Prediction and evaluation method and device for early-stage ovarian insufficiency

Abstract

The invention discloses a method and a device for predicting and evaluating early-stage ovarian insufficiency. The method comprises the following steps: acquiring clinical and image parameters of a patient; the ovarian image data of the patient is input into an ovarian matrix extraction network model, morphological texture features are obtained, and the ovarian matrix extraction network model comprises an MRI sequence input layer, a slics scribing layer, a matrix feature fusion layer and a feedback network composed of a plurality of convolution layers and fusion layers; the morphological texture features comprise a gray-level co-occurrence matrix, a gray-level run-length matrix, a gray-level dependent matrix and a neighborhood gray-level difference matrix; inputting the clinical parameter data and the morphological texture features into an early-stage ovarian insufficiency prediction model, and outputting a prediction result of the early-stage ovarian insufficiency; the early ovarian insufficiency prediction model comprises a plurality of internal residual networks, an EMP model, an LMP model and a VGG-16 network. Through the ovarian matrix extraction network model and the early ovarian insufficiency prediction model, the early ovarian insufficiency prediction precision is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical auxiliary diagnosis, and in particular to a method and device for predicting and evaluating early ovarian insufficiency, and a computing device. Background Art

[0002] In the diagnosis and prediction of early ovarian insufficiency, traditional methods mainly rely on the patient's clinical history, symptom observation, and hormone level testing. Although it can provide a basis for diagnosis to a certain extent, it has the following major defects:

[0003] Highly subjective: Diagnosis based on clinical symptoms and medical history often relies on the doctor's experience and subjective judgment, and there are large diagnostic differences between different doctors.

[0004] Low sensitivity: The symptoms of early ovarian insufficiency are often not obvious, and the changes in hormone levels are not significant enough, making it difficult for traditional detection methods to accurately identify the disease in the early stages.

[0005] Lack of comprehensive evaluation: Traditional methods often only focus on a single dimension of information (such as hormone levels) and ignore other factors that affect ovarian function (such as age, physical fitness, lifestyle habits, etc.).

[0006] In addition, although imaging technology is widely used in medical diagnosis, in the prediction of early ovarian insufficiency, traditional imaging methods (such as B-ultrasound) can usually only provide morphological information of the ovaries, and it is difficult to deeply analyze the microstructure and functional status of the ovaries. Therefore, it is difficult to achieve a high-accuracy diagnosis in the early stage of the disease.

[0007] Therefore, in order to solve the above problems, the present invention applies deep learning technology in medical image processing to perform comprehensive analysis, extract the rich information contained in ovarian image data, and provide a new perspective for the prediction of early ovarian insufficiency. Summary of the invention

[0008] In view of the above problems, the present invention provides a method and device for predicting and evaluating early ovarian insufficiency, and a computing device.

[0009] According to one aspect of the present invention, a method for predicting and evaluating early ovarian insufficiency is provided, comprising:

[0010] Obtaining clinical parameters of the patient, including but not limited to age, body mass index, reproductive history, sports habits, history of tobacco and alcohol use, weight loss history, and female endocrine hormone levels;

[0011] The patient's ovarian imaging data is input into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slices layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to a grayscale co-occurrence matrix, a grayscale run matrix, a grayscale dependency matrix, and a neighborhood grayscale difference matrix;

[0012] The clinical parameter data and the morphological texture features are input into a pre-trained early ovarian insufficiency prediction model to output a prediction result of early ovarian insufficiency; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, an EMP model, an LMP model and a VGG-16 network.

[0013] In an optional manner, the MRI sequence input layer is used to receive T2-weighted imaging and apparent diffusion coefficient sequence images of MRI, and stack the coefficient sequence images according to channel dimensions to form multi-channel three-dimensional MRI input data;

[0014] The Slices layer is used to select two-dimensional slices covering the ovarian region from the three-dimensional MRI input data, and balance the amount of calculation and information retention according to a preset slice sampling interval.

