients of various candidate radiomic features and various candidate clinical features. Then, the probability of postoperative lymph node abnormality can be predicted based on each feature weight coefficient, each third feature information and each fourth feature information to obtain the prediction result under the current feature weight coefficient.

[0104] Subsequently, the difference between the predicted result and the postoperative lymph node metastasis information of the sample object is determined, and the weight coefficients of each feature are summed to obtain the sum of the coefficients. The loss value is obtained based on the difference and the sum of the coefficients. In some examples, the sum of the difference and the sum of the coefficients can be used as the loss value.

[0105] Furthermore, at least part of the feature weight coefficients are adjusted downward according to the loss value. That is, by adjusting at least part of the feature weight coefficients downward, the prediction result is made closer to the postoperative lymph node metastasis information of the sample object, and the sum of the coefficients is reduced. In some exemplary embodiments, the optimization objective function includes the difference between the prediction result and the postoperative lymph node metastasis information of the sample object and the sum of the coefficients, and also includes a penalty term, which can be L1 regularization. By adding the penalty term, weight sparseness can be achieved.

[0106] After lowering the feature weight coefficient, the step of determining the current feature weight coefficients of various candidate radiomic features and various candidate clinical features may be returned to execute until the iteration end condition is met. For example, the iteration end condition may be determined to be met when the loss value reaches a minimum value or the number of iterations reaches a threshold.

[0107] Then, according to the current weight coefficients of each feature at the end of the iteration, key radiomics features and key clinical features whose correlations meet the correlation conditions can be determined from various candidate radiomics features and various candidate clinical features.

[0108] In this embodiment, the loss value is determined based on the difference between the predicted result and the postoperative lymph node metastasis information of the sample object, and the sum of the coefficients of each feature weight coefficient, and the feature weight coefficient is iteratively lowered according to the loss value, so that the predicted result can be made as close as possible to the actual postoperative pathological detection result of the sample object, while the feature weight coefficients of multiple features are made to be 0 or close to 0. Subsequently, by selecting the features corresponding to the non-zero feature weight coefficients, the features with better predictive ability for the target variable are screened out, thereby avoiding collinearity and overfitting, achieving feature dimensionality reduction, and improving model generalization.

[0109] In one embodiment, according to the current feature weight coefficients, determining the key radiomics features and key clinical features whose correlations meet the correlation conditions from various candidate radiomics features and various candidate clinical features may include the following steps:

[0110] The coefficient values ​​of the current feature weight coefficients are compared, and based on the comparison results, multiple key features whose coefficient values ​​are greater than any other features are determined from various candidate radiomics features and various candidate clinical features; based on the multiple key features, key radiomics features and key clinical features whose correlations meet the correlation conditions are obtained.

[0111] Other features are other features among various candidate radiomics features and various candidate clinical features except the key features.

[0112] In some exemplary embodiments, various candidate radiomic features and various candidate clinical features can be arranged in descending order according to the coefficient values ​​of the current feature weight coefficients, and the features with the first K (K is a positive integer greater than or equal to 2) numbers of features can be used as key features. For example, the first 10 features with the first coefficient values ​​can be used as key features, wherein the multiple key features include radiomic features and clinical features. Then, the multiple key features obtained by screening can be determined as key radiomic features and key clinical features whose correlations meet the correlation conditions.

[0113] For example, when screening key features, input data and output data can be preloaded. In this example, the input data is the radiomics feature data of the lymph node drainage area and the clinical indicator parameter data of the sample object, and the output data is the pathological examination result of the lymph node drainage area (1 or 0). After all the data is loaded, the input data can be first subjected to standard normalization processing, and then the processing method provided in the previous embodiment can be used to fit the input data and the output data. During the fitting process, the value of the penalty term alpha, that is, the weight value of the L1 regularization term, can be pre-set. Finally, the 10 indicators that are most relevant to the output data are selected from thousands of indicators as the input of the diagnostic model.

[0114] In this embodiment, after the iteration is completed, multiple key features that are more relevant to the prediction of postoperative lymph node abnormality information can be quickly determined through the current feature weight coefficient, effectively improving the screening efficiency of key features.

