An endometrial benign and malignant disease auxiliary prediction model construction method and system

Abstract

The application relates to an endometrial benign and malignant disease auxiliary prediction model construction method and system, which comprises the following steps: obtaining first historical image information of a non-target part and second historical image information of a target part; respectively performing pretreatment and respectively constructing corresponding pre-training sample data sets and training sample data sets; constructing a convolutional neural network model, and inputting the pre-training sample data set into the convolutional neural network model for comparative learning pre-training; inputting the training sample data set into the pre-trained convolutional neural network model for training until the training is completed, so that an endometrial benign and malignant disease auxiliary prediction model is obtained. By using the comparative learning technology, the recognition ability of the convolutional neural network model to subtle lesion characteristics is enhanced, the model generalization ability is improved, the sensitivity and specificity of diagnosis are significantly improved, the misdiagnosis rate is reduced, the recognition accuracy of the model is greatly improved, the possibility of misdiagnosis and unnecessary biopsy is reduced, and the perception deviation and visual fatigue of an endoscope doctor are helped to be relieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical auxiliary diagnosis, and particularly relates to a method and system for constructing an endometrial benign and malignant disease auxiliary prediction model. BACKGROUND

[0002] Endometrial cancer (EC) is one of the common gynecological malignancies worldwide, and early diagnosis and treatment are crucial to improve the survival rate of patients. Atypical endometrial hyperplasia (AEH) is a precancerous lesion with a high risk of progressing to endometrial cancer. Traditional hysteroscopy and diagnostic curettage methods are prone to misdiagnosis in identifying atypical endometrial hyperplasia (AEH) and endometrial cancer (EC). Even experienced gynecological endoscopic physicians may experience visual fatigue due to long working hours, increasing the risk of misdiagnosis or missed diagnosis, which greatly disturbs the actual clinical work of doctors. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a method and system for constructing an endometrial benign and malignant disease auxiliary prediction model to solve the above technical problems of the prior art.

[0004] The technical solution for solving the above technical problem is as follows: a method for constructing an endometrial benign and malignant disease auxiliary prediction model, comprising the following steps:

[0005] Obtaining first historical image information of a non-target part and second historical image information of a target part;

[0006] Preprocessing the first historical image information and the second historical image information, respectively, and constructing corresponding pre-training sample data sets and training sample data sets, respectively;

[0007] Constructing a convolutional neural network model, and inputting the pre-training sample data set into the convolutional neural network model for comparative learning pre-training;

[0008] Inputting the training sample data set into the pre-trained convolutional neural network model for training until the training is completed, and obtaining an endometrial benign and malignant disease auxiliary prediction model.

[0009] The uterus endometrial benign and malignant disease auxiliary prediction model construction method has the beneficial effects that: the pre-training sample data set constructed by the pre-processed first historical image information of the non-target part is used for pre-training of the convolutional neural network model through contrastive learning, the contrastive learning technology is used to enhance the recognition ability of the convolutional neural network model to subtle lesion characteristics, and the model generalization ability is improved. Especially in the case that malignant and benign lesions are difficult to distinguish, the sensitivity and specificity of diagnosis are significantly improved, the misdiagnosis rate is reduced, the performance of the convolutional neural network model under the condition of limited labeled data is enhanced, and atypical endometrial hyperplasia (AEH), endometrial cancer (EC) and benign lesions can be accurately distinguished. The training sample data set constructed by the pre-processed second historical image information of the target part is used for training of the convolutional neural network model, the recognition accuracy of the model is greatly improved, the possibility of misdiagnosis and unnecessary biopsy is reduced, and the perception bias and visual fatigue commonly encountered by endoscope doctors are helped to be relieved.

[0010] On the basis of the above technical scheme, the application can also be improved as follows:

[0011] Further: the pre-processing of the first historical image information and the second historical image information specifically includes the following steps:

[0012] The first historical image information and the second historical image information are respectively subjected to standardization processing and normalization processing, wherein the standardization processing includes scaling processing, translation processing, rotation processing and flip processing, and a unified format and size image is obtained.

[0013] The beneficial effects of the above further scheme are that: through the standardization processing and normalization processing of the first historical image information and the second historical image information, overfitting can be prevented, and the image is normalized and adjusted to a unified pixel, which facilitates the recognition and extraction processing of the convolutional neural network model, and improves the performance of the model.

[0014] Further: the pre-training of the pre-training sample data set into the convolutional neural network model specifically includes the following steps:

[0015] The convolutional neural network model performs feature extraction processing on the first historical image information in the pre-training sample data set, and maps the extracted pre-training features to a low-dimensional space;

[0016] In the low-dimensional space, a contrastive loss function is constructed, and the cosine similarity between the feature vectors corresponding to the extracted pre-training features is calculated according to the contrastive loss function, and the calculation formula of the contrastive loss function is:

[0017]

[0018] Wherein, zi and z j represents cosine similarity between the positive feature vector z i and z j , τ is a temperature parameter, and l i represents a loss function between the positive feature vector z j and z i,j , N is the number of samples; i j The distribution concentration of the loss function is adjusted based on the cosine similarity until the change value of the contrast loss function is within a preset range, and the pre-training is completed.

[0019] The distribution concentration of the loss function is adjusted based on the cosine similarity until the change value of the contrast loss function is within a preset range, and the pre-training is completed.

[0020] The above further scheme has the beneficial effects that: by constructing a contrast loss function, the cosine similarity between feature vectors can be accurately calculated, and the cross-entropy is adjusted and scaled using a temperature parameter, so that the change value of the contrast loss function tends to be stable, thereby greatly improving the feature extraction capability of the convolutional neural network model, and enabling the convolutional neural network model to distinguish subtle differences in medical images.