[0015] In an optional manner, the matrix feature fusion layer is used to simulate the direction and distance parameters in the gray-level co-occurrence matrix through a convolution layer with multiple convolution kernels, and to enable the output feature map to capture the gray-level dependency in the gray-level co-occurrence matrix through an activation function and a pooling operation;

[0016] And, it is used to simulate the run effect of the grayscale run-length matrix by combining the convolution layer and the pooling layer, and capture the neighborhood grayscale difference of the grayscale run-length matrix through the local binary pattern.

[0017] In an optional manner, the internal residual network includes a ResNet-v2 network, a plurality of convolutional layers, a ResNet-50 network or a ResNet-101 network, a BN network layer, and a cross-layer residual connection, wherein the ResNet-v2 network is placed before the plurality of convolutional layers as a pre-activated residual network;

[0018] The EMP model includes a dilated convolutional network, a CBAM attention network, and a digital twin network;

[0019] The LMP model includes an anchor box generation network and a bounding box regression network.

[0020] In an optional manner, the weighted residual connection of the internal residual network is dynamically weighted by introducing a Squeeze-and-Excitation module of the SE-Net network;

[0021] The calculation formula of the weight is:

[0022] α=σ(W2.ReLU(W1.GlobalAvgPool(H(x)⊕x)))

[0023] Among them, ⊕ is the concatenation operation of the feature map; GlobalAvgPool is the global average pooling operation, which is used to compress the spatial dimension of the feature map to 1x1; H(x) performs a convolution transformation on the input x; W1 and W2 are learnable weight matrices; ReLU is the activation function; σ is the Sigmoid activation function.

[0024] In an optional manner, the slices slicing layer adopts a slice sampling algorithm based on the adaptive shape of the ovarian region, and the slice sampling algorithm further includes:

[0025] Using a U-Net network to identify the contour of the ovarian region, and calculating the major axis, minor axis, and main axis direction of the minimum circumscribed ellipse of the contour;

[0026] The positions of each slice are obtained by sampling at non-uniform intervals in the main axis direction. The calculation formula of each slice position is:

[0027]

[0028] Among them, 2a is the pre-coefficient, which represents the amplitude or proportional factor of the entire expression; is a linear transformation that maps the slice index i to an angle related to N and π; β is an adjustment coefficient that adjusts the amplitude of the sine function; x c is a constant term used to adjust the average value or baseline of the output signal; i is the slice index; s i is the i-th slice position; θ is the phase direction.

[0029] In an optional manner, the anchor box generation network in the LMP model uses the K-means++ algorithm for clustering, and during the clustering process, the candidate boxes on the feature maps of different scales are distributed to jointly determine the initial position of the cluster center;

[0030] The IOU loss function of the candidate box is extended to a weighted IOU loss, and different weights are given to candidate boxes of different scales and overlaps to reflect the matching between the anchor box and the real ovarian region;

[0031] The bounding box regression network adopts an FPN network, which includes a separable convolutional layer and a multi-head attention layer.

[0032] In an optional manner, the method further includes:

[0033] Visualize and interactively display the prediction results of early ovarian insufficiency, where the visualization includes but is not limited to comparison charts of prediction probability, risk level, and key clinical parameters; interactive operations support adjustment of prediction model parameters to explore the impact of different parameter settings on prediction results;

[0034] A comprehensive assessment report is generated based on the predictions, the patient's clinical parameters, and ovarian imaging data.

[0035] According to another aspect of the present invention, there is provided a device for predicting and evaluating early ovarian insufficiency, comprising:

[0036] A clinical parameter acquisition module is used to acquire clinical parameters of the patient, including but not limited to age, body mass index, reproductive history, sports habits, history of tobacco and alcohol use, weight loss history, and female endocrine hormone levels;

[0037] A morphological texture feature extraction module is used to input the patient's ovarian image data into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slice layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to a grayscale co-occurrence matrix, a grayscale run matrix, a grayscale dependency matrix, and a neighborhood grayscale difference matrix;

[0038] The result prediction module is used to input the clinical parameter data and the morphological texture features into a pre-trained early ovarian insufficiency prediction model, and output the prediction result of early ovarian insufficiency; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, EMP models, LMP models and VGG-16 networks.

[0039] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus.