[0115] In one embodiment, in step S203, using the third feature information of the sample object under the key radiomics feature and the fourth feature information under the key clinical feature to train the diagnostic model may include the following steps:

[0116] Identify multiple machine learning models; for each machine learning model, use the third feature information of the sample objects under the key imaging genomics features and the fourth feature information under the key clinical features to train the machine learning model, and obtain the model training effect corresponding to the machine learning model; from the multiple machine learning models, screen out the machine learning model whose model training effect meets the preset training effect screening conditions as the trained diagnostic model.

[0117] Among them, the model training effect characterizes the accuracy of the prediction results of the machine learning model using the third feature information and the fourth feature information to predict the probability of postoperative lymph node abnormality.

[0118] In practical applications, after determining the key radiomics features and key clinical features, a variety of machine learning models can be trained, for example, a variety of machine learning models of different types such as random forest models and deep learning models can be trained. For each machine learning model, the third feature information of the sample object under the key radiomics features and the fourth feature information under the key clinical features can be used to train the machine learning model, and the model training effect corresponding to the machine learning model can be obtained at the end of the training. In some embodiments, the key features can be used to predict whether the sample objects in the test set have postoperative lymph node metastasis, and compared with the actual pathological metastasis results of the sample objects in the test set to obtain a quantitative indicator of the model training effect.

[0119] After obtaining multiple machine learning models, the model training effects of the various machine learning models can be compared, and then the machine learning models whose model training effects meet the preset training effect screening conditions can be screened out from the multiple machine learning models as the trained diagnostic models. For example, the machine learning model with the best model training effect (for example, the highest prediction result accuracy) can be used as the trained diagnostic model.

[0120] In some embodiments, Figure 2 As shown, clinicians can subsequently evaluate the prediction effect of the model to determine whether the diagnostic model meets the clinical usability requirements; if not, they can continue to optimize and adjust in the key feature selection stage until the clinical usability requirements are met; if so, the diagnostic model can be deployed to the built-in program of the prediction device to assist doctors in using the automatic delineation model of the lymph node drainage area for segmentation during the treatment process, and then further predict whether lymph node metastasis will occur in the segmented area, to assist doctors in making individualized treatment decisions and follow-up strategies for patients.

[0121] In this embodiment, the prognosis of lymph node metastasis of the sample object can be predicted based on multiple machine learning algorithms, and then the machine learning model with better model training effect can be screened out as the diagnostic model. The training effects of multiple models can be integrated to accurately and effectively select the diagnostic model with better prediction effect, thereby improving the accuracy of subsequent predictions.

[0122] In an exemplary embodiment, in step S101, determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to a target object may include the following steps: inputting the preoperative medical image into a lymph node delineation model, and outputting the lymph node delineation area.

[0123] In traditional technology, the delineation and evaluation of lymph nodes in preoperative medical images mainly rely on the visual observation and subjective judgment of professional physicians. This method is not only time-consuming and labor-intensive, but also highly dependent on the physician's personal experience, which may lead to inconsistent delineation and evaluation results due to differences in subjective judgment and clinical knowledge.

[0124] In this regard, in this embodiment, a lymph node delineation model can be pre-trained. After obtaining a preoperative medical image of a target object containing lymph nodes, the preoperative medical image can be input into the trained lymph node delineation model, and the lymph node delineation model performs feature extraction on the preoperative medical image, and the model output result is determined according to the feature extraction result. Among them, the model output result includes one or more lymph node delineation areas delineated by the lymph node delineation model, and each lymph node delineation area can include one or more lymph nodes. In some examples, the lymph node delineation model can delineate each lymph node separately to obtain multiple lymph node delineation areas, and one lymph node delineation area corresponds to one lymph node.

[0125] In some optional embodiments, each lymph node delineation area can be directly obtained according to the model output result of the lymph node delineation model, that is, each area delineated by the lymph node delineation model is directly used as the final lymph node delineation area; in other embodiments, the staff can still manually adjust the delineation results automatically output by the lymph node delineation model, obtain each lymph node delineation area based on the manually adjusted results, and then perform subsequent processing. For example, Figure 3 As shown, a diagnostic CT image of the target object can be obtained before surgery and then input into a pre-trained lymph node delineation model to obtain multiple automatic lymph node delineation contours output by the model. For example, the model can automatically delineate dozens of tiny lymph node areas. After manual adjustment or confirmation by the doctor, the first feature information of the multiple delineated lymph nodes under key imaging genomics features can be extracted, and the second feature information of the target object under key clinical features can be extracted. The first feature information and the second feature information are then input together into the pre-trained diagnostic model to obtain postoperative lymph node abnormality information output by the model. The doctor can subsequently determine an individualized treatment strategy or / and an individualized follow-up strategy based on the postoperative lymph node abnormality information.

