Endometrial hyperplasia detection method and device, storage medium and electronic equipment
By combining endoscopic images and case information with a neural network model, the problems of misdiagnosis and missed diagnosis in the classification of endometrial hyperplasia have been solved, and more accurate identification of endometrial hyperplasia and cancer risk has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU HAIKANG HUIYING TECH CO LTD
- Filing Date
- 2023-08-15
- Publication Date
- 2026-04-21
AI Technical Summary
Visual classification of endometrial hyperplasia is challenging, and existing endoscopic techniques are prone to misdiagnosis or missed diagnosis, especially under different endoscopic lighting conditions and pathological features.
By combining endometrial images obtained from endoscopic imaging with patient case information using a neural network model, and through feature extraction and multimodal classification, the type of endometrial hyperplasia and the binary classification information and probability information of the presence of endometrial cancer were determined.
It improves the accuracy of endometrial hyperplasia detection, provides more classification and probability information, and assists doctors in making accurate diagnoses.
Smart Images
Figure CN117011610B_ABST
Abstract
Description
Technical Field
[0001] This application relates to image processing technology, and in particular to a method, apparatus, storage medium, and electronic device for detecting endometrial hyperplasia. Background Technology
[0002] With the advancement of image processing technology, it has been increasingly widely used in the processing of medical images.
[0003] Endometrial hyperplasia refers to an excessive growth of the uterine lining caused by inflammation, endocrine disorders, or stimulation from certain medications. Endometrial hyperplasia can occur at any age, including puberty, reproductive years, perimenopause, and postmenopause. It is classified into atypia and atypical endometrial hyperplasia, both of which carry a risk of developing into endometrial cancer. Therefore, early detection and identification of polyp types are crucial for cancer prevention and treatment.
[0004] Endoscopes are commonly used medical instruments consisting of a beam guide structure and a set of lenses. After entering the target object through natural openings or small incisions, the endoscope can acquire raw images of the object's interior and be used to examine it. However, the visual classification of endometrial hyperplasia is challenging; different endoscopic lighting conditions and varying pathological features can easily lead to misdiagnosis or missed diagnosis. Summary of the Invention
[0005] This application provides a method, apparatus, storage medium, and electronic device for detecting endometrial hyperplasia, which can improve the accuracy of endometrial hyperplasia classification.
[0006] To achieve the above objectives, this application adopts the following technical solution:
[0007] A method for detecting endometrial hyperplasia, comprising:
[0008] Acquire a target image obtained through endoscopic imaging and the patient's case information corresponding to the target image; wherein, the target image includes endometrial tissue;
[0009] The case information and the preprocessed target image are input into the hyperplasia recognition model for processing. The hyperplasia category corresponding to the target image is obtained through the first classification of the hyperplasia recognition model, and / or the binary classification information of whether the target image corresponds to endometrial cancer and the probability information of corresponding endometrial cancer are obtained through the second classification of the hyperplasia recognition model.
[0010] The hyperplasia categories include no endometrial hyperplasia, endometrial hyperplasia without atypical endometrial hyperplasia, and atypical endometrial hyperplasia.
[0011] Preferably, the processing of the proliferation recognition model includes:
[0012] Feature extraction is performed based on the target image and the case information to obtain a multimodal feature vector;
[0013] The multimodal feature vector is subjected to a first classification identification to determine the proliferation category corresponding to the target image;
[0014] The multimodal feature vector is subjected to a second classification identification to determine the binary classification information and the probability information.
[0015] Preferably, the step of extracting multimodal feature vectors based on the target image and the case information includes:
[0016] Image feature extraction is performed on the target image to obtain an image feature vector;
[0017] Feature extraction is performed on the case information to obtain a case feature vector;
[0018] The image feature vector and the case feature vector are fused and then feature extraction is performed to obtain a multimodal feature vector.
[0019] Preferably, determining the probability information that the target image corresponds to endometrial cancer includes: determining the confidence level of the target image corresponding to endometrial cancer, and using the confidence level as the probability information.
[0020] Preferably, the method further includes: outputting the proliferation category, the binary classification information, and the probability information.
