Prediction method, device and equipment for endometrial cancer based on pathological deep learning and storage medium

Abstract

The invention provides an endometrial cancer prediction method, device and equipment based on pathological deep learning, and a storage medium, and the method comprises the steps: obtaining the image data of endometrium, carrying out the spatial feature extraction of the image data of endometrium through employing a convolutional neural network, carrying out the analysis of the lesion form, tissue density and abnormal region of the spatial features, and obtaining the prediction result of the endometrial cancer. Generating an image vector; the method comprises the following steps: acquiring a historical clinical diagnosis text, and extracting a text vector associated with endometrial cancer by using a Transform model; obtaining candidate genes associated with endometrial cancer in the gene data, analyzing the candidate genes by using PCA (Principal Component Analysis) to extract mutation site information and variation types, and generating gene vectors according to the mutation site information and the variation types; and fusing the image vector, the text vector and the gene vector to generate a fusion feature, predicting the fusion feature by using a pre-trained multi-layer perceptron, and outputting a prediction result of the endometrial cancer. Early prediction of endometrial cancer is realized, and invasive surgical operations are reduced.

Description

Technical Field

[0001] The present invention relates to the field of pathology deep learning, and in particular to a prediction method, device, equipment and storage medium for endometrial cancer based on pathology deep learning. Background Art

[0002] Diagnostic curettage is usually performed after symptoms become apparent, and has a high diagnosis rate. However, because patients are often already in the middle or late stages of the disease when they are discovered, many cases have missed the best time for treatment before diagnosis.

[0003] In view of this, this application is filed. Summary of the invention

[0004] The present invention discloses a prediction method, device, equipment and storage medium for endometrial cancer based on pathological deep learning, aiming to achieve early prediction of endometrial cancer and reduce invasive surgical operations.

[0005] The first embodiment of the present invention provides a method for predicting endometrial cancer based on pathological deep learning, comprising: Acquiring endometrial imaging data, and using a convolutional neural network to extract spatial features from the endometrial imaging data, and analyzing the spatial features in terms of lesion morphology, tissue density, and abnormal regions, so as to generate a high-dimensional feature vector of the imaging data; Obtain historical clinical diagnosis texts and use the Transformer model to extract text embedding vectors associated with endometrial cancer; Obtaining candidate genes associated with endometrial cancer in the gene data, and analyzing the candidate genes using PCA to extract mutation site information and variation typing, and generating a gene data feature vector based on the mutation site information and variation typing; The high-dimensional feature vector of the image data, the text embedding vector, and the feature vector of the gene data are fused to generate a fused feature, and the fused feature is predicted using a pre-trained multi-layer perceptron to output a prediction result of endometrial cancer, wherein the multi-layer perceptron uses a cross entropy loss function and an Adam optimizer to update the model parameters during the training process.

[0006] Preferably, the image data of the endometrium is acquired, and a convolutional neural network is used to extract spatial features from the image data of the endometrium, and the spatial features are analyzed for lesion morphology, tissue density and abnormal area to generate a high-dimensional feature vector of the image data, specifically: Acquiring image data and preprocessing the image data, wherein the image data is one or more of ultrasonic images, CT images, and MRI images; Calling a pre-trained convolutional neural network, performing a convolution operation on the pre-processed image data through a set of trained convolution kernels in a convolution layer to extract local feature information, and down-sampling the local feature information through a pooling layer to extract spatial features; The lesion morphology, tissue density and abnormal area are analyzed based on the spatial features, and a high-dimensional feature vector of the image data is generated, wherein the high-dimensional feature vector of the image data can represent the pathological features in the endometrial image.

[0007] Preferably, the method of obtaining candidate genes associated with endometrial cancer in the gene data, analyzing the candidate genes using PCA to extract mutation site information and variation typing, and generating a gene data feature vector according to the mutation site information and variation typing is specifically as follows: Obtaining candidate genes associated with endometrial cancer in the gene data, and performing standardization processing on the candidate genes; Mapping the standardized candidate genes from high-dimensional features to a low-dimensional space through PCA, and extracting mutation site information from each candidate gene, wherein the mutation site information includes mutation position and mutation type; The gene is typed based on the mutation type and a typing result is generated, and the mutation site information and the typing result are converted into a gene data feature vector, wherein the gene data feature vector includes a binary feature, a mutation frequency feature, a mutation type ratio, a function prediction score, and a variation typing code.

