A medical data intelligent recognition method and system based on deep learning

Abstract

The embodiments of the present disclosure provide a method and system for intelligent identification of medical data based on deep learning. The method includes determining the data type of the medical data in response to the medical data received in real time; substituting the medical data of different data types into the encryption and decryption model, so that at least one user corresponding to the medical data and the current state of the medical data are determined according to the medical data, and the medical data is selectively encrypted or decrypted to obtain the processed medical data; substituting the processed medical data into the intelligent identification model based on deep learning to obtain the identification result matching the medical data, wherein the identification result includes the abnormal mark in the medical data. It ensures that the medical data matches the user, and ensures the data security and data integrity between the system and each acquisition device, so as to realize real-time detection and abnormal identification of medical data.

Description

Technical Field

[0001] The present disclosure relates to the technical field of medical data intelligent recognition based on deep learning, and in particular to a medical data intelligent recognition method and system based on deep learning. Background Art

[0002] As the population ages and the number of patients with chronic conditions increases, remote health monitoring becomes increasingly important for improving patient care and reducing healthcare costs.

[0003] However, in the process of simultaneously processing remote health monitoring data of multiple users, the medical health monitoring system in the prior art often suffers from multi-user data confusion and data misidentification, resulting in low data processing efficiency and untimely feedback on monitoring issues, and even delays in the user's condition.

[0004] Accordingly, this field needs a new deep learning-based intelligent medical data recognition solution to solve the above problems. Summary of the invention

[0005] The present invention provides a method and system for intelligent identification of medical data based on deep learning, which solves the technical problems in the prior art that, when the medical health monitoring system simultaneously processes the remote health monitoring data of multiple users, the data of multiple users are often easily confused and misidentified, resulting in low data processing efficiency and untimely feedback on monitoring problems, and even delays in the user's condition.

[0006] According to a first aspect of the present disclosure, a method for intelligent identification of medical data based on deep learning is provided. The method is applied to a medical data intelligent identification system, wherein the system interacts with multiple acquisition devices, and the method comprises:

[0007] In response to the medical data received in real time, determining a data type of the medical data;

[0008] Substituting medical data of different data types into the encryption and decryption model, so as to determine at least one user corresponding to the medical data and the current state of the medical data according to the medical data, and selectively encrypting or decrypting the medical data to obtain processed medical data;

[0009] Substitute the processed medical data into an intelligent recognition model based on deep learning to obtain a recognition result that matches the medical data, wherein the recognition result includes abnormal markers in the medical data.

[0010] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein in response to the medical data received in real time, determining the data type of the medical data includes:

[0011] In response to the medical data received in real time, determining device information of at least one acquisition device corresponding to the medical data;

[0012] The data type of the medical data is determined based on device information of at least one acquisition device corresponding to the medical data and the medical data.

[0013] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the medical data of different data types are substituted into the encryption and decryption model, so that at least one user corresponding to the medical data and the current state of the medical data are determined according to the medical data, and the medical data is selectively encrypted or decrypted, and the processed medical data includes:

[0014] Determining at least one user corresponding to the medical data based on the medical data of different data types and device information of at least one acquisition device corresponding to the medical data;

[0015] Based on the medical data and at least one user corresponding to the medical data, determine a current state of the medical data corresponding to each user, wherein the current state is an encrypted state or an original state;

[0016] Based on the current state of the medical data corresponding to each user, each medical data is selectively encrypted or decrypted to obtain processed medical data, including:

[0017] If the current state of the medical data corresponding to the user is the original state, encrypting the medical data to obtain encrypted medical data;

[0018] or,

[0019] If the current state of the medical data corresponding to the user is an encrypted state, decryption processing is performed on the medical data to selectively obtain the decrypted medical data.

[0020] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein encrypting the medical data to obtain the encrypted medical data comprises:

[0021] Preprocessing the medical data to obtain screened medical data;

[0022] The screened medical data is encrypted in sections to obtain data-encrypted medical data;

[0023] Performing key encryption on the encrypted medical data to obtain secondary encrypted medical data;

[0024] Based on the medical data, generating an original hash value corresponding to the medical data;

[0025] The encrypted medical data is composed of the twice-encrypted medical data and the original hash value corresponding to the medical data.

