Ovarian cancer early diagnosis device based on AI image recognition

Abstract

The invention relates to the technical field of image processing, and provides an early ovarian cancer diagnosis device based on AI image recognition, which is used for improving the accuracy of early ovarian cancer diagnosis. The device is characterized in that an image acquisition module is in communication connection with an ovarian cancer recognition processor, and the ovarian cancer recognition processor is used for acquiring patient image data acquired by the image acquisition module; preprocessing the patient image data, and identifying the preprocessed patient image data to mark abnormal position points; based on the abnormal position point, acquiring a local magnified image corresponding to the abnormal position point; inputting the locally amplified image and the image data into an ovarian cancer recognition model to obtain an ovarian cancer prediction result; the ovarian cancer recognition model can be obtained according to sample image data and ovarian cancer label training corresponding to the sample image data, and the sample image data comprises the sample image data and a plurality of locally amplified sample images corresponding to the sample image data.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an early diagnostic device for ovarian cancer based on AI image recognition. Background Technology

[0002] Ovarian cancer is a malignant tumor of the ovary, referring to malignant tumors that grow on the ovary. 90%–95% of these are primary ovarian cancers, while the remaining 5%–10% are metastases from primary cancers in other sites. Because early-stage ovarian cancer often lacks symptoms and screening methods are limited, early diagnosis is difficult. By the time patients seek medical attention, 60%–70% are already at an advanced stage, and treatment outcomes for late-stage cases are often poor. Although the incidence of ovarian cancer is lower than that of cervical and endometrial cancers, its mortality rate exceeds the combined mortality rates of cervical and endometrial cancers, ranking first among gynecological cancers and posing a significant threat to women's health.

[0003] Currently, ovarian cancer is diagnosed based on the individual professional experience of physicians. However, relying solely on physicians' diagnostic results may lead to misdiagnosis. Therefore, there is an urgent need for an early diagnostic device for ovarian cancer to assist physicians in diagnosing ovarian cancer and improve the accuracy of ovarian cancer diagnosis. Summary of the Invention

[0004] This application provides an AI-based image recognition-based early diagnosis device for ovarian cancer, which improves the accuracy of early diagnosis of ovarian cancer.

[0005] This invention provides an AI-based image recognition-based early diagnosis device for ovarian cancer. The device includes an image acquisition module and an ovarian cancer recognition processor. The image acquisition module is communicatively connected to the ovarian cancer recognition processor, and the ovarian cancer recognition processor is used to perform:

[0006] The image acquisition module acquires patient image data, including upper abdominal image data, pelvic image data, and laparoscopic imaging data of ovarian cancer patients.

[0007] The patient image data is preprocessed, and the preprocessed patient image data is identified to mark abnormal locations.

[0008] Based on the abnormal location point, obtain the local magnified image corresponding to the abnormal location point;

[0009] The magnified local image and the image data are input into the ovarian cancer recognition model to obtain ovarian cancer prediction results.

[0010] The ovarian cancer identification model is obtained by training based on sample image data and its corresponding ovarian cancer labels. The sample image data includes sample image data and its corresponding multiple magnified sample images.

[0011] In an optional embodiment of the present invention, the preprocessing of the patient image data includes:

[0012] Noise and interference in the patient image data are filtered out, and contrast processing and edge enhancement processing are performed on the patient image data.

[0013] In an optional embodiment of the present invention, the step of identifying and marking abnormal location points in the preprocessed patient image data includes:

[0014] The preprocessed patient image data is input into the first neural network model and the second neural network model to obtain the first abnormal location point and the second abnormal location point;

[0015] The first abnormal location point and the second abnormal location point are merged, and the merged first abnormal location point and second abnormal location point are clustered according to unit regions to obtain the cluster center point corresponding to each unit region.

[0016] Abnormal locations are marked using the cluster center points.

[0017] In an optional embodiment of the present invention, before marking outlier locations using the cluster center points, the method further includes:

[0018] Obtain the distance values ​​between multiple cluster centroids within adjacent unit regions;

[0019] Determine whether the distance between any two cluster centers is less than a preset value, where the preset value is less than the boundary length of the unit region;

[0020] The step of marking outlier locations using the cluster center points includes:

[0021] If the distance between any two cluster centers is less than the preset value, then the target center point corresponding to multiple cluster centers in the adjacent unit area is calculated, and the abnormal location point is marked by the target center point.