[0126] In this embodiment, on the one hand, an end-to-end learning framework for automatic lymph node delineation, automatic prediction, individualized treatment and follow-up strategy formulation is provided, which can achieve the purpose of individualized treatment of multiple lymph node drainage areas of different target objects in the same framework based on the input of automated imaging genomics indicators and manual input of clinical indicators of the target object, thereby effectively improving the treatment efficiency; on the other hand, the prediction device provided by this application is a non-invasive and automated tool that can not only assist clinicians in automatically delineating tiny lymph node areas, but also help doctors understand the patient's lymph node recurrence after surgical resection, so as to achieve rapid, personalized, precise and adaptive treatment of the target object.

[0127] In one embodiment, the lymph node delineation model is trained by the following steps:

[0128] S1011, obtaining a preoperative medical image sample corresponding to the sample object, lymph node annotation information corresponding to the preoperative medical image sample, and a lymph node pathology detection result.

[0129] In some embodiments, preoperative medical images of the sample subject may be obtained, e.g. Figure 4As shown, a diagnostic CT image of a sample object can be obtained. For the sake of distinction, the preoperative medical image is also referred to as a preoperative medical image sample. The preoperative medical image sample includes lymph node image information corresponding to the lymph node, such as the lymph node drainage area. Then, the staff can outline and annotate the lymph nodes of the preoperative medical image sample to obtain lymph node annotation information, wherein the lymph node annotation information can characterize the area where each lymph node is located in the preoperative medical image sample.

[0130] On the other hand, since the sample object has undergone surgery to remove lymph nodes within a preset range (such as a preset range centered on the lesion), a pathological examination can be performed on the lymph nodes after surgical removal to obtain the lymph node pathology test results corresponding to the labeled lymph nodes in the preoperative medical imaging sample. The result can indicate whether the lymph node is benign or malignant. In some examples, the lymph node pathology test results can indicate whether it is benign or malignant through labels 0 or 1.

[0131] In some embodiments, the original preoperative medical image of the sample object may include some redundant or irrelevant information. For example, for the acquired abdominal CT image, only the designated area information of the abdominal CT image is required when generating the preoperative medical image sample. To this end, the image of the required area in the diagnostic CT image and the corresponding manual gold standard outline can be extracted through a preprocessing process to obtain a preoperative medical image sample carrying lymph node annotation information. Then, the original format of the preoperative medical image sample file (such as a dicom format file) can be converted into a format file required for lymph node outline model input (such as a nii format file).

[0132] S1012, obtaining prompt information for excluding the correlation between the lymph node delineation area and the lymph node delineation area that does not meet the preset conditions, and inputting the prompt information and the preoperative medical image sample into the lymph node delineation model to be trained, and the lymph node delineation model extracts the image features corresponding to the candidate delineation area in the preoperative medical image sample according to the prompt information, and outputs the lymph node delineation results and the predicted pathological detection results according to the image features.

[0133] In the related art, computer algorithms can be used to automatically identify the location and range of lymph nodes by analyzing the density, shape, size and other features in CT images. However, this technology can often only process specific types of lymph nodes, and may cause false detection or missed detection of other types of lymph nodes. In some possible optimization methods, although the number of data samples can be increased to enable the computer algorithm to learn more information, if the preoperative medical image samples are directly input into the lymph node delineation model to be trained for lymph node delineation, due to the lack of training, the model may delineate lymph nodes in irrelevant areas. For example, for the lymph nodes to be delineated near the lungs in the preoperative medical image samples, the delineated structure output by the lymph node delineation model may be located in the stomach or within a preset range of the stomach. Only after a large amount of training can the lymph node delineation accuracy of the lymph node delineation model be improved, which takes a lot of time.

