A system and method for processing rare disease consultation data

Abstract

The present application provides a system and method for processing rare disease consultation data, which relate to the field of medical technology. The processing system includes a rare disease analysis terminal and multiple medical department clients. The rare disease analysis terminal is used to receive a rare disease MDT collaborative analysis request sent by any target medical department client; extract a semantic feature vector corresponding to the disease prediction result of the subject expert, and extract an image feature vector based on the medical imaging image of the target patient; input the semantic feature vector and the image feature vector into a pre-trained disease classifier to obtain a comprehensive disease prediction result output by the disease classifier; enable the medical department client corresponding to the comprehensive disease prediction result to determine the disease probability of the comprehensive disease prediction result based on all medical test data and medical imaging images; and determine the final disease prediction result of the target patient based on the disease probability and the comprehensive disease prediction result output by the disease classifier.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular, to a system and method for processing rare disease consultation data. Background Art

[0002] According to the World Health Organization, rare diseases are those that affect 0.65‰ to 1‰ of the total population. Many rare diseases are caused by genetics and are usually chronic, progressive, degenerative and life-threatening.

[0003] Unlike common diseases, the diagnosis and treatment of rare diseases often requires collaborative diagnosis between multiple different departments. For example, in the prior art, there are already a variety of remote MDT consultation systems used in the diagnosis and treatment of rare diseases, allowing doctors from multiple different medical institutions and disciplines to jointly diagnose and treat patients with rare diseases, but this also leads to disagreements. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a system and method for processing rare disease consultation data, which is used to integrate the opinions of experts from multiple disciplines and the medical data of patients through deep learning, and further analyze the types of diseases suffered by patients with rare diseases to assist doctors in diagnosis.

[0005] In a first aspect, the present application provides a system for processing rare disease consultation data, the processing system comprising a rare disease analysis end and multiple medical department clients, the rare disease analysis end being communicatively connected to the multiple medical department clients; the rare disease analysis end being used to receive a rare disease MDT collaborative analysis request sent by any target medical department client, the MDT collaborative analysis request comprising medical imaging images of a target patient, medical test data, and subject expert disease prediction results corresponding to subject experts of multiple medical department clients; for the subject expert disease prediction results of each subject expert obtained for the target patient, the rare disease analysis end extracts a semantic feature vector corresponding to the subject expert disease prediction result, and extracts a picture feature vector based on the medical imaging images of the target patient; the rare disease analysis end inputs the semantic feature vector and the picture feature vector into A pre-trained disease classifier is used to obtain the comprehensive disease prediction results output by the disease classifier; based on the comprehensive disease prediction results, the rare disease analysis end obtains other medical test data and medical imaging pictures of the target patient required to confirm the comprehensive disease prediction results; the rare disease analysis end sends all the medical test data and medical imaging pictures of the target patient to the medical department client corresponding to the comprehensive disease prediction results, so that the medical department client corresponding to the comprehensive disease prediction results determines the disease probability of the comprehensive disease prediction results based on all the medical test data and medical imaging pictures; the rare disease analysis end receives the disease probability of the comprehensive disease prediction results sent by the medical department client corresponding to the comprehensive disease prediction results, and determines the final disease prediction results of the target patient according to the disease probability and the comprehensive disease prediction results output by the disease classifier.

[0006] Preferably, for each subject expert corresponding to the medical department client, the rare disease analysis terminal obtains the semantic feature vector corresponding to all subject expert disease prediction results issued by the subject expert for the current treatment stage of the target patient in the following manner:

[0007] The disease prediction results of each subject expert issued by the subject expert for the current treatment stage of the target patient are input into the embedding model in sequence to obtain the embedding vector corresponding to each subject expert disease prediction result output by the embedding model; all embedding vectors are input into the bidirectional long short-term memory network in sequence to obtain the semantic feature vector output by the bidirectional long short-term memory network.

[0008] Preferably, the rare disease analysis terminal extracts the image feature vector of the medical image in the following manner:

[0009] The medical image is transformed into grayscale to obtain an enhanced medical image; the enhanced medical image is input into a convolutional neural network to obtain an image feature vector output by the convolutional neural network.

[0010] Preferably, the rare disease analysis terminal obtains the comprehensive disease prediction results output by the disease classifier in the following manner:

[0011] Based on the semantic feature vector and the image feature vector, a comprehensive feature vector is fused and generated; the comprehensive feature vector is input into a pre-trained disease classifier to obtain the comprehensive disease prediction result output by the rare disease classification model.

