Preoperative prediction model training method and device for ovarian cancer tumor reduction
By acquiring text and image data from ovarian cancer patients and training a prediction model using a neural network model, the problem of insufficient prediction accuracy for ovarian cancer cytoreductive surgery in existing technologies has been solved, achieving efficient preoperative assessment.
Patent Information
- Application Number
- CN202410109766.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-25
- Publication Date
- 2026-02-06
AI Technical Summary
In existing technologies, the accuracy of preoperative prediction for ovarian cancer cytoreductive surgery relies on the physician's personal experience, resulting in inaccurate predictions and difficulty in effectively guiding the surgery.
By acquiring textual and image data from ovarian cancer patients, a first and second risk value were determined, respectively. A prediction model was trained using a neural network model, and preoperative assessment was conducted by combining multiple data types.
It improves the accuracy of preoperative prediction for ovarian cancer cytoreductive surgery and provides more reliable preoperative assessment guidance.
Smart Images
Figure CN121483493A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical data processing technology, and in particular to a training method and apparatus for a predictive model before ovarian cancer debulking surgery. Background Technology
[0002] Ovarian cancer is a malignant tumor of the ovary, referring to malignant tumors that grow on the ovary. 90%–95% of these are primary ovarian cancers, while the remaining 5%–10% are metastases from primary cancers in other parts of the body. Because early-stage ovarian cancer often lacks symptoms and screening methods are limited, early diagnosis is relatively difficult.
[0003] For patients with advanced ovarian cancer, cytoreductive surgery is commonly used. This technique involves first removing the entire Douglas fossa as a pseudocyst for the tumor, and then removing other metastatic lesions and affected organs in the abdominal cavity. A comprehensive evaluation is necessary before this surgery to ensure its successful execution.
[0004] Currently, relying on physicians' personal work experience for preoperative assessment results in low accuracy of predictions, making it difficult to provide effective guidance for subsequent surgeries. Summary of the Invention
[0005] This application provides a method and apparatus for training a prediction model for ovarian cancer cytoreductive surgery, which is used to improve the accuracy of preoperative prediction for ovarian cancer cytoreductive surgery.
[0006] This invention provides a method for training a predictive model before ovarian cancer cytoreductive surgery, the method comprising:
[0007] Obtain sample data of ovarian cancer patients and their corresponding assessment labels; the assessment labels are risk assessment values before ovarian cancer cytoreductive surgery.
[0008] First sample data and second sample data are extracted from the sample data. The first sample data is text data of ovarian cancer patients; the second sample data is image data of ovarian cancer patients.
[0009] The first risk value and the second risk value are determined based on the first sample data and the second sample data, respectively.
[0010] The first sample data is input into the first network model to extract text data features, and the second sample data is input into the second network model to extract image data features.
[0011] A first data feature is determined by using the first risk value and the text data features, and a second data feature is determined by using the second risk value and the image data features;
[0012] The neural network model is trained based on the first data feature, the second data feature, and the evaluation label to obtain a prediction model, which is then used to evaluate and predict the outcome of ovarian cancer cytoreductive surgery.
[0013] In an optional embodiment of the present invention, determining the first risk value and the second risk value based on the first sample data and the second sample data respectively includes:
[0014] Calculate the similarity between the first sample data and each text data item in the first database; calculate the matching degree between the second sample data and each image data item in the second database.
[0015] The risk value corresponding to the text data with the highest similarity in the first database is determined as the first risk value, and the risk value corresponding to the image data with the highest matching degree in the second database is determined as the second risk value;
[0016] The first database stores the risk values corresponding to each type of text data, and the second database stores the risk values corresponding to each type of image data.
[0017] In an optional embodiment of the present invention, determining the first risk value and the second risk value based on the first sample data and the second sample data respectively includes:
[0018] The first sample data is input into a pre-fitted model to obtain the corresponding first risk value;
[0019] The second sample data is subjected to image recognition technology to obtain an image recognition result, and the second risk value corresponding to the image recognition result is determined according to a preset mapping table.
[0020] In an optional embodiment of the present invention, determining the first data feature using the first risk value and the text data feature, and determining the second data feature using the second risk value and the image data feature, includes:
[0021] The first data feature is obtained by concatenating the first risk value and the text data feature, and the second data feature is obtained by concatenating the second risk value and the image data feature.
