A medical data inference and prediction method based on homomorphic encryption and related equipment

Abstract

The application provides a medical data inference prediction method based on homomorphic encryption and related equipment, encrypts original remote medical data of a target patient through homomorphic encryption to obtain encrypted remote medical data; inputs the encrypted remote medical data into an inference prediction model to perform inference prediction, and obtains a medical prediction result of the target patient; the inference prediction model is obtained by training a convolutional neural network model optimized through a polynomial acceleration algorithm; compared with the prior art, the application uses homomorphic encryption technology to ensure that the original remote medical data of the target patient is always in an encrypted state, effectively protects personal privacy, and at the same time, the convolutional neural network model is optimized through the polynomial acceleration algorithm, the inference speed of the model on the remote medical data is improved, and the real-time performance of the remote medical service is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart medical treatment, and in particular to a medical data inference and prediction method based on homomorphic encryption and related equipment. BACKGROUND

[0002] With the progress of Internet of Things medical technology and cryptography technology, remote medical services have become more popular, which brings new challenges to patient privacy and data security. The convolutional neural network on the edge side has become the basis of advanced machine learning due to its excellent performance and scalability, but directly processing convolutional neural network inference in plaintext has privacy and inference efficiency problems. Therefore, homomorphic encryption provides an alternative solution that can perform inference on encrypted data without exposing any original information. Convolutional neural networks using homomorphic encryption can directly perform calculations on encrypted data, bringing advantages such as single-round communication, reduced bandwidth, and non-interactive computation.

[0003] In remote medical applications, the reliance of these homomorphic encryption-based models on specific medical data sets further hinders the development of standardized and universally applicable network architectures. Different medical institutions may collect and use different types of patient data, which poses a challenge to the universality of the model. In addition, the computational differences between different model layers in homomorphic encryption convolutional neural networks also pose a significant challenge, especially when performing homomorphic calculations on encrypted medical data. The inherent characteristics of homomorphic encryption further exacerbate these challenges, as they bring a huge computational and memory burden, which results in the need for a large allocation of computing resources even for basic medical diagnosis or prediction operations, thereby limiting the practical application of this method in edge devices or limited computing resources, and time-critical remote medical scenarios. For example, intelligent devices deployed in patients' homes may not be able to bear the overhead of complex homomorphic calculations, thereby affecting the real-time nature of remote diagnosis and treatment. SUMMARY

[0004] The present application provides a medical data inference and prediction method based on homomorphic encryption and related equipment, which aims to improve the real-time nature of remote medical services by improving inference speed.

[0005] To achieve the above purpose, the present application provides a medical data inference and prediction method based on homomorphic encryption, comprising:

[0006] Obtaining the original remote medical data of the target patient, and encrypting the original remote medical data through homomorphic encryption to obtain encrypted remote medical data;

[0007] Inputting the encrypted remote medical data into the inference and prediction model for inference and prediction to obtain the medical prediction result of the target patient; the medical prediction result includes one or more physiological indicator data;

[0008] The inference prediction model is obtained by training a convolutional neural network model optimized by a polynomial acceleration algorithm.

[0009] Further, the original telemedicine data of the target patient is acquired, and the original telemedicine data is encrypted by homomorphic encryption to obtain encrypted telemedicine data, including:

[0010] The original telemedicine data of the target patient is acquired by the acquisition device;

[0011] The original telemedicine data is preprocessed to obtain preprocessed telemedicine data;

[0012] The preprocessed telemedicine data is encrypted by homomorphic encryption to obtain encrypted telemedicine data.

[0013] Further, the encrypted telemedicine data is input into the inference prediction model for inference prediction, including:

[0014] An initialized convolutional neural network model is selected, and the initialized convolutional neural network model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer connected in sequence;

[0015] The initialized convolutional neural network model is optimized by a polynomial acceleration algorithm to obtain an optimized convolutional neural network model;

[0016] The optimized convolutional neural network model is trained to obtain the inference prediction model;

[0017] The encrypted telemedicine data is input into the inference prediction model for inference prediction.

[0018] Further, the initialized convolutional neural network model is optimized by a polynomial acceleration algorithm, including:

[0019] The input layer of the initialized convolutional neural network model is optimized to enable the input layer to process the encrypted telemedicine data in parallel;

[0020] Each homomorphic encryption operation type is assigned a respective implementation strategy to optimize the data processing process of the initialized convolutional neural network model;

[0021] The model parameters are stored in the register of the image processor and the shared memory of the image processor is used for data exchange and transmission to optimize the data structure of the initialized convolutional neural network model.