[0012] Preferably, the rare disease analysis terminal generates a comprehensive feature vector by fusing in the following manner:

[0013] The semantic feature vector and the image feature vector are concatenated to generate a comprehensive feature vector; or the semantic feature vector and the image feature vector are weightedly fused to generate a comprehensive feature vector.

[0014] Preferably, the disease classifier is generated by training in the following manner: obtaining multiple training samples in a sample training set, wherein each training sample includes a semantic feature vector corresponding to the predicted disease type of the sample patient by experts in different disciplines and a picture feature vector of a preset medical imaging picture of the patient; for each training sample, the semantic feature vector and the picture feature vector in the training sample are annotated with positive labels and negative labels; the comprehensive feature vectors corresponding to the annotated training samples are sequentially input into the original classifier so that the original classifier outputs the confidence corresponding to the disease type; the loss value of the original classifier is calculated based on the loss function to optimize the hyperparameters of the original classifier to obtain the disease classifier, wherein the loss function is:

[0015] ;

[0016] in, , , , is the weight coefficient, , is a positive value, is the difference between the actual disease type and the predicted result of the classifier for the positive label disease type, is the difference between the actual disease type and the predicted result of the classifier for the negative label disease type, For the The prediction results of the disease types corresponding to the positive label training samples, is the number of positive label training samples, For the The prediction results of the disease types corresponding to the negative label training samples, is the number of negative label training samples.

[0017] Preferably, the rare disease analysis end obtains the embedding vector corresponding to each disease prediction result of the subject expert output by the embedding model in the following manner:

[0018] Each subject expert's disease prediction result issued by the subject expert for the current treatment stage of the target patient is input into the embedding model in turn to obtain the word embedding vector and character embedding vector corresponding to the subject expert's disease prediction result; for each subject expert's disease prediction result, the word embedding vector and character embedding vector corresponding to the subject expert's disease prediction result are input into the bidirectional long short-term memory network to obtain the semantic feature vector corresponding to the subject expert's disease prediction result output by the bidirectional long short-term memory network.

[0019] Preferably, the comprehensive disease prediction result includes at least one rare disease entity, and the medical department client determines the disease probability of the comprehensive disease prediction result in the following manner:

[0020] Construct a medical data knowledge graph of the target patient based on all medical test data and medical imaging images; determine the similarity between the medical data knowledge graph and the standard knowledge graph corresponding to each rare disease entity indicated by the comprehensive disease prediction result as the disease probability recorded in the comprehensive disease prediction.

[0021] Preferably, the rare disease analysis end determines the final disease prediction result of the target patient in the following manner: all rare disease entity types are arranged in order of probability of disease from large to small, so as to determine at least one rare disease entity as the final disease prediction result.

[0022] In a second aspect, the present application provides a method for processing rare disease consultation data, the method comprising:

[0023] Receive a rare disease MDT collaborative analysis request sent by any target medical department client, the MDT collaborative analysis request includes the medical imaging pictures of the target patient, medical test data, and the subject expert disease prediction results corresponding to the subject experts of multiple medical department clients; extract the semantic feature vector corresponding to the subject expert disease prediction result of each subject expert obtained for the target patient, and extract the image feature vector based on the medical imaging pictures of the target patient; input the semantic feature vector and the image feature vector into a pre-trained disease classifier to obtain the comprehensive disease prediction result output by the disease classifier; based on the comprehensive Disease prediction results, obtain other medical test data and medical imaging pictures of the target patient required for confirming the comprehensive disease prediction results; send all medical test data and medical imaging pictures of the target patient to the medical department client corresponding to the comprehensive disease prediction results, so that the medical department client corresponding to the comprehensive disease prediction results determines the disease probability of the comprehensive disease prediction results based on all medical test data and medical imaging pictures; receive the disease probability of the comprehensive disease prediction results sent by the medical department client corresponding to the comprehensive disease prediction results, and determine the final disease prediction results of the target patient according to the disease probability and the comprehensive disease prediction results output by the disease classifier.