[0040] According to the scheme provided by the present invention, the clinical parameters of the patient are obtained, and the clinical parameters include but are not limited to age, body mass index, fertility history, sports habits, history of tobacco and alcohol use, history of weight loss, and female endocrine hormone levels; the ovarian image data of the patient is input into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slices layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to grayscale co-occurrence matrix, grayscale run matrix, grayscale dependency matrix, and neighborhood grayscale difference matrix; the clinical parameter data and the morphological texture features are input into a pre-trained early ovarian insufficiency prediction model, and the prediction result of early ovarian insufficiency is output; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, EMP models, LMP models, and VGG-16 networks. The present invention improves the diagnostic accuracy of early ovarian insufficiency by combining the morphological texture features of the patient's clinical parameters and ovarian image data, and using a deep learning model for comprehensive evaluation. Not only clinical parameters such as the patient's age and body mass index are considered, but also the morphological and texture features of the ovarian imaging data are extracted, which can provide personalized assessments based on the specific conditions of different patients. In the processing of ovarian imaging data, adaptive slice sampling is performed based on the shape of the ovarian region to retain key information. The early ovarian insufficiency prediction model integrates multiple internal residual networks, EMP models, LMP models, and VGG-16 networks, and can maintain high prediction accuracy and stability for different patients and different types of ovarian imaging data.

[0041] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0043] Figure 1 A schematic flow chart of a method for predicting and evaluating early ovarian insufficiency according to an embodiment of the present invention is shown;

[0044] Figure 2 A schematic diagram showing the structure of an ovarian matrix extraction network model according to an embodiment of the present invention is shown;

[0045] Figure 3A schematic diagram showing the structure of a model for predicting early ovarian insufficiency according to an embodiment of the present invention is shown;

[0046] Figure 4 A schematic diagram of the framework of a device for predicting and evaluating early ovarian insufficiency according to an embodiment of the present invention is shown;

[0047] Figure 5 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0048] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0049] Figure 1 FIG. 2 is a flow chart showing a method for predicting and evaluating early ovarian insufficiency according to an embodiment of the present invention. Specifically, Figure 1 As shown, the following steps are included:

[0050] Step S101, obtaining clinical parameters of the patient, including but not limited to age, body mass index, reproductive history, sports habits, history of tobacco and alcohol use, weight loss history, and female endocrine hormone levels.

[0051] In this embodiment, the female endocrine hormone levels include follicle stimulating hormone (FSH), anti-Mullerian hormone (AMH) and estrogen (E2), etc. The above clinical parameters combine the patient's physiology, living habits and endocrine status to provide doctors with a comprehensive patient portrait, which helps to more accurately assess the patient's health status.

[0052] Step S102, inputting the patient's ovarian image data into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slices layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to a grayscale co-occurrence matrix, a grayscale run matrix, a grayscale dependency matrix, and a neighborhood grayscale difference matrix.

[0053] In this embodiment, through various matrix methods such as grayscale co-occurrence matrix, grayscale run matrix, grayscale dependency matrix and neighborhood grayscale difference matrix, various morphological texture features such as grayscale co-occurrence matrix and grayscale run matrix show that there are abnormal texture changes in the patient's ovarian images, such as uneven grayscale distribution and abnormal run length.

[0054] Specifically, Figure 2 Therefore, the MRI sequence data is input into the MRI sequence input layer of the network model, and the MRI sequence data is divided into multiple slices. Each slice represents a level of the image data, which helps to convert the three-dimensional image data into two-dimensional data. The matrix feature fusion layer further processes the slices to extract morphological texture features such as grayscale co-occurrence matrix, grayscale run matrix, grayscale dependency matrix, and neighborhood grayscale difference matrix. The features are convolved through multiple convolutional layers to extract deeper features. The convolutional layers contain structures such as pooling layers to reduce the feature dimension. The fusion layer fuses features at different levels to form a more representative feature representation.

[0055] In an optional manner, the MRI sequence input layer is used to receive T2-weighted imaging and apparent diffusion coefficient sequence images of MRI, and stack the coefficient sequence images according to channel dimensions to form multi-channel three-dimensional MRI input data;

[0056] The Slices layer is used to select two-dimensional slices covering the ovarian region from the three-dimensional MRI input data, and balance the amount of calculation and information retention according to a preset slice sampling interval.