[0021] Preferably, the step of outputting the hyperplasia category, the binary classification information, and the probability information includes: outputting the hyperplasia category, and when the hyperplasia category is atypical endometrial hyperplasia, outputting the binary classification information and the probability information;
[0022] or,
[0023] The method further includes: when the hyperplasia category is no endometrial hyperplasia or endometrial hyperplasia without atypical features, and the binary classification information is endometrial cancer corresponding to the target image, outputting an error message, or canceling the output of the hyperplasia category, the binary classification information, and the probability information, or outputting a prompt message to prompt the doctor to re-determine the category of endometrial hyperplasia.
[0024] A method for detecting endometrial hyperplasia, comprising:
[0025] Acquire a target image obtained through endoscopic imaging and the patient's case information corresponding to the target image; wherein, the target image includes endometrial tissue;
[0026] By combining the feature information of the target image and the case information, the type of endometrial hyperplasia in the target image is identified;
[0027] Output a prompt message indicating the type of proliferation.
[0028] Preferably, after identifying the type of endometrial hyperplasia in the target image, the method further includes:
[0029] When the hyperplasia category indicates that the endometrial hyperplasia has a risk of becoming cancerous, the probability information of the endometrial hyperplasia developing into endometrial cancer is identified;
[0030] Output the probability information.
[0031] A device for detecting endometrial hyperplasia includes: an acquisition unit and a hyperplasia identification and processing unit;
[0032] The acquisition unit is used to acquire a target image obtained by endoscopic imaging and the patient's case information corresponding to the target image; the target image includes endometrial tissue.
[0033] The hyperplasia recognition processing unit is used to input the case information and the preprocessed target image into the hyperplasia recognition model for processing, and to obtain the hyperplasia category corresponding to the target image through the first classification recognition of the hyperplasia recognition model, and / or to obtain the binary classification information of whether the target image corresponds to endometrial cancer and the probability information of corresponding endometrial cancer through the second classification recognition of the hyperplasia recognition model.
[0034] The hyperplasia categories include no endometrial hyperplasia, endometrial hyperplasia without atypical endometrial hyperplasia, and atypical endometrial hyperplasia.
[0035] Preferably, the proliferation recognition process includes an input subunit, a feature extraction subunit, a first classification recognition subunit, and a second classification recognition subunit;
[0036] The input subunit is used to input the target image and the case information into the feature extraction subunit;
[0037] The feature extraction subunit is used to extract features based on the target image and the case information to obtain a multimodal feature vector;
[0038] The first classification and recognition subunit is used to perform a first classification and recognition on the multimodal feature vector to determine the proliferation category corresponding to the target image;
[0039] The second classification and recognition subunit is used to perform a second classification and recognition on the multimodal feature vector to determine the binary classification information and the probability information.
[0040] Preferably, the feature extraction subunit includes an image feature extraction module, a medical record feature extraction module, and a multimodal feature extraction module;
[0041] The image feature extraction module is used to extract image features from the target image to obtain an image feature vector;
[0042] The medical record feature extraction module is used to extract features from the case information to obtain a case feature vector;
[0043] The multimodal feature extraction module is used to fuse the image feature vector and the case feature vector and then extract the features to obtain a multimodal feature vector.
[0044] Preferably, in the first classification and recognition subunit, determining the probability information of the target image corresponding to endometrial cancer includes: determining the confidence level of the target image corresponding to endometrial cancer, and using the confidence level as the probability information.
[0045] Preferably, the device further includes an output unit for outputting the proliferation category, the binary classification information, and the probability information.
[0046] Preferably, the output unit outputs the hyperplasia category, and when the hyperplasia category is atypical endometrial hyperplasia, it outputs the binary classification information and the probability information.
[0047] or,
[0048] In the output unit, when the hyperplasia category is no endometrial hyperplasia or endometrial hyperplasia without atypical features, and the binary classification information is endometrial cancer corresponding to the target image, an error message is output, or the output of the hyperplasia category, the binary classification information, and the probability information is canceled, or a prompt message is output to prompt the doctor to re-determine the category of endometrial hyperplasia.
[0049] Preferably, the device further includes a model training unit for training and generating the proliferation recognition model;
[0050] In the model training unit, the first classification recognition and the second classification recognition are jointly trained.
[0051] Preferably, in the model training unit, the joint training includes:
[0052] The weighted sum of the loss functions of the first classification and the second classification is used as the comprehensive loss function; the parameters of the proliferation recognition model are updated based on the value of the comprehensive loss function.