[0024] The present application provides a system and method for processing rare disease consultation data, wherein the processing system includes a rare disease analysis terminal and multiple medical department clients, and the rare disease analysis terminal is communicatively connected with the multiple medical department clients; the rare disease analysis terminal is used to receive a rare disease MDT collaborative analysis request sent by any target medical department client, and the MDT collaborative analysis request includes medical imaging pictures of the target patient, medical test data, and subject expert disease prediction results corresponding to subject experts of multiple medical department clients; for the subject expert disease prediction results of each subject expert obtained for the target patient, the rare disease analysis terminal extracts the semantic feature vector corresponding to the subject expert disease prediction result, and extracts the image feature vector based on the medical imaging pictures of the target patient; the rare disease analysis terminal inputs the semantic feature vector and the image feature vector into Pre-trained disease classifiers are used to obtain the comprehensive disease prediction results output by the disease classifier; the rare disease analysis end obtains other medical test data and medical imaging images of the target patient required to confirm the comprehensive disease prediction results based on the comprehensive disease prediction results; the rare disease analysis end sends all medical test data and medical imaging images of the target patient to the medical department client corresponding to the comprehensive disease prediction results, so that the medical department client corresponding to the comprehensive disease prediction results determines the probability of the comprehensive disease prediction results based on all medical test data and medical imaging images; the rare disease analysis end receives the probability of the comprehensive disease prediction results sent by the medical department client corresponding to the comprehensive disease prediction results, and determines the final disease prediction results of the target patient based on the probability of the disease and the comprehensive disease prediction results output by the disease classifier. Through deep learning, the opinions of experts from multiple disciplines and the medical data of patients are integrated and learned, and the diseases of rare disease patients are further analyzed to assist doctors in diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 A schematic diagram of the structure of a system for processing rare disease consultation data provided in an embodiment of the present application;

[0027] Figure 2 A communication flow chart of a rare disease consultation data processing system provided in an embodiment of the present application;

[0028] Figure 3 A schematic diagram of the partial structure of a knowledge graph provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] First, the application scenario of this application is explained. The technical solution of this application can be applied to the integration of disease prediction results of doctors from multiple different medical institutions and disciplines after the MDT consultation of rare disease patients, and combined with the medical data of rare disease patients, to analyze the rare diseases that rare disease patients are most likely to suffer from, and then assist in completing the final diagnosis and treatment judgment for rare disease patients.

[0030] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0031] Figure 1 A schematic diagram of the structure of a system for processing rare disease consultation data provided in an embodiment of the present application. Figure 2 A communication flow chart of a system for processing rare disease consultation data provided in an embodiment of the present application.

[0032] like Figure 1 , Figure 2 As shown, an embodiment of the present application provides a system for processing rare disease consultation data, the processing system includes a rare disease analysis terminal and multiple medical department clients, and the rare disease analysis terminal is communicatively connected with the multiple medical department clients.

[0033] Among them, the medical department client can be a client deployed in different medical institutions and different medical departments, and there is no limit on the number of medical department clients. Figure 1 The medical department B1 and the medical department C1 in can be the same medical department or different medical departments.

[0034] The processing system analyzes rare diseases through the following steps:

[0035] S1. The rare disease analysis terminal receives a rare disease MDT collaborative analysis request sent by any target medical department client. The MDT collaborative analysis request includes medical imaging images of the target patient, medical test data, and disease prediction results of subject experts corresponding to subject experts of multiple medical department clients.

[0036] Here, when the subject experts corresponding to the target medical department client disagree with the diagnosis results of the rare disease of the target patient, or there are different opinions among the subject experts corresponding to multiple medical department clients, the target medical department client can initiate a rare disease MDT collaborative analysis request. The rare disease MDT collaborative analysis request may include medical images taken during the historical diagnosis and treatment of the target patient, medical test data, and the disease prediction results of the subject experts corresponding to the subject experts of multiple medical department clients.

[0037] For example, the medical image may be an X-ray image, a CT scan image, a positron emission tomography scan image, an ultrasound image, a magnetic resonance imaging image, an electrocardiogram, an electroencephalogram, etc., or may be an ordinary image of a patient, but the imaging content of the medical image may indicate disease information such as a disease phenotype or characteristic of the target patient, such as an ordinary image of the hand or face of a patient with acromegaly or an electrocardiogram of a patient with Smith-Lemli-Opitz syndrome, etc.

[0038] The medical test data here may be test data such as blood test, urine test or DNA test data of the target patient.