[0022] In an optional embodiment of the present invention, training the neural network model to obtain a prediction model based on the first data feature, the second data feature, and the evaluation label includes:
[0023] The first data feature and the second data feature are input into the neural network model to obtain the first prediction evaluation value and the second prediction evaluation value;
[0024] Calculate the first loss value of the first predicted evaluation value and the evaluation label, and the second loss value of the second predicted evaluation value and the evaluation label;
[0025] The target loss value is obtained by fusing the first loss value and the second loss value and their corresponding weight values.
[0026] If the target loss value is determined to be less than the preset value, then the training of the neural network model is completed and a prediction model is obtained;
[0027] If the target loss value is determined to be greater than or equal to a preset value, then the neural network model continues to be trained until the target loss value is less than the preset value.
[0028] In an optional embodiment of the present invention, after determining that the target loss value is greater than or equal to a preset value, the method further includes:
[0029] Based on the difference between the first loss value and the second loss value;
[0030] Update the weight values corresponding to the first loss value and the second loss value according to the difference.
[0031] In an optional embodiment of the present invention, the step of inputting the first data feature and the second data feature into the neural network model to obtain a first prediction evaluation value and a second prediction evaluation value includes:
[0032] The first data feature is input into the first prediction module of the neural network model to obtain the first prediction evaluation value; the second data feature is input into the second prediction module of the neural network model to obtain the second prediction evaluation value.
[0033] In an optional embodiment of the present invention, the step of fusing the first loss value and the second loss value and their corresponding weight values to obtain the target loss value includes:
[0034] The target loss value is calculated using the following formula:
[0035] loss = αloss1 + βloss2;
[0036] Where loss1 is the first loss value, loss2 is the second loss value, α is the weight value of the first loss value, and β is the weight value of the second loss value.
[0037] In an optional embodiment of the present invention, the text data includes basic data and clinical test data of ovarian cancer patients; the image data includes upper abdominal image data, pelvic image data, and laparoscopic exploration image data of ovarian cancer patients.
[0038] This invention provides a training device for a predictive model before ovarian cancer cytoreductive surgery, the device comprising:
[0039] The acquisition module is used to acquire sample data of ovarian cancer patients and their corresponding assessment labels; the assessment labels are risk assessment values before ovarian cancer cytoreductive surgery.
[0040] The extraction module is used to extract first sample data and second sample data from the sample data, wherein the first sample data is text data of ovarian cancer patients; and the second sample data is image data of ovarian cancer patients.
[0041] The determination module is used to determine a first risk value and a second risk value based on the first sample data and the second sample data, respectively.
[0042] The extraction module is also used to input the first sample data into the first network model to extract text data features, and to input the second sample data into the second network model to extract image data features.
[0043] The determining module is further configured to determine a first data feature by using the first risk value and the text data feature, and to determine a second data feature by using the second risk value and the image data feature;
[0044] The training module is used to train a neural network model based on the first data features, the second data features, and the evaluation labels to obtain a prediction model, so as to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer through the prediction model.
[0045] This application provides a training method and apparatus for a predictive model for ovarian cancer cytoreductive surgery. First, sample data of ovarian cancer patients and their corresponding assessment labels are acquired. Then, first sample data and second sample data are extracted from the sample data, and a first risk value and a second risk value are determined based on the first and second sample data, respectively. The first sample data is input into a first network model to extract text data features, and the second sample data is input into a second network model to extract image data features. A first data feature is determined using the first risk value and text data features, and a second data feature is determined using the second risk value and image data features. Finally, the neural network model is trained based on the first and second data features and the assessment labels to obtain a predictive model for preoperative assessment and prediction of ovarian cancer cytoreductive surgery. Since the first sample data in this application consists of text data from ovarian cancer patients, and the second sample data consists of image data from ovarian cancer patients, the predictive model trained based on the first and second sample data can combine multiple data sources for preoperative assessment of ovarian cancer cytoreductive surgery, thereby improving the accuracy of preoperative prediction for ovarian cancer cytoreductive surgery. Attached Figure Description
[0046] Figure 1 A flowchart illustrating the training method for a preoperative predictive model for ovarian cancer cytoreductive surgery provided in this application;
[0047] Figure 2 A flowchart of a prediction model training process provided in this application;
[0048] Figure 3 This is a structural diagram of a training device for a preoperative prediction model of ovarian cancer cytoreductive surgery provided in this application. Detailed Implementation
[0049] To better understand the above technical solutions, the technical solutions of the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this application and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this application, rather than limitations on the technical solutions of this application. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other.