[0022] If the distance between any two cluster centers is greater than or equal to the preset value, then the abnormal location points are marked by the cluster centers in the adjacent unit area.

[0023] In an optional embodiment provided by the present invention, the training process of the first neural network model and the second neural network model is as follows:

[0024] Acquire patient image sample data and their corresponding annotated abnormal location points;

[0025] The patient image sample data is input into a first neural network model and a second neural network model respectively to obtain a first predicted abnormal location point and a second predicted abnormal location point; the first neural network model and the second neural network model adopt different network structures;

[0026] Calculate the first positional difference value and the second positional difference value between the first predicted anomaly location point and the second predicted anomaly location point and the marked anomaly location point, respectively;

[0027] The model parameters of the first neural network model and the second neural network model are updated based on the first position difference value and the second position difference value.

[0028] In an optional embodiment of the present invention, updating the model parameters of the first neural network model and the second neural network model based on the first positional difference value and the second positional difference value includes:

[0029] Determine whether both the first position difference value and the second position difference value are less than a preset loss value;

[0030] If both are less than the preset loss value, then training of the first neural network model and the second neural network model is stopped.

[0031] If the loss is not less than the preset loss value, the model parameters of the first neural network model and the second neural network model are updated according to the magnitude of the first position difference value and the second position difference value, and the first neural network model and the second neural network model are trained again using other patient image sample data.

[0032] In an optional embodiment of the present invention, updating the model parameters of the first neural network model and the second neural network model based on the magnitudes of the first positional difference value and the second positional difference value includes:

[0033] If the first position difference value is less than the second position difference value, then the model parameters of the second neural network model are updated using the model parameters of the first neural network model;

[0034] If the first position difference value is greater than the second position difference value, then the model parameters of the first neural network model are updated using the model parameters of the second neural network model.

[0035] In an optional embodiment of the present invention, inputting the magnified local image and the image data into the ovarian cancer recognition model to obtain the ovarian cancer prediction result includes:

[0036] Determine the positional relationship matrix of each of the magnified local images in the image data;

[0037] The positional relationship matrix, the corresponding magnified local image, and the image data are input into the ovarian cancer recognition model to obtain the ovarian cancer prediction result.

[0038] In an optional embodiment provided by the present invention, the training process of the ovarian cancer identification model is as follows:

[0039] Acquire sample image data and its corresponding ovarian cancer label, wherein the sample image data includes sample image data and its corresponding multiple magnified local sample images;

[0040] Determine the positional relationship matrix of each magnified local sample image in the sample image data;

[0041] The positional relationship matrix, the corresponding local magnified sample image, and the sample image data are input into the ovarian cancer recognition model to obtain the ovarian cancer prediction result.

[0042] The model parameters of the ovarian cancer identification model are updated based on the ovarian cancer prediction results, the ovarian cancer labels, and the location relationship matrix.

[0043] In an optional embodiment of the present invention, updating the model parameters of the ovarian cancer identification model based on the ovarian cancer prediction result, the ovarian cancer label, and the positional relationship matrix includes:

[0044] Determine the positional relationships of the ovarian cancer prediction results and the ovarian cancer labels in the positional relationship matrix;

[0045] The loss value is calculated based on the location relationship, the ovarian cancer prediction result, and the ovarian cancer label.

[0046] If the loss value is less than the target value, then training of the ovarian cancer identification model is stopped.

[0047] This invention provides an ovarian cancer detection processor, the ovarian cancer detection processor comprising:

[0048] The acquisition module is used to acquire patient image data acquired by the image acquisition module; the patient image data includes upper abdominal image data, pelvic image data, and laparoscopic exploration image data of ovarian cancer patients;

[0049] The annotation module is used to preprocess the patient image data and identify abnormal locations in the preprocessed patient image data.