[0039] The disease prediction results of subject experts are the disease predictions for the target patient after the MDT consultation of each subject expert. For example, subject expert A believes that the target patient has Alzheimer's disease or Niemann-Pick disease type C (NPC), and subject expert B believes that the target patient has Niemann-Pick disease type C (NPC) or Sanfilippo syndrome. Subject expert C believes that the target patient does not have a rare disease but common epilepsy (chronic brain disease).

[0040] S2. Based on the disease prediction results of each subject expert obtained for the target patient, the rare disease analysis end extracts the semantic feature vector corresponding to the disease prediction results of the subject expert, and extracts the image feature vector based on the medical imaging images of the target patient.

[0041] Here, the rare disease analysis end can use deep neural networks to extract semantic feature vectors corresponding to the disease prediction results of subject experts and image feature vectors corresponding to medical images.

[0042] S3. The rare disease analysis end inputs the semantic feature vector and the image feature vector into a pre-trained disease classifier to obtain the comprehensive disease prediction results output by the disease classifier.

[0043] The disease classifier here can generate a comprehensive feature vector based on the semantic feature vector and the image feature vector. The comprehensive feature vector is input into the pre-trained disease classifier to obtain the comprehensive disease prediction results output by the rare disease classification model. By fusing the semantic feature vector of the disease type with the image feature vector, the relationship between the disease type and the specific manifestation of the disease can be better captured, thereby improving the accuracy of the disease classifier results.

[0044] The input of the disease classifier is a comprehensive feature vector that integrates the semantics of the diagnosis results of subject experts and the characteristics of the medical imaging images of the target patients. The output of the disease classifier is the type of disease that the target patient may have and the confidence level of the disease type. The disease type here can be one or more.

[0045] S4. Based on the comprehensive disease prediction results, the rare disease analysis end obtains other medical test data and medical imaging images of the target patients required to confirm the comprehensive disease prediction results.

[0046] After analyzing the comprehensive disease prediction results based on deep learning, medical data of special projects can be collected for target patients in a targeted manner to further help subject experts determine the type of disease the target patients suffer from.

[0047] S5. The rare disease analysis terminal sends all medical test data and medical imaging images of the target patient to the medical department client corresponding to the comprehensive disease prediction result, so that the medical department client corresponding to the comprehensive disease prediction result determines the probability of suffering from the comprehensive disease prediction result based on all medical test data and medical imaging images.

[0048] The rare disease analysis end uses the learning model again to re-predict the prevalence of each comprehensive disease in patients based on more complete and targeted data.

[0049] S6. The rare disease analysis terminal receives the disease probability of the comprehensive disease prediction result sent by the medical department client corresponding to the comprehensive disease prediction result, and determines the final disease prediction result of the target patient based on the disease probability and the comprehensive disease prediction result output by the disease classifier.

[0050] Specifically, the rare disease analysis end determines the final disease prediction result of the target patient in the following ways:

[0051] All rare disease entities are arranged in descending order according to the probability of disease occurrence to determine at least one rare disease entity as the final disease prediction result.

[0052] In this way, subject experts can assist in determining whether the target patient suffers from a rare disease based on one or two rare disease entities with the highest probability of suffering from the disease.

[0053] The present application provides a system for processing rare disease consultation data. It integrates the opinions of experts from multiple disciplines and the medical data of patients through deep learning, and uses a multimodal disease classifier to perform a first round of analysis on the diseases of rare disease patients. After supplementing the medical data, a second round of targeted analysis is performed to obtain the final disease prediction results. The prediction results are more accurate and contribute greatly to the doctor's diagnosis.

[0054] In one embodiment of the present application, for each subject expert corresponding to the medical department client, the rare disease analysis terminal can obtain the semantic feature vector corresponding to all subject expert disease prediction results issued by the subject expert for the current treatment stage of the target patient in the following manner:

[0055] The disease prediction results of each subject expert issued by the subject expert for the current treatment stage of the target patient are input into the embedding model in sequence to obtain the embedding vector corresponding to each disease prediction result of the subject expert output by the embedding model. All embedding vectors are input into the bidirectional long short-term memory network in sequence to obtain the semantic feature vector output by the bidirectional long short-term memory network.