[0053] Preferably, in the model training unit, when the actual classification of the training sample image is atypical endometrial hyperplasia, the second classification identification participates in the training of the hyperplasia identification model;
[0054] When the actual classification of the training sample image is no endometrial hyperplasia or endometrial hyperplasia without atypical features, the loss function of the second classification recognition is set to zero.
[0055] A computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, can implement the method for detecting endometrial hyperplasia as described above.
[0056] An electronic device, comprising at least a computer-readable storage medium and a processor;
[0057] The processor is configured to read executable instructions from the computer-readable storage medium and execute the instructions to implement the method for detecting endometrial hyperplasia as described above.
[0058] As can be seen from the above technical solution, this application acquires the target image obtained through endoscopic imaging and the corresponding patient's case information; the case information and the preprocessed target image are input into a hyperplasia recognition model for processing. A first classification method identifies the hyperplasia category corresponding to the target image, and a second classification method identifies whether the target image corresponds to endometrial cancer and the probability information of corresponding endometrial cancer. Through this processing, medical record information strongly correlated with the occurrence of uterine hyperplasia and endometrial cancer is introduced into the endometrial hyperplasia detection method. Combining medical record information and the target image as the detection basis and using a neural network model for hyperplasia detection can effectively improve the accuracy of endometrial hyperplasia detection. Simultaneously, two different classification methods are performed in the neural network model: one classification method identifies the endometrial hyperplasia category, and the other classification method identifies whether it corresponds to endometrial cancer and the probability information of corresponding endometrial cancer, thus providing more classification and probability information for the classification of endometrial hyperplasia. Attached Figure Description
[0059] Figure 1 This is a schematic diagram of the basic process of a method for detecting endometrial hyperplasia in this application;
[0060] Figure 2 This is a schematic diagram of the basic process of another method for detecting endometrial hyperplasia in this application;
[0061] Figure 3 This is a schematic diagram of the specific process of the endometrial hyperplasia detection method in a specific embodiment;
[0062] Figure 4 This is a schematic diagram of the basic structure of a detection device for endometrial hyperplasia according to this application;
[0063] Figure 5 This is a schematic diagram of another endometrial hyperplasia detection device in this application;
[0064] Figure 6 This is a schematic diagram of the basic structure of the electronic device in this application. Detailed Implementation
[0065] To make the objectives, technical means, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings.
[0066] Figure 1 This is a schematic diagram of the basic process of a method for detecting endometrial hyperplasia according to this application. Figure 1 As shown, the method includes:
[0067] Step 101: Obtain the target image obtained through endoscopic imaging and the patient's case information corresponding to the target image.
[0068] This step is used to obtain target images including endometrial tissue and the patient's medical record information. Specifically, the target image is obtained through endoscopic imaging; this image includes tissue located at the endometrial site and is used for subsequent detection and processing of endometrial hyperplasia.
[0069] Given the strong correlation between the probability of endometrial hyperplasia and endometrial cancer and certain basic medical information of patients, this application, when performing endometrial hyperplasia detection, requires obtaining the patient's basic medical information in addition to endometrial images. Specifically, information strongly correlated with the probability of developing endometrial hyperplasia and endometrial cancer includes: the patient's height, weight, age, and history of diabetes. Therefore, the patient's medical information obtained in this application may include, but is not limited to: the patient's height, weight, age, and / or history of diabetes.
[0070] Step 102: Combining the feature information of the target image and case information, identify the type of endometrial hyperplasia in the target image.
[0071] The feature information of the target image and the feature information of the medical record are extracted. Based on these two types of feature information, the type of endometrial hyperplasia in the target image is identified.
[0072] Step 103: Output a prompt message to indicate the type of proliferation.
[0073] After the hyperplasia category is determined in step 102, in step 103, the corresponding hyperplasia category information is output to provide a reference for the doctor's diagnosis. Alternatively, after identifying the hyperplasia category in step 102, if the hyperplasia category indicates that endometrial hyperplasia has a risk of becoming cancerous, the probability information of endometrial hyperplasia developing into endometrial cancer can be further identified, and in step 103, the probability information of endometrial hyperplasia developing into endometrial cancer can be further output.
[0074] At this point, Figure 1 The illustrated method flow is now complete. In the above method, considering the strong correlation between medical record information and the probability of developing endometrial hyperplasia and endometrial cancer, steps 101 and 102 incorporate medical record information and its characteristic information when identifying hyperplasia categories, thereby improving the accuracy of hyperplasia category identification.