[0050] The acquisition module is further configured to acquire a magnified local image corresponding to the abnormal location point based on the abnormal location point;

[0051] The recognition module is used to input the magnified local image and the image data into the ovarian cancer recognition model to obtain ovarian cancer prediction results.

[0052] The ovarian cancer identification model is obtained by training based on sample image data and its corresponding ovarian cancer labels. The sample image data includes sample image data and its corresponding multiple magnified sample images.

[0053] This application provides an AI-based image recognition-based early diagnosis device for ovarian cancer. The device includes an image acquisition module and an ovarian cancer recognition processor. The image acquisition module and the ovarian cancer recognition processor are communicatively connected. The ovarian cancer recognition processor performs the following actions: acquiring patient image data collected by the image acquisition module; the patient image data includes upper abdominal image data, pelvic image data, and laparoscopic imaging data of ovarian cancer patients; preprocessing the patient image data and identifying abnormal locations within the preprocessed image data; finally, based on the abnormal locations, acquiring magnified images of the corresponding areas; and inputting the magnified images and image data into an ovarian cancer recognition model to obtain ovarian cancer prediction results. Currently, existing technologies rely on physicians' personal professional experience to diagnose ovarian cancer. This application, however, can obtain ovarian cancer prediction results based on collected patient image data and an ovarian cancer recognition model. The entire process does not require the involvement of a professional physician, thus improving the efficiency of ovarian cancer prediction. Furthermore, the ovarian cancer recognition model in this application is trained based on sample image data and its corresponding ovarian cancer labels. This sample image data includes sample image data and its corresponding multiple magnified local sample images. Therefore, the ovarian cancer prediction results can be obtained through the ovarian cancer recognition model, thereby improving the accuracy of early ovarian cancer diagnosis. Attached Figure Description

[0054] Figure 1 A flowchart of an execution method within an ovarian cancer recognition processor is provided in this application;

[0055] Figure 2 A flowchart of an abnormal location point annotation method provided in this application;

[0056] Figure 3 This is an internal structural diagram of an ovarian cancer recognition processor provided in this application. Detailed Implementation

[0057] To better understand the above technical solutions, the technical solutions of the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this application and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this application, rather than limitations on the technical solutions of this application. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other.

[0058] Please see Figure 1 As shown in the figure, an AI-based image recognition-based early diagnosis device for ovarian cancer is provided in an embodiment of the present invention. The device includes: an image acquisition module and an ovarian cancer recognition processor. The image acquisition module is communicatively connected to the ovarian cancer recognition processor, and the ovarian cancer recognition processor is used to perform the following steps:

[0059] S101, acquire patient image data acquired by the image acquisition module.

[0060] The patient imaging data includes at least upper abdominal imaging data, pelvic imaging data, and laparoscopic imaging data of ovarian cancer patients, but this embodiment does not specifically limit these.

[0061] S102, preprocess the patient image data, and identify abnormal locations in the preprocessed patient image data.

[0062] In one optional embodiment provided in this application, the preprocessing of the patient image data includes: filtering out noise and interference in the patient image data, and performing contrast processing and edge enhancement processing on the patient image data. Preprocessing the patient image data in this embodiment can improve the accuracy of patient image data identification, thereby improving the accuracy of abnormal location point labeling.

[0063] like Figure 2 As shown, in an optional embodiment provided in this application, the step of identifying and marking abnormal location points in the preprocessed patient image data includes:

[0064] S1021, The preprocessed patient image data is input into the first neural network model and the second neural network model to obtain the first abnormal location point and the second abnormal location point.

[0065] The first neural network model and the second neural network model are two pre-trained network models with different structures. The abnormal location points corresponding to the patient's image data can be obtained through the first neural network model and the second neural network model. That is, the two neural network models will mark the abnormal location points in the original patient image data, thereby marking the location of possible lesions, so that the corresponding local magnified image can be extracted based on the abnormal location points in subsequent steps.

[0066] In one optional embodiment provided in this application, the training process of the first neural network model and the second neural network model is as follows:

[0067] S201, Obtain patient image sample data and its corresponding anomaly location points.

[0068] The abnormal location points are manually marked locations in the patient image sample data that may indicate lesions. There may be one or more abnormal location points, and this embodiment does not make a specific limitation on this.