[0026] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the screened medical data is encrypted in sections to obtain encrypted medical data, including:

[0027] Segmenting the screened medical data to obtain segmented medical data, wherein the segmented medical data includes a plurality of medical data segments;

[0028] Based on the multiple medical data segments in the segmented medical data, respectively generate an AES key corresponding to each medical data segment;

[0029] Based on the AES key corresponding to each medical data segment, each medical data segment is encrypted respectively to obtain data-encrypted medical data.

[0030] According to the above aspects and any possible implementation manner, an implementation manner is further provided, wherein the encrypted medical data is encrypted with a key to obtain the secondary encrypted medical data, including:

[0031] Update the preset RSA public key;

[0032] Based on the preset RSA public key, the AES key corresponding to each medical data segment in the data-encrypted medical data is encrypted to obtain the encrypted AES key corresponding to each medical data segment;

[0033] The encrypted AES key corresponding to each medical data segment and each encrypted medical data segment constitute the secondary encrypted medical data.

[0034] According to the above aspects and any possible implementation, an implementation is further provided, wherein the medical data in an encrypted state includes an original hash value corresponding to the medical data, a plurality of encrypted medical data segments, and an encrypted AES key corresponding to each medical data segment, and the encryption and decryption model further stores a preset user database, wherein the preset user database includes a periodically updated preset RSA public key, identity information of each user, device information of a collection device matched to each user, and a periodically updated RSA private key of each user; the medical data is decrypted to selectively obtain the decrypted medical data, including:

[0035] Based on the user corresponding to the medical data and a preset user database, obtaining the user's latest RSA private key;

[0036] Based on the user's latest RSA private key, decrypt each encrypted AES key in the medical data to obtain an AES key corresponding to each encrypted AES key;

[0037] Based on each AES key and the encrypted medical data segment corresponding to each AES key, decrypt each encrypted medical data segment to obtain each medical data segment, and sort each medical data segment to form decrypted medical data;

[0038] Based on the decrypted medical data, generating a decrypted hash value corresponding to the decrypted medical data;

[0039] Based on the decrypted hash value corresponding to the decrypted medical data and the original hash value corresponding to the medical data in the encrypted medical data, the decrypted medical data is selectively used to form decrypted medical data for output.

[0040] According to the above aspects and any possible implementation, an implementation is further provided, wherein the network structure of the deep learning-based intelligent recognition model is a UDC-DNN network structure, specifically including: an input layer, a feature extraction module, a self-attention mechanism module, a deep residual convolutional neural network, a bidirectional long short-term memory network, a global average pooling layer, a fully connected layer and a regularization module, and an output layer; wherein,

[0041] The input layer is used to receive medical data, and the shape is (None, 64, 64, 3), representing multiple 64x64 RGB image sequences;

[0042] The feature extraction module includes: a convolution layer and a maximum pooling layer; the convolution layer uses a Conv2D layer for convolution operation, the convolution kernel size is 3x3, the activation function is ReLU, and the filling method is 'same'; the maximum pooling layer uses a MaxPooling2D layer for downsampling, and the pooling window size is 2x2;

[0043] The self-attention mechanism module uses a multi-head attention mechanism to perform self-attention calculations on features and extract more important features;

[0044] The deep residual convolutional neural network includes a residual block; the residual block contains two convolutional layers, the convolution kernel size is 3x3, the activation function is ReLU, the padding mode is 'same', and the input information is retained and activated by adding a skip connection;

[0045] The bidirectional long short-term memory network includes: a time distribution flattening layer and a bidirectional LSTM layer; the time distribution flattening layer is used to flatten each frame of the input sequence; the bidirectional LSTM layer uses 128 units, processes the sequence data bidirectionally, and returns the final state of the sequence;

[0046] The fully connected layer and regularization module includes a first fully connected layer, a Dropout layer, and a second fully connected layer; the first fully connected layer contains 256 units, and the activation function is ReLU; the Dropout layer uses a dropout rate of 0.5; the last layer of the second fully connected layer contains 10 units, and the activation function is Softmax, which is used for classification tasks;

[0047] The output layer is used to output prediction results.