[0057] In this embodiment, the MRI sequence input layer receives the T2-weighted imaging and apparent diffusion coefficient (ADC) sequence images of MRI, which respectively provide structural information and functional information of ovarian tissue. Stacking them according to the channel dimension to form multi-channel three-dimensional MRI input data helps to more comprehensively reflect the state of the ovaries and improve the accuracy of diagnosis. The Slices layer selects two-dimensional slices covering the ovarian area from the three-dimensional MRI input data, and balances the amount of calculation and information retention according to the preset slice sampling interval. It not only reduces unnecessary calculations, but also retains key diagnostic information, which helps to detect ovarian lesions at an early stage.

[0058] Specifically, in the MRI sequence input layer, T2-weighted imaging and ADC sequence images are stacked according to the channel dimension, and T2-weighted imaging is used as a channel, and each diffusion direction or diffusion weight of the ADC sequence image is used as a channel to form multi-channel three-dimensional MRI input data. After receiving the three-dimensional MRI input data, the Slices layer uses the preset ovarian region positioning algorithm to annotate the information and determine the approximate position of the ovary in three-dimensional space. Then, according to the preset slice sampling interval (such as selecting a layer every certain number of layers or a certain thickness), a two-dimensional slice covering the ovarian region is extracted from the three-dimensional data to retain as much diagnostic information of the ovarian region as possible.

[0059] In an optional manner, the matrix feature fusion layer is used to simulate the direction and distance parameters in the gray-level co-occurrence matrix through a convolution layer with multiple convolution kernels, and to enable the output feature map to capture the gray-level dependency in the gray-level co-occurrence matrix through an activation function and a pooling operation;

[0060] And, it is used to simulate the run effect of the grayscale run-length matrix by combining the convolution layer and the pooling layer, and capture the neighborhood grayscale difference of the grayscale run-length matrix through the local binary pattern.

[0061] In this embodiment, in the matrix feature fusion layer, a convolution layer with multiple convolution kernels is first used to simulate the direction and distance parameters in GLCM. Different convolution kernels capture grayscale dependencies in different directions and distances. Next, nonlinearity is increased by activation functions (such as ReLU) to enable the model to learn more complex feature representations. Then, a pooling layer (such as maximum pooling or average pooling) is used to downsample the feature map to reduce the dimension of the feature map. For the simulation of GLRLM, a combination of convolution layers and pooling layers is also used. However, in the design of the convolution layer, more emphasis is placed on capturing the run effect of continuous grayscale values. Local binary pattern (LBP) is used as an auxiliary tool in the above process to capture the neighborhood grayscale difference in GLRLM by comparing the grayscale value difference between the center pixel and its neighborhood pixels. For example, the radiomics library of python is used to extract morphological texture features such as grayscale co-occurrence matrix, grayscale run matrix, grayscale dependency matrix, and neighborhood grayscale difference matrix as the gold standard. During training, the learning rate is set to 0.001, batch_size is set to 8, epoch is set to 100, and optimizers such as Adam or RMSProp are used. The mean square error is used as the loss function, and finally a training model with the minimum or optimal loss value is obtained. Then, the gray-level co-occurrence matrix, gray-level run matrix, gray-level dependency matrix, and neighborhood gray-level difference matrix are inferred through the training model to obtain morphological texture features.

[0062] In an optional manner, the slices slicing layer adopts a slice sampling algorithm based on the adaptive shape of the ovarian region, and the slice sampling algorithm further includes:

[0063] Using a U-Net network to identify the contour of the ovarian region, and calculating the major axis, minor axis, and main axis direction of the minimum circumscribed ellipse of the contour;

[0064] The positions of each slice are obtained by sampling at non-uniform intervals in the main axis direction. The calculation formula of each slice position is:

[0065]

[0066] Among them, 2a is the pre-coefficient, which represents the amplitude or proportional factor of the entire expression; is a linear transformation that maps the slice index i to an angle related to N and π; β is an adjustment coefficient that adjusts the amplitude of the sine function; x c is a constant term used to adjust the average value or baseline of the output signal; i is the slice index; s i is the i-th slice position; θ is the phase direction.