[0069] S202, the patient image sample data is input into the first neural network model and the second neural network model respectively to obtain the first predicted abnormal location point and the second predicted abnormal location point.

[0070] The first neural network model and the second neural network model employ different network structures. For example, the first neural network model may be a convolutional neural network and the second neural network model may be a radial neural network; or the first neural network model may be a recurrent neural network and the second neural network model may be a long short-term memory network. This embodiment does not impose specific limitations on these structures.

[0071] S203, calculate the first positional difference value and the second positional difference value between the first predicted abnormal position point and the second predicted abnormal position point and the marked abnormal position point, respectively.

[0072] In this embodiment, after obtaining the first predicted abnormal location point and the second predicted abnormal location point through the first neural network model and the second neural network model, the first position difference value and the second position difference value can be calculated based on the pixel position points corresponding to the predicted abnormal location point and the labeled abnormal location point in the patient image data. Then, the current accuracy of the two neural network models is verified based on the calculated position difference value.

[0073] It should be noted that this embodiment can also calculate the first position difference value and the second position difference value using a loss function, for example, by calculating the position difference value using the following loss function:

[0074]

[0075] Where loss is the calculated location difference value, i.e., the first location difference value and the second location difference value, n is the number of predicted outlier locations, and x i Let y be the coordinates of the i-th predicted abnormal location point in the patient image sample data, m be the number of labeled abnormal location points, and y be the coordinates of the i-th predicted abnormal location point. j Let x be the coordinates of the j-th labeled abnormal location point in the patient image sample data, min(x) i yj The positional difference between the i-th predicted anomaly location and the j-th labeled anomaly location among the m labeled anomaly locations is minimized.

[0076] For example, the patient image sample data obtained through the first neural network model corresponds to two first predicted abnormal location points x1 and x2, and the patient image sample data corresponds to two labeled abnormal location points y1 and y2. The location difference value calculated based on the above loss function is: (x1, y1) + (x2, y2), that is, the distance between the first predicted abnormal location point x1 and the labeled abnormal location point y1 plus the distance between the first predicted abnormal location point x2 and the labeled abnormal location point y2. Wherein, the distance between the first predicted abnormal location point x1 and the labeled abnormal location point y1 is closer than the distance between the first predicted abnormal location point x1 and the labeled abnormal location point y2, and the distance between the first predicted abnormal location point x2 and the labeled abnormal location point y2 is closer than the distance between the first predicted abnormal location point x2 and the labeled abnormal location point y1.

[0077] S204, update the model parameters of the first neural network model and the second neural network model according to the first position difference value and the second position difference value.

[0078] Specifically, updating the model parameters of the first neural network model and the second neural network model based on the first position difference value and the second position difference value includes:

[0079] S2041, determine whether both the first position difference value and the second position difference value are less than a preset loss value.

[0080] The preset loss value is a value pre-set according to the model's prediction accuracy requirements. In this embodiment, after determining the first position difference value and the second position difference value, it is necessary to determine whether the first position difference value and the second position difference value are less than the preset loss value, so as to determine whether the two neural network models meet the convergence condition, that is, whether the prediction accuracy of the neural network models meets the requirements.

[0081] S2042, if both are less than the preset loss value, then stop training the first neural network model and the second neural network model.

[0082] In this embodiment, if both the first position difference value and the second position difference value are less than the preset value, it indicates that the model accuracy of the first neural network model and the second neural network model meets the requirements, and the training of the first neural network model and the second neural network model can be terminated at this time.

[0083] S2043, if the loss is not less than the preset loss value, then update the model parameters of the first neural network model and the second neural network model according to the magnitude of the first position difference value and the second position difference value, and continue to train the first neural network model and the second neural network model using other patient image sample data.

[0084] If one or both of the first position difference value and the second position difference value are less than the preset value, this embodiment needs to update the model parameters of the two neural network models according to the first position difference value and the second position difference value, and then continue to train the first neural network model and the second neural network model using other patient image sample data until the position difference value of the two neural network models is less than the preset loss value.