[0048] According to the above aspects and any possible implementation, an implementation is further provided, wherein before substituting the processed medical data into the intelligent recognition model based on deep learning to obtain a recognition result matching the medical data, the method further trains the intelligent recognition model through the following steps:

[0049] Using multiple groups of processed medical data as multiple groups of training sample sets;

[0050] Preprocess each group of training sample sets to obtain each group of filtered training sample sets;

[0051] Perform feature extraction on each group of training sample sets after screening to obtain multiple groups of feature values ​​corresponding to each group of training sample sets;

[0052] Identify and classify multiple sets of feature values ​​corresponding to each set of training samples, and selectively mark each set of training samples as abnormal, so as to obtain recognition results that match each set of training samples;

[0053] Based on the abnormal labeling of each set of training samples, the training results of the intelligent recognition model are determined.

[0054] According to a second aspect of the present disclosure, a medical data intelligent recognition system based on deep learning is provided. The system interacts with multiple acquisition devices, the system includes an encryption and decryption model, an intelligent recognition model, and a control device, the encryption and decryption model also stores a preset user database, the control device includes a processor and a memory, the memory is suitable for storing multiple program codes, and the program code is suitable for being loaded and run by the processor to execute the above-mentioned medical data intelligent recognition method based on deep learning.

[0055] The above one or more technical solutions disclosed in the present invention have at least one or more of the following beneficial effects:

[0056] By using the medical data received in real time, the data type of the medical data is determined, and then the medical data of different data types are substituted into the encryption and decryption model, so that the medical data can be selectively encrypted or decrypted to obtain the processed medical data, thereby ensuring that the medical data matches the user and that the data security and integrity between the system and each acquisition device are ensured. The processed medical data is then substituted into an intelligent recognition model based on deep learning to obtain recognition results that match the medical data, so as to realize real-time detection and abnormality recognition of medical data, thereby avoiding the technical problems in the prior art that the medical health monitoring system, when simultaneously processing the remote health monitoring data of multiple users, often has the problem of easy confusion of multi-user data and misidentification of data, resulting in low data processing efficiency and untimely feedback on monitoring problems, and even delays in the user's condition.

[0057] In implementing the technical solution disclosed herein, the data type of the medical data is determined jointly by the medical data and the device information of the acquisition device corresponding to the medical data, thereby ensuring the accuracy of the data type of the medical data, thereby improving the encryption and decryption model's processing capability and efficiency for the medical data, and reducing the encryption and decryption model's processing error rate for the medical data.

[0058] In the technical solution implemented in the present disclosure, in the encryption processing of medical data by the encryption and decryption model, the interference values ​​and / or noise data in the medical data are identified and removed by preprocessing the medical data, so as to facilitate the subsequent further processing of the screened medical data. The screened medical data is segmented to obtain segmented medical data, so as to facilitate the subsequent independent encryption of the segmented medical data, thereby improving the security of the medical data, regenerating the AES key corresponding to the segmented medical data, and encrypting the medical data segment according to the AES key to obtain data-encrypted medical data, thereby realizing data encryption of the medical data.

[0059] In the technical solution implemented in the present invention, in the encryption processing of medical data by the encryption and decryption model, the AES key of each medical data segment is encrypted by the latest preset RSA public key to obtain the encrypted AES key, thereby forming the secondary encrypted medical data, thereby realizing the key encryption of the AES key of the medical data, thereby realizing the secondary encryption of the medical data, ensuring the security of the medical data, and then generating the original hash value for the medical data so that during the decryption operation, it is possible to determine whether the data has been tampered with based on the original hash value, thereby ensuring the integrity of the medical data.