[0075] Figure 2 This is a schematic diagram of the basic process of another method for detecting endometrial hyperplasia in this application. Figure 2 As shown, the method includes:
[0076] Step 201: Obtain the target image obtained through endoscopic imaging and the patient's case information corresponding to the target image.
[0077] The process in step 201 is the same as that in step 101, so it will not be repeated here.
[0078] Step 202: Input the case information and the preprocessed target image into the hyperplasia recognition model.
[0079] The target image undergoes preprocessing, such as downsampling and dehazing, to improve its quality and standardize it for better recognition. This image, along with medical record information, is then used as input to the endometrial hyperplasia recognition model. The hyperplasia recognition model is a pre-trained neural network model used to identify different types of endometrial hyperplasia.
[0080] Step 203: The hyperplasia recognition model processes the image to obtain the hyperplasia category corresponding to the target image through the first classification of the hyperplasia recognition model, and / or to obtain the binary classification information of whether the target image corresponds to endometrial cancer and the probability information of corresponding endometrial cancer through the second classification of the hyperplasia recognition model.
[0081] The hyperplasia identification model processes the input target image and medical record information to identify the type of endometrial hyperplasia. The type identification is divided into two categories. The first category determines the overall type of endometrial hyperplasia, including no endometrial hyperplasia, endometrial hyperplasia without atypical features, and atypical endometrial hyperplasia. The second category determines whether the target image corresponds to endometrial cancer and provides the probability information of the target image corresponding to endometrial cancer. Specifically, when determining whether the target image corresponds to endometrial cancer, binary classification information is directly provided, and the probability of corresponding endometrial cancer is used to provide more accurate quantitative information. In this step of the hyperplasia identification model, one or both categories can be selected for identification processing to obtain the corresponding identification information.
[0082] At this point, Figure 2 The illustrated detection procedure for endometrial hyperplasia is now complete. This detection method, on the one hand, incorporates medical record information strongly correlated with endometrial hyperplasia and endometrial cancer, improving detection accuracy; on the other hand, by utilizing a neural network model—a hyperplasia identification model—and the two classification results within this model, it provides not only hyperplasia category information but also information on whether it corresponds to endometrial cancer and the probability of such a diagnosis, thus offering more classification and quantitative information for the classification of endometrial hyperplasia.
[0083] In addition, the aforementioned Figure 1 In the method shown, the processing in steps 102-103 can be specifically carried out as described above. Figure 2 The processing method for steps 202-203.
[0084] The above will be illustrated below through specific embodiments. Figure 2 The specific implementation of the endometrial hyperplasia detection method shown.
[0085] Figure 3 This is a schematic diagram illustrating the specific process of the endometrial hyperplasia detection method in the embodiments of this application. Figure 3 As shown, the method includes:
[0086] Step 301: Obtain the target image obtained through endoscopic imaging.
[0087] Images of endometrial tissue are obtained through endoscopic imaging, and these images are used as target images for subsequent endometrial hyperplasia detection.
[0088] Step 302: Preprocess the target image.
[0089] The target image can be preprocessed according to actual needs. This application does not limit the specific preprocessing method. For example, it can be performed according to the environment of endoscopic imaging and the image requirements during subsequent detection.
[0090] Step 303: Obtain the medical record information of the patient corresponding to the target image.
[0091] As mentioned earlier, endometrial hyperplasia and endometrial cancer may have a strong correlation with certain aspects of a patient's medical record. Therefore, this application requires obtaining the patient's medical record information as one of the bases for detecting endometrial hyperplasia. Specifically, the obtained medical record information may include, but is not limited to, structured data such as height, weight, age, and / or history of diabetes, as well as past medical history.
[0092] Step 304: Input the preprocessed target image and medical record information into the hyperplasia recognition model.
[0093] Step 305: Based on the target image and medical record information, feature extraction is performed to obtain a multimodal feature vector.
[0094] The feature extraction part of the hyperplasia recognition model includes an image feature extraction module, a medical record feature extraction module, and a multimodal feature extraction module.
[0095] The image feature extraction module is a network model used to extract image features from a target image. Specifically, it can use models such as ResNet-50 to construct the image feature extraction network model. The preprocessed target image is input into the image feature extraction module, and after processing, it outputs an image feature vector.