[0050] Please see Figure 1 The diagram illustrates a method for training a preoperative predictive model for ovarian cancer cytoreductive surgery, as provided in an embodiment of the present invention. The execution flow of this method is as follows:
[0051] S101, Obtain sample data of ovarian cancer patients and their corresponding assessment labels; the assessment labels are the risk assessment values before ovarian cancer cytoreductive surgery.
[0052] The risk assessment value is an artificially calibrated value used to represent the risk of ovarian cancer cytoreductive surgery. It can be represented by a value from 1 to N, with a higher value indicating a greater risk; or by the letter AN. This embodiment does not make a specific limitation on this.
[0053] It should be noted that the sample data of ovarian cancer patients in this embodiment includes detection data, image data, clinical data, etc., which can be used to assess the risk of ovarian cancer cytoreductive surgery.
[0054] S102, extract the first sample data and the second sample data from the sample data.
[0055] The first sample data consists of text data from ovarian cancer patients; the second sample data consists of imaging data from ovarian cancer patients. Specifically, the text data includes basic data and clinical test data of ovarian cancer patients; the imaging data includes upper abdominal imaging data, pelvic imaging data, and laparoscopic imaging data of ovarian cancer patients. Basic data may include age, height, weight, ASA classification, CA125, etc., but this embodiment does not specifically limit these.
[0056] In this embodiment, the sample data is divided into first sample data and second sample data according to their different characteristics, so that the risk value can be determined in subsequent steps based on the divided first sample data and second sample data, and the risk of ovarian cancer cytoreductive surgery can be assessed based on the determined risk value.
[0057] S103, determine the first risk value and the second risk value based on the first sample data and the second sample data, respectively.
[0058] The first risk value and the second risk value are determined based on the assessment label, that is, based on the risk assessment value before ovarian cancer cytoreductive surgery. In the process of determining the assessment label corresponding to the sample data, the first risk value and the second risk value need to be set. The first risk value and the second risk value can represent the risk value of the first sample data and the second sample data for assessing the ovarian cancer cytoreductive surgery. In this embodiment, the sum of the first risk value and the second risk value can be the risk assessment value before ovarian cancer cytoreductive surgery.
[0059] In one optional embodiment provided in this application, determining the first risk value and the second risk value based on the first sample data and the second sample data respectively includes: calculating the similarity between the first sample data and each text data in the first database; calculating the matching degree between the second sample data and each image data in the second database; determining the risk value corresponding to the text data with the highest similarity in the first database as the first risk value, and determining the risk value corresponding to the image data with the highest matching degree in the second database as the second risk value.
[0060] The first database stores risk values corresponding to each type of text data. In this embodiment, after obtaining the first sample data, the similarity between the first sample data and each type of sample data in the first database is calculated. Then, the risk value corresponding to the text data with the highest similarity in the first database is determined as the first risk value. That is, this embodiment determines the first risk value corresponding to the first sample data through similarity. The second database stores risk values corresponding to each type of image data. After obtaining the second sample data, the matching degree between the second sample data and each type of image data in the second database is calculated. Then, the risk value corresponding to the image data with the highest matching degree in the second database is determined as the second risk value.
[0061] In another optional embodiment provided in this application, determining the first risk value and the second risk value based on the first sample data and the second sample data respectively includes: inputting the first sample data into a preset fitting model to obtain the corresponding first risk value; performing image recognition on the second sample data using image recognition technology to obtain an image recognition result; and determining the second risk value corresponding to the image recognition result according to a preset mapping table.
[0062] The pre-set fitting model is obtained by fitting a large amount of known data. The pre-set fitting model is as follows:
[0063] y = ax1 + bx2
[0064] Where x1 represents the baseline data of ovarian cancer patients, x2 represents the clinical test data of ovarian cancer patients, a and b are the fitting coefficients, and y is the first risk value. That is, after obtaining the first sample data, the first sample data is input into the above-mentioned pre-set fitting model, and the first risk value can be obtained by calculating this formula.