[0067] In this embodiment, the U-Net network is used to identify the contour of the ovarian region, and the ovarian boundary can be accurately extracted. In the slices layer, the U-Net network is first used to identify the contour of the ovarian region, and the major axis, minor axis and main axis direction of its minimum circumscribed ellipse are calculated. Then, according to the preset slice sampling algorithm parameters (such as N = 50, β = 0.5, θ = 0°, 2a = 100, x c = x value of the center coordinate of the ovarian region), 50 slice positions are calculated. These slice positions are unevenly distributed along the main axis of the ovary, with higher slice density in areas that may contain lesions (such as the two ends of the long axis) and lower slice density in other areas. Subsequently, two-dimensional images of these slice positions are extracted from the MRI images and input into subsequent network layers for processing.

[0068] Step S103, inputting the clinical parameter data and the morphological texture features into a pre-trained early ovarian insufficiency prediction model, and outputting a prediction result of early ovarian insufficiency; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, an EMP model, an LMP model and a VGG-16 network.

[0069] In an optional manner, the internal residual network includes a ResNet-v2 network, a plurality of convolutional layers, a ResNet-50 network or a ResNet-101 network, a BN network layer, and a cross-layer residual connection, wherein the ResNet-v2 network is placed before the plurality of convolutional layers as a pre-activated residual network;

[0070] The EMP model includes a dilated convolutional network, a CBAM attention network, and a digital twin network;

[0071] The LMP model includes an anchor box generation network and a bounding box regression network.

[0072] In this embodiment, Figure 3As shown in the figure, ResNet-v2, as a pre-activated residual network, improves the information flow through the pre-activated function and improves the model's ability to extract deep features. ResNet-50 and ResNet-101 networks enhance the accuracy of feature extraction by introducing residual connections. The CBAM attention network in the EMP model adaptively adjusts the importance of different channels and spatial positions in the feature map through channel attention and spatial attention mechanisms, thereby improving the model's attention to key features. The dilated convolutional network is used in the EMP model to capture multi-scale contextual information and more comprehensively capture the complex structure in the ovarian image. The digital twin network is used to construct a virtual representation of the ovary.

[0073] In an optional manner, the weighted residual connection of the internal residual network is dynamically weighted by introducing a Squeeze-and-Excitation module of the SE-Net network;

[0074] The calculation formula of the weight is:

[0075] α=σ(W2.ReLU(W1.GlobalAvgPool(H(x)⊕x)))

[0076] Among them, ⊕ is the concatenation operation of the feature map; GlobalAvgPool is the global average pooling operation, which is used to compress the spatial dimension of the feature map to 1x1; H(x) performs a convolution transformation on the input x; W1 and W2 are learnable weight matrices; ReLU is the activation function; σ is the Sigmoid activation function.

[0077] In this embodiment, SE-Net can adaptively recalibrate the feature response in the channel direction by explicitly modeling the interdependence between channels, enhance the sensitivity to useful features, and suppress useless features, thereby improving the overall representation ability of the network.

[0078] Specifically, the input feature map x( is concatenated (⊕) with the feature map after the convolution transformation H(x) to obtain the concatenated feature map. For example, the shape of the feature map is [N, C, H, W], where N is the batch size, C is the number of channels, and H and W are the height and width of the feature map, respectively.

[0079] A global average pooling operation is performed on the concatenated feature map to compress the spatial dimension of the feature map to 1x1, and a channel descriptor is obtained, which embeds the global distribution information of the channel feature response.

[0080] The first fully connected layer inputs the channel descriptor after global average pooling into the first fully connected layer and performs nonlinear transformation through the ReLU activation function.

[0081] The output of the first fully connected layer is input into the second fully connected layer, and the weight α of each channel is generated through the Sigmoid activation function (σ).

[0082] Finally, according to the calculated weight α, the feature map after convolution transformation is weighted and adjusted to obtain the final output feature map.

