WO2022110719A1

WO2022110719A1 - Federated transfer learning-based neurodegenerative disease model building device, and related apparatus

Info

Publication number: WO2022110719A1
Application number: PCT/CN2021/096645
Authority: WO
Inventors: 李泽远; 王健宗
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-11-24
Filing date: 2021-05-28
Publication date: 2022-06-02
Also published as: CN112447298A

Abstract

A federated transfer learning-based neurodegenerative disease model building device and a related apparatus, pertaining to the technical field of artificial intelligence, and applicable to smart hospital systems. The device comprises: a data set acquisition unit (101) used to acquire a medical history data set A and a medical history data set B of type-I neurodegenerative disease patients; an intersection selection unit (102) used to select intersection data between the medical history data set A and the medical history data set B, and use the intersection data as training set samples; a model training unit (103) used to train an original neural network model on the basis of training set samples; and a federated transfer unit (104) used to perform federated transfer learning on the original neural network model on the basis of training set samples of multiple types of neurodegenerative diseases, so as to obtain a final neural network model of predicting progression of neurodegenerative disease patients. The invention solves the problem of isolated data islands in the model building process, and achieves effective prediction of disease progression of patients.

Description

Neurodegenerative disease modeling device and related equipment based on federated transfer learning

This application claims the priority of the Chinese patent application filed on November 24, 2020 with the application number 202011329754.3 and the invention titled "Neurodegenerative Disease Modeling Apparatus and Related Equipment Based on Federated Transfer Learning", all of which The contents are incorporated herein by reference.

technical field

The present application relates to the field of artificial intelligence technology and the field of digital medicine, specifically to the field of AI + medical care, and in particular to a neurodegenerative disease modeling device and related equipment based on federated transfer learning.

Background technique

Neurodegenerative diseases are caused by the loss of function of neurons and their myelin sheaths and can worsen over time with various dysfunctions. With the increasing number of aging population in my country, more and more elderly people in our country suffer from neurodegenerative diseases. With the continuous development of the level of medical research in various countries in the world, the research methods of neurodegenerative diseases in various medical institutions are becoming more and more perfect. The research on neurodegenerative diseases in the industry now focuses on the research on the causes of different diseases, and there are also many new ideas for treatment. Current research results suggest that neurodegenerative diseases are often caused by oxidative stress, mitochondrial dysfunction, excitotoxins, and immune inflammation. The condition is diverse and includes aspects such as cognitive and behavioral disorders. At the same time, the pathological changes are irreversible, and the course of the disease is often in the middle and late stages when patients have cognitive impairment. At this time, treatment can only delay the development of the disease, but cannot fundamentally reverse the damage to the neural network. Therefore, neurological diseases should be diagnosed and treated as soon as possible.

In terms of pathological research, the industry focuses more on the pathological research of a single disease, because the patient's test information is private information, which is inconvenient to directly obtain and use; at the same time, the uneven regional distribution of patients with neurodegenerative diseases leads to Data collection is inconvenient. In addition, in terms of disease treatment, most doctors can only encounter a few patients in their years of practice, which makes it impossible for scholars to conduct detailed and complete research on the specific etiology of neurodegenerative diseases through a large amount of disease data, which is not conducive to symptomatic diagnosis and treatment. . Various medical institutions have put forward many ideas in the smart medical scenario, but the inventor realized that due to the insufficient amount of data available for research in each institution, the investment in medical data preprocessing and labeling is huge, and the intelligent diagnosis system in the current medical scenario is still limited. It is not perfect, and many assumptions about the prediction of the prevalence of neurodegenerative diseases have not yet been implemented.

Application content

The purpose of this application is to provide a neurodegenerative disease modeling device and related equipment based on federated transfer learning, aiming to solve the problem that the existing neurodegenerative disease disease degree prediction device cannot effectively model under the premise of ensuring privacy.