[0008] Preferably, the expression of the fusion feature is:

[0009] in, , , is the weight parameter, , , They are high-dimensional feature vectors of image data, text embedding vectors, and feature vectors of genetic data. , , are the number of features of each data modality; The expression of multi-layer perceptron is:

[0010] in, is the predicted probability of the lesion, is the Sigmoid activation function, and are the weights and biases of the multilayer perceptron.

[0011] The second embodiment of the present invention provides a prediction device for endometrial cancer based on pathological deep learning, comprising: An image data processing unit, used to obtain image data of the endometrium, and use a convolutional neural network to extract spatial features of the image data of the endometrium, and analyze the spatial features in terms of lesion morphology, tissue density and abnormal area to generate a high-dimensional feature vector of the image data; A diagnostic text processing unit that obtains historical clinical diagnosis text and uses the Transformer model to extract text embedding vectors associated with endometrial cancer; A gene data processing unit, used to obtain candidate genes associated with endometrial cancer in the gene data, and use PCA to analyze the candidate genes to extract mutation site information and variation typing, and generate a gene data feature vector according to the mutation site information and variation typing; A prediction unit is used to fuse the high-dimensional feature vector of the image data, the text embedding vector, and the feature vector of the gene data to generate a fused feature, and use a pre-trained multi-layer perceptron to predict the fused feature, and output a prediction result of endometrial cancer, wherein the multi-layer perceptron uses a cross entropy loss function and an Adam optimizer to update model parameters during the training process.

[0012] Preferably, the image data processing unit is specifically used for: Acquiring image data and preprocessing the image data, wherein the image data is one or more of ultrasonic images, CT images, and MRI images; Calling a pre-trained convolutional neural network, performing a convolution operation on the pre-processed image data through a set of trained convolution kernels in a convolution layer to extract local feature information, and down-sampling the local feature information through a pooling layer to extract spatial features; The lesion morphology, tissue density and abnormal area are analyzed based on the spatial features, and a high-dimensional feature vector of the image data is generated, wherein the high-dimensional feature vector of the image data can represent the pathological features in the endometrial image.

[0013] Preferably, the gene data processing unit is specifically used for: Obtaining candidate genes associated with endometrial cancer in the gene data, and performing standardization processing on the candidate genes; Mapping the standardized candidate genes from high-dimensional features to a low-dimensional space through PCA, and extracting mutation site information from each candidate gene, wherein the mutation site information includes mutation position and mutation type; The gene is typed based on the mutation type and a typing result is generated, and the mutation site information and the typing result are converted into a gene data feature vector, wherein the gene data feature vector includes a binary feature, a mutation frequency feature, a mutation type ratio, a function prediction score, and a variation typing code.

[0014] Preferably, the expression of the fusion feature is:

[0015] in, , , is the weight parameter, , , They are high-dimensional feature vectors of image data, text embedding vectors, and feature vectors of genetic data. , , are the number of features of each data modality; The expression of multi-layer perceptron is:

[0016] in, is the predicted probability of the lesion, is the Sigmoid activation function, and are the weights and biases of the multilayer perceptron.

[0017] The third embodiment of the present invention provides a prediction device for endometrial cancer based on pathological deep learning, including a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a prediction method for endometrial cancer based on pathological deep learning as described in any one of the above.

[0018] A fourth embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer-readable storage medium is located, so as to implement a method for predicting endometrial cancer based on pathological deep learning as described in any one of the above items.

[0019] A prediction method, device, equipment and storage medium for endometrial cancer based on pathological deep learning provided by the present invention are provided. The method first obtains image data of the endometrium, and uses a convolutional neural network to extract spatial features of the image data of the endometrium, and analyzes the spatial features in terms of lesion morphology, tissue density and abnormal area to generate a high-dimensional feature vector of the image data; then, historical clinical diagnosis texts are obtained, and a Transformer model is used to extract text embedding vectors associated with endometrial cancer; then, candidate genes associated with endometrial cancer in gene data are obtained, and PCA is used to analyze the candidate genes to extract mutation site information and variation typing, and a gene data feature vector is generated according to the mutation site information and variation typing; finally, the high-dimensional feature vector of the image data, the text embedding vector, and the gene data feature vector are fused to generate a fused feature, and a pre-trained multilayer perceptron is used to predict the fused feature, and a prediction result of endometrial cancer is output, wherein the multilayer perceptron uses a cross entropy loss function and an Adam optimizer to update model parameters during training. Achieve early prediction of endometrial cancer and reduce invasive surgical procedures. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flowchart of a method for predicting endometrial cancer based on pathological deep learning provided by the first embodiment of the present invention; Figure 2 This is a module schematic diagram of a device for predicting endometrial cancer based on pathological deep learning provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In order to better understand the technical solution of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] It should be clear that the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0025] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0026] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0027] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0028] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0029] The present invention discloses a prediction method, device, equipment and storage medium for endometrial cancer based on pathological deep learning, aiming to achieve early prediction of endometrial cancer and reduce invasive surgical operations.