[0096] The medical record feature extraction module is a network model used to extract features from medical record information. Various feature extraction models can be used; for example, a medical record feature extraction network model can be constructed using multiple fully connected layers. After being organized (e.g., various types of medical record information can be concatenated into a one-dimensional vector), the medical record information is input into the medical record feature extraction module, and after processing, it outputs a medical record feature vector.
[0097] The multimodal feature extraction module is used to fuse and extract features from image feature vectors and medical record feature vectors. First, the image feature vectors and medical record feature vectors output by the image feature extraction module and the medical record feature extraction module are fused. The specific fusion method can be set according to needs; for example, the two feature vectors can be concatenated or weighted summed. Then, feature extraction is performed on the fused feature vector to obtain the multimodal feature vector. Various feature extraction network models can be used to process the feature extraction of the fused feature vector.
[0098] Step 306: Perform first classification recognition on the multimodal feature vectors to determine the proliferation category corresponding to the target image.
[0099] In this embodiment, the multimodal feature vectors are classified twice. Accordingly, the classification and recognition part of the proliferation recognition model may include a first classification and recognition module and a second classification and recognition module.
[0100] This step involves processing the first classification and recognition module to determine the hyperplasia category corresponding to the target image. Specifically, there are three hyperplasia categories: no endometrial hyperplasia, endometrial hyperplasia without atypical features, and atypical endometrial hyperplasia. The first classification and recognition module can employ a typical classification network to classify the multimodal feature vectors.
[0101] Step 307: Perform a second classification recognition on the multimodal feature vector to determine whether the target image corresponds to the binary classification information of endometrial cancer and the probability information of the corresponding endometrial cancer.
[0102] This step involves processing by the second classification and recognition module to determine whether the target image corresponds to endometrial cancer and the probability information of such correspondence. The classification result indicates whether it corresponds to endometrial cancer; therefore, the second classification and recognition process can employ a typical binary classification network. Simultaneously, the second classification and recognition module can determine the confidence level of the target image corresponding to endometrial cancer, and this confidence level is used as the probability information of corresponding endometrial cancer.
[0103] Step 308: Output the detection results.
[0104] The detection results of steps 307 and 308 are output. Considering that endometrial cancer is typically only a possibility when the endometrial hyperplasia is classified as atypical endometrial hyperplasia, optionally, the hyperplasia category can be output when the detection results are output. If the hyperplasia category is atypical endometrial hyperplasia, binary classification information indicating whether it corresponds to endometrial cancer and the corresponding probability information can be output. Alternatively, if the first classification determines that the hyperplasia category is no endometrial hyperplasia or endometrial hyperplasia without atypicality, and the second classification determines that the target image corresponds to endometrial cancer, an error message can be output, or the output of the results of steps 307 and 308 can be canceled. Alternatively, a prompt message can be output to remind the doctor to reassess the category of endometrial hyperplasia.
[0105] The output method of the test results can be selected according to actual needs, such as display screen prompts, endoscope host prompts, sound prompts, text prompts, output of test reports, remote prompts from the terminal, etc.
[0106] At this point, Figure 3 The procedure for detecting endometrial hyperplasia shown is now complete.
[0107] In the endometrial hyperplasia detection method of this application, a trained hyperplasia recognition model is used to classify and identify specific endometrial hyperplasia. The training process of the hyperplasia recognition model is described below. The structure of the hyperplasia recognition model includes: an image feature extraction module, a medical record feature extraction module, a multimodal feature extraction module, a first classification recognition module, and a second classification recognition module. The training process of the hyperplasia recognition model determines the network model weights for each of the above modules. Specific processing includes:
[0108] (1) Collect training data;
[0109] Collect raw endometrial training images and patient medical records, label the type of endometrial hyperplasia, and if it is atypical hyperplasia, label whether it is endometrial cancer; the raw endometrial training images, medical records, endometrial type, and endometrial cancer label constitute a set of training data; the training set Ω can include multiple sets of training data;
[0110] (2) Configure network parameters for the initial network model;
[0111] Configure the initial parameters of the network model, i.e., the parameter set is θ0. This can include the initial weights of the network model for the image feature extraction module, medical record feature extraction module, multimodal feature extraction module, first classification recognition module, and second classification recognition module, and set the high-level parameters related to training (such as learning rate, gradient descent algorithm parameters, etc.). The values of the initial parameters can be set based on experience.