[0085] Specifically, updating the model parameters of the first neural network model and the second neural network model based on the magnitude of the first position difference value and the second position difference value includes: if the first position difference value is less than the second position difference value, then updating the model parameters of the second neural network model through the model parameters of the first neural network model; if the first position difference value is greater than the second position difference value, then updating the model parameters of the first neural network model through the model parameters of the second neural network model.

[0086] In this embodiment, if the first position difference value is less than the second position difference value, it indicates that the accuracy of the current first neural network model is higher than that of the second neural network model. At this time, the model parameters of the second neural network model can be updated based on the model parameters of the first neural network model. Conversely, if the first position difference value is greater than the second position difference value, it indicates that the accuracy of the current second neural network model is higher than that of the first neural network model. At this time, the model parameters of the first neural network model can be updated based on the model parameters of the second neural network model. This enables the two neural network models to learn from each other's training results at any time during the training process, thereby improving the training efficiency and accuracy of the neural network models.

[0087] S1022, merge the first abnormal location point and the second abnormal location point, and cluster the merged first abnormal location point and the second abnormal location point according to the unit region to obtain the cluster center point corresponding to each unit region.

[0088] In this embodiment, after obtaining the first abnormal location point and the second abnormal location point, the first abnormal location point and the second abnormal location point are merged in the patient image data. Then, the patient image data is divided into n unit regions according to the unit regions. Then, the merged first abnormal location point and the second abnormal location point are clustered according to the unit regions to obtain the cluster center point corresponding to each unit region.

[0089] The unit area is a square area set according to actual needs. The size of the unit area can be determined based on the resolution of the image pixels. For example, the unit area is a square with a side length of 100 pixels.

[0090] S1023, mark the abnormal location points using the cluster center points.

[0091] In an optional embodiment provided in this application, before marking abnormal location points using the cluster center points, the method further includes: obtaining distance values ​​between multiple cluster center points within adjacent unit regions; determining whether the distance value between any two cluster center points is less than a preset value, wherein the preset value is less than the boundary length of the unit region. In this embodiment, after determining the cluster center points, it is determined whether the surrounding adjacent unit regions (8 unit regions) contain other cluster center points or abnormal location points based on the unit region where the cluster center point is located. If so, it is necessary to calculate the distance values ​​between other cluster center points or abnormal location points adjacent to the cluster center point, and then calculate whether the distance values ​​are less than the preset value. After that, the cluster center points are processed accordingly based on the comparison results.

[0092] Accordingly, the step of marking abnormal location points through the cluster center points includes: if the distance between any two cluster center points is less than the preset value, then calculating the target center point corresponding to multiple cluster center points in the adjacent unit area and marking the abnormal location point through the target center point; if the distance between any two cluster center points is greater than or equal to the preset value, then marking the abnormal location point through each cluster center point in the adjacent unit area.

[0093] In this embodiment, abnormal location points are marked based on cluster center points, thereby ensuring that the local magnified image determined based on abnormal location points in subsequent steps contains lesion content (the location of the abnormal location point) centered in the image.

[0094] S103, Based on the abnormal location point, obtain the local magnified image corresponding to the abnormal location point.

[0095] S104. Input the magnified local image and image data into the ovarian cancer recognition model to obtain the ovarian cancer prediction result.

[0096] The ovarian cancer identification model is obtained by training based on sample image data and its corresponding ovarian cancer labels. The sample image data includes sample image data and its corresponding multiple magnified sample images.

[0097] In one optional embodiment provided in this application, the step of inputting the magnified local image and the image data into the ovarian cancer recognition model to obtain the ovarian cancer prediction result includes: determining the positional relationship matrix of each of the magnified local images in the image data; and inputting the positional relationship matrix and the corresponding magnified local image and the image data into the ovarian cancer recognition model to obtain the ovarian cancer prediction result.

[0098] In one optional embodiment provided in this application, the training process of the ovarian cancer recognition model is as follows: acquiring sample image data and its corresponding ovarian cancer labels, wherein the sample image data includes sample image data and its corresponding multiple magnified sample images; determining the positional relationship matrix of each magnified sample image in the sample image data; inputting the positional relationship matrix, the corresponding magnified sample images, and the sample image data into the ovarian cancer recognition model to obtain ovarian cancer prediction results; and updating the model parameters of the ovarian cancer recognition model according to the ovarian cancer prediction results, the ovarian cancer labels, and the positional relationship matrix.