[0030] The first embodiment of the present invention provides a method for predicting endometrial cancer based on pathological deep learning, which can be performed by a prediction device for endometrial cancer based on pathological deep learning (hereinafter referred to as a prediction device), and in particular, by one or more processors in the prediction device to implement at least the following steps: S101, acquiring image data of the endometrium, and using a convolutional neural network to extract spatial features of the image data of the endometrium, and analyzing the spatial features in terms of lesion morphology, tissue density, and abnormal area, so as to generate a high-dimensional feature vector of the image data; In this embodiment, the prediction device can be a terminal with data processing capabilities such as a desktop computer or a laptop computer. The prediction device can be installed with a corresponding operating system and application software, and the functions required by this embodiment can be realized through the combination of the operating system and the application software.

[0031] In a specific embodiment of the present invention, endometrial image data collected by one or more devices including ultrasound, CT, and MRI are obtained. It should be understood that different types of image data provide different perspectives and details, and can capture the morphological characteristics and potential lesion areas of the endometrium from multiple angles.

[0032] After acquiring the image data, it is first preprocessed. The preprocessing steps include operations such as removing noise, enhancing contrast, and standardizing image size to ensure the consistent quality of the image data input to the convolutional neural network (CNN) and avoid interference caused by inconsistent data. Next, the pre-trained convolutional neural network is used to extract spatial features from the preprocessed image data. The convolution layer of the convolutional neural network performs convolution operations on the image data using the trained convolution kernel to extract local features from the image, such as basic features such as edges, textures, and colors. Through multiple convolution operations, the network can automatically learn complex patterns in the image, including tiny lesions or abnormal structures of the endometrium. The features extracted by the convolutional layer are usually high-dimensional and contain a large amount of local information, especially for small lesions or subtle changes in tissue structures.

[0033] After the convolution layer extracts local features, these local features are downsampled through the pooling layer to reduce the amount of calculation. The pooling operation helps retain the most important spatial features in the image while reducing unimportant noise and details, thereby focusing more on large-scale tissues and lesion areas.

[0034] Based on the spatial features extracted by the convolutional network, the lesion morphology, tissue density and abnormal areas are further analyzed in depth. In the lesion morphology analysis, the convolutional neural network can accurately identify abnormal morphology in the endometrium, including irregular hyperplasia, tumor boundaries, etc., and can even judge the nature of the lesion from tiny morphological changes. By analyzing the tissue density, it can be found that the density changes in the lesion area compared to normal tissue can be found. In particular, through CT and MRI images, the difference in tissue density can be clearly seen, further indicating possible pathological changes.

[0035] In addition, the analysis of abnormal areas is particularly important in image processing. By analyzing spatial features, areas in the endometrium that are different from normal tissues can be automatically identified, especially those with asymmetric morphology and heterogeneous distribution, which are often potential signs of lesions. Using convolutional neural networks, these abnormal areas can be quickly and accurately separated from normal tissues, thereby providing more accurate diagnostic basis for clinicians. These spatial features will be aggregated into a high-dimensional feature vector. This feature vector integrates the analysis results of lesion morphology, tissue density, and abnormal areas, and can accurately express the pathological characteristics in endometrial images.

[0036] S102, obtaining historical clinical diagnosis texts and using the Transformer model to extract text embedding vectors associated with endometrial cancer; It should be noted that in this embodiment, the goal is to extract valuable information related to endometrial cancer by acquiring historical clinical diagnosis text data. These historical text data come from the hospital's electronic medical records, diagnostic reports, examination records, etc. These text records contain important information such as the patient's medical history, symptom description, clinical examination results, and diagnosis and treatment recommendations. Since the early symptoms of endometrial cancer are usually hidden and the clinical manifestations of different patients vary to a certain extent, the details recorded in the historical clinical diagnosis text data may provide key clues for the early identification of the disease.