[0022] Further, the input layer of the initialized convolutional neural network model is optimized, including:

[0023] The plurality of encrypted remote medical data are allocated thread blocks in the input layer of the initialized convolutional neural network model, and the thread blocks are processed in parallel.

[0024] Further, the implementation strategy includes spatial operations and non-spatial operations.

[0025] For the spatial operation, a local feature map and an edge feature map are extracted on the spatial domain of the encrypted remote medical data.

[0026] For the non-spatial operation, a one-dimensional feature vector is extracted by performing a non-linear transformation on the local feature map and the edge feature map.

[0027] Further, the local feature map and the edge feature map are extracted on the spatial domain of the encrypted remote medical data, including:

[0028] The parallel-processed remote medical data are input into a convolutional layer, and the convolutional layer extracts features by sliding a two-dimensional convolution kernel on the spatial domain of the encrypted remote medical data to obtain a local feature map.

[0029] The local feature map is input into a pooling layer, and the pooling layer performs down-sampling on the local feature map on the spatial domain of the encrypted remote medical data to obtain an edge feature map.

[0030] The application also provides a medical data inference and prediction device based on homomorphic encryption, comprising:

[0031] An acquisition module is configured to acquire original remote medical data of a target patient, encrypt the original remote medical data through homomorphic encryption, and obtain encrypted remote medical data.

[0032] An inference and prediction module is configured to input the encrypted remote medical data into an inference and prediction model to perform inference and prediction, and obtain a medical prediction result of the target patient.

[0033] The inference and prediction model is obtained by training a convolutional neural network model optimized through a polynomial acceleration algorithm.

[0034] The application also provides a computer-readable storage medium storing a computer program, and the computer program is executed by a processor to implement the medical data inference and prediction method based on homomorphic encryption.

[0035] The application also provides a terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the medical data inference and prediction method based on homomorphic encryption when executing the computer program.

[0036] The above-mentioned scheme of the application has the following advantages:

[0037] The present application encrypts the obtained original remote medical data of the target patient through homomorphic encryption, and obtains encrypted remote medical data; the encrypted remote medical data is input into an inference prediction model for inference prediction, and the medical prediction result of the target patient is obtained; the inference prediction model is obtained by training a convolutional neural network model optimized through a polynomial acceleration algorithm; compared with the prior art, the present application uses homomorphic encryption technology to ensure that the original remote medical data of the target patient is always in an encrypted state, effectively protecting personal privacy, and at the same time, the convolutional neural network model is optimized through a polynomial acceleration algorithm, the inference speed of the model on the remote medical data is improved, and the real-time performance of the remote medical service is improved.

[0038] Other benefits of the present application will be described in detail in the subsequent specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The flowchart of the embodiment of the present application is shown;

[0040] Figure 2 The structural diagram of the medical data inference prediction device provided by the embodiment of the present application is shown;

[0041] Figure 3 The structural diagram of the terminal device provided by the embodiment of the present application is shown. DETAILED DESCRIPTION

[0042] In order to make the technical problems, technical solutions and advantages of the present application clearer, the following will be described in detail in combination with the drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0043] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore it cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third" are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0044] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting" should be understood in a broad sense, for example, can be locking connection, can also be detachable connection, or integrally connected, can be mechanical connection, can also be electrical connection, can be directly connected, can also be indirectly connected through an intermediate medium, and can be internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0045] In addition, the technical features involved in the different embodiments of the application described below can be combined with each other as long as there is no conflict.

[0046] The present application aims at the existing problems, and provides a medical data reasoning and prediction method based on homomorphic encryption and related equipment.

[0047] As shown in Figure 1 The embodiment of the present application provides a medical data reasoning and prediction method based on homomorphic encryption, which comprises the following steps:

[0048] Obtaining the original remote medical data of the target patient, and encrypting the original remote medical data through homomorphic encryption to obtain encrypted remote medical data;

[0049] Inputting the encrypted remote medical data into a reasoning and prediction model for reasoning and prediction to obtain a medical prediction result of the target patient; the medical prediction result comprises one or more physiological index data, such as heart rate, respiratory rate, electromyogram, etc.