[0065] In this embodiment of the application, the risk values corresponding to the first sample data and the second sample data can be determined by using a pre-set fitting model and image recognition technology, so that the model can be trained according to the risk value in subsequent steps, so that the trained model can be used to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer.
[0066] S104, input the first sample data into the first network model to extract text data features, and input the second sample data into the second network model to extract image data features.
[0067] In this embodiment, the first network model and the second network model can be network models with the same structure or network models with different structures. The text data features and image data features are the output data of the feature extraction layers of the first network model and the second network model, respectively. That is, the relevant data features can be extracted through the first network model and the second network model, without the need to perform result prediction on the input data.
[0068] S105, determine the first data feature by using the first risk value and text data features, and determine the second data feature by using the second risk value and image data features.
[0069] In this embodiment, after obtaining text data features and image data features through the first network model and the second network model, in order to improve the accuracy of training the neural network model in subsequent steps, this embodiment needs to determine the first data feature by the first risk value and the text data features, and determine the second data feature by the second risk value and the image data features. Then, the neural network model is trained using the first data feature and the second data feature, thereby using the trained neural network model to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer.
[0070] In one optional embodiment provided in this application, determining the first data feature by the first risk value and the text data feature, and determining the second data feature by the second risk value and the image data feature, includes: concatenating the first risk value and the text data feature to obtain the first data feature, and concatenating the second risk value and the image data feature to obtain the second data feature.
[0071] S106, The neural network model is trained based on the first data features, the second data features, and the evaluation label to obtain a prediction model, so as to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer.
[0072] In this embodiment, the first data feature and the second data feature can be input into a neural network model. Then, the evaluation prediction value is obtained through the neural network model. The loss value is calculated based on the loss function corresponding to the evaluation prediction value and the evaluation label. It is determined whether the loss value is less than a preset value. If it is not less than the preset value, the neural network model is trained again using other sample data. If it is less than the preset value, it means that the accuracy of the neural network model meets the requirements. At this time, the trained prediction model can be used to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer.
[0073] Specifically, in this embodiment, the data to be predicted from ovarian cancer patients is obtained. Then, through steps S101 to S1015, the data to be predicted is converted into corresponding first data features and second data features. After that, the first data features and second data features are input into the trained prediction model. The prediction model can then be used to obtain the preoperative evaluation results of ovarian cancer cytoreductive surgery.
[0074] It should be noted that in this embodiment, the first and second data features corresponding to the data to be predicted can be text data features, image data features, or combinations of text data features, image data features and their corresponding risk values.
[0075] like Figure 2 As shown, in an optional embodiment provided in this application, training the neural network model to obtain a prediction model based on the first data features, the second data features, and the evaluation label includes:
[0076] S1061, the first data feature and the second data feature are input into the neural network model to obtain the first prediction evaluation value and the second prediction evaluation value.
[0077] Specifically, the neural network model in this embodiment includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The output layer includes a first output layer and a second output layer. The first output layer includes a first sub-output layer and a second sub-output layer. The first sub-output layer outputs a first predicted evaluation value, and the second sub-output layer outputs the same first predicted evaluation value. In this embodiment, the input layer, convolutional layer, pooling layer, and fully connected layer are shared layers. That is, the first data feature and the second data feature are sequentially input into the input layer, convolutional layer, pooling layer, and fully connected layer. Then, the first data feature is input into the fully connected layer and then into the first sub-output layer to obtain the first predicted evaluation value; similarly, the second data feature is input into the fully connected layer and then into the second sub-output layer to obtain the second predicted evaluation value.
[0078] In one optional embodiment provided in this application, the step of inputting the first data feature and the second data feature into the neural network model to obtain a first prediction evaluation value and a second prediction evaluation value includes: inputting the first data feature into a first prediction module in the neural network model to obtain the first prediction evaluation value; and inputting the second data feature into a second prediction module in the neural network model to obtain the second prediction evaluation value. The first prediction module is disposed within a first sub-output layer, and the second prediction module is disposed within a second sub-output layer.
[0079] S1062, calculate the first loss value of the first predicted evaluation value and the evaluation label, and the second loss value of the second predicted evaluation value and the evaluation label.
[0080] In this embodiment, different loss functions are used to calculate the first loss value and the second loss value.