In a first aspect, the embodiments of the present application provide a neurodegenerative disease modeling device based on federated transfer learning, including:

A data set acquisition unit, configured to acquire a medical record data set A and a medical record data set B of the first type of neurodegenerative disease patient, the medical record data set A includes a plurality of pieces of data described by the first feature item set, the medical record data Set B includes multiple pieces of data described by the second feature item set; the first feature item set includes the feature items used to describe the patient's identity and the feature item used to describe the patient's disease state; the second feature item set includes the feature items used to describe Characteristics of the patient's identity and characteristics used to describe the patient's clinical presentation data;

An intersection selection unit, for selecting intersection data from between the case data set A and the case data set B, using the intersection data as a training set sample;

a model training unit for training the original neural network model based on the training set samples;

The federated transfer unit is used to perform federated transfer learning on the original neural network model based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, and obtain a result for predicting the incidence of neurodegenerative diseases. The final neural network model of disease severity.

In a second aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program The following steps are implemented at the time of obtaining the first type of neurodegenerative disease patient medical record data set A and medical record data set B, the medical record data set A includes a plurality of pieces of data described by the first feature item set, and the medical record data set B It includes multiple pieces of data described by the second feature item set; the first feature item set includes the feature items used to describe the patient's identity and the feature item used to describe the patient's disease state; the second feature item set includes the feature items used to describe the patient's identity The characteristic item of and the characteristic item used to describe the clinical performance data of the patient; Select the intersection data from the case data set A and the case data set B, and use the intersection data as the training set sample; Based on the training set sample, train the original neural network model; based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative disease, perform federated transfer learning on the original neural network model, and obtain a model for predicting neurodegenerative diseases The final neural network model of disease prevalence.

In a third aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when executed by a processor, the computer program causes the processor to perform the following steps: Obtain a medical record data set A and a medical record data set B of a patient with a first type of neurodegenerative disease, the medical record data set A includes a plurality of pieces of data described by the first feature item set, and the medical record data set B includes a second feature set B Multiple pieces of data described by the item set; the first feature item set includes the feature item used to describe the patient's identity and the feature item used to describe the patient's disease state; the second feature item set includes the feature item used to describe the patient's identity and the The feature items used to describe the clinical performance data of the patient; select the intersection data from the case data set A and the case data set B, and use the intersection data as a training set sample; based on the training set sample, the original neural network model Carry out training; based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, perform federated transfer learning on the original neural network model, and obtain a model for predicting the prevalence of neurodegenerative diseases. The final neural network model.

The embodiments of the present application provide a neurodegenerative disease modeling device and related equipment based on federated transfer learning. The device includes: a data set acquisition unit configured to acquire a medical record data set A and a medical record data set of the first type of neurodegenerative disease patients B; an intersection selection unit for selecting intersection data from the case data set A and the case data set B, and using the intersection data as a training set sample; a model training unit for selecting the intersection data based on the training set sample The original neural network model is trained; the federated transfer unit is used to perform federated transfer learning on the original neural network model based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, and obtain the The ultimate neural network model for predicting the prevalence of neurodegenerative diseases. The embodiments of the present application solve the problem of isolated data islands in the modeling process of the predictive model, and can effectively predict the degree of neurodegenerative diseases of patients.

Description of drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. For those of ordinary skill, other drawings can also be obtained from these drawings without any creative effort.

FIG. 1 is a schematic block diagram of a neurodegenerative disease modeling apparatus based on federated transfer learning provided by an embodiment of the present application;

FIG. 2 is a schematic block diagram of subunits of the apparatus for modeling neurodegenerative diseases based on federated transfer learning according to an embodiment of the present application;

FIG. 3 is a schematic block diagram of another subunit of the apparatus for modeling neurodegenerative diseases based on federated transfer learning according to an embodiment of the present application;

FIG. 4 is a schematic block diagram of another subunit of the apparatus for modeling neurodegenerative diseases based on federated transfer learning provided in an embodiment of the present application;

FIG. 5 is a schematic block diagram of another subunit of the apparatus for modeling neurodegenerative diseases based on federated transfer learning according to an embodiment of the present application;

FIG. 6 is a schematic block diagram of another subunit of the apparatus for modeling neurodegenerative diseases based on federated transfer learning according to an embodiment of the present application;

FIG. 7 is a schematic block diagram of a computer device provided by an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

It is to be understood that, when used in this specification and the appended claims, the terms "comprising" and "comprising" indicate the presence of the described features, integers, steps, operations, elements and/or components, but do not exclude one or The presence or addition of a number of other features, integers, steps, operations, elements, components, and/or sets thereof.