[0056] Furthermore, the rare disease analysis end obtains the embedding vector corresponding to each disease prediction result of the subject expert output by the embedding model in the following way:

[0057] Each disease prediction result issued by the subject expert for the current treatment stage of the target patient is input into the embedding model in sequence to obtain the word embedding vector and character embedding vector corresponding to the disease prediction result of the subject expert. For each disease prediction result of the subject expert, the word embedding vector and character embedding vector corresponding to the disease prediction result of the subject expert are input into the bidirectional long short-term memory network to obtain the semantic feature vector corresponding to the disease prediction result of the subject expert output by the bidirectional long short-term memory network.

[0058] Here, for each disease type indicated by the disease type prediction result of each subject expert, the word embedding vector and character embedding vector are extracted through the word2vec model. For example, if the disease type is "Marfan syndrome", the word embedding vector corresponding to "Marfan syndrome" is extracted through the word2vec model. , and "horse" 、"Fan" , "Comprehensive" ,"combine" ,"disease" The corresponding word embedding vectors.

[0059] Next, for each disease type indicated by the disease type prediction result of each subject expert, the corresponding word embedding vector and character embedding vector are concatenated into a vector and input into the bidirectional long short-term memory network to extract the semantic feature vector corresponding to the disease type prediction result of the subject expert. The bidirectional long short-term memory network here includes a forward long short-term memory network and a reverse long short-term memory network.

[0060] For example, the word embedding vector and the character embedding vector can be concatenated as .

[0061] The concatenated word embedding vectors are passed through the forward LSTM network and the reverse LSTM network to extract two sub-semantic feature vectors:

[0062]

[0063] .

[0064] Then the two sub-semantic feature vectors are fused to generate the final semantic feature vector .

[0065] Next, the rare disease analysis end can extract the image feature vector of the medical image in the following ways:

[0066] The medical image is gray-scaled to obtain an enhanced medical image. The enhanced medical image is input into a convolutional neural network to obtain an image feature vector output by the convolutional neural network.

[0067] The convolutional neural network here can specifically be a VGG residual neural network.

[0068] Before inputting the medical image into the convolutional neural network, the medical image can be preprocessed by denoising, cropping, etc. Then the medical image is grayscaled and the image feature vector is extracted. , the image feature vector extracted in this way can better reflect the texture.

[0069] Here the image feature vector and semantic feature vector The dimensions between can be the same.

[0070] Next, the rare disease analysis end can generate a comprehensive feature vector by fusing in the following ways:

[0071] The semantic feature vector and the image feature vector are concatenated to generate a comprehensive feature vector. Alternatively, the semantic feature vector and the image feature vector are weightedly fused to generate a comprehensive feature vector.

[0072] It should be noted that when weighted fusion is used to generate a comprehensive feature vector, the weight of the image feature vector needs to be greater than the weight of the semantic feature vector. In addition, a fusion neural network can also be used to fuse the semantic feature vector and the image feature vector, which is not limited here.

[0073] The rare disease analysis end inputs the comprehensive feature vector into a pre-trained disease classifier to obtain the comprehensive disease prediction results output by the rare disease classification model.

[0074] The disease classifier here is used to output the confidence level corresponding to each disease type based on the comprehensive feature vector of the disease type prediction results corresponding to all subject experts input.

[0075] For example, if subject expert A believes that the target patient has Alzheimer's disease or Niemann-Pick disease type C (NPC), the first comprehensive feature vector corresponding to Alzheimer's disease and the second comprehensive feature vector corresponding to Niemann-Pick disease type C (NPC) can be obtained respectively. Then, the first comprehensive feature vector, the second comprehensive feature vector, and the Nth comprehensive feature vector corresponding to all subject experts are input into the disease classifier in parallel to obtain the first N diseases indicated by the comprehensive disease prediction result output by the disease classifier, and at least one rare disease is included.

[0076] It should be noted that when the disease type with the highest confidence output by the disease classifier is not a rare disease, it is necessary to search for the next disease type until a rare disease type is found, then stop and determine it as the current comprehensive disease prediction result.

[0077] In a specific embodiment, the disease classifier is generated by training in the following manner:

[0078] Acquire multiple training samples in the sample training set, wherein each training sample includes a semantic feature vector corresponding to the disease predicted by experts in different disciplines for the sample patient and an image feature vector of a medical imaging image of a preset patient;

[0079] For each training sample, the semantic feature vector and image feature vector in the training sample are labeled with positive labels and negative labels;

[0080] The comprehensive feature vectors corresponding to the labeled training samples are sequentially input into the original classifier, so that the original classifier outputs the confidence level corresponding to the disease type;

[0081] The loss value of the original classifier is calculated based on the loss function to optimize the hyperparameters of the original classifier to obtain the disease classifier, wherein the loss function is:

[0082] ;

[0083] in, , , , is the weight coefficient, , is a positive value, is the difference between the actual disease type and the predicted result of the classifier for the positive label disease type, is the difference between the actual disease type and the predicted result of the classifier for the negative label disease type, For the The prediction results of the disease types corresponding to the positive label training samples, is the number of positive label training samples, For the The prediction results of the disease types corresponding to the negative label training samples, is the number of negative label training samples.