[0081] Specifically, a first loss function is used to calculate a first loss value, and a second loss function is used to calculate a second loss value. The first and second loss functions are as follows:
[0082]
[0083]
[0084] Where loss1 is the first loss value, loss2 is the second loss value, N is the number of sample data, and y i x is the first predicted evaluation value. i This is the second predicted evaluation value. To assess the preoperative risk assessment values for ovarian cancer cytoreductive surgery corresponding to the label.
[0085] S1063, the target loss value is obtained by fusing the first loss value and the second loss value and their corresponding weight values.
[0086] In an optional embodiment provided in this application, the step of fusing the first loss value and the second loss value and their corresponding weight values to obtain the target loss value includes: calculating the target loss value using the following formula:
[0087] loss = αloss1 + βloss2;
[0088] Where loss1 is the first loss value, loss2 is the second loss value, α is the weight value of the first loss value, and β is the weight value of the second loss value.
[0089] It should be noted that the initial weight values of the first loss value and the second loss value in this embodiment can be set as needed, or they can be set to equal values, such as setting the initial weight values of the first loss value and the second loss value to 0.5.
[0090] S1064, If it is determined that the target loss value is less than the preset value, then the training of the neural network model is completed to obtain the prediction model.
[0091] S1065, if it is determined that the target loss value is greater than or equal to the preset value, then continue training the neural network model until the target loss value is less than the preset value.
[0092] In this step, S1065 is a parallel step to S1064. If the target loss value is determined to be greater than or equal to the preset value, it means that the accuracy of the neural network model has not yet met the requirements, and the neural network model needs to be trained again until the target loss value is less than the preset value. If the target loss value is less than the preset value, the training task of the neural network model is completed and a prediction model is obtained. The trained prediction model can be used to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer.
[0093] Furthermore, after determining that the target loss value is greater than or equal to a preset value, the method further includes: updating the weight values corresponding to the first loss value and the second loss value based on the difference between the first loss value and the second loss value. Specifically, the first loss value and the second loss value are inversely proportional to their corresponding weight values; that is, the larger the first loss value and the second loss value, the smaller their corresponding weight values; and the smaller the first loss value and the second loss value, the larger their corresponding weight values. The two weight values are added together to equal 1.
[0094] Specifically, when the difference is positive, it means that the first loss value is greater than the second loss value. In order to make the target loss value smaller, the weight of the first loss value needs to be increased and the weight of the second loss value needs to be decreased. Conversely, when the difference is negative, it means that the first loss value is less than the second loss value. In order to make the target loss value smaller, the weight of the first loss value needs to be decreased and the weight of the second loss value needs to be increased.
[0095] In this embodiment, the weight values corresponding to the first and second loss values are adjusted according to the relationship between the calculated first loss value and the second loss value, so as to ensure that the target loss value becomes smaller, thereby making the neural network model gradually converge, thus completing the training of the neural network model and obtaining the prediction model.
[0096] This invention provides a method for training a predictive model for ovarian cancer cytoreductive surgery. First, sample data of ovarian cancer patients and their corresponding assessment labels are acquired. Then, first sample data and second sample data are extracted from the sample data, and a first risk value and a second risk value are determined based on the first and second sample data, respectively. The first sample data is input into a first network model to extract text data features, and the second sample data is input into a second network model to extract image data features. A first data feature is determined using the first risk value and text data features, and a second data feature is determined using the second risk value and image data features. Finally, the neural network model is trained based on the first and second data features and the assessment labels to obtain a predictive model, which is then used to assess and predict ovarian cancer cytoreductive surgery. Since the first sample data in this application consists of text data from ovarian cancer patients, and the second sample data consists of image data from ovarian cancer patients, the predictive model trained based on the first and second sample data can combine multiple types of data to perform preoperative assessment of ovarian cancer cytoreductive surgery, thereby improving the accuracy of intelligent predictive model training. Therefore, the predictive model in this application can improve the accuracy of preoperative prediction for ovarian cancer cytoreductive surgery.
[0097] In one embodiment, a training device for a predictive model prior to ovarian cancer cytoreductive surgery is provided. This training device for the predictive model prior to ovarian cancer cytoreductive surgery corresponds one-to-one with the training method for the aforementioned predictive model prior to ovarian cancer cytoreductive surgery. For example... Figure 3 As shown, the device includes:
[0098] The acquisition module 31 is used to acquire sample data of ovarian cancer patients and their corresponding assessment labels; the assessment labels are risk assessment values before ovarian cancer cytoreductive surgery.