[0050] The reasoning and prediction model is obtained by training a convolutional neural network model optimized through a polynomial acceleration algorithm.

[0051] Specifically, obtaining the original remote medical data of the target patient, and encrypting the original remote medical data through homomorphic encryption to obtain encrypted remote medical data, comprising:

[0052] Obtaining the original remote medical data of the target patient through a collection device;

[0053] Preprocessing the original remote medical data to obtain preprocessed remote medical data;

[0054] Encrypting the preprocessed remote medical data through homomorphic encryption to obtain encrypted remote medical data.

[0055] In the embodiment of the present application, the collection device can be a mobile phone or a palm computer of the target patient or other devices with camera function, and the target patient can obtain the original remote medical data by shooting or uploading the original medical data through the mobile phone or the palm computer. The original remote medical data includes but is not limited to electrocardiogram, electroencephalogram and other original medical examination sheets.

[0056] The original remote medical data is preprocessed through a series of means such as data cleaning, denoising and standardization to obtain preprocessed remote medical data.

[0057] The preprocessed remote medical data is encrypted by the collection device using homomorphic encryption technology to obtain encrypted remote medical data.

[0058] In the embodiment of the application, the monitoring service provider generates a key for the collection device using a homomorphic encryption algorithm, and the collection device encrypts the preprocessed remote medical data through the key to obtain encrypted remote medical data.

[0059] Specifically, the encrypted remote medical data is input into the inference prediction model for inference prediction, including:

[0060] An initialized convolutional neural network model is selected, and the initialized convolutional neural network model includes an input layer for receiving data, a convolutional layer for extracting local features of the data, a pooling layer for extracting edge features of the data and a fully connected layer for fusing features connected in sequence.

[0061] The initialized convolutional neural network model is optimized through a polynomial acceleration algorithm to obtain an optimized convolutional neural network model.

[0062] The optimized convolutional neural network model is trained to obtain the inference prediction model.

[0063] The encrypted remote medical data is input into the inference prediction model for inference prediction.

[0064] In the embodiment of the application, the initialized convolutional neural network model is a conventional convolutional neural network model, and the application does not involve improving the structure of the model, so the structure will not be described one by one; the optimized convolutional neural network model is also trained in a conventional training manner, so the training process will not be described one by one.

[0065] Specifically, the initialized convolutional neural network model is optimized through a polynomial acceleration algorithm, including:

[0066] The input layer of the initialized convolutional neural network model is optimized to enable the input layer to process the encrypted remote medical data in parallel;

[0067] Each homomorphic encryption operation type is assigned a respective implementation strategy to optimize the data processing process of the initialized convolutional neural network model.

[0068] The model parameters are stored through the registers of the image processor, and data exchange and transmission are performed by using the shared memory of the image processor, so that the data structure of the initialized convolutional neural network model is optimized.

[0069] Specifically, the intermediate data of the convolution operation is first stored in the shared memory of the image processor (GPU), the model parameters stored in the shared memory of the GPU are read into the registers, and operations such as number theory transformation (NTT) are performed in the registers, and then the data results are transmitted to the shared memory of the GPU for subsequent operation, and the shared memory of the GPU reads other model parameters to be subjected to number theory transformation into the shared memory of the GPU again to complete the data exchange, so as to enter the next NTT data operation cycle.

[0070] Specifically, the input layer of the initialized convolutional neural network model is optimized, including:

[0071] In the input layer of the initialized convolutional neural network model, a plurality of encrypted remote medical data are allocated to thread blocks, and the thread blocks are processed in parallel.

[0072] In the embodiment of the application, each thread block contains 128 threads, which are used to realize fine-grained high parallel processing, and the batch processing of thread blocks is used to further improve the parallel degree.

[0073] In the embodiment of the application, the homomorphic encryption operation types include element addition (EleAdd), element multiplication (EleMult), convolution (Conv) and number theory transformation operation.

[0074] Specifically, the implementation strategy includes spatial operation and non-spatial operation.

[0075] For spatial operation, local feature maps and edge feature maps are extracted in the spatial domain of encrypted remote medical data.

[0076] For non-spatial operation, the local feature maps and edge feature maps are subjected to nonlinear transformation to extract one-dimensional feature vectors.