[0021] Further, the training sample data set is input into the pre-trained convolutional neural network model for training until the training is completed, and an endometrial benign and malignant disease auxiliary prediction model is obtained, which specifically includes the following steps:

[0022] The pre-trained convolutional neural network model classifies and identifies the second historical image information in the training sample data set to obtain case classification information.

[0023] A focal loss function is constructed based on the case classification information and the corresponding true label value, and the specific formula is:

[0024] focal loss(p t )=(1-p t ) 0·1 CE(p,y) (2)

[0025] CE(p,y)= -log (p t ) (3)

[0026]

[0027] Wherein, p is the probability that the sample is correctly predicted and classified by the convolutional neural network model, y is the true label value of the sample, p t is the predicted label value of the model, and CE(p,y) is the cross-entropy loss value between the true label value and the predicted label value, focal loss(p t ​) is a focal loss value;

[0028] The model parameter minimizing the loss value of the focal loss function is taken as a target parameter of the convolutional neural network model, training is completed, and an endometrial benign and malignant disease auxiliary prediction model is obtained.

[0029] The beneficial effect of the further scheme is that the second historical image information in the training sample data set is classified and recognized by the convolutional neural network model, the case classification information is recognized, then the focal loss function is constructed based on the case classification information recognized by the convolutional neural network model and the corresponding true label value, the focal loss value between the two is calculated, the sensitivity of the endometrial benign and malignant disease auxiliary prediction model to the endometrial image is greatly increased by introducing the adjustment factor on the basis of the cross-entropy loss function, and the recognition accuracy of the endometrial benign and malignant disease auxiliary prediction model is ensured.

[0030] Further, the endometrial benign and malignant disease auxiliary prediction model construction method further includes the following steps:

[0031] The endometrial benign and malignant disease auxiliary prediction model obtained after training is evaluated, and an evaluation result is obtained, and the specific calculation formula is:

[0032]

[0033] Wherein, TP, TN, FP and FN represent true positive, true negative, false positive and false negative respectively, PPV represents positive predictive value, and NPV represents negative predictive value.

[0034] The beneficial effect of the further scheme is that the accuracy, sensitivity, specificity, positive predictive value, negative predictive value and F1 score of the convolutional neural network model can be calculated, which can represent the classification efficiency and classification accuracy of the trained convolutional neural network model, and the prediction score of each lesion can be calculated according to the prediction result.

[0035] The application also provides an endometrial benign and malignant disease auxiliary prediction model construction system, which comprises an acquisition module, a preprocessing module and a model training and recognition module.

[0036] The acquisition module is used for acquiring first historical image information of a non-target part and second historical image information of a target part.

[0037] The preprocessing module is used for preprocessing the first historical image information and the second historical image information respectively, and constructing corresponding pre-training sample data sets and training sample data sets respectively.

[0038] The model training identification module is configured to construct a convolutional neural network model and input the pre-training sample data set into the convolutional neural network model for contrastive learning pre-training.

[0039] The model training identification module is further configured to input the training sample data set into the pre-trained convolutional neural network model for training until the training is completed, thereby obtaining an endometrial benign and malignant disease auxiliary prediction model.

[0040] On the basis of the above technical solution, the application can be further improved as follows:

[0041] Further, the specific implementation of the model training identification module inputting the pre-training sample data set into the convolutional neural network model for contrastive learning pre-training is as follows:

[0042] The convolutional neural network model performs feature extraction processing on the first historical image information in the pre-training sample data set and maps the extracted pre-training features to a low-dimensional space;

[0043] In the low-dimensional space, a contrastive loss function is constructed, and the cosine similarity between the feature vectors corresponding to the extracted pre-training features is calculated according to the contrastive loss function, and the calculation formula of the contrastive loss function is as follows:

[0044]

[0045] wherein, z i and z j represent positive feature vectors, sim(z i ,z j ) represents the cosine similarity between the positive feature vectors z i and z j , τ is a temperature parameter, l i,j represents the loss function between the positive feature vectors z i and z j , and N is the sample quantity.

[0046] The distribution concentration of the loss function is adjusted based on the cosine similarity until the change value of the contrastive loss function is within a preset range, and the pre-training is completed.

[0047] The above further solution has the following beneficial effects: by constructing the contrastive loss function, the cosine similarity between the feature vectors can be accurately calculated, and the cross-entropy is adjusted and scaled by using the temperature parameter, so that the change value of the contrastive loss function tends to be stable, thereby greatly improving the feature extraction capability of the convolutional neural network model, and enabling the convolutional neural network model to distinguish subtle differences in medical images.

[0048] Further, the model training identification module inputs the training sample data set into the pre-trained convolutional neural network model for training until the training is completed, and the specific implementation of the endometrial benign and malignant disease auxiliary prediction model is:

[0049] The pre-trained convolutional neural network model performs classification and identification processing on the second historical image information in the training sample data set to obtain case classification information.

[0050] A focal loss function is constructed based on the case classification information and the corresponding true label value, and the specific formula is:

[0051] focal loss(p t )=(1-p t ) 0·1 CE(p,y) (2)

[0052] CE(p,y)= -log (p t ) (3)

[0053]

[0054] Wherein, p is the probability that the sample is correctly predicted and classified by the convolutional neural network model, y is the true label value of the sample, p t is the predicted label value of the model, and CE(p,y) is the cross-entropy loss value between the true label value and the predicted label value, and focal loss(p t ) is the focal loss value.