[0134] In this regard, when training the lymph node delineation model, this embodiment can obtain prompt information, which is used to prompt the model to exclude areas whose correlation with the lymph node delineation area does not meet the preset conditions. For example, the prompt information can indicate the target area that is irrelevant to the lymph node delineation area or whose correlation is lower than the threshold, and prompt the model to perform lymph node delineation at other locations outside the target area. Then, the prompt information and the preoperative medical image sample can be input into the lymph node delineation model to be trained. The lymph node delineation model excludes the interference area in the preoperative medical image sample that is irrelevant to the delineation area according to the prompt information, determines the candidate delineation area related to the lymph node delineation, and then extracts the image features corresponding to the candidate delineation area. The lymph node delineation result is output according to the image features. At the same time, the benign or malignant condition of the delineated lymph node can be predicted according to the extracted image features to obtain the predicted pathological detection result.

[0135] In some optional embodiments, the lymph node delineation model to be trained may be a deep learning model, such as a nnUNet network, a U-Net network, a V-Net network, or any other model used for medical image segmentation.

[0136] S1013, according to the difference between the lymph node delineation result and the lymph node annotation information and the difference between the predicted pathological result and the lymph node pathological detection result, the model parameters of the lymph node delineation model are adjusted to obtain a trained lymph node delineation model.

[0137] Then, on the one hand, the region where the lymph nodes identified by the model are located can be determined based on the lymph node delineation results, and the difference between the region and the pre-acquired lymph node annotation information can be determined. On the other hand, the difference between the lymph node pathology detection result and the predicted pathology result can be determined, and then the model loss value can be determined based on the above difference, and the model parameters of the lymph node delineation model can be adjusted. If the training end condition is not met at present, the step S1011 can be returned to iteratively adjust the model parameters until the training end condition is met to obtain a trained lymph node delineation model.

[0138] In some embodiments, a data set containing multiple preoperative medical image samples may be collected and divided into a training set and a test set, for example, in a ratio of 4:1; wherein the training set data is used to train the lymph node delineation model, and the test set data is used to perform a delineation test on the pre-trained lymph node delineation model, and the clinician determines whether the currently trained lymph node delineation model meets the clinical use requirements, such as Figure 4 As shown, if it is determined that the current lymph node delineation model has been trained, it can be deployed to the prediction device for use, for example, deployment refers to the built-in program of the prediction device to assist staff in calling the lymph node delineation model for segmentation at any time during the processing of medical tasks, thereby improving clinical treatment efficiency and reducing doctors' workload; if it is determined that the current lymph node delineation model has not been trained, a data set containing multiple preoperative medical imaging samples and the corresponding manual gold standard delineation can be re-collected for retraining and adjustment of the model.

[0139] In this embodiment, on the one hand, by inputting prompt information used to exclude areas whose correlation with lymph node delineation areas does not meet preset conditions into the lymph node delineation model together with preoperative medical image samples, the area in which the lymph node delineation model delineates lymph nodes can be effectively limited, thereby improving the model training efficiency and the accuracy of the lymph node delineation results. When the model is trained using a large amount of sample data containing different types of lymph nodes, this method can effectively improve the model training speed and model training effect; on the other hand, when delineating lymph nodes in preoperative medical image samples, the lymph node delineation model correspondingly predicts the pathological detection results of the delineated lymph nodes, which can help the lymph node delineation model learn the diversity characteristics of lymph nodes under different states, thereby improving the generalization and delineation accuracy of the lymph node delineation model, thereby enabling the lymph node delineation model to efficiently delineate multiple types of lymph nodes in medical images.

[0140] In one embodiment, the preoperative medical image sample includes the target structure tissue, the lymph nodes within the preset range of the target structure tissue, and the adjacent structure tissue other than the target organ structure tissue, that is, the preoperative medical image sample includes not only the target structure tissue and the lymph node drainage area near it, but also the target area, and the surrounding endangered organs that may be negatively imaged during the surgical operation. For example, for a target subject with lung cancer, the preoperative medical image sample will include not only the lungs and lymph nodes near the lungs, but also one or more adjacent structure tissues other than the lungs, such as the heart, spleen, etc. If the preoperative medical images containing multiple organs are directly provided to the lymph node delineation model, it may be that the structure tissue associated with the delineated lymph nodes is not the target structure tissue. In this regard, the prediction device in this embodiment can obtain prompt information through the following steps:

[0141] Acquire clinical prior information indicating abnormal conditions of target structural tissue of the sample object, and acquire anatomical structure restriction information of adjacent structural tissue; obtain prompt information based on the clinical prior information and the anatomical structure restriction information.