[0112] (3) Perform forward processing using the initial network model;
[0113] Based on a network model with parameters θ0, forward computation is performed on the training set Ω to obtain the predicted endometrial lining category and the predicted endometrial cancer result. Based on the predicted endometrial lining category and the predicted endometrial cancer result, and the endometrial lining category and endometrial cancer label in the training set Ω, the loss value is determined.
[0114] For the first classification recognition module, existing classification loss functions, such as softmax, can be used; for the second classification recognition module, existing binary classification loss functions, such as cross-entropy loss function, can be used.
[0115] Optionally, the first classification recognition module and the second classification recognition module can be jointly trained. In this case, the overall loss function of the entire network model can be determined based on the loss function of the first classification recognition module and the loss function of the second classification recognition module, for example, it can be a weighted sum of the two loss functions.
[0116] (4) The loss value based on the network model is adjusted using the backpropagation algorithm to obtain θ0.i ;
[0117] After processing in step (3), the loss value of the network model is obtained. Based on this loss value, the parameters of the current network model are updated, and the updated parameters are labeled as θ. i .
[0118] (5) Repeat steps (3) to (4) until the network model converges and the output parameter θ is obtained. final And based on parameter θ final The network model that is formed is the trained network model—the proliferation category recognition model.
[0119] Furthermore, during the training process of the aforementioned hyperplasia category recognition model, considering that endometrial cancer is usually only possible when the endometrial hyperplasia category is atypical endometrial hyperplasia, the second classification recognition module can optionally participate in the training when the endometrial label of the training sample is atypical endometrial hyperplasia, i.e., the loss function of the second classification module is calculated normally; while when the endometrial label of the training sample is no endometrial hyperplasia or endometrial hyperplasia without atypicality, the loss function of the second classification recognition module is set to zero, i.e., the second classification recognition module does not participate in the model training of that sample.
[0120] The above describes the specific implementation of the detection method for endometrial hyperplasia in this application. Using this method, target images of endometrial tissue obtained through endoscopic imaging and the patient's medical record information are used to classify endometrial hyperplasia based on a deep learning neural network model. This accurately categorizes endometrial hyperplasia, further assesses atypical hyperplasia types, and provides the probability of endometrial cancer. This can assist doctors in determining endometrial hyperplasia and reduce their workload.
[0121] This application also provides a device for detecting endometrial hyperplasia, which can be used to perform the above-mentioned procedures. Figure 1 The detection method shown. Figure 4 This is a schematic diagram of the basic structure of the device for detecting endometrial hyperplasia. Figure 4 As shown, the device includes: an acquisition unit, a proliferation recognition processing unit, and an output unit.
[0122] The acquisition unit is used to acquire the target image obtained by endoscopic imaging and the patient's case information corresponding to the target image;
[0123] The hyperplasia recognition processing unit is used to combine feature information from the target image and case information to identify the type of endometrial hyperplasia in the target image;
[0124] The output unit is used to output prompts indicating the type of proliferation.
[0125] Optionally, the hyperplasia identification processing unit can be further used to identify the probability information of endometrial hyperplasia developing into endometrial cancer when the hyperplasia category indicates that endometrial hyperplasia has a risk of becoming cancerous;
[0126] Correspondingly, the output unit can further output probability information.
[0127] Figure 5 This is a schematic diagram of the basic structure of another endometrial hyperplasia detection device according to this application. This detection device can be used to implement the above-mentioned... Figure 2 The method for detecting endometrial hyperplasia is shown. (For example...) Figure 5 As shown, the device includes an acquisition unit and a proliferation recognition processing unit.
[0128] The acquisition unit is used to acquire the target image obtained by endoscopic imaging and the patient's case information corresponding to the target image; the target image includes endometrial tissue.
[0129] The hyperplasia recognition processing unit is used to input case information and preprocessed target images into the hyperplasia recognition model for processing. The hyperplasia category corresponding to the target image is obtained through the first classification of the hyperplasia recognition model, and / or the binary classification information of whether the target image corresponds to endometrial cancer and the probability information of corresponding endometrial cancer are obtained through the second classification of the hyperplasia recognition model.
[0130] The hyperplasia categories include no endometrial hyperplasia, endometrial hyperplasia without atypical endometrial hyperplasia, and atypical endometrial hyperplasia.