[0055] The model parameter with the minimum loss value of the focal loss function is taken as the target parameter of the convolutional neural network model, the training is completed, and the endometrial benign and malignant disease auxiliary prediction model is obtained.

[0056] The beneficial effects of the above further scheme are: the convolutional neural network model is used to perform classification and identification processing on the second historical image information in the training sample data set to identify the case classification information, then a focal loss function is constructed based on the case classification information identified by the convolutional neural network model and the corresponding true label value, the focal loss value between them is calculated, and by introducing the adjustment factor on the basis of the cross-entropy loss function, the sensitivity of the endometrial benign and malignant disease auxiliary prediction model to the endometrial image is greatly increased, and the recognition accuracy of the endometrial benign and malignant disease auxiliary prediction model is ensured.

[0057] The application also provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the endometrial benign and malignant disease auxiliary prediction model construction method.

[0058] The application further provides an endometrial benign and malignant disease auxiliary prediction model construction device, characterized by comprising a communication interface, a memory, a communication bus and a processor, wherein the processor, the communication interface and the memory communicate with each other through the communication bus.

[0059] The memory is used for storing a computer program.

[0060] The processor is used for executing the program stored in the memory to realize the steps of the endometrial benign and malignant disease auxiliary prediction model construction method. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 It is a flowchart of an endometrial benign and malignant disease auxiliary prediction model construction method according to an embodiment of the application.

[0062] Figure 2 It is a comparison diagram of the accuracy of an endometrial benign and malignant disease auxiliary prediction model and different experience levels of experts according to an embodiment of the application.

[0063] Figure 3 It is a visualization diagram of an endometrial benign and malignant disease auxiliary prediction model predicting atypical endometrial hyperplasia and endometrial cancer according to an embodiment of the application.

[0064] Figure 4 It is a structural diagram of an endometrial benign and malignant disease auxiliary prediction model construction system according to an embodiment of the application. DETAILED DESCRIPTION

[0065] The principles and characteristics of the application are described below in combination with the drawings, and the examples are only used to explain the application and are not used to limit the scope of the application.

[0066] As shown in the drawings, an endometrial benign and malignant disease auxiliary prediction model construction method comprises the following steps: Figure 1

[0067] S1: obtaining first historical image information of a non-target part and second historical image information of a target part;

[0068] S2: respectively pre-processing the first historical image information and the second historical image information, and respectively constructing corresponding pre-training sample data sets and training sample data sets;

[0069] S3: constructing a convolutional neural network model, and inputting the pre-training sample data set into the convolutional neural network model for comparative learning pre-training;

[0070] ​S4: inputting the training sample data set into the pre-trained convolutional neural network model for training until the convolutional neural network model meets a preset condition, until the training is completed, and an endometrial benign and malignant disease auxiliary prediction model is obtained.

[0071] The endometrial benign and malignant disease auxiliary prediction model construction method of the present application pre-trains the convolutional neural network model by using the pre-processed first historical image information of the non-target part to construct the pre-training sample data set, uses the contrast learning technology to enhance the recognition ability of the convolutional neural network model to subtle lesions, and improves the generalization ability of the model. Especially in the case where malignant and benign lesions are difficult to distinguish, the sensitivity and specificity of diagnosis are significantly improved, the misdiagnosis rate is reduced, the performance of the convolutional neural network model under the condition of limited labeled data is enhanced, and atypical endometrial hyperplasia (AEH), endometrial cancer (EC) and benign lesions can be accurately distinguished. The training sample data set constructed by the pre-processed second historical image information of the target part is used to train the convolutional neural network model, which greatly improves the recognition accuracy of the model, reduces the possibility of misdiagnosis and unnecessary biopsy, and helps to alleviate the perceptual bias and visual fatigue often encountered by endoscopists.

[0072] In one or more embodiments of the present application, in order to enhance the learning ability of the convolutional neural network model, contrast learning pre-training is used. Specifically, the first historical image information of the non-target part is obtained and pre-processed to obtain a pre-training sample data set. Here, the pre-training sample data set includes four data sets: CP-CHILD17, PolypGen18-20, IPCL21 and Hyper Kvasir22. The CP-CHILD data set contains pediatric colonoscopy images, which are divided into CP-CHILD-A (8,000 images) and CP-CHILD-B (1,500 images); the PolypGen data set includes 8,037 frames of images from six hospitals for polyp segmentation and detection; the IPCL21 data set contains ME-NBI video frames from 114 patients, mainly focusing on magnifying endoscope sub-sequences; Hyper Kvasir22 is the largest large intestine image public data set, containing 110,079 images and 373 videos. These diverse data sets ensure strong and discriminative feature learning, thereby improving the performance of the convolutional neural network model.

[0073] The training sample data set of the application was collected in three third-grade hospitals, a total of 1394 cases were included, the total number of images was 55874 PNG format hysteroscopy images, the number of cases and images in the training set and the test set is shown in Table 1. The images were taken by one of the three high-resolution devices (Olympus OTV-S190, Japan; Karl Storz 26105FA or 26120BA, Germany). All images were confirmed by two experts WW.W. and W.M. The control group category (benign lesions) includes endometrial polyps, submucosal uterine leiomyoma, atypical endometrial hyperplasia and normal uterine cavity.