[0142] In some embodiments, Figure 4 As shown, on the one hand, clinical prior information can be obtained, wherein the clinical prior information can be prior information obtained by testing the sample object in advance, and the clinical prior information can indicate the abnormality of the target structure tissue in the sample object, such as indicating the specific structure tissue where the lesion exists or the location of the lesion. By providing clinical prior information, it helps the lymph node delineation model to avoid searching for lymph nodes in irrelevant or interfering areas far away from the target structure tissue. For example, for the test result of "esophageal cancer" with clinical prior information, by providing this information to the model, the model can avoid delineating lymph nodes near the heart or lungs.

[0143] On the other hand, the anatomical structural restriction information of the adjacent structural tissues can also be obtained, wherein the anatomical structural restriction information of the adjacent structural tissues can be used to inform the model of one or more information such as the anatomical position relationship, morphological structure and functional connection of other structural tissues adjacent to the target structural tissue. Specifically, the structural tissues in the object have their specific anatomical positions, morphologies and sizes, and have certain adjacent relationships with other structural tissues. By providing the anatomical structural restriction information to the model, the model can be aware of the positional relationship between the target structural tissue and other adjacent structural tissues, and avoid outlining in areas adjacent to the target structural tissue but unrelated to the lymph nodes.

[0144] It is understandable that, during the actual reasoning process of the lymph node delineation model, at least one of clinical prior information and anatomical structure restriction information may also be input into the lymph node delineation model, thereby improving the delineation speed and accuracy of the lymph node delineation model during application.

[0145] In this embodiment, clinical prior information and anatomical structure restriction information can be combined to quickly exclude interference areas in preoperative medical images that are not related to lymph node delineation, which helps to improve model training efficiency and efficiently complete the task of automatically delineating multiple lymph node drainage areas.

[0146] In one embodiment, the present application also provides a prediction method. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The implementation scheme for solving the problem provided by the method is similar or identical to the implementation scheme recorded in the above method, so the specific limitations in one or more prediction method embodiments provided below can refer to the limitations on the prediction system above, and will not be repeated here.

[0147] In one embodiment, the method comprises the following steps:

[0148] S401, determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object.

[0149] S402, extracting features of the lymph node delineation area to obtain first feature information corresponding to radiomics features, and obtaining second feature information corresponding to clinical features of the target object.

[0150] S403: Input the first characteristic information and the second characteristic information into a diagnosis model to obtain postoperative lymph node diagnosis information.

[0151] In one embodiment, the diagnostic model is trained by the following steps:

[0152] Obtaining a lymph node delineation result of a preoperative medical image sample corresponding to the sample object, extracting third feature information of the lymph node delineation result under multiple candidate imaging genomics features, and obtaining fourth feature information of the sample object under multiple candidate clinical features;

[0153] According to the combination of multiple feature information of each of the third feature information and each of the fourth feature information, multiple predictions are made on the postoperative lymph node abnormality information, and according to the multiple prediction results, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features;

[0154] The diagnostic model is trained using the third feature information of the sample object under the key imaging features and the fourth feature information under the key clinical features.

[0155] In one embodiment, the postoperative lymph node abnormality information is predicted multiple times according to the combination of multiple feature information of each of the third feature information and each of the fourth feature information, and key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features according to the multiple prediction results, including:

[0156] Determine current feature weight coefficients of various candidate imaging features and various candidate clinical features, and predict the probability of postoperative lymph node abnormality based on each of the feature weight coefficients, each of the third feature information, and each of the fourth feature information to obtain a prediction result;

[0157] Determine a loss value according to the difference between the prediction result and the postoperative lymph node metastasis information of the sample object, and the sum of the coefficients of each of the feature weight coefficients;

[0158] According to the loss value, at least part of each of the feature weight coefficients is adjusted downward, and the step of determining current feature weight coefficients of various candidate radiomics features and various candidate clinical features is returned to be executed until an iteration end condition is met;

[0159] According to the current feature weight coefficients, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features.

[0160] In one embodiment, the determining of key radiomics features and key clinical features whose correlations satisfy the correlation conditions from various candidate radiomics features and various candidate clinical features according to the current feature weight coefficients includes:

[0161] Comparing the coefficient values ​​of the current feature weight coefficients, and determining, according to the comparison results, from various candidate imaging features and various candidate clinical features, a plurality of key features whose coefficient values ​​are greater than any other features; the plurality of key features include imaging features and clinical features, and the other features are other features of various candidate imaging features and various candidate clinical features except the key features;

[0162] Based on the multiple key features, key radiomics features and key clinical features whose correlations meet correlation conditions are obtained.