[0037] Next, in order to efficiently process and analyze these text data, the Transformer model in modern natural language processing technology was used. As a deep learning model based on the self-attention mechanism architecture, the Transformer model can automatically extract potential semantic information from the text. By calling the trained Transformer model, the historical clinical diagnosis text is first processed. After preprocessing steps such as word segmentation, stop word removal and standardization, the original text data enters the input layer of the Transformer model. The model uses the self-attention mechanism to weight each word in the text, and assigns different weights according to the importance of the word in the context, so as to effectively capture the deep semantics and potential relationships in the text. Unlike traditional rule-based methods, Transformer can automatically learn the language rules in the text through a large amount of corpus training, especially in the expression of disease diagnosis-related terms, symptom descriptions and clinical characteristics, and can identify and extract the most diagnostically significant information.

[0038] After the encoding process of the Transformer model, the model can generate a text embedding vector for each piece of text. These vectors are high-dimensional numerical representations that can capture the core information and semantic structure in the text. For text related to endometrial cancer, Transformer can identify keywords related to the disease, such as "irregular bleeding", "endometrial thickening", "lumps", etc., and can even associate this information through contextual relationships to provide a more comprehensive disease risk assessment. It should be understood that the generated text embedding vector is not just a simple word frequency statistic, but a multi-dimensional, deep representation that can reflect the deep semantics of the text.

[0039] In a possible implementation of the present invention, a multi-task learning mechanism can be introduced to configure multiple related tasks for different clinical characteristics of endometrial cancer (such as symptoms, diagnosis, treatment, etc.), and multi-task joint training can be performed using shared feature representation. This allows the model to not only extract basic information about endometrial cancer, but also better capture the relationship between different clinical factors, thereby improving the model's ability in early disease prediction and generation of personalized diagnosis and treatment plans.

[0040] S103, obtaining candidate genes associated with endometrial cancer in the gene data, and using PCA to analyze the candidate genes to extract mutation site information and variation typing, and generating a gene data feature vector according to the mutation site information and variation typing; In this embodiment, candidate genes are first screened out. These candidate genes have been confirmed to be closely related to endometrial cancer in multiple independent studies, and show significant differences in gene mutation, expression level and functional impact.

[0041] After obtaining the candidate genes, these data are standardized to ensure data consistency between different samples, thereby eliminating deviations caused by external factors such as the detection platform and sequencing depth. Next, the principal component analysis (PCA) method is used to map the high-dimensional candidate gene data to a low-dimensional space, which not only effectively reduces data redundancy, but also retains the main variation information that best reflects the characteristics of endometrial cancer. In this process, this embodiment introduces an adaptive weight adjustment strategy based on variation information, so that mutation information that is closely related to the disease is given a higher retention weight during the dimensionality reduction process.

[0042] Through PCA, the preset device can accurately extract the mutation site information in each candidate gene, including the specific mutation location and mutation type, thereby forming a preliminary description of the variation characteristics. Subsequently, based on the extracted mutation type information, the prediction device further types the candidate genes. This typing not only depends on the classification of mutation types, but also comprehensively considers the frequency of occurrence of each mutation in different samples and the functional prediction scores calculated by functional prediction tools (such as SIFT and PolyPhen), thereby generating more accurate and biologically meaningful typing results.

[0043] Finally, these mutation site information and typing results are converted into a multi-dimensional genetic data feature vector, which contains binary features (used to indicate whether a mutation site exists), mutation frequency features, mutation type ratios, functional prediction scores, and variant typing codes. This feature vector not only fully reflects the key role of candidate genes in the pathogenesis of endometrial cancer, but also greatly improves the prediction accuracy of the overall multimodal diagnostic system by integrating with other modal data (such as medical images and clinical texts).

[0044] S104, fusing the high-dimensional feature vector of the image data, the text embedding vector, and the feature vector of the gene data to generate a fused feature, and using a pre-trained multi-layer perceptron to predict the fused feature, and outputting a prediction result of endometrial cancer, wherein the multi-layer perceptron uses a cross entropy loss function and an Adam optimizer to update model parameters during the training process.