[0099] In this embodiment, updating the model parameters of the ovarian cancer identification model based on the ovarian cancer prediction result, the ovarian cancer label, and the positional relationship matrix includes: determining the positional relationship between the ovarian cancer prediction result and the ovarian cancer label in the positional relationship matrix; calculating a loss value based on the positional relationship, the ovarian cancer prediction result, and the ovarian cancer label; and stopping the training of the ovarian cancer identification model if the loss value is less than a target value.

[0100] This application provides an AI-based image recognition-based early diagnosis device for ovarian cancer. The device includes an image acquisition module and an ovarian cancer recognition processor. The image acquisition module and the ovarian cancer recognition processor are communicatively connected. The ovarian cancer recognition processor performs the following actions: acquiring patient image data collected by the image acquisition module; the patient image data includes upper abdominal image data, pelvic image data, and laparoscopic imaging data of ovarian cancer patients; preprocessing the patient image data and identifying abnormal locations in the preprocessed patient image data; finally, based on the abnormal locations, acquiring magnified images corresponding to the abnormal locations; and inputting the magnified images and image data into an ovarian cancer recognition model to obtain ovarian cancer prediction results. Currently, existing technologies rely on physicians' personal professional experience to diagnose ovarian cancer. This application, however, can obtain ovarian cancer prediction results based on collected patient image data and an ovarian cancer recognition model. The entire process does not require the involvement of a professional physician, thus improving the efficiency of ovarian cancer prediction. Furthermore, the ovarian cancer recognition model in this application is trained based on sample image data and its corresponding ovarian cancer labels. This sample image data includes sample image data and its corresponding multiple magnified local sample images. Therefore, the ovarian cancer prediction results can be obtained through the ovarian cancer recognition model, thereby improving the accuracy of early ovarian cancer diagnosis.

[0101] In one embodiment, an ovarian cancer detection processor is provided, such as Figure 3 As shown, the ovarian cancer detection processor includes:

[0102] The acquisition module 31 is used to acquire patient image data acquired by the image acquisition module; the patient image data includes upper abdominal image data, pelvic image data, and laparoscopic exploration image data of ovarian cancer patients;

[0103] The annotation module 32 is used to preprocess the patient image data and identify abnormal location points in the preprocessed patient image data.

[0104] The acquisition module 31 is further configured to acquire a magnified local image corresponding to the abnormal location point based on the abnormal location point;

[0105] The recognition module 33 is used to input the magnified local image and the image data into the ovarian cancer recognition model to obtain the ovarian cancer prediction result;

[0106] The ovarian cancer identification model is obtained by training based on sample image data and its corresponding ovarian cancer labels. The sample image data includes sample image data and its corresponding multiple magnified sample images.

[0107] In an optional embodiment provided by the present invention, the annotation module 32 is further configured to filter out noise and interference in the patient image data, and to perform contrast processing and edge enhancement processing on the patient image data.

[0108] In an optional embodiment provided by the present invention, the annotation module 32 is specifically used for:

[0109] The preprocessed patient image data is input into the first neural network model and the second neural network model to obtain the first abnormal location point and the second abnormal location point;

[0110] The first abnormal location point and the second abnormal location point are merged, and the merged first abnormal location point and second abnormal location point are clustered according to unit regions to obtain the cluster center point corresponding to each unit region.

[0111] Abnormal locations are marked using the cluster center points.

[0112] In an optional embodiment provided by the present invention, the annotation module 32 is specifically used for:

[0113] Obtain the distance values ​​between multiple cluster centroids within adjacent unit regions;

[0114] Determine whether the distance between any two cluster centers is less than a preset value, where the preset value is less than the boundary length of the unit region;

[0115] If the distance between any two cluster centers is less than the preset value, then the target center point corresponding to multiple cluster centers in the adjacent unit area is calculated, and the abnormal location point is marked by the target center point.