[0074] Table 1. Baseline characteristics of training and test data sets

[0075]

[0076] 1.AEH / EC: Atypical endometrial hyperplasia / endometrial carcinoma

[0077] The training data set was collected from January 2008 to December 2017 from the Maternal and Child Hospital of Hubei Province (MCH), using Olympus OTV-S190 (Japan) and Karl Storz 26105FA or 26120BA (Germany). The internal test data set consists of hysteroscopy images collected at MCH using the same equipment from January 2018 to June 2019. The external test data set contains data collected from January 2019 to December 2019 from the Tongji Hospital of Huazhong University of Science and Technology (TJH) and the Second Affiliated Hospital of Zhengzhou University (ZZSH). The AEH / EC category includes cases of endometrial atypical hyperplasia and endometrial carcinoma, and the external test data set is mainly collected using Olympus OTV-S190 (Japan) equipment, and there is no overlap between the training set and the test set.

[0078] In the embodiments of the application, two additional test data sets are also included: the first contains 3419 images from 23 AEH / EC cases and 62 control cases diagnosed at MCH from January 2018 to June 2019, and the second contains 2809 images from 16 AEH / EC cases and 89 control cases diagnosed at TJH / ZZSH from January 2019 to December 2019. The former is an internal test data set, and the latter is an external test data set.

[0079] The case classification information corresponding to the training sample data set includes atypical endometrial hyperplasia (AEH), endometrial carcinoma (EC) and benign lesions.

[0080] In one or more embodiments of the present application, the pre-processing of the first historical image information and the second historical image information specifically includes the following steps:

[0081] S21: respectively performing standardization processing and normalization processing on the first historical image information and the second historical image information, wherein the standardization processing includes scaling processing, translation processing, rotation processing and flipping processing, and a unified image in format and size is obtained.

[0082] By performing standardization processing and normalization processing on the first historical image information and the second historical image information, overfitting can be prevented, and the image is normalized and adjusted to a unified pixel, facilitating the recognition and extraction processing of the convolutional neural network model, and improving the performance of the model. In an embodiment of the present application, all sample data are adjusted to 224x224 pixels to facilitate input of the convolutional neural network for recognition analysis.

[0083] In addition, in an embodiment of the present application, an oversampling technique is used to compensate for the influence of data imbalance in the training data set.

[0084] In one or more embodiments of the present application, the inputting of the pre-training sample data set into the convolutional neural network model for contrastive learning pre-training specifically includes the following steps:

[0085] S31: the convolutional neural network model performs feature extraction processing on the first historical image information in the pre-training sample data set, and maps the extracted pre-training features to a low-dimensional space;

[0086] Here, the present application uses ResNet-50 as the backbone network, and ResNet-50 is a CNN with a depth of 50 layers, which is famous for its residual connection. This connection helps to alleviate the gradient vanishing problem and supports the training of deeper networks. In the present application, a projection head is added to the ResNet-50 network, which is composed of three fully connected layers, aiming to map the features extracted by ResNet-50 to a lower-dimensional space, in which contrastive learning is applied for pre-training.

[0087] S32: in the low-dimensional space, a contrastive loss function is constructed, and the cosine similarity between the feature vectors corresponding to the extracted pre-training features is calculated according to the contrastive loss function, and the calculation formula of the contrastive loss function is:

[0088]

[0089] wherein z i and z j represent positive feature vectors, sim(z i ,z jrepresents the cosine similarity between the feature vector z i and z j , τ is a temperature parameter, l i,j represents the loss function between the feature vector z i and z j , N is the number of samples;

[0090] The contrastive loss function scales the cross-entropy by a temperature parameter and adjusts the sharpness of the resulting distribution using temperature scaling. This contrastive learning pre-training method significantly improves the feature extraction capability of the convolutional neural network model, enabling the convolutional neural network model to distinguish subtle differences in medical images, such as the difference between atypical endometrial hyperplasia (AEH) and endometrial cancer (EC). The application of contrastive learning pre-training is crucial for improving the overall performance of the convolutional neural network model.

[0091] S33: Adjust the concentration of the distribution of the loss function based on the cosine similarity until the change value of the contrastive loss function is within a preset range, complete the training, and obtain an endometrial benign and malignant disease auxiliary prediction model.

[0092] By constructing a contrastive loss function, the cosine similarity between feature vectors can be accurately calculated, and the cross-entropy can be adjusted and scaled using a temperature parameter, so that the change value of the contrastive loss function tends to be stable, thereby greatly improving the feature extraction capability of the convolutional neural network model, enabling the endometrial benign and malignant disease auxiliary prediction model to distinguish subtle differences in medical images.

[0093] In the pre-training phase, the embodiment of the present application uses contrastive training learning to enable the convolutional neural network model to extract more meaningful and reliable features. Contrastive training learning is a self-supervised learning method that optimizes the differences between the same data points, enhances the similarity between positive pairs (the picture itself and its preprocessed picture), and minimizes the differences between different data points, reducing the similarity between negative pairs (the picture and other preprocessed pictures), to improve feature learning. This technology has shown significant potential in medical image tasks, especially in improving diagnostic accuracy and model robustness.