It should also be understood that the terminology used in the specification of the application herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural unless the context clearly dictates otherwise.

It should also be further understood that, as used in this specification and the appended claims, the term "and/or" refers to and including any and all possible combinations of one or more of the associated listed items .

Please refer to FIG. 1 . FIG. 1 is a schematic block diagram of a neurodegenerative disease modeling apparatus 100 based on federated transfer learning provided by an embodiment of the present application. The apparatus includes a data set acquisition unit 101 , an intersection selection unit 102 , and a model training unit 103 and Federal Migration Unit 104:

A data set obtaining unit 101 is configured to obtain a medical record data set A and a medical record data set B of a first-type neurodegenerative disease patient, where the medical record data set A includes a plurality of pieces of data described by a first feature item set, and the medical record The data set B includes a plurality of pieces of data described by the second feature item set; the first feature item set includes the feature items used to describe the patient's identity and the feature items used to describe the patient's disease state; the second feature item set includes Features that describe the patient's identity and features that describe the patient's clinical presentation data;

The intersection selection unit 102 is used to select intersection data from between the case data set A and the case data set B, and the intersection data is used as a training set sample;

A model training unit 103, configured to train the original neural network model based on the training set samples;

The federated transfer unit 104 is configured to perform federated transfer learning on the original neural network model based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, to obtain a model for predicting neurodegenerative diseases The final neural network model of disease severity.

In this embodiment, in the data set obtaining unit 101, both the medical record data set A and the medical record data set B include multiple pieces of data, and the data types of the medical record data set A and the medical record data set B overlap. There are also parts that do not intersect. Specifically, the medical record data set A includes multiple pieces of data described by a first feature item set, and the first feature item set includes feature items used to describe the patient's identity (for example, Zhang San, male, 65 years old, etc.) Items that describe the patient's disease state (eg, 20% Parkinson's disease). The medical record data set B includes multiple pieces of data described by the second feature item set, wherein the second feature item set includes the feature items used to describe the patient's identity (which are the same as the feature items corresponding to the first feature item set, such as Zhang. Three, male, 65 years old, etc.) and characteristic terms used to describe patient clinical presentation data (eg, less sleep, decreased mobility, intermittent mania, etc.).

The first group of neurodegenerative diseases may be any one of neurodegenerative diseases, such as Parkinson's disease, obsessive lateral sclerosis or Huntington's syndrome. In the embodiment of the present application, Parkinson's syndrome is preferably used, because the first type of neurodegenerative disease needs to be used for modeling and training of the original model, so some basic data volume is required, so the use of Parkinson's syndrome is beneficial to the original model. training.

In the intersection selection unit 102, intersection data is selected from the case data set A and the case data set B. Due to different data sources, there may be missing or incomplete data. The intersection data is the data of a patient recorded in the case that records both the clinical manifestations and the disease state.

In this embodiment of the present application, the intersection data of the medical record data set A and the medical record data set B can be taken as the training set sample, so that the model can maximize the learning of the common features of the medical record data set A and the medical record data set B, and also avoid sample differentiation. impact on the model.

In the model training unit 103, the original neural network model is trained by using the training set samples obtained above. This unit trains the original neural network model first, and then generalizes the trained original neural network model to predict the prevalence of other neurodegenerative diseases in subsequent units.

In one embodiment, as shown in FIG. 2 , the model training unit 103 includes:

The data input unit 201 is used to use the data of the characteristic item for describing the patient's diseased state in the training set sample as the label of the original neural network model; and the data used to describe the characteristic item of the patient's clinical performance data data, as the input data of the original neural network model;

The model training subunit 202 is configured to train the original neural network model based on the input data to obtain the trained original neural network model.

In this embodiment, the data used to describe the characteristic item of the patient's disease state is used as the label of the original neural network model, and the data used to describe the characteristic item of the patient's clinical performance data is used as the original neural network model. The input data of the network model can make a corresponding relationship between the patient's performance and the disease state. The original neural network model predicts the patient's disease degree based on the patient's behavior and its characteristics. In the model training subunit 202, the original neural network model can be trained by means of federated learning, thereby obtaining the original neural network model for predicting the degree of the first type of neurodegenerative disease.