[0077] Specifically, the local feature maps and edge feature maps are extracted in the spatial domain of encrypted remote medical data, including:

[0078] The parallel-processed remote medical data are input into a convolution layer, and the convolution layer extracts features by sliding a two-dimensional convolution kernel in the spatial domain of encrypted remote medical data to obtain local feature maps.

[0079] The local feature maps are input into a pooling layer, and the pooling layer performs down-sampling on the local feature maps in the spatial domain of encrypted remote medical data to obtain edge feature maps.

[0080] In the embodiments of the present application, the spatial operation refers to an operation involving two-dimensional spatial processing of images or signals in the deep learning model. This includes convolution operations, which use a two-dimensional convolution kernel to slide over the spatial domain of the input data to extract local features; and pooling operations, which down-sample the feature map in the spatial domain to extract more abstract edge features. These operations preserve the spatial structure of the input data, enabling the model to learn and understand the spatial hierarchical relationships in the input; for spatial operations, first, in the input layer, the encrypted remote medical data is packaged and processed in batches in a parallel computing manner to optimize the calculation of weight multiplication and bias addition; next, the weights of the convolution layer are encoded as plaintext, and the convolution operation is performed in the encrypted domain; by multiplying the corresponding plaintext and encrypted remote medical data and aggregating, the convolution result of each channel can be obtained; the pooling layer is optimized by interleaved batch processing in the embodiments of the present application, which is different from the traditional method of sequentially storing each receptive field. The process of this method is as follows:

[0081] First, extract all receptive fields of size FxF from the encrypted remote medical data and map them to F 2 messages;

[0082] Next, perform a permutation operation on all receptive fields, which changes the position of each receptive field but does not affect the convolution operation itself, because the calculation is performed in a single instruction multiple data (SIMD) batch processing manner;

[0083] Then, perform 4 homomorphic rotations and addition operations on the permuted receptive fields to calculate the average of the 16 values required by the pooling layer.

[0084] This interleaved batch processing allows the pooling layer to be executed directly after the batch convolution operation without intermediate storage and data transmission.

[0085] In the embodiments of the present application, for non-spatial operations, i.e. in the fully connected layer, each neuron is connected to all neurons of the previous layer, regardless of their spatial position relationship and spatial structure. The activation function is used to independently perform non-linear transformation on each input value, with the purpose of optimizing the activation function, as follows:

[0086] Select the activation function: use the ReLU activation function because of its computational efficiency and ease of implementation;

[0087] Approximate optimization: since using the ReLU activation function in the homomorphic encryption environment will bring complexity problems, the embodiments of the present application propose to use a simpler quadratic function as the ReLU activation function, i.e. ReLU(x)≈x2 ;

[0088] Computational efficiency optimization: Quadratic function approximation can be implemented through single homomorphic multiplication, greatly improving computational efficiency, which is very critical for homomorphic encryption scenarios;

[0089] Trade-off: The embodiment of the present application balances the approximation accuracy and computational efficiency, and finally selects the quadratic function with higher computational efficiency as the ReLU activation function. Although it lacks the sparsity feature of the ordinary ReLU activation function, in the homomorphic encryption environment, the calculation mode is not affected by the plaintext value, so the sparsity feature of the ordinary ReLU activation function will not bring significant performance improvement.

[0090] Specifically, by assigning each homomorphic encryption operation type a respective implementation strategy, the data processing process of initializing the convolutional neural network model is optimized, and the optimization process is as follows:

[0091] The polynomial acceleration algorithm simplifies the overall calculation process by decomposing complex homomorphic encryption operations into basic residual number system operations through hierarchical reconstruction; the specific process of hierarchical reconstruction includes:

[0092] Thread allocation: The embodiment of the present application allocates a dedicated thread block to each residual, each thread block is configured with 128 threads, and these thread blocks are batch loaded into each kernel to realize concurrent processing of residuals;

[0093] Element-wise kernel operation: For element-wise operations such as element addition (EleAdd), element multiplication (EleMult), and convolution (Conv), each thread is responsible for processing a residual coefficient, performing weighting, comparison, and optimization operations to obtain different levels (grades) of residual coefficients corresponding to different runs (convergence loop times);

[0094] Implementation of number theory transformation: The hierarchical method is used to implement number theory operations, following the latest methodology, and implementing the number theory transformation for N = 2 14 Optimized, involving 14 processing levels;

[0095] For complex number theory transformations, they are divided into two kernels for processing, the first kernel processes the first 8 levels of number theory transformation, and the second kernel processes the last 6 levels of number theory transformation. In each kernel, each thread manages 8 coefficients, greatly reducing the computational complexity.