[0094] In one or more embodiments of the present application, the training sample data set is input into the pre-trained convolutional neural network model for training until the training is completed, and an endometrial benign and malignant disease auxiliary prediction model is obtained, which specifically includes the following steps:

[0095] S41: the pre-trained convolutional neural network model performs classification and recognition processing on the second historical image information in the training sample dataset to obtain case classification information. Here, a binary classification recognition algorithm is adopted, and the obtained classification information is atypical endometrial hyperplasia (AEH) or endometrial cancer (EC), and benign lesions;

[0096] S42: Construct a focal loss function based on the case classification information and the corresponding true label value. The specific formula is:

[0097] focal loss(p t )=(1-p t ) 0·1 CE(p,y) (2)

[0098] CE(p,y)= -log (p t ) (3)

[0099]

[0100] Wherein, p is the probability that the sample is correctly predicted and classified by the convolutional neural network model, and y is the true label value of the sample. In this embodiment, when the sample is atypical endometrial hyperplasia (AEH) and endometrial cancer (EC), y = 1, and in other cases y = 0, p t is the predicted label value of the model, CE(p,y) is the cross entropy loss value between the true label value and the predicted label value, focal loss(p t ) is the focus loss value;

[0101] In order to overcome the data imbalance problem caused by the small number of malignant cases (such as AEH / EC), the convolutional neural network model proposed in this paper adopts the focal loss method, which increases the sensitivity to misclassified AEH / EC cases by introducing a regulation factor in the cross-entropy loss function (Formula (2) (3)).

[0102] S43: The model parameter with the smallest loss value of the focal loss function is used as the target parameter of the convolutional neural network model to complete the training and obtain an auxiliary prediction model for benign and malignant endometrial diseases.

[0103] In practice, the training method of the present application combines multiple key deep learning libraries such as PyTorch, torchvision, timm, etc. In order to prevent overfitting and accelerate network training, the present application "freezes" the first 10 layers in the convolutional neural network model, which is achieved by setting the learning rate to zero. In the present application, the stochastic gradient descent (SGD) optimizer is used, the learning rate is set to 0.01, the momentum is 0.9, the learning rate is decayed once every 10 epochs (decay rate is 0.1), and the number of training rounds is 30.

[0104] The second historical image information in the training sample data set is classified and recognized by the convolutional neural network model, and the case classification information is recognized. Then, a focal loss function is constructed based on the case classification information recognized by the convolutional neural network model and the corresponding true label value, and the focal loss value between them is calculated. By introducing an adjustment factor on the basis of the cross-entropy loss function, the sensitivity of the endometrial benign and malignant disease auxiliary prediction model to endometrial images is greatly increased, and the recognition accuracy of the endometrial benign and malignant disease auxiliary prediction model is ensured.

[0105] In the embodiment of the present application, after the training is completed, the target endometrial image information can be input into the endometrial benign and malignant disease auxiliary prediction model obtained after the training, and the corresponding endometrial benign and malignant disease type is obtained.

[0106] In one or more embodiments of the present application, the endometrial benign and malignant disease auxiliary prediction model construction method further comprises the following steps:

[0107] S5: evaluating the endometrial benign and malignant disease auxiliary prediction model obtained after training to obtain an evaluation result, and the specific calculation formula is:

[0108]

[0109]

[0110] Wherein, TP, TN, FP, FN represent true positive, true negative, false positive and false negative respectively, PPV represents positive predictive value, and NPV represents negative predictive value.

[0111] By calculating the accuracy, sensitivity, specificity, positive predictive value, negative predictive value and F1 score of the endometrial benign and malignant disease auxiliary prediction model, the classification efficiency and classification accuracy of the endometrial benign and malignant disease auxiliary prediction model obtained after training can be represented, and the prediction score of each lesion can be calculated according to the prediction result.

[0112] In order to verify the recognition accuracy of the auxiliary prediction model for benign and malignant endometrial diseases obtained after contrastive learning pre-training in the present invention, in this embodiment, the recognition results of the auxiliary prediction model for benign and malignant endometrial diseases without contrastive learning pre-training and the auxiliary prediction model for benign and malignant endometrial diseases after contrastive learning pre-training were tested. The results are as follows:

[0113] The test sample dataset collected for Hubei Maternal and Child Hospital (MCH):

[0114] For the auxiliary prediction model for benign and malignant endometrial diseases without contrastive learning (CL), its AUC (area under the ROC curve and the coordinate axis) value was 0.969 (95% CI: 0.928-0.999), the accuracy was 91.8% (95% CI: 85.9-97.6%), the sensitivity was 96.8% (95% CI: 91.7-100%), the specificity was 78.3% (95% CI: 60.9-92.9%), and the F1 score was 0.945 (95% CI: 0.902-0.979). In contrast, the performance of the auxiliary prediction model for benign and malignant endometrial diseases using contrastive learning (CL) was slightly improved, with an AUC value of 0.979 (95% CI: 0.942-1.000), an accuracy of 94.1% (95% CI: 89.1-99.1%), a sensitivity of 95.2% (95% CI: 89.5-100%), a specificity of 91.3% (95% CI: 78.2-100%), and an F1 score of 0.959 (95% CI: 0.920-0.992).

[0115] The test sample dataset was collected for Tongji Hospital of Huazhong University of Science and Technology (TJH) / The Second Affiliated Hospital of Zhengzhou University (ZZSH):

[0116] For the endometrial benign and malignant disease auxiliary prediction model without contrast learning (CL), the AUC (area under the ROC curve and the coordinate axis) value was 0.891 (95% CI: 0.810-0.964), the accuracy was 89.5% (95% CI: 83.7-95.4%), the sensitivity was 94.4% (95% CI: 89.4-98.8%), the specificity was 62.5% (95% CI: 40.0-86.7%), and the F1 score was 0.939 (95% CI: 0.898-0.973). In contrast, the endometrial benign and malignant disease auxiliary prediction model with contrast learning (CL) significantly improved the performance, with an AUC value of 0.975 (95% CI: 0.942-0.998), an accuracy of 93.3% (95% CI: 88.6-98.1%), a sensitivity of 92.1% (95% CI: 86.4-96.8%), a specificity of 100% (95% CI: 100-100%), and an F1 score of 0.959 (95% CI: 0.925-0.988).