[0060] In the technical solution implemented in the present disclosure, in the decryption processing of medical data by the encryption and decryption model, the user's latest RSA private key is obtained through the user corresponding to the medical data and the preset user database, and the encrypted AES key corresponding to the user is decrypted according to the user's latest RSA private key to obtain each AES key, thereby realizing one-time decryption of the encrypted medical data, and then the encrypted medical data segment corresponding to each AES key is decrypted according to the AES key to obtain each medical data segment and constitute the decrypted medical data, thereby realizing secondary decryption of the encrypted medical data and ensuring the security of the medical data, and then the corresponding decrypted hash value is generated according to the decrypted medical data, and whether the medical data has been tampered with is determined according to the decrypted hash value and the original hash value, and then the decrypted medical data is selectively used to constitute the decrypted medical data for output, thereby realizing verification of the integrity of the medical data.

[0061] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which:

[0063] Figure 1 A flowchart of a method for intelligent identification of medical data based on deep learning according to an embodiment of the present disclosure is shown;

[0064] Figure 2 A schematic diagram showing a model architecture of an encryption and decryption model of a medical data intelligent identification method based on deep learning according to an embodiment of the present disclosure;

[0065] Figure 3 A schematic diagram showing a model architecture of an intelligent recognition model of a medical data intelligent recognition method based on deep learning according to an embodiment of the present disclosure is shown;

[0066] Figure 4 A schematic diagram of the network structure of a deep learning network UDC-DNN according to an embodiment of the present disclosure is shown;

[0067] Figure 5 A block diagram of a medical data intelligent recognition system based on deep learning according to an embodiment of the present disclosure is shown.

[0068] List of reference numerals:

[0069] 200: medical data intelligent recognition system; 201: control device; 2011: processor; 2012: memory; 2013: program code; 202: encryption and decryption model; 2021: user database; 203: intelligent recognition model. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0071] In the description of the present disclosure, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as program codes, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable medium that can store program codes, such as a magnetic disk, a hard disk, an optical disk, a flash memory, a read-only memory, a random access memory, and the like. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0072] See attached Figure 1 , Figure 1 FIG. 1 is a flowchart of a method for intelligently identifying medical data based on deep learning according to an embodiment of the present disclosure. Figure 1 As shown, the medical data intelligent recognition method based on deep learning in the embodiment of the present disclosure is applied to a medical data intelligent recognition system based on deep learning, and the system interacts with multiple energy storage power stations. The method mainly includes the following steps S101-step S103.

[0073] Step S101: determining the data type of the medical data in response to the medical data received in real time;

[0074] Specifically, in some embodiments, in response to the medical data received in real time, determining the data type of the medical data includes:

[0075] In response to the medical data received in real time, determining device information of at least one acquisition device corresponding to the medical data;

[0076] The data type of the medical data is determined based on device information of at least one acquisition device corresponding to the medical data and the medical data.

[0077] Specifically, in some embodiments, the acquisition device may be a front-end sensor network, and the IoT device is equipped with the front-end sensor network framework, wherein the front-end sensor network may include a MAX30100 sensor for measuring blood oxygen level and heart rate; an AD8232 ECG sensor module for ECG signal data; and an MLX90614 sensor for a non-contact infrared sensor for detecting body temperature and collecting physiological data to achieve the acquisition of medical data, and the data type of the medical data may include blood oxygen level, heart rate, body temperature and electrocardiogram signal. The selection of the acquisition device and the data type of the medical data here is only an exemplary description. In actual testing, those skilled in the art can make a selection according to actual needs, as long as it is possible to obtain medical data of different data types through the acquisition device, and no further details will be given here.

[0078] Specifically, the acquisition devices are in direct or indirect contact with the body of a user wearing the acquisition devices, and are used for real-time monitoring and collection of medical data.

[0079] Specifically, in some embodiments, the communication method for the system to implement information interaction with multiple acquisition devices may be the MQTT protocol. The choice of communication method here is only an exemplary description. In actual testing, those skilled in the art can make a choice based on actual needs, which will not be repeated here.

[0080] In the above embodiment, the data type of the medical data is determined jointly by the medical data and the device information of the acquisition device corresponding to the medical data, thereby ensuring the accuracy of the data type of the medical data, thereby improving the processing capability and efficiency of the encryption and decryption model for the medical data, and reducing the error rate of the encryption and decryption model in processing the medical data.