[0083] In an optional manner, the anchor box generation network in the LMP model uses the K-means++ algorithm for clustering, and during the clustering process, the candidate boxes on the feature maps of different scales are distributed to jointly determine the initial position of the cluster center;

[0084] The IOU loss function of the candidate box is extended to a weighted IOU loss, and different weights are given to candidate boxes of different scales and overlaps to reflect the matching between the anchor box and the real ovarian region;

[0085] The bounding box regression network adopts an FPN network, which includes a separable convolutional layer and a multi-head attention layer.

[0086] In this embodiment, compared with the traditional K-means algorithm, K-means++ reduces the possibility of the algorithm falling into the local optimum by intelligently selecting the initial cluster center, thereby generating a more accurate and representative anchor box. The distribution of candidate boxes on feature maps of different scales is used to jointly determine the initial position of the cluster center, better capture targets of different sizes and shapes, and improve the adaptability of the anchor box. By assigning different weights to candidate boxes of different scales and overlaps, the weighted IOU loss function can more accurately reflect the degree of match between the anchor box and the real ovarian area, and pay more attention to the candidate box that has a greater impact on the detection performance. The separable convolution layer and the multi-head attention layer further enhance the feature extraction capability of the FPN network, making the regression of the bounding box more accurate.

[0087] In an optional manner, the method further includes:

[0088] Visualize and interactively display the prediction results of early ovarian insufficiency, where the visualization includes but is not limited to comparison charts of prediction probability, risk level, and key clinical parameters; interactive operations support adjustment of prediction model parameters to explore the impact of different parameter settings on prediction results;

[0089] A comprehensive assessment report is generated based on the predictions, the patient's clinical parameters, and ovarian imaging data.

[0090] In this embodiment, in the visualization interface, the doctor can clearly see the patient's predicted probability, risk level, and comparison chart of key clinical parameters. The parameters of the prediction model can be further adjusted through interactive operations, and a comprehensive evaluation report can be generated based on the prediction results, the patient's clinical parameters, and ovarian imaging data.

[0091] According to the scheme provided by the present invention, the clinical parameters of the patient are obtained, and the clinical parameters include but are not limited to age, body mass index, fertility history, sports habits, history of tobacco and alcohol use, history of weight loss, and female endocrine hormone levels; the ovarian image data of the patient is input into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slices layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to grayscale co-occurrence matrix, grayscale run matrix, grayscale dependency matrix, and neighborhood grayscale difference matrix; the clinical parameter data and the morphological texture features are input into a pre-trained early ovarian insufficiency prediction model, and the prediction result of early ovarian insufficiency is output; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, EMP models, LMP models, and VGG-16 networks. The present invention improves the diagnostic accuracy of early ovarian insufficiency by combining the morphological texture features of the patient's clinical parameters and ovarian image data, and using a deep learning model for comprehensive evaluation. Not only clinical parameters such as the patient's age and body mass index are considered, but also the morphological and texture features of the ovarian imaging data are extracted, which can provide personalized assessments based on the specific conditions of different patients. In the processing of ovarian imaging data, adaptive slice sampling is performed based on the shape of the ovarian region to retain key information. The early ovarian insufficiency prediction model integrates multiple internal residual networks, EMP models, LMP models, and VGG-16 networks, and can maintain high prediction accuracy and stability for different patients and different types of ovarian imaging data.

[0092] Figure 4 The schematic diagram of the framework of the early ovarian insufficiency prediction and evaluation device of the embodiment of the present invention is shown. The early ovarian insufficiency prediction and evaluation device comprises:

[0093] A clinical parameter acquisition module is used to acquire clinical parameters of the patient, including but not limited to age, body mass index, reproductive history, sports habits, history of tobacco and alcohol use, weight loss history, and female endocrine hormone levels;

[0094] A morphological texture feature extraction module is used to input the patient's ovarian image data into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slice layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to a grayscale co-occurrence matrix, a grayscale run matrix, a grayscale dependency matrix, and a neighborhood grayscale difference matrix;

[0095] The result prediction module is used to input the clinical parameter data and the morphological texture features into a pre-trained early ovarian insufficiency prediction model, and output the prediction result of early ovarian insufficiency; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, EMP models, LMP models and VGG-16 networks.

