Method and device for medical data privacy protection and resource utilization based on continuous learning

Abstract

The present invention provides a method and device for medical data privacy protection and resource utilization based on continuous learning. The method includes: initializing the data required for a medical task, where the data comes from n institutions, n being an integer greater than 1, and the data of these n institutions is relatively independent. Then, initializing the deep model for the medical task, where the deep model includes a feature sampling model, a data review model, and a task expression model. Finally, training the deep model successively according to the data from the n institutions. This method can overcome the data sharing barrier, and an intelligent medical system for medical knowledge sharing will greatly improve the existing medical technology level in the future, solving the technical problems of the deep learning method for knowledge sharing without data sharing conditions in the prior art.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and deep learning, and particularly to a method and device for medical data privacy protection and resource utilization based on continuous learning. Background Art

[0002] With the increasing maturity and wide use of deep learning technology, intelligent healthcare has gradually become a key project for the industrial implementation of deep learning. Many related studies involve the collection, analysis of medical data, and the use of various types of medical data such as medical images, case reports, and test indicators for disease detection, pathological analysis, assisted treatment, and risk assessment. However, most of the studies use a small amount of data, and the number of source institutions, the breadth of data distribution, and the resulting model generalization ability are very limited. Since medical data involves patient privacy and is difficult to share, even if a large amount of human and material resources are spent to collect a sufficient amount of medical data, it is only limited to a few studies by participating scientific research and medical institutions, and it is difficult to promote the development of the field through data and experience sharing.

[0003] Continuous learning is a deep learning technology proposed in recent years. There are many types of application scenarios it aims to solve, but they all have the following commonalities: gradually iterating the same deep learning model during continuous data input and training processes; dividing data into multiple tasks according to the acquisition time, and there are differences between tasks, such as training objectives, data distributions, model outputs, etc. may all change; when training the model for a new task, the data of the previous task is difficult to obtain again or only partially obtained. The goal of continuous learning is to not forget the knowledge of old tasks while continuously learning new tasks.

[0004] Therefore, in terms of medical data with strict case privacy protection, continuous learning can provide a deep learning method for knowledge sharing without data sharing. With the continuous deepening of people's understanding of a healthy life, an intelligent medical system that can overcome the data sharing barrier for medical knowledge sharing will greatly improve the existing medical technology level in the future, improve the problem of unbalanced medical resources, and improve people's quality of life. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems in the related technologies to some extent.

[0006] To this end, the first object of the present invention is to propose a method for medical data privacy protection and resource utilization based on continuous learning to realize an intelligent medical system for medical knowledge sharing that overcomes the data sharing barrier.

[0007] The second object of the present invention is to propose a device for medical data privacy protection and resource utilization based on continuous learning.

[0008] The third object of the present invention is to provide a non - transitory computer - readable storage medium.

[0009] The fourth object of the present invention is to provide a computer program product.

[0010] To achieve the above object, an embodiment of the first aspect of the present invention provides a method, including:

[0011] Initializing the data required for a medical task, where the data comes from n institutions, n is an integer greater than 1, and the data of the n institutions is relatively independent;

[0012] Initializing a deep model for the medical task, where the deep model includes a feature sampling model, a data review model, and a task expression model;

[0013] Training the deep model sequentially according to the data from n institutions.

[0014] Optionally, in an embodiment of the present application, the training of the deep model according to the data from n institutions includes:

[0015] Determining institution i, where i <= n and i is a positive integer;

[0016] When i is 1, training the deep model using the data of institution i;

[0017] When i is not 1, training the deep model using the data of institution i, and training the hidden space of the deep model according to multi - scale review sample features;

[0018] Until the data of all n institutions has been trained.

[0019] Optionally, in an embodiment of the present application, before training the hidden space of the deep model according to multi - scale review sample features, it further includes:

[0020] Uniformly sampling a plurality of sample encodings in the data distribution space of institutions 1 to i - 1 using the feature sampling model of institution i - 1;

[0021] Passing the plurality of sample encodings through the data review model of institution i - 1 to generate the multi - scale review sample features.

[0022] Optionally, in an embodiment of the present application, uniformly sampling a plurality of sample encodings using the feature sampling model includes:

[0023] Extracting a feature vector for each data sample of each institution through the feature sampling model, and the feature vectors of all data samples of an institution span a feature space for characterizing the overall feature of the institution;

[0024] A feature space pool is formed according to the overall feature spaces of all institutions.

[0025] The feature vectors of each data sample within the current institution are linearly constrained using cosine similarity, and the feature vectors of each data sample within the current institution are orthogonally constrained with the overall feature spaces of all previous institutions within the current institution using orthogonality.