[0045] Preferably, the expression of the fusion feature is:

[0046] in, , , The weight parameters need to be optimized through back propagation. , , They are high-dimensional feature vectors of image data, text embedding vectors, and feature vectors of genetic data. , , are the number of features of each data modality; The expression of multi-layer perceptron is:

[0047] in, is the predicted probability of the lesion, is the Sigmoid activation function, and are the weights and biases of the multilayer perceptron.

[0048] The expression of the cross entropy loss function is: ; in, is the true label, is the predicted probability of the lesion.

[0049] It should be noted that in this embodiment, different features are first extracted from the multimodal data. Specifically, the image data will undergo a preprocessing step to remove noise and unify the data format. This process provides standardized input for the subsequent deep learning model, ensuring that the model can accurately learn the important features in the image data. After that, the extracted image data features will be input into a multi-layer perceptron and fused with the basic data features to generate fused features. This fusion process is learned and optimized through multi-layer neural networks, which can not only retain the spatial features from the image data, but also effectively capture the potential laws in other modal data (such as genetic data or clinical texts).

[0050] In order to further process and learn these fused features, a multi-layer perceptron structure based on back propagation is adopted. Through the repeated adjustment of convolution kernels and weight parameters during the training process, the neural network can automatically identify which features are most important for the final prediction task and continuously optimize the weights of these features. After each feature is processed by the multi-layer neural network, the output prediction results can effectively reflect the relationship between different modalities.

[0051] In the training process of the multilayer perceptron, the Adam optimizer is used, which can efficiently update network parameters and quickly converge to the optimal solution. Specifically, the Adam optimizer combines the momentum method and the adaptive learning rate adjustment mechanism, which greatly improves the stability of the training process and avoids overfitting by correcting and adjusting the gradient each time.

[0052] See also Figure 2 The second embodiment of the present invention provides a prediction device for endometrial cancer based on pathological deep learning, comprising: An image data processing unit 201 is used to obtain image data of the endometrium, and use a convolutional neural network to extract spatial features of the image data of the endometrium, and analyze the spatial features in terms of lesion morphology, tissue density and abnormal area to generate a high-dimensional feature vector of the image data; A diagnosis text processing unit 202, used to obtain historical clinical diagnosis texts and extract text embedding vectors associated with endometrial cancer using a Transformer model; A gene data processing unit 203 is used to obtain candidate genes associated with endometrial cancer in the gene data, and use PCA to analyze the candidate genes to extract mutation site information and variation typing, and generate a gene data feature vector according to the mutation site information and variation typing; The prediction unit 204 is used to fuse the high-dimensional feature vector of the image data, the text embedding vector, and the gene data feature vector to generate a fused feature, and use a pre-trained multi-layer perceptron to predict the fused feature, and output a prediction result of endometrial cancer, wherein the multi-layer perceptron uses a cross entropy loss function and an Adam optimizer to update the model parameters during the training process.

[0053] Preferably, the image data processing unit is specifically used for: Acquiring image data and preprocessing the image data, wherein the image data is one or more of ultrasonic images, CT images, and MRI images; Calling a pre-trained convolutional neural network, performing a convolution operation on the pre-processed image data through a set of trained convolution kernels in a convolution layer to extract local feature information, and down-sampling the local feature information through a pooling layer to extract spatial features; The lesion morphology, tissue density and abnormal area are analyzed based on the spatial features, and a high-dimensional feature vector of the image data is generated, wherein the high-dimensional feature vector of the image data can represent the pathological features in the endometrial image.

[0054] Preferably, the gene data processing unit is specifically used for: Obtaining candidate genes associated with endometrial cancer in the gene data, and performing standardization processing on the candidate genes; Mapping the standardized candidate genes from high-dimensional features to a low-dimensional space through PCA, and extracting mutation site information from each candidate gene, wherein the mutation site information includes mutation position and mutation type; The gene is typed based on the mutation type and a typing result is generated, and the mutation site information and the typing result are converted into a gene data feature vector, wherein the gene data feature vector includes a binary feature, a mutation frequency feature, a mutation type ratio, a function prediction score, and a variation typing code.