[0084] Specifically, training samples are first collected based on the historical diagnosis and treatment data of other patients diagnosed through MDT consultation. Each training sample includes the semantic feature vector corresponding to the predicted disease type of a patient by experts in different disciplines and the image feature vector of the patient's medical imaging image. Among them, the semantic feature vector and the image feature vector are marked with positive and negative labels. The positive label is used to indicate that the disease type corresponding to the semantic feature vector is the actual disease type or the medical imaging image corresponding to the image feature vector is the manifestation of the actual disease type, and the negative label is the opposite.

[0085] For each training sample, the semantic feature vector and the image feature vector are fused to obtain the comprehensive feature vector corresponding to each training sample.

[0086] Each comprehensive feature vector is input into the original classifier so that the original classifier outputs the confidence of each disease type.

[0087] In each round of training, the loss function can be calculated in the following way to optimize the hyperparameters of the original classifier to obtain the disease classifier:

[0088] ;

[0089] in, , , , is the weight coefficient, , is a positive value, is the difference between the actual disease type and the predicted result of the classifier for the positive label disease type, is the difference between the actual disease type and the predicted result of the classifier for the negative label disease type, For the The prediction results of the disease types corresponding to the positive label training samples, is the number of positive label training samples, For the The prediction results of the disease types corresponding to the negative label training samples, is the number of negative label training samples.

[0090] The original classifier here can be built based on algorithms such as logistic regression, support vector machine, decision tree, etc.

[0091] In one embodiment of the present application, after the comprehensive disease prediction result is determined, the comprehensive disease prediction result can be sent to the target medical department client corresponding to the comprehensive disease prediction result, and the medical data of the target patient can be expanded in a targeted manner according to the comprehensive disease prediction result. For example, blood tests and local CT scans can be performed based on the phenotypes and symptoms of different diseases to further distinguish different diseases in the comprehensive disease prediction results.

[0092] The comprehensive disease prediction results include at least one rare disease entity. The medical department client determines the disease probability of the comprehensive disease prediction results in the following ways:

[0093] The medical data knowledge graph of the target patient is constructed based on all medical test data and medical imaging images. The similarity between the medical data knowledge graph and the standard knowledge graph corresponding to each rare disease entity indicated by the comprehensive disease prediction result is determined as the disease probability recorded in the comprehensive disease prediction.

[0094] Specifically, Figure 3 As shown, the medical data knowledge graph here can be set with disease entity, medical image entity, medical data entity, disease clinical manifestation entity, therapeutic drug entity, etc. The entity value of the medical data knowledge graph of the target patient is determined based on the target patient's case, medical test data, medical imaging image and other data.

[0095] The structure of the standard knowledge graph is the same as that of the medical data knowledge graph, but the entity values ​​are extracted and determined based on relevant standard medical literature, open source case databases, rare disease clinical cohorts, and medical textbooks.

[0096] Furthermore, the similarity S between the medical data knowledge graph and the standard knowledge graph corresponding to each rare disease entity indicated by the comprehensive disease prediction result can be calculated in the following way:

[0097] S= ;

[0098] in, Entities in the medical data knowledge graph Entities in standard knowledge graphs The distance value between Entities in the medical data knowledge graph The corresponding weight factor is is the number of entities, .

[0099] By calculating the similarity between the medical data knowledge graph and the standard knowledge graph corresponding to each rare disease entity indicated by the comprehensive disease prediction result, the probability of suffering from the comprehensive disease prediction can be used as the basis for judging whether the target patient suffers from the rare disease based on one or two rare disease entities with the highest probability of suffering from the disease.

[0100] In one embodiment of the present application, a method for processing rare disease consultation data is provided, the method comprising:

[0101] Receive a rare disease MDT collaborative analysis request sent by any target medical department client. The MDT collaborative analysis request includes medical imaging images of the target patient, medical test data, and disease prediction results of subject experts corresponding to subject experts of multiple medical department clients.