[0117] The recognition results of the pre-trained and trained endometrial benign and malignant disease auxiliary prediction model were compared with the judgment results of endoscopic doctors, and the results were as follows:

[0118] For the test sample data set collected by the Maternal and Child Health Hospital (MCH) in Hubei Province:

[0119] When comparing the average performance of junior, intermediate, and senior endoscopic doctors (4 people in each group) with the endometrial benign and malignant disease auxiliary prediction model, the deep learning endometrial benign and malignant disease auxiliary prediction model was always superior to artificial experts. When using contrast learning (CL), the AUC value (0.979 vs. 0.952), accuracy (94.1% vs. 87.0%), sensitivity (95.2% vs. 92.3%), and F1 score (0.959 vs. 0.770) of the endometrial benign and malignant disease auxiliary prediction model were all superior to the average results of senior endoscopic doctors, highlighting the advantages of the endometrial benign and malignant disease auxiliary prediction model in diagnostic accuracy and consistency, especially when using contrast learning (CL).

[0120] For the test sample data set collected by the Tongji Hospital (TJH) of Huazhong University of Science and Technology (HUST) and the Second Affiliated Hospital of Zhengzhou University (ZZSH):

[0121] The endometrial benign and malignant disease auxiliary prediction model, especially when using contrast learning (CL), consistently demonstrated superior diagnostic accuracy compared to the average performance of the experts (4 per group) at the Tongji Hospital of Huazhong University of Science and Technology (TJH). After integrating contrast learning (CL), the endometrial benign and malignant disease auxiliary prediction model had an AUC value of 0.975, which was significantly higher than the average value of 0.862 for the experts. Similarly, the endometrial benign and malignant disease auxiliary prediction model outperformed the experts in terms of accuracy (93.3% vs. 80.2%) and sensitivity (92.1% vs. 71.9%). Furthermore, the endometrial benign and malignant disease auxiliary prediction model had an F1 score of 0.959, which was significantly higher than the experts' score of 0.530, highlighting the consistency and precision of the endometrial benign and malignant disease auxiliary prediction model in diagnostic performance.

[0122] The comparison results of the endometrial benign and malignant disease auxiliary prediction model and the endoscopic physician in terms of sensitivity, specificity, negative predictive value, and Kappa coefficient, among other evaluation indicators, are also listed in Tables 2 and 3.

[0123] Table 2. Comparison of endoscopic physician and endometrial benign and malignant disease auxiliary prediction model diagnostic performance for each patient in the MCH test data set

[0124]

[0125] Table 3. Comparison of endoscopic physician and endometrial benign and malignant disease auxiliary prediction model diagnostic performance for each patient in the TJH / ZZSH test data set

[0126]

[0127] For the comparison of the endometrial benign and malignant disease auxiliary prediction model and experts with different experience levels in terms of accuracy, see Figure 2 wherein, Figure 2 (a) is a schematic diagram of the test result accuracy of the endometrial benign and malignant disease auxiliary prediction model of the present application and experts with different experience levels based on the test sample data set collected by the Maternal and Infant Hospital of Hubei Province (MCH), respectively, Figure 2 (b) is a schematic diagram of the test result accuracy of the endometrial benign and malignant disease auxiliary prediction model of the present application and experts with different experience levels based on the test sample data set collected by the Tongji Hospital of Huazhong University of Science and Technology (TJH) / Zhengzhou University Second Affiliated Hospital (ZZSH).

[0128] It can be seen that the endometrial benign and malignant disease auxiliary prediction model, especially after combining contrast learning (CL), can accurately identify AEH / EC patients on test datasets from different medical centers, outperforming endoscopists with different levels of experience. The endometrial benign and malignant disease auxiliary prediction model performs well in terms of sensitivity and specificity, especially in distinguishing AEH / EC from other benign lesions such as polyps, uterine fibroids, endometrial hyperplasia without atypical hyperplasia, and normal uterine cavity. By integrating the endometrial benign and malignant disease auxiliary prediction model with the hysteroscopy system, the diagnostic process can be accelerated, ensuring balanced performance regardless of the endoscopist's experience. In addition, as a reliable auxiliary tool, the endometrial benign and malignant disease auxiliary prediction model reduces the likelihood of misdiagnosis and unnecessary biopsy, helping to alleviate the perceptual bias and visual fatigue commonly encountered by endoscopists.

[0129] In the embodiments of the present application, the effectiveness of contrast learning comes from its ability to enhance feature extraction, which helps the model better distinguish samples, especially in the case of limited and imbalanced labeled data. The present application uses colonoscopy datasets, which have similarities with hysteroscopy datasets, to pretrain a convolutional neural network model. Therefore, the endometrial benign and malignant disease auxiliary prediction model can more effectively capture the subtle differences between AEH / EC and benign lesions. This leads to significant improvements in diagnostic accuracy and robustness, even on external datasets from different medical centers.

[0130] The present application uses the Grad-CAM algorithm to identify the key areas used by the endometrial benign and malignant disease auxiliary prediction model to predict AEH / EC, as shown in Figure 3 As shown in the figure, the first column is the hysteroscopy image, the second column is the visualization heat map of the endometrial benign and malignant disease auxiliary prediction model without contrast learning (CL) for predicting atypical endometrial hyperplasia (AEH) and endometrial cancer (EC), and the third column is the visualization heat map of the endometrial benign and malignant disease auxiliary prediction model with contrast learning (CL) for predicting atypical endometrial hyperplasia (AEH) and endometrial cancer (EC). The heat map in the figure highlights the areas in the hysteroscopy image that may contain important morphological and vascular features associated with different pathological patterns of atypical endometrial hyperplasia (AEH) and endometrial cancer (EC), such as obvious deformation of the endometrial cavity, local necrosis, consistency of friability, and atypical blood vessels. Compared with the model without contrast learning, the model combined with contrast learning focuses more effectively on these key areas. This indicates that contrast learning significantly improves the ability of the endometrial benign and malignant disease auxiliary prediction model to identify and distinguish AEH / EC, enhancing its sensitivity to these key pathological features.