[0055] Preferably, the expression of the fusion feature is:

[0056] in, , , is the weight parameter, , , They are high-dimensional feature vectors of image data, text embedding vectors, and feature vectors of genetic data. , , are the number of features of each data modality; The expression of multi-layer perceptron is:

[0057] in, is the predicted probability of the lesion, is the Sigmoid activation function, and are the weights and biases of the multilayer perceptron.

[0058] The third embodiment of the present invention provides a prediction device for endometrial cancer based on pathological deep learning, including a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a prediction method for endometrial cancer based on pathological deep learning as described in any one of the above.

[0059] A fourth embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer-readable storage medium is located, so as to implement a method for predicting endometrial cancer based on pathological deep learning as described in any one of the above items.

[0060] A prediction method, device, equipment and storage medium for endometrial cancer based on pathological deep learning provided by the present invention are provided. The method first obtains image data of the endometrium, and uses a convolutional neural network to extract spatial features of the image data of the endometrium, and analyzes the spatial features in terms of lesion morphology, tissue density and abnormal area to generate a high-dimensional feature vector of the image data; then, historical clinical diagnosis texts are obtained, and a Transformer model is used to extract text embedding vectors associated with endometrial cancer; then, candidate genes associated with endometrial cancer in gene data are obtained, and PCA is used to analyze the candidate genes to extract mutation site information and variation typing, and a gene data feature vector is generated according to the mutation site information and variation typing; finally, the high-dimensional feature vector of the image data, the text embedding vector, and the gene data feature vector are fused to generate a fused feature, and a pre-trained multilayer perceptron is used to predict the fused feature, and a prediction result of endometrial cancer is output, wherein the multilayer perceptron uses a cross entropy loss function and an Adam optimizer to update model parameters during training. Achieve early prediction of endometrial cancer and reduce invasive surgical procedures.

[0061] Exemplarily, the computer program described in the third and fourth embodiments of the present invention may be divided into one or more modules, and the one or more modules are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the device for implementing a prediction of endometrial cancer based on pathological deep learning. For example, the device described in the second embodiment of the present invention.

[0062] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the method for predicting endometrial cancer based on pathological deep learning, and uses various interfaces and lines to connect the various parts of the method for predicting endometrial cancer based on pathological deep learning.

[0063] The memory can be used to store the computer program and / or module, and the processor realizes various functions of a prediction method for endometrial cancer based on pathological deep learning by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, a text conversion function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, text message data, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0064] Wherein, if the implemented module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0065] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0066] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A prediction method for endometrial cancer based on pathological deep learning, characterized in that: include: Acquiring endometrial imaging data, and using a convolutional neural network to extract spatial features from the endometrial imaging data, and analyzing the spatial features in terms of lesion morphology, tissue density, and abnormal regions, so as to generate a high-dimensional feature vector of the imaging data; Obtain historical clinical diagnosis texts and use the Transformer model to extract text embedding vectors associated with endometrial cancer; Obtaining candidate genes associated with endometrial cancer in the gene data, and analyzing the candidate genes using PCA to extract mutation site information and variation typing, and generating a gene data feature vector based on the mutation site information and variation typing; The high-dimensional feature vector of the image data, the text embedding vector, and the feature vector of the gene data are fused to generate a fused feature, and the fused feature is predicted using a pre-trained multi-layer perceptron to output a prediction result of endometrial cancer, wherein the multi-layer perceptron uses a cross entropy loss function and an Adam optimizer to update the model parameters during the training process.

2. The method for predicting endometrial cancer based on pathological deep learning according to claim 1, characterized in that: The method of acquiring the image data of the endometrium, extracting the spatial features of the image data of the endometrium using a convolutional neural network, and analyzing the spatial features in terms of lesion morphology, tissue density, and abnormal area to generate a high-dimensional feature vector of the image data is as follows: Acquiring image data and preprocessing the image data, wherein the image data is one or more of ultrasonic images, CT images, and MRI images; Calling a pre-trained convolutional neural network, performing a convolution operation on the pre-processed image data through a set of trained convolution kernels in a convolution layer to extract local feature information, and down-sampling the local feature information through a pooling layer to extract spatial features; The lesion morphology, tissue density and abnormal area are analyzed based on the spatial features, and a high-dimensional feature vector of the image data is generated, wherein the high-dimensional feature vector of the image data can represent the pathological features in the endometrial image.