[0131] Optionally, case information includes, but is not limited to: height, weight, age, and / or past medical history such as diabetes;
[0132] Optionally, the proliferation recognition process may include an input subunit, a feature extraction subunit, a first classification recognition subunit, and a second classification recognition subunit;
[0133] The input subunit is used to input the target image and the case information into the feature extraction subunit.
[0134] The feature extraction subunit is used to extract features based on the target image and the case information to obtain a multimodal feature vector;
[0135] The first classification and recognition subunit is used to perform first classification and recognition on the multimodal feature vector to determine the growth category corresponding to the target image;
[0136] The second classification recognition subunit is used to perform second classification recognition on multimodal feature vectors to determine binary classification information and probability information.
[0137] Optionally, the feature extraction subunit includes an image feature extraction module, a medical record feature extraction module, and a multimodal feature extraction module;
[0138] The image feature extraction module is used to extract image features from the target image to obtain an image feature vector;
[0139] The medical record feature extraction module is used to extract features from the case information to obtain a case feature vector;
[0140] The multimodal feature extraction module is used to fuse the image feature vector and the case feature vector and then extract the features to obtain a multimodal feature vector.
[0141] Optionally, in the first classification and identification subunit, determining the probability information of the target image corresponding to endometrial cancer includes: determining the confidence level of the target image corresponding to endometrial cancer, and using the confidence level as the probability information.
[0142] Optionally, the device may further include an output unit for outputting proliferation category, binary classification information, and probability information.
[0143] Optionally, in the output unit, the hyperplasia category is output, and when the hyperplasia category is atypical endometrial hyperplasia, binary classification information and the probability information are output;
[0144] or,
[0145] In the output unit, when the hyperplasia category is no endometrial hyperplasia or endometrial hyperplasia without atypical features, and the binary classification information is endometrial cancer corresponding to the target image, an error message is output, or the output of the hyperplasia category, binary classification information, and probability information is canceled, or a prompt message is output to prompt the doctor to re-determine the category of endometrial hyperplasia.
[0146] Optionally, the device further includes a model training unit for training a model to generate an invasiveness recognition model;
[0147] In the model training unit, the first classification recognition and the second classification recognition are jointly trained.
[0148] Optionally, during joint training in the model training unit, the weighted sum of the loss functions of the first classification recognition and the second classification recognition is used as the comprehensive loss function; the parameters of the augmentation recognition model are updated based on the value of the comprehensive loss function.
[0149] Optionally, in the model training unit, when the actual classification of the training sample image is atypical endometrial hyperplasia, the second classification identification participates in the training of the hyperplasia identification model;
[0150] When the actual classification of the training sample image is no endometrial hyperplasia or endometrial hyperplasia without atypical features, the loss function for the second classification is set to zero.
[0151] This application also provides a computer-readable storage medium that stores instructions, which, when executed by a processor, can perform the steps in the method for detecting endometrial hyperplasia as described above. In practical applications, the computer-readable medium may be included in the devices / apparatus / systems of the above embodiments, or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium stores instructions, which, when executed by a processor, can perform the steps in the method for detecting endometrial hyperplasia as described above.
[0152] According to the embodiments disclosed in this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof, but not intended to limit the scope of protection of this application. In the embodiments disclosed in this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0153] Figure 6 An electronic device is also provided for this application. For example... Figure 6 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:
[0154] The electronic device may include a processor 601 with one or more processing cores, a memory 602 with one or more computer-readable storage media, and a computer program stored in the memory and executable on the processor. When the program in the memory 602 is executed, a method for detecting endometrial hyperplasia can be implemented.
[0155] Specifically, in practical applications, this electronic device may also include components such as a power supply 603 and an input / output unit 604. Those skilled in the art will understand that... Figure 6 The structure of the electronic device shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:
[0156] The processor 601 is the control center of the electronic device. It connects various parts of the electronic device through various interfaces and lines. By running or executing software programs and / or modules stored in the memory 602, and calling data stored in the memory 602, it performs various functions of the server and processes data, thereby monitoring the electronic device as a whole.
[0157] Memory 602 can be used to store software programs and modules, i.e., the aforementioned computer-readable storage medium. Processor 601 executes various functional applications and data processing by running the software programs and modules stored in memory 602. Memory 602 may primarily include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created according to the use of the server, etc. In addition, memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, memory 602 may also include a memory controller to provide processor 601 with access to memory 602.