[0099] Extraction module 32 is used to extract first sample data and second sample data from the sample data, wherein the first sample data is text data of ovarian cancer patients; and the second sample data is image data of ovarian cancer patients.
[0100] The determination module 33 is used to determine a first risk value and a second risk value based on the first sample data and the second sample data, respectively.
[0101] The extraction module 32 is also used to input the first sample data into the first network model to extract text data features, and input the second sample data into the second network model to extract image data features;
[0102] The determining module 33 is further configured to determine a first data feature by using the first risk value and the text data feature, and to determine a second data feature by using the second risk value and the image data feature;
[0103] Training module 34 is used to train a neural network model based on the first data features, the second data features, and the evaluation label to obtain a prediction model, so as to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer through the prediction model.
[0104] In an optional embodiment provided by the present invention, the determining module 33 is specifically used for:
[0105] Calculate the similarity between the first sample data and each text data item in the first database; calculate the matching degree between the second sample data and each image data item in the second database.
[0106] The risk value corresponding to the text data with the highest similarity in the first database is determined as the first risk value, and the risk value corresponding to the image data with the highest matching degree in the second database is determined as the second risk value;
[0107] The first database stores the risk values corresponding to each type of text data, and the second database stores the risk values corresponding to each type of image data.
[0108] In an optional embodiment provided by the present invention, the determining module 33 is specifically used for:
[0109] The first sample data is input into a pre-fitted model to obtain the corresponding first risk value;
[0110] The second sample data is subjected to image recognition technology to obtain an image recognition result, and the second risk value corresponding to the image recognition result is determined according to a preset mapping table.
[0111] In an optional embodiment provided by the present invention, the determining module 33 is specifically used for:
[0112] The first data feature is obtained by concatenating the first risk value and the text data feature, and the second data feature is obtained by concatenating the second risk value and the image data feature.
[0113] In an optional embodiment provided by the present invention, the training module 34 is specifically used for:
[0114] The first data feature and the second data feature are input into the neural network model to obtain the first prediction evaluation value and the second prediction evaluation value;
[0115] Calculate the first loss value of the first predicted evaluation value and the evaluation label, and the second loss value of the second predicted evaluation value and the evaluation label;
[0116] The target loss value is obtained by fusing the first loss value and the second loss value and their corresponding weight values.
[0117] If the target loss value is determined to be less than the preset value, then the training of the neural network model is completed and a prediction model is obtained;
[0118] If the target loss value is determined to be greater than or equal to a preset value, then the neural network model continues to be trained until the target loss value is less than the preset value.
[0119] In an optional embodiment provided by the present invention, the training module 34 is further configured to:
[0120] Based on the difference between the first loss value and the second loss value;
[0121] Update the weight values corresponding to the first loss value and the second loss value according to the difference.
[0122] The first data feature is input into the first prediction module of the neural network model to obtain the first prediction evaluation value; the second data feature is input into the second prediction module of the neural network model to obtain the second prediction evaluation value.
[0123] In an optional embodiment provided by the present invention, the training module 34 is specifically used for:
[0124] The target loss value is calculated using the following formula:
[0125] loss = αloss1 + βloss2;
[0126] Where loss1 is the first loss value, loss2 is the second loss value, α is the weight value of the first loss value, and β is the weight value of the second loss value.
[0127] In an optional embodiment of the present invention, the text data includes basic data and clinical test data of ovarian cancer patients; the image data includes upper abdominal image data, pelvic image data, and laparoscopic exploration image data of ovarian cancer patients.
[0128] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A method for training a predictive model before ovarian cancer cytoreductive surgery, characterized in that, The method includes: Obtain sample data of ovarian cancer patients and their corresponding assessment labels; the assessment labels are risk assessment values before ovarian cancer cytoreductive surgery. First sample data and second sample data are extracted from the sample data. The first sample data is text data of ovarian cancer patients; the second sample data is image data of ovarian cancer patients. The first risk value and the second risk value are determined based on the first sample data and the second sample data, respectively. The first sample data is input into the first network model to extract text data features, and the second sample data is input into the second network model to extract image data features. A first data feature is determined by using the first risk value and the text data features, and a second data feature is determined by using the second risk value and the image data features; The neural network model is trained based on the first data feature, the second data feature, and the evaluation label to obtain a prediction model, which is then used to evaluate and predict the outcome of ovarian cancer cytoreductive surgery.