[0131] As Figure 4The application also provides an endometrial benign and malignant disease auxiliary prediction model construction system, which comprises an acquisition module, a preprocessing module and a model training and recognition module.

[0132] The acquisition module is configured to acquire first historical image information of a non-target part and second historical image information of a target part.

[0133] The preprocessing module is configured to preprocess the first historical image information and the second historical image information respectively, and construct corresponding pre-training sample data sets and training sample data sets.

[0134] The model training and recognition module is configured to construct a convolutional neural network model, and input the pre-training sample data set into the convolutional neural network model for comparative learning pre-training.

[0135] The model training and recognition module is further configured to input the training sample data set into the pre-trained convolutional neural network model for training until the convolutional neural network model meets a preset condition and the training is completed.

[0136] In one or more embodiments of the application, the specific implementation of the model training and recognition module inputting the pre-training sample data set into the convolutional neural network model for comparative learning pre-training is as follows:

[0137] The convolutional neural network model performs feature extraction processing on the first historical image information in the pre-training sample data set, and maps the extracted pre-training features to a low-dimensional space.

[0138] In the low-dimensional space, a contrastive loss function is constructed, and the cosine similarity between the feature vectors corresponding to the extracted pre-training features is calculated according to the contrastive loss function, and the calculation formula of the contrastive loss function is as follows:

[0139]

[0140] wherein z i and z j represent positive pair feature vectors, sim(z i ,z j ) represents the cosine similarity between the positive pair feature vectors z i and z j , tau is a temperature parameter, l i,j represents the loss function between the positive pair feature vectors z i and z j , and N is the sample quantity.

[0141] The distribution concentration of the loss function is adjusted based on the cosine similarity until the change value of the contrastive loss function is within a preset range, and the pre-training is completed.

[0142] By constructing a contrast loss function, the cosine similarity between the feature vectors can be accurately calculated, and the temperature parameter is used to adjust and scale the cross-entropy, so that the change value of the contrast loss function tends to be stable, thereby greatly improving the feature extraction capability of the convolutional neural network model, and enabling the convolutional neural network model to distinguish subtle differences in medical images.

[0143] In one or more embodiments of the present application, the model training identification module inputs the training sample data set into the pre-trained convolutional neural network model for training, and the specific implementation is as follows:

[0144] The pre-trained convolutional neural network model classifies and identifies the second historical image information in the training sample data set to obtain case classification information;

[0145] A focal loss function is constructed based on the case classification information and the corresponding true label value, and the specific formula is as follows:

[0146] focal loss(p t )=(1-p t ) 0·1 CE(p,y) (2)

[0147] CE(p,y)= -log (p t ) (3)

[0148]

[0149] Wherein, p is the probability that the sample is correctly predicted by the convolutional neural network model, y is the true label value of the sample, p t is the predicted label value of the model, CE(p,y) is the cross-entropy loss value between the true label value and the predicted label value, and focal loss(p t ) is the focal loss value.

[0150] The model parameter with the minimum loss value of the focal loss function is taken as the target parameter of the convolutional neural network model, and the training is completed, and the endometrial benign and malignant disease auxiliary prediction model.

[0151] The second historical image information in the training sample data set is classified and identified by the convolutional neural network model to identify the case classification information, and then a focal loss function is constructed based on the case classification information identified by the convolutional neural network model and the corresponding true label value, and the focal loss value between them is calculated. By introducing an adjusting factor on the basis of the cross-entropy loss function, the sensitivity of the convolutional neural network model to endometrial images is greatly increased, and the recognition accuracy of the convolutional neural network model is ensured.

[0152] The application further provides a computer readable storage medium, which stores a computer program, and the computer program, when executed by a processor, realizes the auxiliary prediction model construction method for endometrial benign and malignant diseases.

[0153] The application further provides an auxiliary prediction model construction device for endometrial benign and malignant diseases, which is characterized by comprising a communication interface, a memory, a communication bus and a processor, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus.

[0154] The memory is used for storing a computer program.

[0155] The processor is used for executing the program stored on the memory, and realizes the steps of the auxiliary prediction model construction method for endometrial benign and malignant diseases.