In one embodiment, the data input unit 201 includes:

The encryption unit is used for encrypting the data in the training set samples by using the homomorphic encryption technology.

That is, during training, homomorphic encryption technology can be used to encrypt sample data to protect data privacy. Then, the encrypted data of the feature items used to describe the patient's disease state is used as the label of the original neural network model; and the encrypted data of the feature items used to describe the patient's clinical performance data is used as the input of the original neural network model data.

In one embodiment, as shown in FIG. 3 , the model training unit 103 further includes:

The test set selection unit 301 is used to use the non-intersection data between the case data set A and the case data set B as a test set sample;

The testing unit 302 is configured to test the original neural network model based on the test set samples.

In this embodiment, since the condition data of neurodegenerative diseases is relatively small, in order to make full use of the existing data, the non-intersection data between the case data set A and the case data set B is also obtained in the embodiment of the present application, and the The non-intersection data is used as a test set sample, and the trained original neural network model is tested by using the test set sample, so that the prediction effect of the original neural network model is more accurate.

In one embodiment, as shown in FIG. 4 , the federated migration unit 104 includes:

A model obtaining unit 401, configured to obtain the trained original neural network model;

The federated transfer learning unit 402 is configured to perform federated transfer learning on the original neural network model using the training set samples of various types of neurodegenerative diseases stored in each node;

The gradient updating unit 403 is configured to update the original neural network model by using the gradient values obtained by training of each node to obtain a final neural network model.

In the embodiment of the present application, the trained original neural network model can be obtained first, and then the original neural network model can be generalized to the training of various neurodegenerative diseases.

That is, in the federated transfer learning unit, the training set samples of various neurodegenerative diseases stored in each node are used for federated transfer learning. The three diseases share commonalities in etiology, clinical manifestations, and patient population. For example, patients with the three diseases all have oxidative damage to the nerve tissue and the phenomenon of increased mitochondria in the brain, and some patients have the clinical manifestations of the three diseases at the same time. Federated transfer learning can complete the training of disease detection models under the premise of ensuring patient privacy, and at the same time generalize the trained models to the detection of other neurodegenerative diseases, and finally train a model that can detect the prevalence of neurodegenerative diseases. Model.

Specifically, the node may be the local end of the hospital, and the local end of each hospital has different training set samples. Maybe node A has the training set samples of Parkinson's syndrome, node B has the training set samples of Huntington's syndrome, and node A has the training set samples of Parkinson's syndrome. C has the training set samples of mandatory lateral sclerosis, so the original neural network model can be trained locally on each node, so that the data of each node will not be leaked, and the privacy of the data is protected. Then, the original neural network model is updated by using the gradient values obtained by training each node to obtain the final neural network model.

In one embodiment, as shown in FIG. 5 , the federated transfer learning unit 402 includes:

The weight judgment unit 501 is used to obtain the model weight of each node in the training process of each round of federated transfer learning, and judge whether the model weight of each node is equal;

The weight updating unit 502 is configured to update the model weight of each node if the model weight of each node is not equal, so as to be used in the training process of this round of federated transfer learning.

In this embodiment, in each round of federated transfer learning training process, if the model weights of different nodes are not equal, the model weights of each node need to be updated, so that each node can still be used in the case of differences in samples. have a consistent learning effect.

In one embodiment, the weight update unit 502 includes:

The weight update subunit is used to update the model weight of each node according to the following formula:

ω ^k+1 ←ω ^k +μF _k (ω _k )

ω ^k represents the model weight of the current node in the current training round, ω ^k+1 represents the updated model weight of the current node, μ is the learning rate, F _k represents the training data of the current node accounts for the total training data of all nodes. proportion.

Carry out model training after the weight update according to the above update method, and then continue to judge whether the model weights are equal in the training process of the next round of federated transfer learning, until the model weights are equal, and the training stops. State-of-the-art encryption technology is used to encrypt data to ensure data security.