[0163] In one embodiment, the using the third feature information of the sample object under the key radiomics feature and the fourth feature information under the key clinical feature to train the diagnostic model includes:

[0164] Identify multiple machine learning models;

[0165] For each of the machine learning models, the third feature information of the sample object under the key imaging genomics feature and the fourth feature information under the key clinical feature are used to train the machine learning model, and a model training effect corresponding to the machine learning model is obtained; the model training effect represents the accuracy of the prediction result of the machine learning model predicting the probability of postoperative lymph node abnormality using the third feature information and the fourth feature information;

[0166] From the multiple machine learning models, a machine learning model whose model training effect meets the preset training effect screening conditions is screened out as the trained diagnostic model.

[0167] In one embodiment, determining the lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object includes:

[0168] The preoperative medical image is input into the lymph node delineation model, and the lymph node delineation area is output.

[0169] In one embodiment, the prediction device obtains the lymph node delineation model by training through the following steps:

[0170] Obtaining a preoperative medical image sample corresponding to the sample object, lymph node annotation information corresponding to the preoperative medical image sample, and a lymph node pathology detection result;

[0171] Acquire prompt information for excluding the correlation between the candidate delineated areas and the lymph node delineation regions that does not meet the preset conditions, and input the prompt information and the preoperative medical image sample into the lymph node delineation model to be trained, and the lymph node delineation model extracts the image features corresponding to the candidate delineated areas in the preoperative medical image sample according to the prompt information, and outputs the lymph node delineation results and the predicted pathological detection results according to the image features;

[0172] According to the difference between the lymph node delineation result and the lymph node annotation information and the difference between the predicted pathological result and the lymph node pathological detection result, the model parameters of the lymph node delineation model are adjusted to obtain a trained lymph node delineation model.

[0173] In one embodiment, the preoperative medical image sample includes the target structure tissue, lymph nodes within a preset range of the target structure tissue, and adjacent structure tissues outside the target structure tissue;

[0174] Obtain the prompt information through the following steps:

[0175] Acquiring clinical prior information indicating abnormal conditions of target structural tissue of the sample object, and acquiring anatomical structural restriction information of the adjacent structural tissue;

[0176] The prompt information is obtained according to the clinical prior information and the anatomical structure restriction information.

[0177] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0178] Based on the same inventive concept, the embodiment of the present application also provides a prediction device for implementing the above-mentioned prediction method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned prediction system, so the specific limitations in one or more prediction device embodiments provided below can refer to the limitations of the prediction system above, and will not be repeated here.

[0179] In an exemplary embodiment, Figure 5 As shown, a prediction device is provided, comprising:

[0180] A lymph node delineation module 501 is used to determine a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object;

[0181] A feature information extraction module 502 is used to extract features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtain second feature information corresponding to the clinical feature of the target object;

[0182] The prediction module 503 is used to input the first characteristic information and the second characteristic information into a diagnosis model to obtain postoperative lymph node diagnosis information.

[0183] In one embodiment, the prediction device further includes a diagnosis model training module, and the diagnosis model training module is used to:

[0184] Obtaining a lymph node delineation result of a preoperative medical image sample corresponding to the sample object, extracting third feature information of the lymph node delineation result under multiple candidate imaging genomics features, and obtaining fourth feature information of the sample object under multiple candidate clinical features;

[0185] According to the combination of multiple feature information of each of the third feature information and each of the fourth feature information, multiple predictions are made on the postoperative lymph node abnormality information, and according to the multiple prediction results, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features;

[0186] The diagnostic model is trained using the third feature information of the sample object under the key imaging features and the fourth feature information under the key clinical features.

[0187] In one embodiment, the diagnostic model training module is used to:

[0188] Determine current feature weight coefficients of various candidate imaging features and various candidate clinical features, and predict the probability of postoperative lymph node abnormality based on each of the feature weight coefficients, each of the third feature information, and each of the fourth feature information to obtain a prediction result;

[0189] Determine a loss value according to the difference between the prediction result and the postoperative lymph node metastasis information of the sample object, and the sum of the coefficients of each of the feature weight coefficients;

[0190] According to the loss value, at least part of each of the feature weight coefficients is adjusted downward, and the step of determining current feature weight coefficients of various candidate radiomics features and various candidate clinical features is returned to be executed until an iteration end condition is met;

[0191] According to the current feature weight coefficients, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features.