[0096] Specifically, the image processor's register stores model parameters and uses the image processor's shared memory for data exchange and transmission to optimize the data structure of initializing the convolutional neural network model, and the optimization process is as follows:

[0097] 8 coefficients managed by each thread are stored to the register of the image processor, the memory access efficiency is improved, then the transformation is executed using the shared memory of the image processor, after the transformation is completed, the next round of transformation is performed, and the access to the global memory is reduced.

[0098] In conclusion, the embodiment of the present application encrypts the obtained original remote medical data of the target patient through homomorphic encryption to obtain encrypted remote medical data; inputs the encrypted remote medical data into an inference prediction model to perform inference prediction to obtain a medical prediction result of the target patient; the medical prediction result includes one or more physiological index data; the inference prediction model is obtained by training a convolutional neural network model optimized through a polynomial acceleration algorithm; compared with the prior art, the present application uses homomorphic encryption technology to ensure that the original remote medical data of the target patient is always in an encrypted state, effectively protecting personal privacy, and at the same time, the convolutional neural network model is optimized through a polynomial acceleration algorithm, the calculation process of complex operations is simplified, the inference speed of the model on remote medical data is improved, and the real-time performance of remote medical services is further improved.

[0099] As shown in Figure 2 The embodiment of the present application also provides a medical data inference prediction device 100 based on homomorphic encryption, which comprises:

[0100] An acquisition module 101 is configured to acquire original remote medical data of a target patient, and encrypt the original remote medical data through homomorphic encryption to obtain encrypted remote medical data.

[0101] An inference prediction module 102 is configured to input the encrypted remote medical data into an inference prediction model to perform inference prediction to obtain a medical prediction result of the target patient; the medical prediction result includes one or more physiological index data.

[0102] The inference prediction model is obtained by training a convolutional neural network model optimized through a polynomial acceleration algorithm.

[0103] In some embodiments, the acquisition module 101 is specifically configured to acquire the original remote medical data of the target patient through a collection device; pre-process the original remote medical data to obtain pre-processed remote medical data; and encrypt the pre-processed remote medical data through homomorphic encryption to obtain the encrypted remote medical data.

[0104] The acquisition module 101 comprises:

[0105] An acquisition sub-module is configured to acquire the original remote medical data of the target patient through a collection device.

[0106] The preprocessing submodule is configured to preprocess the original remote medical data to obtain preprocessed remote medical data.

[0107] The encryption submodule is configured to encrypt the preprocessed remote medical data by homomorphic encryption to obtain encrypted remote medical data.

[0108] In some embodiments, the inference prediction module 102 is configured to select an initialized convolutional neural network model, the initialized convolutional neural network model comprising an input layer, a convolutional layer, a pooling layer, and a fully connected layer; optimize the initialized convolutional neural network model by a polynomial acceleration algorithm to obtain an optimized convolutional neural network model; train the optimized convolutional neural network model to obtain an inference prediction model; and input the encrypted remote medical data into the inference prediction model for inference prediction to obtain a disease risk index of the target patient.

[0109] The inference prediction submodule 102 comprises:

[0110] The selection submodule is configured to select an initialized convolutional neural network model, the initialized convolutional neural network model comprising an input layer, a convolutional layer, a pooling layer, and a fully connected layer connected in sequence.

[0111] The optimization submodule is configured to optimize the initialized convolutional neural network model by a polynomial acceleration algorithm to obtain an optimized convolutional neural network model.

[0112] The training submodule is configured to train the optimized convolutional neural network model to obtain an inference prediction model.

[0113] The inference prediction submodule is configured to input the encrypted remote medical data into the inference prediction model for inference prediction to obtain a medical prediction result of the target patient; the medical prediction result comprising one or more physiological indicator data, such as heart rate, respiratory rate, and muscle electricity.