In one embodiment, as shown in FIG. 6 , the gradient update unit 403 includes:

A weighted averaging unit 601, configured to perform a weighted average of the gradient values obtained by training each node according to the model weight of each node to obtain an average gradient;

The gradient update subunit 602 is configured to update the original neural network model by using the average gradient to obtain a final neural network model.

In this embodiment, for the gradient values obtained after each node is trained, the model weights of the corresponding nodes are used for weighted average to obtain the average gradient, which represents the contribution provided by the sample training of each node. The original neural network model is updated to obtain the final neural network model.

After the final neural network model is obtained, the data to be measured in each medical record data set can be sequentially input into the final neural network model for calculation, and the obtained results are the respective corresponding disease states.

The device provided by the embodiment of the present application solves the problem of data island in the process of modeling the prediction model, and can effectively predict the degree of the patient's neurodegenerative disease.

The above-mentioned neurodegenerative disease modeling apparatus 100 based on federated transfer learning can be implemented in the form of a computer program, and the computer program can be executed on a computer device as shown in FIG. 7 .

Please refer to FIG. 7 , which is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 700 is a server, and the server may be an independent server or a server cluster composed of multiple servers.

Referring to FIG. 7 , the computer device 700 includes a processor 702 , a memory and a network interface 705 connected by a system bus 701 , wherein the memory may include a non-volatile storage medium 703 and an internal memory 704 .

The nonvolatile storage medium 703 can store an operating system 7031 and a computer program 7032 .

The processor 702 is used to provide computing and control capabilities to support the operation of the entire computer device 700 .

The internal memory 704 provides an environment for the execution of the computer program 7032 in the non-volatile storage medium 703 .

The network interface 705 is used for network communication, such as providing transmission of data information. Those skilled in the art can understand that the structure shown in FIG. 7 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device 700 to which the solution of the present application is applied. The specific computer device 700 may include more or fewer components than shown, or combine certain components, or have a different arrangement of components.

Wherein, the processor 702 is configured to run the computer program 7032 stored in the memory, so as to realize the following functions: acquiring the medical record data set A and the medical record data set B of the first type of neurodegenerative disease patients, the medical record data set A It includes multiple pieces of data described by the first feature item set, and the medical record data set B includes multiple pieces of data described by the second feature item set; the first feature item set includes the feature items used to describe the patient's identity and the The characteristic item of the patient's disease state; the second characteristic item set includes the characteristic item used to describe the patient's identity and the characteristic item used to describe the clinical manifestation data of the patient; select the intersection between the case data set A and the case data set B data, the intersection data is used as a training set sample; based on the training set sample, the original neural network model is trained; based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, Federated transfer learning is performed on the original neural network model to obtain a final neural network model for predicting the prevalence of neurodegenerative diseases.

In one embodiment, when performing the step of training the original neural network model based on the training set samples, the processor 702 performs the following operation: using the training set samples for describing the patient's disease state The data of the characteristic item is used as the label of the original neural network model; and the data of the characteristic item used to describe the clinical performance data of the patient is used as the input data of the original neural network model; The neural network model is trained to obtain the trained original neural network model.

In one embodiment, when the processor 702 performs the step of training the original neural network model based on the training set samples, the processor 702 further performs the following operation: converting the data between the case data set A and the case data set B The non-intersecting set of data is used as a test set sample; the original neural network model is tested based on the test set sample.

In one embodiment, the processor 702 performs federated transfer learning on the original neural network model when executing the training set samples based on multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, to obtain In the step of predicting the final neural network model of the degree of neurodegenerative disease, the following operations are also performed: obtaining the trained original neural network model; using the training sets of various types of neurodegenerative diseases stored in each node The sample performs federated transfer learning on the original neural network model; the original neural network model is updated by using the gradient values obtained from the training of each node to obtain the final neural network model.

In one embodiment, when the processor 702 performs the step of performing federated transfer learning on the original neural network model using the training set samples of various types of neurodegenerative diseases stored in each node, the processor 702 performs the following operations: During the training process of the federated transfer learning round, the model weights of each node are obtained to determine whether the model weights of each node are equal; if the model weights of each node are not equal, the model weights of each node are updated for the current round of federation. The training process of transfer learning.