[0116] If the distance between any two cluster centers is greater than or equal to the preset value, then the abnormal location points are marked by the cluster centers in the adjacent unit area.

[0117] In an optional embodiment of the present invention, the ovarian cancer recognition processor further includes a training module 34, which is used for:

[0118] Acquire patient image sample data and their corresponding annotated abnormal location points;

[0119] The patient image sample data is input into a first neural network model and a second neural network model respectively to obtain a first predicted abnormal location point and a second predicted abnormal location point; the first neural network model and the second neural network model adopt different network structures;

[0120] Calculate the first positional difference value and the second positional difference value between the first predicted anomaly location point and the second predicted anomaly location point and the marked anomaly location point, respectively;

[0121] The model parameters of the first neural network model and the second neural network model are updated based on the first position difference value and the second position difference value.

[0122] In an optional embodiment provided by the present invention, the training module 34 is specifically used for:

[0123] Determine whether both the first position difference value and the second position difference value are less than a preset loss value;

[0124] If both are less than the preset loss value, then training of the first neural network model and the second neural network model is stopped.

[0125] If the loss is not less than the preset loss value, the model parameters of the first neural network model and the second neural network model are updated according to the magnitude of the first position difference value and the second position difference value, and the first neural network model and the second neural network model are trained again using other patient image sample data.

[0126] In an optional embodiment provided by the present invention, the training module 34 is specifically used for:

[0127] If the first position difference value is less than the second position difference value, then the model parameters of the second neural network model are updated using the model parameters of the first neural network model;

[0128] If the first position difference value is greater than the second position difference value, then the model parameters of the first neural network model are updated using the model parameters of the second neural network model.

[0129] In an optional embodiment provided by the present invention, the identification module 33 is specifically used for:

[0130] Determine the positional relationship matrix of each of the magnified local images in the image data;

[0131] The positional relationship matrix, the corresponding magnified local image, and the image data are input into the ovarian cancer recognition model to obtain the ovarian cancer prediction result.

[0132] In an optional embodiment provided by the present invention, the training module 34 is specifically used for:

[0133] Acquire sample image data and its corresponding ovarian cancer label, wherein the sample image data includes sample image data and its corresponding multiple magnified local sample images;

[0134] Determine the positional relationship matrix of each magnified local sample image in the sample image data;

[0135] The positional relationship matrix, the corresponding local magnified sample image, and the sample image data are input into the ovarian cancer recognition model to obtain the ovarian cancer prediction result.

[0136] The model parameters of the ovarian cancer identification model are updated based on the ovarian cancer prediction results, the ovarian cancer labels, and the location relationship matrix.

[0137] In an optional embodiment provided by the present invention, the training module 34 is specifically used for:

[0138] Determine the positional relationships of the ovarian cancer prediction results and the ovarian cancer labels in the positional relationship matrix;

[0139] The loss value is calculated based on the location relationship, the ovarian cancer prediction result, and the ovarian cancer label.

[0140] If the loss value is less than the target value, then training of the ovarian cancer identification model is stopped.

[0141] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An early diagnostic device for ovarian cancer based on AI image recognition, characterized in that, The device includes: an image acquisition module and an ovarian cancer recognition processor, wherein the image acquisition module is communicatively connected to the ovarian cancer recognition processor, and the ovarian cancer recognition processor is used to perform: The image acquisition module acquires patient image data, including upper abdominal image data, pelvic image data, and laparoscopic imaging data of ovarian cancer patients. The patient image data is preprocessed, and the preprocessed patient image data is identified to mark abnormal locations. Based on the abnormal location point, obtain the local magnified image corresponding to the abnormal location point; The magnified local image and the image data are input into the ovarian cancer recognition model to obtain ovarian cancer prediction results. The ovarian cancer identification model is obtained by training based on sample image data and its corresponding ovarian cancer labels. The sample image data includes sample image data and its corresponding multiple magnified sample images.

2. The apparatus according to claim 1, characterized in that, The preprocessing of the patient image data includes: Noise and interference in the patient image data are filtered out, and contrast processing and edge enhancement processing are performed on the patient image data.