[0026] A latent variable is sampled in the Gaussian space, projected onto the overall feature spaces of all previous institutions before the current institution, obtaining a latent variable containing the previous institutions before the current institution, and the projected latent variable is mapped into a sample code through an encoding mapping network.

[0027] To achieve the above object, an embodiment of the second aspect of the present invention proposes a medical data privacy protection and resource utilization device based on continual learning, including:

[0028] A data initialization module, configured to initialize the data required for a medical task, where the data comes from n institutions, where n is an integer greater than 1, and the data of the n institutions is relatively independent.

[0029] A model initialization module, configured to initialize a deep model for a medical task, where the deep model includes a feature sampling model, a data review model, and a task expression model.

[0030] A training module, configured to sequentially train the deep model according to the data from n institutions.

[0031] To achieve the above object, an embodiment of the third aspect of the present invention proposes a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the computer program is executed by the processor, the method described in the embodiment of the first aspect of the present application is implemented.

[0032] To achieve the above object, an embodiment of the fourth aspect of the present invention proposes a non-volatile storage medium, on which computer-readable instructions are stored, and when the computer-readable instructions are executable by a processor, the method described in the embodiment of the first aspect of the present application is implemented.

[0033] In summary, the method, device, computer device, and non-volatile storage medium for cross-center medical data privacy protection and resource utilization based on continual learning proposed by the present invention can effectively utilize the medical data of multiple institutions while effectively ensuring patient privacy, thereby contributing to promoting the sharing of medical institution experiences, improving the social medical level, and safeguarding the physical health of the general public. Description of the Drawings

[0034] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of embodiments in conjunction with the accompanying drawings, where:

[0035] Figure 1 It is a schematic flowchart of a method for medical data privacy protection and resource utilization based on continuous learning provided by an embodiment of the present invention;

[0036] Figure 2 It is a flowchart of a feature sampling model provided by an embodiment of the present invention;

[0037] Figure 3 It is a general flowchart of a method provided by an embodiment of the present invention;

[0038] Figure 4 It is a schematic structural diagram of a device for medical data privacy protection and resource utilization based on continuous learning provided by an embodiment of the present invention. Detailed Embodiments

[0039] The embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, and should not be construed as limiting the present invention.

[0040] The method and device for medical data privacy protection and resource utilization based on continuous learning according to the embodiments of the present invention will be described below with reference to the accompanying drawings.

[0041] Figure 1 It is a schematic flowchart of a method for medical data privacy protection and resource utilization based on continuous learning provided by an embodiment of the present invention.

[0042] As Figure 1 shown, the method for medical data privacy protection and resource utilization based on continuous learning includes the following steps:

[0043] Step S1: Initialize the data required for the medical task. The data comes from n institutions, where n is an integer greater than 1, and the data of the n institutions is relatively independent.

[0044] Step S2: Initialize the deep model of the medical task. The deep model includes a feature sampling model, a data review model, and a task expression model.

[0045] The data review model and the task expression model have different structures due to different medical tasks applied. Here are several common model examples: The data review model includes, but is not limited to, GAN (Generative Adversarial Networks), VAE (Variational Auto-Encoder), etc.; The task expression model includes, but is not limited to, medical image detection and segmentation models, diagnostic report classification language models, case survival rate risk prediction models, etc.

[0046] Step S3: Train the deep model according to the data from n institutions in sequence.

[0047] In one embodiment of the present invention, training the deep model according to the data from n institutions includes:

[0048] Determine institution i, where i <= n and i is a positive integer;

[0049] When i is 1, use the data of institution i to train the deep model;

[0050] When i is not 1, use the data of institution i to train the deep model, and train the hidden space of the deep model according to the multi-scale review sample features;

[0051] Until the data of all n institutions are all trained.

[0052] In one embodiment of the present invention, before training the hidden space of the deep model according to the multi-scale review sample features, it further includes:

[0053] Sample multiple sample encodings in the data distribution space of institutions 1 to i - 1 using the feature sampling model of institution i - 1;

[0054] Pass the multiple sample encodings through the data review model of institution i - 1 to generate multi-scale review sample features.

[0055] In one embodiment of the present invention, sampling multiple sample encodings using the feature sampling model includes:

[0056] Extract a feature vector for each data sample of each institution through the feature sampling model. The feature vectors of all data samples of an institution span a feature space, which is used to characterize the overall feature space of the institution;

[0057] Form a feature space pool according to the overall feature spaces of all institutions;

[0058] Linearly constrain the feature vectors of each data sample within the current institution using cosine similarity, and use orthogonality to orthogonally constrain the feature vectors of each data sample within the current institution with the overall feature space of all previous institutions of the current institution;

[0059] Sample a latent variable in the Gaussian space, project the sampled latent variable onto the overall feature space of all previous institutions of the current institution to obtain a latent variable containing the previous institutions of the current institution, and map the projected latent variable to a sample code through an encoding mapping network.