[0114] In some embodiments, the optimization submodule is configured to optimize the input layer of the initialized convolutional neural network model to enable the input layer to process the encrypted remote medical data in parallel; optimize the data processing process of the initialized convolutional neural network model by assigning respective implementation strategies to each homomorphic encryption operation type; and optimize the data structure of the initialized convolutional neural network model by storing model parameters in the register of the image processor and exchanging and transmitting data by using the shared memory of the image processor.

[0115] The optimization submodule comprises:

[0116] The first optimization unit is configured to optimize an input layer of the initialized convolutional neural network model to enable the input layer to process the encrypted remote medical data in parallel.

[0117] The second optimization unit is configured to optimize a data processing procedure of the initialized convolutional neural network model by assigning a respective implementation strategy to each homomorphic encryption operation type.

[0118] The third optimization unit is configured to optimize a data structure of the initialized convolutional neural network model by storing model parameters in registers of the image processor and exchanging and transmitting data using shared memory of the image processor.

[0119] In some embodiments, the first optimization unit is specifically configured to assign thread blocks to the plurality of encrypted remote medical data in the input layer of the initialized convolutional neural network model and process the thread blocks in parallel.

[0120] In some embodiments, the implementation strategy includes a spatial operation and a non-spatial operation.

[0121] For the spatial operation, local feature maps and edge feature maps are extracted in a spatial domain of the encrypted remote medical data.

[0122] For the non-spatial operation, the local feature maps and the edge feature maps are subjected to a nonlinear transformation to extract one-dimensional feature vectors.

[0123] In some embodiments, extracting the local feature maps and the edge feature maps in the spatial domain of the encrypted remote medical data includes:

[0124] The processed remote medical data is input into a convolutional layer, which extracts features from the encrypted remote medical data in the spatial domain by sliding a two-dimensional convolution kernel to obtain local feature maps.

[0125] The local feature maps are input into a pooling layer, which down-samples the local feature maps in the spatial domain of the encrypted remote medical data to obtain edge feature maps.

[0126] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or software. In addition, the specific name of each functional unit and module is only for easy distinction, and does not limit the protection scope of the application. The specific working process of the unit and module in the above system can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

[0127] The application also provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method for medical data inference and prediction based on homomorphic encryption is realized.

[0128] If the integrated module is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above method embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium. When the processor executes the computer program, the steps of each method embodiment can be realized. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium at least includes any entity or device that can carry the computer program code to the construction device / terminal equipment, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium. For example, U disk, mobile hard disk, magnetic disk or optical disk, etc. In some jurisdictions, according to legislation and patent practice, the computer readable medium cannot be an electrical carrier signal and a telecommunication signal.

[0129] As shown in Figure 3 The application also provides a terminal device D10, which includes at least one processor D100 Figure 3The terminal device comprises at least one processor D100, a memory D101, and a computer program D102 stored in the memory D101 and capable of running on the at least one processor D100. The processor D100 implements the steps in any of the method embodiments described above when executing the computer program D102. Alternatively, the processor D100 implements the functions of each module / unit in the apparatus embodiments described above when executing the computer program D102.

[0130] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0131] The memory can be an internal storage unit of the terminal device in some embodiments, for example, a hard disk or a memory of the terminal device. The memory can also be an external storage device of the terminal device in other embodiments, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory can include both an internal storage unit and an external storage device of the terminal device. The memory is used to store an operating system, application programs, a boot loader, data, and other programs, for example, program codes of the computer program, etc. The memory can also be used to temporarily store data that has been output or will be output.

[0132] It should be noted that the information interaction and execution process between the above apparatuses / modules are based on the same concept as the method embodiments of the present application, and the specific functions and technical effects brought by the above apparatuses / modules can be referred to the method embodiments part, which will not be described here.