[0192] In one embodiment, the diagnostic model training module is used to:

[0193] Comparing the coefficient values ​​of the current feature weight coefficients, and determining, according to the comparison results, from various candidate imaging features and various candidate clinical features, a plurality of key features whose coefficient values ​​are greater than any other features; the plurality of key features include imaging features and clinical features, and the other features are other features of various candidate imaging features and various candidate clinical features except the key features;

[0194] Based on the multiple key features, key radiomics features and key clinical features whose correlations meet correlation conditions are obtained.

[0195] In one embodiment, the diagnostic model training module is used to:

[0196] Identify multiple machine learning models;

[0197] For each of the machine learning models, the third feature information of the sample object under the key imaging genomics feature and the fourth feature information under the key clinical feature are used to train the machine learning model, and a model training effect corresponding to the machine learning model is obtained; the model training effect represents the accuracy of the prediction result of the machine learning model predicting the probability of postoperative lymph node abnormality using the third feature information and the fourth feature information;

[0198] From the multiple machine learning models, a machine learning model whose model training effect meets the preset training effect screening conditions is screened out as the trained diagnostic model.

[0199] In one embodiment, the lymph node delineation module 501 is used to:

[0200] The preoperative medical image is input into the lymph node delineation model, and the lymph node delineation area is output.

[0201] In one embodiment, the prediction device further comprises a lymph node delineation model training module, wherein the lymph node delineation model training module is used to:

[0202] Obtaining a preoperative medical image sample corresponding to the sample object, lymph node annotation information corresponding to the preoperative medical image sample, and a lymph node pathology detection result;

[0203] Acquire prompt information for excluding the correlation between the candidate delineated areas and the lymph node delineation regions that does not meet the preset conditions, and input the prompt information and the preoperative medical image sample into the lymph node delineation model to be trained, and the lymph node delineation model extracts the image features corresponding to the candidate delineated areas in the preoperative medical image sample according to the prompt information, and outputs the lymph node delineation results and the predicted pathological detection results according to the image features;

[0204] According to the difference between the lymph node delineation result and the lymph node annotation information and the difference between the predicted pathological result and the lymph node pathological detection result, the model parameters of the lymph node delineation model are adjusted to obtain a trained lymph node delineation model.

[0205] In one embodiment, the preoperative medical image sample includes the target structure tissue, lymph nodes within a preset range of the target structure tissue, and adjacent structure tissues outside the target structure tissue;

[0206] The lymph node delineation model training module is used for:

[0207] Acquiring clinical prior information indicating abnormal conditions of target structural tissue of the sample object, and acquiring anatomical structural restriction information of the adjacent structural tissue;

[0208] The prompt information is obtained according to the clinical prior information and the anatomical structure restriction information.

[0209] Each module in the above prediction device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0210] In an exemplary embodiment, a prediction device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store preoperative medical imaging data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a prediction method is implemented.

[0211] In an exemplary embodiment, a prediction device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a prediction method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0212] Those skilled in the art will understand that Figure 6 and Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0213] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0214] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0215] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0216] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0217] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0218] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0219] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A prediction device, The prediction device is configured to perform the following steps: Determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object; Extracting features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtaining second feature information corresponding to the clinical feature of the target object; The first characteristic information and the second characteristic information are input into a diagnosis model to obtain postoperative lymph node diagnosis information.

2. The prediction device according to claim 1, characterized in that The prediction device is trained to obtain the diagnosis model through the following steps: Obtaining a lymph node delineation result of a preoperative medical image sample corresponding to the sample object, extracting third feature information of the lymph node delineation result under multiple candidate imaging genomics features, and obtaining fourth feature information of the sample object under multiple candidate clinical features; According to the combination of multiple feature information of each of the third feature information and each of the fourth feature information, multiple predictions are made on the postoperative lymph node abnormality information, and according to the multiple prediction results, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features; The diagnostic model is trained using the third feature information of the sample object under the key imaging features and the fourth feature information under the key clinical features.