[0096] Figure 5 The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0097] like Figure 5 As shown, the computing device may include: a processor (processor) 502 , a communications interface (Communications Interface) 504 , a memory (memory) 506 , and a communication bus 508 .

[0098] The processor 502, the communication interface 504, and the memory 506 communicate with each other via a communication bus 508. The communication interface 504 is used to communicate with other devices such as a client or other server network elements. The processor 502 is used to execute a program 510, which can specifically execute the relevant steps in the above-mentioned early ovarian insufficiency prediction and evaluation method embodiment.

[0099] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0100] The processor 502 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0101] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0102] According to the scheme provided by the present invention, the clinical parameters of the patient are obtained, and the clinical parameters include but are not limited to age, body mass index, fertility history, sports habits, history of tobacco and alcohol use, history of weight loss, and female endocrine hormone levels; the ovarian image data of the patient is input into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slices layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to grayscale co-occurrence matrix, grayscale run matrix, grayscale dependency matrix, and neighborhood grayscale difference matrix; the clinical parameter data and the morphological texture features are input into a pre-trained early ovarian insufficiency prediction model, and the prediction result of early ovarian insufficiency is output; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, EMP models, LMP models, and VGG-16 networks. The present invention improves the diagnostic accuracy of early ovarian insufficiency by combining the morphological texture features of the patient's clinical parameters and ovarian image data, and using a deep learning model for comprehensive evaluation. Not only clinical parameters such as the patient's age and body mass index are considered, but also the morphological and texture features of the ovarian imaging data are extracted, which can provide personalized assessments based on the specific conditions of different patients. In the processing of ovarian imaging data, adaptive slice sampling is performed based on the shape of the ovarian region to retain key information. The early ovarian insufficiency prediction model integrates multiple internal residual networks, EMP models, LMP models, and VGG-16 networks, and can maintain high prediction accuracy and stability for different patients and different types of ovarian imaging data.

[0103] Those skilled in the art will appreciate that the modules in the devices in the embodiments can be adaptively changed and set in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition they can be divided into multiple submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner can be combined in any combination. Unless otherwise explicitly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) can be replaced by an alternative feature that provides the same, equivalent or similar purpose. In addition, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments means being within the scope of the present invention and forming different embodiments. For example, in the following claims, any one of the claimed embodiments can be used in any combination. The present invention can be implemented by hardware including several different elements and by appropriately programmed computers. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The steps in the above embodiments should not be understood as limiting the execution order unless otherwise specified.

Claims

1. A method for predicting and evaluating early ovarian insufficiency, characterized in that: include: Obtaining clinical parameters of the patient, including but not limited to age, body mass index, reproductive history, sports habits, history of tobacco and alcohol use, weight loss history, and female endocrine hormone levels; The patient's ovarian imaging data is input into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slices layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to a grayscale co-occurrence matrix, a grayscale run matrix, a grayscale dependency matrix, and a neighborhood grayscale difference matrix; The clinical parameter data and the morphological texture features are input into a pre-trained early ovarian insufficiency prediction model to output a prediction result of early ovarian insufficiency; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, an EMP model, an LMP model and a VGG-16 network.

2. The method for predicting and evaluating early ovarian insufficiency according to claim 1, characterized in that: The MRI sequence input layer is used to receive T2-weighted imaging and apparent diffusion coefficient sequence images of MRI, and stack the coefficient sequence images according to channel dimensions to form multi-channel three-dimensional MRI input data; The Slices layer is used to select two-dimensional slices covering the ovarian region from the three-dimensional MRI input data, and balance the amount of calculation and information retention according to a preset slice sampling interval.

3. The method for predicting and evaluating early ovarian insufficiency according to claim 1, characterized in that: The matrix feature fusion layer is used to simulate the direction and distance parameters in the gray-level co-occurrence matrix through a convolution layer with multiple convolution kernels, and to enable the output feature map to capture the gray-level dependency in the gray-level co-occurrence matrix through activation functions and pooling operations; And, it is used to simulate the run effect of the grayscale run-length matrix by combining the convolution layer and the pooling layer, and capture the neighborhood grayscale difference of the grayscale run-length matrix through the local binary pattern.