[0133] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically independently, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the embodiments of the present application. The specific working process of the units and modules in the above system can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0134] The above is the preferred embodiment of the present application. It should be noted that for those skilled in the art, without departing from the principles of the present application, several improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A medical data inference and prediction method based on homomorphic encryption, characterized in that, include: The original telemedicine data of the target patient is obtained, and the original telemedicine data is encrypted using homomorphic encryption to obtain encrypted telemedicine data. The encrypted telemedicine data is input into the inference prediction model for inference and prediction to obtain the medical prediction result for the target patient; the medical prediction result includes one or more physiological indicator data. The inference prediction model is obtained by training a convolutional neural network model optimized by a multinomial acceleration algorithm; The encrypted telemedicine data is input into the inference prediction model for inference prediction, including: Select the initialization convolutional neural network model; The initial convolutional neural network model is optimized using a polynomial acceleration algorithm to obtain an optimized convolutional neural network model. The initial convolutional neural network model is optimized using a polynomial acceleration algorithm, including: The input layer of the initial convolutional neural network model is optimized to enable the input layer to process the encrypted remote medical data in parallel. The data processing of the initial convolutional neural network model is optimized by assigning a separate implementation strategy to each homomorphic encryption operation type. Optimizing the input layer of the initialized convolutional neural network model includes: In the input layer of the initial convolutional neural network model, thread blocks are allocated for multiple encrypted remote medical data, and the thread blocks are processed in parallel. The implementation strategy includes spatial operations and non-spatial operations; The spatial operation is used to extract local feature maps and edge feature maps in the spatial domain of the encrypted telemedicine data. The non-spatial operation is used to perform a nonlinear transformation on the local feature map and the edge feature map to extract a one-dimensional feature vector.

2. The medical data inference and prediction method based on homomorphic encryption according to claim 1, characterized in that, The original telemedicine data of the target patient is obtained, and the original telemedicine data is encrypted using homomorphic encryption to obtain encrypted telemedicine data, including: The original remote medical data of the target patient is acquired through the acquisition device; The original telemedicine data is preprocessed to obtain preprocessed telemedicine data; The preprocessed telemedicine data is encrypted using homomorphic encryption to obtain encrypted telemedicine data.

3. The medical data inference and prediction method based on homomorphic encryption according to claim 2, characterized in that, The initialization convolutional neural network model includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer connected sequentially. The encrypted remote medical data is input into the inference and prediction model for inference and prediction. The model also includes: The optimized convolutional neural network model is trained to obtain an inference and prediction model; The encrypted telemedicine data is input into the inference prediction model for inference and prediction.

4. The medical data inference and prediction method based on homomorphic encryption according to claim 1, characterized in that, Optimizing the initial convolutional neural network model using a polynomial acceleration algorithm further includes: The model parameters are stored in the registers of the image processor and data exchange and transmission are performed using the shared memory of the image processor to optimize the data structure of the initialized convolutional neural network model.

5. The medical data inference and prediction method based on homomorphic encryption according to claim 1, characterized in that, Local feature maps and edge feature maps are extracted from the spatial domain of the encrypted telemedicine data, including: The remote medical data after parallel processing is input into the convolutional layer. The convolutional layer extracts features by sliding a two-dimensional convolutional kernel on the spatial domain of the encrypted remote medical data to obtain a local feature map. The local feature map is input into a pooling layer, which downsamples the local feature map in the spatial domain of the encrypted remote medical data to obtain an edge feature map.

6. A medical data inference and prediction device based on homomorphic encryption, characterized in that, include: The acquisition module is used to acquire the original telemedicine data of the target patient and encrypt the original telemedicine data through homomorphic encryption to obtain encrypted telemedicine data. The inference and prediction module is used to input the encrypted remote medical data into the inference and prediction model to perform inference and prediction, and obtain the medical prediction result for the target patient; the medical prediction result includes one or more physiological indicator data. The inference prediction model is obtained by training a convolutional neural network model optimized by a multinomial acceleration algorithm; The inference and prediction module includes: The selection submodule is used to select the initial convolutional neural network model; The optimization submodule is used to optimize the initial convolutional neural network model using a polynomial acceleration algorithm to obtain the optimized convolutional neural network model. The optimization submodule includes: The first optimization unit is used to allocate thread blocks for multiple encrypted remote medical data in the input layer of the initialization convolutional neural network model and to process the thread blocks in parallel. The second optimization unit is used to optimize the data processing of the initial convolutional neural network model by assigning a separate implementation strategy to each homomorphic encryption operation type. The implementation strategy includes spatial operations and non-spatial operations; the spatial operations are used to extract local feature maps and edge feature maps in the spatial domain of the encrypted telemedicine data; the non-spatial operations are used to perform nonlinear transformations on the local feature maps and edge feature maps to extract one-dimensional feature vectors.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the medical data inference and prediction method based on homomorphic encryption as described in any one of claims 1-5.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the medical data inference and prediction method based on homomorphic encryption as described in any one of claims 1-5.

Images