3. The apparatus according to claim 2, characterized in that, The process of identifying and marking abnormal locations in preprocessed patient image data includes: The preprocessed patient image data is input into the first neural network model and the second neural network model to obtain the first abnormal location point and the second abnormal location point; The first abnormal location point and the second abnormal location point are merged, and the merged first abnormal location point and second abnormal location point are clustered according to unit regions to obtain the cluster center point corresponding to each unit region. Abnormal locations are marked using the cluster center points.

4. The apparatus according to claim 3, characterized in that, Before marking outlier locations using the cluster center points, the method further includes: Obtain the distance values ​​between multiple cluster centroids within adjacent unit regions; Determine whether the distance between any two cluster centers is less than a preset value, where the preset value is less than the boundary length of the unit region; The step of marking outlier locations using the cluster center points includes: If the distance between any two cluster centers is less than the preset value, then the target center point corresponding to multiple cluster centers in the adjacent unit area is calculated, and the abnormal location point is marked by the target center point. If the distance between any two cluster centers is greater than or equal to the preset value, then the abnormal location points are marked by the cluster centers in the adjacent unit area.

5. The apparatus according to claim 3, characterized in that, The training process for the first neural network model and the second neural network model is as follows: Acquire patient image sample data and their corresponding annotated abnormal location points; The patient image sample data is input into a first neural network model and a second neural network model respectively to obtain a first predicted abnormal location point and a second predicted abnormal location point; the first neural network model and the second neural network model adopt different network structures; Calculate the first positional difference value and the second positional difference value between the first predicted anomaly location point and the second predicted anomaly location point and the marked anomaly location point, respectively; The model parameters of the first neural network model and the second neural network model are updated based on the first position difference value and the second position difference value.

6. The apparatus according to claim 5, characterized in that, The step of updating the model parameters of the first neural network model and the second neural network model based on the first position difference value and the second position difference value includes: Determine whether both the first position difference value and the second position difference value are less than a preset loss value; If both are less than the preset loss value, then training of the first neural network model and the second neural network model is stopped. If the loss is not less than the preset loss value, the model parameters of the first neural network model and the second neural network model are updated according to the magnitude of the first position difference value and the second position difference value, and the first neural network model and the second neural network model are trained again using other patient image sample data.

7. The apparatus according to claim 6, characterized in that, The step of updating the model parameters of the first neural network model and the second neural network model based on the magnitudes of the first position difference value and the second position difference value includes: If the first position difference value is less than the second position difference value, then the model parameters of the second neural network model are updated using the model parameters of the first neural network model; If the first position difference value is greater than the second position difference value, then the model parameters of the first neural network model are updated using the model parameters of the second neural network model.

8. The apparatus according to any one of claims 1-7, characterized in that, The step of inputting the magnified local image and the image data into the ovarian cancer recognition model to obtain ovarian cancer prediction results includes: Determine the positional relationship matrix of each of the magnified local images in the image data; The positional relationship matrix, the corresponding magnified local image, and the image data are input into the ovarian cancer recognition model to obtain the ovarian cancer prediction result.

9. The apparatus according to claim 8, characterized in that, The training process of the ovarian cancer identification model is as follows: Acquire sample image data and its corresponding ovarian cancer label, wherein the sample image data includes sample image data and its corresponding multiple magnified local sample images; Determine the positional relationship matrix of each magnified local sample image in the sample image data; The positional relationship matrix, the corresponding local magnified sample image, and the sample image data are input into the ovarian cancer recognition model to obtain the ovarian cancer prediction result. The model parameters of the ovarian cancer identification model are updated based on the ovarian cancer prediction results, the ovarian cancer labels, and the location relationship matrix.

10. The apparatus according to claim 9, characterized in that, The step of updating the model parameters of the ovarian cancer identification model based on the ovarian cancer prediction results, the ovarian cancer labels, and the positional relationship matrix includes: Determine the positional relationships of the ovarian cancer prediction results and the ovarian cancer labels in the positional relationship matrix; The loss value is calculated based on the location relationship, the ovarian cancer prediction result, and the ovarian cancer label. If the loss value is less than the target value, then training of the ovarian cancer identification model is stopped.

Images