[0060] To further explain how the feature sampling model works, as Figure 2 shown, is the specific process of the feature sampling model provided by the present invention. During the process of training the data of each institution, the model extracts a feature vector for each data sample of each institution, and the feature vectors of all data samples of an institution span a feature space to characterize the overall features of the institution. The overall feature spaces of all institutions form a feature space pool. Specifically, when extracting the feature space of institution n, use cosine similarity to constrain the features of each sample within institution n to be as linear as possible, and use orthogonality to constrain the feature vector of institution n to be as orthogonal as possible to the feature space spanned by the feature vectors of the first n institutions. When using this feature sampling model for sample code sampling, first sample a latent variable in the Gaussian space to represent the features of the sample. Then project the latent variable onto the specified institution feature space, such as the feature space spanned by the feature vectors of the first n - 1 institutions, to obtain a latent variable that may contain the data information of the first n - 1 institutions. Finally, the projected latent variable is mapped to a sample code through an encoding mapping network to guide the feature generation of the subsequent data review model.

[0061] Based on the above, for easy understanding, as Figure 3 shown, is the overall flowchart of the method provided by the present invention.

[0062] When training institution i (i > 1), first by Figure 2 method, sample a latent variable in the Gaussian space and obtain a sample code using the feature sampling model; then generate multi-scale review sample features through the data review model, and finally train the task expression model using the real data input and the multi-scale review sample features. Among them, the task expression model can give a task output according to the input of a specific task.

[0063] In summary, the present invention proposes a method for cross-center medical data privacy protection and resource utilization based on continuous learning, which can effectively utilize the medical data of multiple institutions across centers while effectively ensuring patient privacy, helps to promote the sharing of medical experience among medical institutions, improve the social medical level, and guarantee the health of the general public.

[0064] Figure 4 This is a schematic structural diagram of a medical data privacy protection and resource utilization device based on continuous learning provided by an embodiment of the present invention.

[0065] As Figure 4 shown, the medical data privacy protection and resource utilization device based on continuous learning includes the following modules:

[0066] A data initialization module, configured to initialize the data required for a medical task. The data comes from n institutions, where n is an integer greater than 1, and the data of the n institutions is relatively independent;

[0067] A model initialization module, configured to initialize a deep model for a medical task. The deep model includes a feature sampling model, a data review model, and a task expression model;

[0068] A training module, configured to train the deep model sequentially according to the data from n institutions.

[0069] In an embodiment of the present disclosure, further, the training module is further configured to:

[0070] Determine institution i, where i <= n and i is a positive integer;

[0071] When i is 1, use the data of institution i to train the deep model;

[0072] When i is not 1, use the data of institution i to train the deep model, and train the hidden space of the deep model according to multi-scale review sample features;

[0073] Until the data of all n institutions is completely trained.

[0074] In an embodiment of the present disclosure, further, the training module is further configured to:

[0075] Uniformly sample multiple sample encodings in the data distribution space of institutions 1 to i - 1 using the feature sampling model of institution i - 1;

[0076] Pass the multiple sample encodings through the data review model of institution i - 1 to generate multi-scale review sample features.

[0077] In an embodiment of the present disclosure, further, the training module is further configured to:

[0078] Extract a feature vector for each data sample of each institution through the feature sampling model. The feature vectors of all data samples of an institution span a feature space, which is used to characterize the overall feature of the institution;

[0079] Compose a feature space pool according to the overall feature spaces of all institutions;

[0080] Use cosine similarity to linearly constrain the feature vectors of each data sample within the current institution, and use orthogonality to orthogonally constrain the feature vectors of each data sample within the current institution with the overall feature spaces of all previous institutions of the current institution;

[0081] Sample a latent variable in the Gaussian space, project the sampled latent variable onto the overall feature spaces of all previous institutions of the current institution to obtain a latent variable containing the previous institutions of the current institution, and map the projected latent variable to a sample code through an encoding mapping network.

[0082] In summary, the present invention proposes a cross-center medical data privacy protection and resource utilization device based on continual learning, which can effectively utilize the medical data of multiple institutions across centers while effectively ensuring patient privacy, helps to promote the experience sharing among medical institutions, improve the social medical level, and safeguard the health of the general public.

[0083] To achieve the above object, an embodiment of the third aspect of the present application proposes a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, the method described in the embodiment of the first aspect of the present application is implemented.

[0084] To achieve the above object, an embodiment of the fourth aspect of the present application proposes a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, the method described in the embodiment of the first aspect of the present application is implemented.