3. The prediction device according to claim 2, characterized in that The method of performing multiple predictions on the postoperative lymph node abnormality information based on the combination of multiple feature information of each of the third feature information and each of the fourth feature information, and determining key radiomics features and key clinical features whose correlations meet the correlation conditions from various candidate radiomics features and various candidate clinical features based on the multiple prediction results, includes: Determine current feature weight coefficients of various candidate imaging features and various candidate clinical features, and predict the probability of postoperative lymph node abnormality based on each of the feature weight coefficients, each of the third feature information, and each of the fourth feature information to obtain a prediction result; Determine a loss value according to the difference between the prediction result and the postoperative lymph node metastasis information of the sample object, and the sum of the coefficients of each feature weight coefficient; According to the loss value, at least part of each of the feature weight coefficients is adjusted downward, and the step of determining current feature weight coefficients of various candidate radiomics features and various candidate clinical features is returned to be executed until an iteration end condition is met; According to the current feature weight coefficients, key radiomics features and key clinical features whose correlations meet the correlation conditions are determined from various candidate radiomics features and various candidate clinical features.

4. The prediction device according to claim 3, characterized in that The step of determining the key radiomics features and key clinical features whose correlations satisfy the correlation conditions from various candidate radiomics features and various candidate clinical features according to the current feature weight coefficients includes: Comparing the coefficient values ​​of the current feature weight coefficients, and determining, according to the comparison results, from various candidate imaging features and various candidate clinical features, a plurality of key features whose coefficient values ​​are greater than any other features; the plurality of key features include imaging features and clinical features, and the other features are other features of various candidate imaging features and various candidate clinical features except the key features; Based on the multiple key features, key radiomics features and key clinical features whose correlations meet correlation conditions are obtained.

5. The prediction device according to claim 2, characterized in that The step of training the diagnostic model by using the third feature information of the sample object under the key radiomics feature and the fourth feature information under the key clinical feature comprises: Identify multiple machine learning models; For each of the machine learning models, the third feature information of the sample object under the key imaging genomics feature and the fourth feature information under the key clinical feature are used to train the machine learning model, and a model training effect corresponding to the machine learning model is obtained; the model training effect represents the accuracy of the prediction result of the machine learning model predicting the probability of postoperative lymph node abnormality using the third feature information and the fourth feature information; From the multiple machine learning models, a machine learning model whose model training effect meets the preset training effect screening conditions is screened out as the trained diagnostic model.

6. The prediction device according to claim 1, characterized in that The step of determining the lymph node delineation area based on the preoperative medical image containing the lymph nodes corresponding to the target object comprises: The preoperative medical image is input into the lymph node delineation model, and the lymph node delineation area is output.

7. The prediction device according to claim 6, characterized in that The prediction device is trained to obtain the lymph node delineation model through the following steps: Obtaining a preoperative medical image sample corresponding to the sample object, lymph node annotation information corresponding to the preoperative medical image sample, and a lymph node pathology detection result; Acquire prompt information for excluding the correlation between the candidate delineated areas and the lymph node delineation regions that does not meet the preset conditions, and input the prompt information and the preoperative medical image sample into the lymph node delineation model to be trained, and the lymph node delineation model extracts the image features corresponding to the candidate delineated areas in the preoperative medical image sample according to the prompt information, and outputs the lymph node delineation results and the predicted pathological detection results according to the image features; According to the difference between the lymph node delineation result and the lymph node annotation information and the difference between the predicted pathological result and the lymph node pathological detection result, the model parameters of the lymph node delineation model are adjusted to obtain a trained lymph node delineation model.

8. The prediction device according to claim 7, characterized in that The preoperative medical image sample includes the target structure tissue, lymph nodes within a preset range of the target structure tissue, and adjacent structure tissues outside the target structure tissue; The prediction device obtains the prompt information through the following steps: Acquiring clinical prior information indicating abnormal conditions of target structural tissue of the sample object, and acquiring anatomical structural restriction information of the adjacent structural tissue; The prompt information is obtained according to the clinical prior information and the anatomical structure restriction information.

9. A prediction method, characterized in that: include: Determining a lymph node delineation area based on a preoperative medical image containing lymph nodes corresponding to the target object; Extracting features of the lymph node delineation area to obtain first feature information corresponding to the radiomics feature, and obtaining second feature information corresponding to the clinical feature of the target object; The first characteristic information and the second characteristic information are input into a diagnosis model to obtain postoperative lymph node diagnosis information.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to claim 9 are implemented.

Images