[0156] The above only describes the preferred embodiments of the application and is not used to limit the application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A method for constructing an auxiliary prediction model for benign and malignant endometrial diseases, characterized in that: The steps include: Acquiring first historical image information of a non-target portion and second historical image information of a target portion; Preprocessing the first historical image information and the second historical image information respectively, and constructing corresponding pre-training sample data sets and training sample data sets respectively; Constructing a convolutional neural network model, and inputting the pre-training sample data set into the convolutional neural network model for comparative learning pre-training; Inputting the training sample data set into the pre-trained convolutional neural network model for training until the training is completed, thereby obtaining an auxiliary prediction model for benign and malignant endometrial diseases; Inputting the pre-training sample data set into the convolutional neural network model for contrastive learning pre-training specifically includes the following steps: The convolutional neural network model performs feature extraction processing on the first historical image information in the pre-training sample data set, and maps the extracted pre-training features into a low-dimensional space; In the low-dimensional space, a contrast loss function is constructed, and the cosine similarity between the feature vectors corresponding to the extracted pre-training features is calculated based on the contrast loss function. The calculation formula of the contrast loss function is: Among them, z i and z j Denotes the positive eigenvector, sim(z i ,z j ) represents the opposite eigenvector z i and z j The cosine similarity between them, τ is the temperature parameter, l i,j Represents the opposite eigenvector z i and z j The loss function between , N is the number of samples; Adjusting the distribution concentration of the loss function based on the cosine similarity until the change value of the contrast loss function is within a preset range, completing the training, and obtaining an auxiliary prediction model for benign and malignant endometrial diseases; Inputting the training sample data set into the pre-trained convolutional neural network model for training until the training is completed to obtain an auxiliary prediction model for benign and malignant endometrial diseases specifically includes the following steps: The pre-trained convolutional neural network model performs classification and recognition processing on the second historical image information in the training sample data set to obtain case classification information; A focal loss function is constructed based on the case classification information and the corresponding true label value. The specific formula is: focal loss(p t )=(1-p t ) 0·1 CE(p,y) (2) CE(p,y)= -log (p t ) (3) Among them, p is the probability that the sample is correctly predicted by the convolutional neural network model, y is the true label value of the sample, and p t is the predicted label value of the model, CE(p,y) is the cross entropy loss value between the true label value and the predicted label value, focal loss(p t ) is the focus loss value; The model parameter with the smallest loss value of the focal loss function is used as the target parameter of the convolutional neural network model to complete the training and form an auxiliary prediction model for benign and malignant endometrial diseases.

2. The method for constructing an auxiliary prediction model for benign and malignant endometrial diseases according to claim 1, characterized in that The pre-processing of the first historical image information and the second historical image information specifically includes the following steps: Standardization and normalization are performed on the first historical image information and the second historical image information respectively, wherein the standardization includes scaling, translation, rotation and flipping, to obtain images with uniform format and size.

3. The method for constructing an auxiliary prediction model for benign and malignant endometrial diseases according to any one of claims 1 to 2, characterized in that: The following steps are also included: The auxiliary prediction model for benign and malignant endometrial diseases obtained after training is evaluated to obtain an evaluation result. The specific calculation formula is: Among them, TP, TN, FP, and FN represent true positive examples, true negative examples, false positive examples, and false negative examples, respectively; PPV represents positive predictive value; and NPV represents negative predictive value.

4. A deep learning-based auxiliary prediction model construction system for benign and malignant endometrial diseases, characterized by: Includes acquisition module, preprocessing module and model training and recognition module; The acquisition module is used to acquire the first historical image information of the non-target part and the second historical image information of the target part; The preprocessing module is used to preprocess the first historical image information and the second historical image information respectively, and to construct corresponding pre-training sample data sets and training sample data sets respectively; The model training recognition module is used to construct a convolutional neural network model and input the pre-training sample data set into the convolutional neural network model for comparative learning pre-training; The model training and identification module is further configured to input the training sample data set into the pre-trained convolutional neural network model for training until the training is completed, thereby obtaining an auxiliary prediction model for benign and malignant endometrial diseases; The specific implementation of the model training recognition module inputting the pre-training sample data set into the convolutional neural network model for contrastive learning pre-training is as follows: The convolutional neural network model performs feature extraction processing on the first historical image information in the pre-training sample data set, and maps the extracted pre-training features into a low-dimensional space; In the low-dimensional space, a contrast loss function is constructed, and the cosine similarity between the feature vectors corresponding to the extracted pre-training features is calculated based on the contrast loss function. The calculation formula of the contrast loss function is: Among them, z i and z j Denotes the positive eigenvector, sim(z i ,z j ) represents the opposite eigenvector z i and z j The cosine similarity between them, τ is the temperature parameter, l i,j Represents the opposite eigenvector z i and z j The loss function between , N is the number of samples; Adjusting the distribution concentration of the loss function based on the cosine similarity until the change value of the contrast loss function is within a preset range, thereby completing the pre-training; The model training recognition module inputs the training sample data set into the pre-trained convolutional neural network model for training until the training is completed, and the specific implementation of the auxiliary prediction model for benign and malignant endometrial diseases is as follows: The pre-trained convolutional neural network model performs classification and recognition processing on the second historical image information in the training sample data set to obtain case classification information; A focal loss function is constructed based on the case classification information and the corresponding true label value. The specific formula is: focal loss(p t )=(1-p t ) 0·1 CE(p,y) (2) CE(p,y)= -log (p t ) (3) Among them, p is the probability that the sample is correctly predicted by the convolutional neural network model, y is the true label value of the sample, and p t is the predicted label value of the model, CE(p,y) is the cross entropy loss value between the true label value and the predicted label value, focal loss(p t ) is the focus loss value; The model parameter with the smallest loss value of the focal loss function is used as the target parameter of the convolutional neural network model to complete the training and obtain an auxiliary prediction model for benign and malignant endometrial diseases.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing an auxiliary prediction model for benign and malignant endometrial diseases according to any one of claims 1 to 3 is implemented.

6. A device for constructing an auxiliary prediction model for benign and malignant endometrial diseases, characterized by: comprising a communication interface, a memory, a communication bus and a processor, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; The memory is used to store computer programs; The processor is configured to implement the steps of the method for constructing an auxiliary prediction model for benign and malignant endometrial diseases according to any one of claims 1 to 3 when executing the program stored in the memory.

Images