4. The method for predicting and evaluating early ovarian insufficiency according to claim 1, characterized in that: The internal residual network includes a ResNet-v2 network, a plurality of convolutional layers, a ResNet-50 network or a ResNet-101 network, a BN network layer, and a cross-layer residual connection, wherein the ResNet-v2 network is placed before the plurality of convolutional layers as a pre-activated residual network; The EMP model includes a dilated convolutional network, a CBAM attention network, and a digital twin network; The LMP model includes an anchor box generation network and a bounding box regression network.

5. The method for predicting and evaluating early ovarian insufficiency according to claim 1, characterized in that: The weighted residual connection of the internal residual network is dynamically weighted by introducing the Squeeze-and-Excitation module of the SE-Net network; The calculation formula of the weight is: α=σ(W2.ReLU(W1.GlobalAvgPool(H(x)⊕x))) Among them, ⊕ is the concatenation operation of the feature map; GlobalAvgPool is the global average pooling operation, which is used to compress the spatial dimension of the feature map to 1x1; H(x) performs a convolution transformation on the input x; W1 and W2 are learnable weight matrices; ReLU is the activation function; σ is the Sigmoid activation function.

6. The method for predicting and evaluating early ovarian insufficiency according to claim 1, characterized in that: The slices slicing layer adopts a slice sampling algorithm based on the adaptive shape of the ovarian region, and the slice sampling algorithm further includes: Using a U-Net network to identify the contour of the ovarian region, and calculating the major axis, minor axis, and main axis direction of the minimum circumscribed ellipse of the contour; The positions of each slice are obtained by sampling at non-uniform intervals in the main axis direction. The calculation formula of each slice position is: Among them, 2a is the pre-coefficient, which represents the amplitude or proportional factor of the entire expression; is a linear transformation that maps the slice index i to an angle related to N and π; β is an adjustment coefficient that adjusts the amplitude of the sine function; x c is a constant term used to adjust the average value or baseline of the output signal; i is the slice index; s i is the i-th slice position; θ is the phase direction.

7. The method for predicting and evaluating early ovarian insufficiency according to claim 4, characterized in that: The anchor box generation network in the LMP model uses the K-means++ algorithm for clustering. During the clustering process, the candidate boxes on the feature maps of different scales are distributed to jointly determine the initial position of the cluster center. The IOU loss function of the candidate box is extended to a weighted IOU loss, and different weights are given to candidate boxes of different scales and overlaps to reflect the matching between the anchor box and the real ovarian region; The bounding box regression network adopts an FPN network, which includes a separable convolutional layer and a multi-head attention layer.

8. The method for predicting and evaluating early ovarian insufficiency according to any one of claims 1 to 7, characterized in that: The method further comprises: Visualize and interactively display the prediction results of early ovarian insufficiency, where the visualization includes but is not limited to comparison charts of prediction probability, risk level, and key clinical parameters; interactive operations support adjustment of prediction model parameters to explore the impact of different parameter settings on prediction results; A comprehensive assessment report is generated based on the predictions, the patient's clinical parameters, and ovarian imaging data.

9. An early ovarian insufficiency prediction and evaluation device, characterized in that: include: A clinical parameter acquisition module is used to acquire clinical parameters of the patient, including but not limited to age, body mass index, reproductive history, sports habits, history of tobacco and alcohol use, weight loss history, and female endocrine hormone levels; A morphological texture feature extraction module is used to input the patient's ovarian image data into the ovarian matrix extraction network model to obtain morphological texture features, wherein the ovarian matrix extraction network model includes an MRI sequence input layer, a slice layer, a matrix feature fusion layer, and a feedback network composed of multiple convolutional layers and fusion layers; the morphological texture features include but are not limited to a grayscale co-occurrence matrix, a grayscale run matrix, a grayscale dependency matrix, and a neighborhood grayscale difference matrix; The result prediction module is used to input the clinical parameter data and the morphological texture features into a pre-trained early ovarian insufficiency prediction model, and output the prediction result of early ovarian insufficiency; wherein the early ovarian insufficiency prediction model includes multiple internal residual networks, EMP models, LMP models and VGG-16 networks.

10. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned early ovarian insufficiency prediction and evaluation method.

Images