[0102] Based on the disease prediction results of each subject expert obtained for the target patient, the semantic feature vector corresponding to the disease prediction result of the subject expert is extracted, and the image feature vector is extracted based on the medical imaging image of the target patient; the semantic feature vector and the image feature vector are input into a pre-trained disease classifier to obtain the comprehensive disease prediction result output by the disease classifier.

[0103] Based on the comprehensive disease prediction results, other medical test data and medical imaging images of the target patients required to confirm the comprehensive disease prediction results are obtained.

[0104] All medical test data and medical imaging images of the target patient are sent to the medical department client corresponding to the comprehensive disease prediction result, so that the medical department client corresponding to the comprehensive disease prediction result determines the probability of disease based on all medical test data and medical imaging images.

[0105] Receive the disease probability of the comprehensive disease prediction results sent by the medical department client corresponding to the comprehensive disease prediction results, and determine the final disease prediction results of the target patient based on the disease probability and the comprehensive disease prediction results output by the disease classifier.

[0106] The principles and technical effects of the method embodiment are similar to those of the system embodiment and will not be described in detail here.

[0107] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0108] In addition, the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0109] Furthermore, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0110] It should be noted that if the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0111] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0112] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A system for processing rare disease consultation data, characterized in that: The processing system includes a rare disease analysis terminal and multiple medical department clients, and the rare disease analysis terminal is communicatively connected with the multiple medical department clients; The rare disease analysis terminal is used to receive a rare disease MDT collaborative analysis request sent by any target medical department client, wherein the MDT collaborative analysis request includes medical imaging pictures of the target patient, medical test data, and disease prediction results of subject experts corresponding to subject experts of multiple medical department clients; The rare disease analysis terminal extracts the semantic feature vector corresponding to the disease prediction result of each subject expert obtained for the target patient, and extracts the image feature vector based on the medical imaging image of the target patient; The rare disease analysis end inputs the semantic feature vector and the image feature vector into a pre-trained disease classifier to obtain a comprehensive disease prediction result output by the disease classifier; The rare disease analysis terminal obtains other medical test data and medical imaging images of the target patient required to confirm the comprehensive disease prediction result based on the comprehensive disease prediction result; The rare disease analysis terminal sends all medical test data and medical imaging pictures of the target patient to the medical department client corresponding to the comprehensive disease prediction result, so that the medical department client corresponding to the comprehensive disease prediction result determines the disease probability of the comprehensive disease prediction result based on all medical test data and medical imaging pictures; The rare disease analysis terminal receives the disease probability of the comprehensive disease prediction result sent by the medical department client corresponding to the comprehensive disease prediction result, and determines the final disease prediction result of the target patient according to the disease probability and the comprehensive disease prediction result output by the disease classifier; The disease classifier is trained and generated in the following manner: Acquire multiple training samples in the sample training set, wherein each training sample includes a semantic feature vector corresponding to the disease predicted by experts in different disciplines for the sample patient and an image feature vector of a medical imaging image of a preset patient; For each training sample, the semantic feature vector and image feature vector in the training sample are labeled with positive labels and negative labels; The comprehensive feature vectors corresponding to the labeled training samples are sequentially input into the original classifier, so that the original classifier outputs the confidence level corresponding to the disease type; The loss value of the original classifier is calculated based on the loss function to optimize the hyperparameters of the original classifier and obtain the disease classifier.

2. The system according to claim 1, characterized in that For the subject experts corresponding to each medical department client, the rare disease analysis terminal obtains the semantic feature vectors corresponding to all subject expert disease prediction results issued by the subject expert for the current treatment stage of the target patient in the following way: Will The subject expert sequentially inputs each subject expert disease prediction result issued by the subject expert for the current treatment stage of the target patient into the embedding model to obtain an embedding vector corresponding to each subject expert disease prediction result output by the embedding model; All embedding vectors are input into a bidirectional long short-term memory network in a sequence form to obtain the semantic feature vector output by the bidirectional long short-term memory network.

3. The system according to claim 2, characterized in that The rare disease analysis end extracts image feature vectors of medical images in the following ways: Performing grayscale transformation on the medical image to obtain an enhanced medical image; The enhanced medical image is input into a convolutional neural network to obtain an image feature vector output by the convolutional neural network.