In one embodiment, the processor 702 performs the following operations when performing the step of updating the model weights of each node for use in the training process of the current round of federated transfer learning if the model weights of the nodes are not equal : Update the model weight of each node according to the following formula: ω ^k+1 ←ω ^k +μF _k (ω _k ), ω ^k represents the model weight of the current node in the current training round, ω ^k+1 represents the current node The updated model weight, μ is the learning rate, and F _k represents the proportion of the training data of the current node to the total training data of all nodes.

In one embodiment, the processor 702 performs the following operations when performing the step of updating the original neural network model by using the gradient values obtained from the training of each node to obtain the final neural network model: according to the model weight of each node A weighted average is performed on the gradient values obtained from the training of each node to obtain an average gradient; the original neural network model is updated by using the average gradient to obtain a final neural network model.

In one embodiment, the processor 702 uses the data of the characteristic items in the training set samples to describe the patient's disease state as the label of the original neural network model; and will be used to describe the clinical symptoms of the patient. When the data representing the characteristic item of the data is used as the input data of the original neural network model, the following operations are performed: using homomorphic encryption technology to encrypt the data in the training set samples.

Those skilled in the art can understand that the embodiment of the computer device shown in FIG. 7 does not constitute a limitation on the specific structure of the computer device. In other embodiments, the computer device may include more or less components than those shown in the drawings. Either some components are combined, or different component arrangements. For example, in some embodiments, the computer device may only include a memory and a processor. In such an embodiment, the structures and functions of the memory and the processor are the same as those of the embodiment shown in FIG. 7 , and details are not repeated here.

It should be understood that, in this embodiment of the present application, the processor 702 may be a central processing unit (Central Processing Unit, CPU), and the processor 702 may also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSP), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Wherein, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor or the like.

In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores a computer program, wherein when the computer program is executed by the processor, the following steps are implemented: obtaining a medical record data set A and a medical record data set B of a patient with a first type of neurodegenerative disease, wherein the medical record data set A includes Multiple pieces of data described by the first feature item set, the medical record data set B includes multiple pieces of data described by the second feature item set; the first feature item set includes the feature items used to describe the patient's identity and The characteristic item of the diseased state; the second characteristic item set includes the characteristic item used to describe the patient's identity and the characteristic item used to describe the clinical manifestation data of the patient; select the intersection data from the case data set A and the case data set B , using the intersection data as a training set sample; based on the training set sample, the original neural network model is trained; based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, The original neural network model is subjected to federated transfer learning to obtain a final neural network model for predicting the prevalence of neurodegenerative diseases.

Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. Interchangeability, the above description has generally described the components and steps of each example in terms of function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each particular application, but such implementations should not be considered beyond the scope of this application.

In the several embodiments provided in this application, it should be understood that the disclosed devices and apparatuses may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the units is only logical function division. In actual implementation, there may be other division methods, or units with the same function may be grouped into one Units, such as multiple units or components, may be combined or may be integrated into another system, or some features may be omitted, or not implemented. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may also be electrical, mechanical or other forms of connection.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solutions of the embodiments of the present application.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware, or may be implemented in the form of software functional units.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a storage medium. Based on this understanding, the technical solutions of the present application are essentially or part of contributions to the prior art, or all or part of the technical solutions can be embodied in the form of software products, and the computer software products are stored in a storage medium , including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a removable hard disk, a read-only memory (ROM, Read-Only Memory), a magnetic disk or an optical disk and other media that can store program codes.