[0158] The electronic device also includes a power supply 603 that supplies power to the various components. This power supply can be logically connected to the processor 601 via a power management system, enabling functions such as charging, discharging, and power consumption management. The power supply 603 may also include one or more DC or AC power supplies, a recharging system, a power fault detection circuit, a power converter or inverter, a power status indicator, or any other components.
[0159] The electronic device may also include an input / output unit 604, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, or optical signal inputs related to user settings and function control. The input unit output 604 can also be used to display information input by the user or information provided to the user, as well as various graphical user interfaces, which can be composed of graphics, text, icons, video, and any combination thereof.
[0160] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for detecting endometrial hyperplasia, characterized in that, include: Acquire a target image obtained through endoscopic imaging and the patient's case information corresponding to the target image; wherein, the target image includes endometrial tissue; The case information and the preprocessed target image are input into the hyperplasia recognition model for processing. The hyperplasia category corresponding to the target image is obtained through the first classification of the hyperplasia recognition model. Furthermore, the binary classification information of whether the target image corresponds to endometrial cancer and the probability information of corresponding endometrial cancer are obtained through the second classification of the hyperplasia recognition model. The hyperplasia categories include no endometrial hyperplasia, endometrial hyperplasia without atypical endometrial hyperplasia, and atypical endometrial hyperplasia.
2. The method according to claim 1, characterized in that, The processing of the proliferation recognition model includes: Feature extraction is performed based on the target image and the case information to obtain a multimodal feature vector; The multimodal feature vector is subjected to a first classification identification to determine the proliferation category corresponding to the target image; The multimodal feature vector is subjected to a second classification identification to determine the binary classification information and the probability information.
3. The method according to claim 2, characterized in that, The step of extracting multimodal feature vectors based on the target image and the case information includes: Image feature extraction is performed on the target image to obtain an image feature vector; Feature extraction is performed on the case information to obtain a case feature vector; The image feature vector and the case feature vector are fused and then feature extraction is performed to obtain a multimodal feature vector.
4. The method according to claim 2, characterized in that, The step of determining the probability information that the target image corresponds to endometrial cancer includes: determining the confidence level of the target image corresponding to endometrial cancer, and using the confidence level as the probability information.
5. The method according to claim 1, characterized in that, The method further includes: outputting the proliferation category, the binary classification information, and the probability information.
6. The method according to claim 5, characterized in that, The step of outputting the hyperplasia category, the binary classification information, and the probability information includes: outputting the hyperplasia category, and when the hyperplasia category is atypical endometrial hyperplasia, outputting the binary classification information and the probability information; or, The method further includes: when the hyperplasia category is no endometrial hyperplasia or endometrial hyperplasia without atypical features, and the binary classification information is endometrial cancer corresponding to the target image, outputting an error message, or canceling the output of the hyperplasia category, the binary classification information, and the probability information, or outputting a prompt message to prompt the doctor to re-determine the category of endometrial hyperplasia.
7. A device for detecting endometrial hyperplasia, characterized in that, include: Acquisition unit and proliferation recognition processing unit; The acquisition unit is used to acquire the target image obtained by endoscopic imaging and the patient's case information corresponding to the target image; The target image includes endometrial tissue; The hyperplasia recognition processing unit is used to input the case information and the preprocessed target image into the hyperplasia recognition model for processing. The hyperplasia category corresponding to the target image is obtained through the first classification recognition of the hyperplasia recognition model, and the binary classification information of whether the target image corresponds to endometrial cancer and the probability information of corresponding endometrial cancer are obtained through the second classification recognition of the hyperplasia recognition model. The hyperplasia categories include no endometrial hyperplasia, endometrial hyperplasia without atypical endometrial hyperplasia, and atypical endometrial hyperplasia.
8. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the instructions are executed by the processor, they can realize the method for detecting endometrial hyperplasia as described in any one of claims 1 to 6.
9. An electronic device, characterized in that, The electronic device includes at least a computer-readable storage medium and a processor; The processor is configured to read executable instructions from the computer-readable storage medium and execute the instructions to implement the method for detecting endometrial hyperplasia according to any one of claims 1 to 6.
Citation Information
Patent Citations
Endometrial cancer cell detection method, system and equipment and storage medium
CN114580558A
Disease classification model training method and device, terminal and readable storage medium
CN116452851A