4. The system according to claim 1, characterized in that The rare disease analysis terminal obtains the comprehensive disease prediction result output by the disease classifier in the following manner: Based on the semantic feature vector and the picture feature vector, a comprehensive feature vector is generated by fusing; The comprehensive feature vector is input into a pre-trained disease classifier to obtain a comprehensive disease prediction result output by the disease classifier.

5. The system according to claim 4, characterized in that The rare disease analysis terminal generates the comprehensive feature vector by fusing in the following manner: Concatenate the semantic feature vector and the image feature vector to generate a comprehensive feature vector; or The semantic feature vector and the image feature vector are weightedly fused to generate a comprehensive feature vector.

6. The system according to claim 1, characterized in that The loss function is: ； in, , , , is the weight coefficient, , is a positive value, is the difference between the actual disease type and the predicted result of the classifier for the positive label disease type, is the difference between the actual disease type and the predicted result of the classifier for the negative label disease type, For the The prediction results of the disease types corresponding to the positive label training samples, is the number of positive label training samples, For the The prediction results of the disease types corresponding to the negative label training samples, is the number of negative label training samples.

7. The system according to claim 2, characterized in that The rare disease analysis end obtains the embedding vector corresponding to each disease prediction result of the subject expert output by the embedding model in the following manner: Input each subject expert's disease prediction result issued by the subject expert for the current treatment stage of the target patient into the embedding model in sequence to obtain the word embedding vector and character embedding vector corresponding to the subject expert's disease prediction result; For each disease prediction result of a subject expert, the word embedding vector and character embedding vector corresponding to the disease prediction result of the subject expert are input into the bidirectional long short-term memory network to obtain the semantic feature vector corresponding to the disease prediction result of the subject expert output by the bidirectional long short-term memory network.

8. The system according to claim 1, characterized in that The comprehensive disease prediction result includes at least one rare disease entity, and the medical department client determines the disease probability of the comprehensive disease prediction result in the following manner: Build a medical data knowledge graph of the target patient based on all medical test data and medical imaging images; Determine the similarity between the medical data knowledge graph and the standard knowledge graph corresponding to each rare disease entity indicated by the comprehensive disease prediction result as the probability of disease occurrence for the comprehensive disease prediction.

9. The system according to claim 8, characterized in that The rare disease analysis end determines the final disease prediction result of the target patient in the following manner: All rare disease entities are arranged in descending order according to the probability of disease occurrence, so as to determine at least one rare disease entity as the final disease prediction result.

10. A method for processing rare disease consultation data, characterized in that: The method comprises: Receive a rare disease MDT collaborative analysis request sent by any target medical department client, wherein the MDT collaborative analysis request includes medical imaging images of the target patient, medical test data, and disease prediction results of subject experts corresponding to subject experts of multiple medical department clients; Extract the semantic feature vector corresponding to the disease prediction result of each subject expert obtained for the target patient, and extract the image feature vector based on the medical imaging image of the target patient; Inputting the semantic feature vector and the image feature vector into a pre-trained disease classifier to obtain a comprehensive disease prediction result output by the disease classifier; Based on the comprehensive disease prediction result, obtaining other medical test data and medical imaging images of the target patient required to confirm the comprehensive disease prediction result; Sending all medical test data and medical imaging pictures of the target patient to the medical department client corresponding to the comprehensive disease prediction result, so that the medical department client corresponding to the comprehensive disease prediction result determines the disease probability of the comprehensive disease prediction result based on all medical test data and medical imaging pictures; Receiving the disease probability of the comprehensive disease prediction result sent by the medical department client corresponding to the comprehensive disease prediction result, and determining the final disease prediction result of the target patient according to the disease probability and the comprehensive disease prediction result output by the disease classifier; The disease classifier is trained and generated in the following manner: Acquire multiple training samples in the sample training set, wherein each training sample includes a semantic feature vector corresponding to the disease predicted by experts in different disciplines for the sample patient and an image feature vector of a medical imaging image of a preset patient; For each training sample, the semantic feature vector and image feature vector in the training sample are labeled with positive labels and negative labels; The comprehensive feature vectors corresponding to the labeled training samples are sequentially input into the original classifier, so that the original classifier outputs the confidence level corresponding to the disease type; The loss value of the original classifier is calculated based on the loss function to optimize the hyperparameters of the original classifier and obtain the disease classifier.

Images