The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalents within the technical scope disclosed in the present application. Modifications or substitutions shall be covered by the protection scope of this application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

A neurodegenerative disease modeling device based on federated transfer learning, comprising:

A data set acquisition unit, configured to acquire a medical record data set A and a medical record data set B of the first type of neurodegenerative disease patient, the medical record data set A includes a plurality of pieces of data described by the first feature item set, the medical record data Set B includes multiple pieces of data described by the second feature item set; the first feature item set includes the feature items used to describe the patient's identity and the feature item used to describe the patient's disease state; the second feature item set includes the feature items used to describe Characteristics of the patient's identity and characteristics used to describe the patient's clinical presentation data;

An intersection selection unit, for selecting intersection data from the case data set A and the case data set B, and using the intersection data as a training set sample;

a model training unit for training the original neural network model based on the training set samples;

The federated transfer unit is used to perform federated transfer learning on the original neural network model based on the training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, and obtain a result for predicting the incidence of neurodegenerative diseases. The final neural network model of disease severity.
The neurodegenerative disease modeling apparatus based on federated transfer learning according to claim 1, wherein the model training unit comprises:

A data input unit, used to use the data of the characteristic item for describing the patient's diseased state in the training set samples as the label of the original neural network model; and the data to be used to describe the characteristic item of the patient's clinical performance data , as the input data of the original neural network model;

A model training subunit, configured to train the original neural network model based on the input data to obtain the trained original neural network model.
The neurodegenerative disease modeling apparatus based on federated transfer learning according to claim 1, wherein the model training unit further comprises:

A test set selection unit, used for taking the non-intersection data between the case data set A and the case data set B as a test set sample;

A test unit, configured to test the original neural network model based on the test set samples.
The neurodegenerative disease modeling apparatus based on federated transfer learning according to claim 1, wherein the federated transfer unit comprises:

a model obtaining unit for obtaining the trained original neural network model;

The federated transfer learning unit is used to perform federated transfer learning on the original neural network model by using the training set samples of various neurodegenerative diseases stored in each node;

The gradient updating unit is used for updating the original neural network model by using the gradient values obtained by training of each node to obtain the final neural network model.
The neurodegenerative disease modeling apparatus based on federated transfer learning according to claim 4, wherein the federated transfer learning unit comprises:

The weight judgment unit is used to obtain the model weight of each node in the training process of each round of federated transfer learning, and judge whether the model weight of each node is equal;

The weight update unit is used to update the model weight of each node if the model weight of each node is not equal, so as to be used for the training process of this round of federated transfer learning.
The neurodegenerative disease modeling apparatus based on federated transfer learning according to claim 5, wherein the weight updating unit comprises:

The weight update subunit is used to update the model weight of each node according to the following formula:

ω k+1 ←ω k +μF k (ω k )

ω k represents the model weight of the current node in the current training round, ω k+1 represents the updated model weight of the current node, μ is the learning rate, F k represents the training data of the current node accounts for the total training data of all nodes. proportion.
The neurodegenerative disease modeling apparatus based on federated transfer learning according to claim 5, wherein the gradient updating unit comprises:

The weighted average unit is used to perform weighted average of the gradient values obtained from the training of each node according to the model weight of each node to obtain the average gradient;

The gradient update subunit is used to update the original neural network model by using the average gradient to obtain the final neural network model.
The neurodegenerative disease modeling apparatus based on federated transfer learning according to claim 2, wherein the data input unit comprises:

The encryption unit is used for encrypting the data in the training set samples by using the homomorphic encryption technology.
A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, when the processor executes the computer program, the following steps are implemented: obtaining a first type of A medical record data set A and a medical record data set B of a neurodegenerative disease patient, the medical record data set A includes a plurality of pieces of data described by the first feature item set, and the medical record data set B Multiple pieces of data; the first feature item set includes the feature items used to describe the patient's identity and the feature item used to describe the patient's disease state; the second feature item set includes the feature items used to describe the patient's identity and the feature items used to describe the patient's clinical Feature items of performance data; select intersection data from the case data set A and case data set B, and use the intersection data as a training set sample; based on the training set sample, the original neural network model is trained; based on The training set samples of multiple types of neurodegenerative diseases, including the first type of neurodegenerative diseases, perform federated transfer learning on the original neural network model to obtain the final neural network model for predicting the prevalence of neurodegenerative diseases .
The computer device according to claim 9, wherein, based on the training set samples, the training of the original neural network model comprises:

The data of the feature items used to describe the patient's diseased state in the training set samples are used as the label of the original neural network model; and the data used to describe the feature items of the patient's clinical performance data are used as the original neural network. The input data of the network model;