[0085] In the description of this specification, the descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0086] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0087] Any process or method description represented in a flowchart or described otherwise herein may be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of the present invention includes additional implementations where functions may be executed not in the order shown or discussed, including in a substantially simultaneous manner according to the functions involved or in a reverse order, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0088] The logic and / or steps represented in a flowchart or described otherwise herein, for example, may be considered as a sequenced list of executable instructions for implementing a logical function and may be specifically embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection portion having one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, as the program can be obtained, for example, by optically scanning the paper or other medium, followed by editing, interpretation, or other appropriate processing as necessary to obtain the program in electronic form and then storing it in a computer memory.

[0089] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0090] Those of ordinary skill in the art can understand that all or part of the steps carried by the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the said program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0091] In addition, each functional unit in various embodiments of the present invention can be integrated in a processing module, or each unit can exist physically alone, or two or more units can be integrated in one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. When the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0092] The above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for medical data privacy protection and resource utilization based on continuous learning, characterized in that, it includes the following steps: Initialize the data required for the medical task, and the data comes from n institutions, where n is an integer greater than 1, and the data of the n institutions is relatively independent; Initialize the deep model of the medical task, and the deep model includes a feature sampling model, a data review model and a task expression model; Train the deep model according to the data from n institutions in sequence; The training of the deep model according to the data from n institutions includes: Determine institution i, where i ≤ n and i is a positive integer; When i is 1, use the data of institution i to train the deep model; When i is not 1, use the data of institution i to train the deep model, and train the hidden space of the deep model according to the multi-scale review sample features; Until the data of all n institutions is completely trained; Before training the hidden space of the deep model according to the multi-scale review sample features, it also includes: Uniformly sample multiple sample encodings in the data distribution space of institutions 1 to i-1 using the feature sampling model of institution i-1; Pass the multiple sample encodings through the data review model of institution i-1 to generate the multi-scale review sample features.

2. The method according to claim 1, characterized in that, The uniform sampling of multiple sample encodings using the feature sampling model includes: Extract a feature vector for each data sample of each institution through the feature sampling model, and the feature vectors of all data samples of an institution span a feature space to characterize the overall features of the institution; Form a feature space pool according to the overall feature spaces of all institutions; Use cosine similarity to linearly constrain the feature vectors of each data sample within the current institution, and use orthogonality to orthogonally constrain the feature vectors of each data sample within the current institution with the overall feature spaces of all institutions before the current institution; Sample a latent variable in the Gaussian space, project the sampled latent variable onto the overall feature spaces of all institutions before the current institution, obtain a latent variable containing the institutions before the current institution, and map the projected latent variable through an encoding mapping network into a sample encoding.

3. A device for medical data privacy protection and resource utilization based on continuous learning, characterized in that, it includes: A data initialization module for initializing the data required for the medical task, and the data comes from n institutions, where n is an integer greater than 1, and the data of the n institutions is relatively independent; A model initialization module for initializing the deep model of the medical task, and the deep model includes a feature sampling model, a data review model and a task expression model; A training module for training the deep model according to the data from n institutions in sequence; The training module is also used for: Determine institution i, where i ≤ n and i is a positive integer; When i is 1, use the data of institution i to train the deep model; When i is not 1, use the data of institution i to train the deep model, and train the hidden space of the deep model according to the multi-scale retrospective sample features; Until the data of all n institutions are all trained; The training module is further configured to: Sample a plurality of sample encodings in the data distribution space of institutions 1 to i-1 using the feature sampling model of institution i-1; Pass the plurality of sample encodings through the data retrospective model of institution i-1 to generate the multi-scale retrospective sample features.

4. The device according to claim 3, wherein, The training module is further configured to: Extract a feature vector for each data sample of each institution through the feature sampling model, and the feature vectors of all data samples of an institution span a feature space for characterizing the overall feature of the institution; Form a feature space pool according to the overall feature spaces of all institutions; Use cosine similarity to linearly constrain the feature vectors of each data sample within the current institution, and use orthogonality to orthogonally constrain the feature vectors of each data sample within the current institution with the overall feature spaces of all institutions before the current institution; Sample a latent variable in the Gaussian space, project the sampled latent variable onto the overall feature spaces of all institutions before the current institution to obtain a latent variable containing the institutions before the current institution, and map the projected latent variable to a sample encoding through an encoding mapping network.

5. A computer device, wherein, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method described in any one of claims 1-2 is implemented.

6. A non-transitory computer-readable storage medium, on which a computer program is stored, wherein, When the computer program is executed by a processor, the method described in any one of claims 1-2 is implemented.

Images