3. The method for predicting endometrial cancer based on pathological deep learning according to claim 1, characterized in that: The step of obtaining candidate genes associated with endometrial cancer in the gene data, analyzing the candidate genes using PCA to extract mutation site information and variation typing, and generating a gene data feature vector according to the mutation site information and variation typing is specifically as follows: Obtaining candidate genes associated with endometrial cancer in the gene data, and performing standardization processing on the candidate genes; Mapping the standardized candidate genes from high-dimensional features to a low-dimensional space through PCA, and extracting mutation site information from each candidate gene, wherein the mutation site information includes mutation position and mutation type; The gene is typed based on the mutation type and a typing result is generated, and the mutation site information and the typing result are converted into a gene data feature vector, wherein the gene data feature vector includes a binary feature, a mutation frequency feature, a mutation type ratio, a function prediction score, and a variation typing code.

4. The method for predicting endometrial cancer based on pathological deep learning according to claim 1, characterized in that: The expression of the fusion feature is: in, , , is the weight parameter, , , They are high-dimensional feature vectors of image data, text embedding vectors, and feature vectors of genetic data. , , are the number of features of each data modality; The expression of multi-layer perceptron is: in, is the predicted probability of the lesion, is the Sigmoid activation function, and are the weights and biases of the multilayer perceptron.

5. A prediction device for endometrial cancer based on pathological deep learning, characterized in that: include: An image data processing unit, used to obtain image data of the endometrium, and use a convolutional neural network to extract spatial features of the image data of the endometrium, and analyze the spatial features in terms of lesion morphology, tissue density and abnormal area to generate a high-dimensional feature vector of the image data; A diagnostic text processing unit that obtains historical clinical diagnosis text and uses the Transformer model to extract text embedding vectors associated with endometrial cancer; A gene data processing unit, used to obtain candidate genes associated with endometrial cancer in the gene data, and use PCA to analyze the candidate genes to extract mutation site information and variation typing, and generate a gene data feature vector according to the mutation site information and variation typing; A prediction unit is used to fuse the high-dimensional feature vector of the image data, the text embedding vector, and the feature vector of the gene data to generate a fused feature, and use a pre-trained multi-layer perceptron to predict the fused feature, and output a prediction result of endometrial cancer, wherein the multi-layer perceptron uses a cross entropy loss function and an Adam optimizer to update model parameters during the training process.

6. The prediction device for endometrial cancer based on pathological deep learning according to claim 5, characterized in that: The image data processing unit is specifically used for: Acquiring image data and preprocessing the image data, wherein the image data is one or more of ultrasonic images, CT images, and MRI images; Calling a pre-trained convolutional neural network, performing a convolution operation on the pre-processed image data through a set of trained convolution kernels in a convolution layer to extract local feature information, and down-sampling the local feature information through a pooling layer to extract spatial features; The lesion morphology, tissue density and abnormal area are analyzed based on the spatial features, and a high-dimensional feature vector of the image data is generated, wherein the high-dimensional feature vector of the image data can represent the pathological features in the endometrial image.

7. The prediction device for endometrial cancer based on pathological deep learning according to claim 5, characterized in that: The gene data processing unit is specifically used for: Obtaining candidate genes associated with endometrial cancer in the gene data, and performing standardization processing on the candidate genes; Mapping the standardized candidate genes from high-dimensional features to a low-dimensional space through PCA, and extracting mutation site information from each candidate gene, wherein the mutation site information includes mutation position and mutation type; The gene is typed based on the mutation type and a typing result is generated, and the mutation site information and the typing result are converted into a gene data feature vector, wherein the gene data feature vector includes a binary feature, a mutation frequency feature, a mutation type ratio, a function prediction score, and a variation typing code.

8. The prediction device for endometrial cancer based on pathological deep learning according to claim 5, characterized in that: The expression of the fusion feature is: in, , , is the weight parameter, , , They are high-dimensional feature vectors of image data, text embedding vectors, and feature vectors of genetic data. , , are the number of features of each data modality; The expression of multi-layer perceptron is: in, is the predicted probability of the lesion, is the Sigmoid activation function, and are the weights and biases of the multilayer perceptron.

9. A prediction device for endometrial cancer based on pathological deep learning, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a method for predicting endometrial cancer based on pathological deep learning as described in any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and the computer program can be executed by a processor of the device where the computer-readable storage medium is located to implement a method for predicting endometrial cancer based on pathological deep learning as described in any one of claims 1 to 4.

Images