The original neural network model is trained based on the input data to obtain the trained original neural network model.
The computer device according to claim 9, wherein, based on the training set samples, the original neural network model is trained, further comprising:

Use the non-intersection data between the case data set A and the case data set B as a test set sample;

The original neural network model is tested based on the test set samples.
The computer device according to claim 9, wherein, based on training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative diseases, federated transfer learning is performed on the original neural network model to obtain a model for The final neural network model for predicting the prevalence of neurodegenerative diseases, including:

Obtain the trained original neural network model;

Use the training set samples of various neurodegenerative diseases stored in each node to perform federated transfer learning on the original neural network model;

The original neural network model is updated using the gradient values obtained by training each node to obtain the final neural network model.
The computer device according to claim 12, wherein, using training set samples of various types of neurodegenerative diseases stored in each node to perform federated transfer learning on the original neural network model, comprising:

In the training process of each round of federated transfer learning, the model weight of each node is obtained, and it is judged whether the model weight of each node is equal;

If the model weights of each node are not equal, the model weights of each node are updated for the training process of this round of federated transfer learning.
The computer device according to claim 13, wherein if the model weights of each node are not equal, updating the model weight of each node for the training process of this round of federated transfer learning, comprising:

The model weights of each node are updated according to the following formula:

ω k+1 ←ω k +μF k (ω k )

ω k represents the model weight of the current node in the current training round, ω k+1 represents the updated model weight of the current node, μ is the learning rate, F k represents the training data of the current node accounts for the total training data of all nodes. proportion.
A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the steps of: obtaining a patient with a first type of neurodegenerative disease medical record data set A and medical record data set B, the medical record data set A includes multiple pieces of data described by the first feature item set, and the medical record data set B includes multiple pieces of data described by the second feature item set; The first feature item set includes feature items used to describe the patient's identity and the feature item used to describe the patient's disease state; the second feature item set includes the feature items used to describe the patient's identity and the feature items used to describe the patient's clinical performance data. Select intersection data from described case data set A and case data set B, and use described intersection data as training set sample; Based on described training set sample, the original neural network model is trained; For training set samples of multiple types of neurodegenerative diseases including degenerative diseases, federated transfer learning is performed on the original neural network model to obtain a final neural network model for predicting the prevalence of neurodegenerative diseases.
The computer-readable storage medium of claim 15, wherein training an original neural network model based on the training set samples comprises:

The data of the feature items used to describe the patient's diseased state in the training set samples are used as the label of the original neural network model; and the data used to describe the feature items of the patient's clinical performance data are used as the original neural network. The input data of the network model;

The original neural network model is trained based on the input data to obtain the trained original neural network model.
The computer-readable storage medium of claim 15, wherein, based on the training set samples, training an original neural network model, further comprising:

Use the non-intersection data between the case data set A and the case data set B as a test set sample;

The original neural network model is tested based on the test set samples.
The computer-readable storage medium of claim 15, wherein federated transfer learning is performed on the original neural network model based on training set samples of multiple types of neurodegenerative diseases including the first type of neurodegenerative disease, Get the final neural network model for predicting the prevalence of neurodegenerative diseases, including:

Obtain the trained original neural network model;

Use the training set samples of various neurodegenerative diseases stored in each node to perform federated transfer learning on the original neural network model;

The original neural network model is updated using the gradient values obtained by training each node to obtain the final neural network model.
The computer-readable storage medium according to claim 18, wherein performing federated transfer learning on the original neural network model using training set samples of various types of neurodegenerative diseases stored by each node, comprising:

In the training process of each round of federated transfer learning, the model weight of each node is obtained, and it is judged whether the model weight of each node is equal;

If the model weights of each node are not equal, the model weights of each node are updated for the training process of this round of federated transfer learning.
The computer-readable storage medium according to claim 19, wherein if the model weights of the nodes are not equal, updating the model weights of the nodes for the training process of this round of federated transfer learning, comprising:

The model weights of each node are updated according to the following formula:

ω k+1 ←ω k +μF k (ω k )

ω k represents the model weight of the current node in the current training round, ω k+1 represents the updated model weight of the current node, μ is the learning rate, F k represents the training data of the current node accounts for the total training data of all nodes. proportion.