2. The method according to claim 1, characterized in that, The step of determining the first risk value and the second risk value based on the first sample data and the second sample data respectively includes: Calculate the similarity between the first sample data and each text data item in the first database; calculate the matching degree between the second sample data and each image data item in the second database. The risk value corresponding to the text data with the highest similarity in the first database is determined as the first risk value, and the risk value corresponding to the image data with the highest matching degree in the second database is determined as the second risk value; The first database stores the risk values corresponding to each type of text data, and the second database stores the risk values corresponding to each type of image data.
3. The method according to claim 1, characterized in that, The step of determining the first risk value and the second risk value based on the first sample data and the second sample data respectively includes: The first sample data is input into a pre-fitted model to obtain the corresponding first risk value; The second sample data is subjected to image recognition technology to obtain an image recognition result, and the second risk value corresponding to the image recognition result is determined according to a preset mapping table.
4. The method according to claim 2 or 3, characterized in that, The step of determining the first data feature using the first risk value and the text data feature, and determining the second data feature using the second risk value and the image data feature, includes: The first data feature is obtained by concatenating the first risk value and the text data feature, and the second data feature is obtained by concatenating the second risk value and the image data feature.
5. The method according to claim 4, characterized in that, The step of training a neural network model to obtain a prediction model based on the first data features, the second data features, and the evaluation label includes: The first data feature and the second data feature are input into the neural network model to obtain the first prediction evaluation value and the second prediction evaluation value; Calculate the first loss value of the first predicted evaluation value and the evaluation label, and the second loss value of the second predicted evaluation value and the evaluation label; The target loss value is obtained by fusing the first loss value and the second loss value and their corresponding weight values. If the target loss value is determined to be less than the preset value, then the training of the neural network model is completed and a prediction model is obtained; If the target loss value is determined to be greater than or equal to a preset value, then the neural network model continues to be trained until the target loss value is less than the preset value.
6. The method according to claim 5, characterized in that, After determining that the target loss value is greater than or equal to a preset value, the method further includes: Based on the difference between the first loss value and the second loss value; Update the weight values corresponding to the first loss value and the second loss value according to the difference.
7. The method according to claim 5, characterized in that, The step of inputting the first data feature and the second data feature into the neural network model to obtain the first prediction evaluation value and the second prediction evaluation value includes: The first data feature is input into the first prediction module of the neural network model to obtain the first prediction evaluation value; the second data feature is input into the second prediction module of the neural network model to obtain the second prediction evaluation value.
8. The method according to claim 7, characterized in that, The step of fusing the first loss value and the second loss value and their corresponding weight values to obtain the target loss value includes: The target loss value is calculated using the following formula: loss = αloss1 + βloss2; Where loss1 is the first loss value, loss2 is the second loss value, α is the weight value of the first loss value, and β is the weight value of the second loss value.
9. The method according to claim 1, characterized in that, The text data includes basic data and clinical test data of ovarian cancer patients; the imaging data includes upper abdominal imaging data, pelvic imaging data, and laparoscopic imaging data of ovarian cancer patients.
10. A training device for a predictive model before ovarian cancer cytoreductive surgery, characterized in that, The device includes: The acquisition module is used to acquire sample data of ovarian cancer patients and their corresponding assessment labels; the assessment labels are risk assessment values before ovarian cancer cytoreductive surgery. The extraction module is used to extract first sample data and second sample data from the sample data, wherein the first sample data is text data of ovarian cancer patients; and the second sample data is image data of ovarian cancer patients. The determination module is used to determine a first risk value and a second risk value based on the first sample data and the second sample data, respectively. The extraction module is also used to input the first sample data into the first network model to extract text data features, and to input the second sample data into the second network model to extract image data features. The determining module is further configured to determine a first data feature by using the first risk value and the text data feature, and to determine a second data feature by using the second risk value and the image data feature; The training module is used to train a neural network model based on the first data features, the second data features, and the evaluation labels to obtain a prediction model, so as to evaluate and predict the preoperative cytoreductive surgery for ovarian cancer through the prediction model.