[0081] Step S102: Substituting medical data of different data types into the encryption and decryption model, so as to determine at least one user corresponding to the medical data and the current state of the medical data according to the medical data, and selectively encrypting or decrypting the medical data to obtain processed medical data;

[0082] Specifically, in some embodiments, the medical data of different data types are substituted into the encryption and decryption model, so that at least one user corresponding to the medical data and the current state of the medical data are determined according to the medical data, and the medical data is selectively encrypted or decrypted to obtain the processed medical data, including:

[0083] Determining at least one user corresponding to the medical data based on the medical data of different data types and device information of at least one acquisition device corresponding to the medical data;

[0084] Based on the medical data and at least one user corresponding to the medical data, determine a current state of the medical data corresponding to each user, wherein the current state is an encrypted state or an original state;

[0085] Based on the current status of the medical data corresponding to each user, each medical data is selectively encrypted or decrypted to obtain processed medical data.

[0086] Specifically, in some embodiments, based on the current state of the medical data corresponding to each user, selectively encrypting or decrypting each medical data to obtain the processed medical data includes:

[0087] If the current state of the medical data corresponding to the user is the original state, encrypting the medical data to obtain encrypted medical data;

[0088] or,

[0089] If the current state of the medical data corresponding to the user is an encrypted state, decryption processing is performed on the medical data to selectively obtain the decrypted medical data.

[0090] Specifically, in some embodiments, encrypting the medical data to obtain the encrypted medical data includes:

[0091] Preprocessing the medical data to obtain screened medical data;

[0092] The screened medical data is encrypted in sections to obtain data-encrypted medical data;

[0093] Performing key encryption on the encrypted medical data to obtain secondary encrypted medical data;

[0094] Based on the medical data, generating an original hash value corresponding to the medical data;

[0095] The encrypted medical data is composed of the twice-encrypted medical data and the original hash value corresponding to the medical data.

[0096] Specifically, in some embodiments, preprocessing the medical data to obtain screened medical data includes:

[0097] Compressing the medical data to obtain compressed medical data;

[0098] Interference values ​​and / or noise data in the compressed medical data are identified and removed to obtain filtered medical data.

[0099] Specifically, in some embodiments, the pre-processing method for the medical data may be through an edge node, so as to realize data compression, identification and screening of the medical data respectively through the edge node, so as to reduce the burden of data transmission and ensure the real-time nature of the data. The choice of processing method here is only an exemplary description. In actual testing, technicians in this field can make a choice according to actual needs, which will not be repeated here.

[0100] Specifically, in some embodiments, the segmented encryption of the screened medical data to obtain encrypted medical data includes:

[0101] Segmenting the screened medical data to obtain segmented medical data, wherein the segmented medical data includes a plurality of medical data segments;

[0102] Based on the multiple medical data segments in the segmented medical data, respectively generate an AES key corresponding to each medical data segment;

[0103] Based on the AES key corresponding to each medical data segment, each medical data segment is encrypted respectively to obtain data-encrypted medical data.

[0104] Specifically, in some embodiments, the segmentation method of the segmentation processing can be the segmentation technology in the prior art, and the segment size of the segmentation processing can be 1024 bytes. The segmentation method of the segmentation processing and the selection of the segment size here are only exemplary. In actual testing, technical personnel in this field can make a selection according to actual needs, which will not be repeated here.

[0105] Specifically, in some embodiments, the method for generating the AES key corresponding to each medical data segment can be generated by using the AES algorithm, and the method for encrypting each medical data segment can be to encrypt each medical data segment using the AES algorithm. The generation method of the AES key of the medical data segment and the selection of the encryption method here are only exemplary descriptions. In actual testing, technical personnel in this field can make a choice according to actual needs, which will not be repeated here.

[0106] Specifically, the encrypted medical data segment in the data-encrypted medical data may include a random number, a ciphertext, and a verification tag to facilitate decryption of the data-encrypted medical data.

[0107] In the above embodiment, in the encryption processing of medical data by the encryption and decryption model, the interference values ​​and / or noise data in the medical data are identified and removed by pre-processing the medical data, so as to facilitate the subsequent further processing of the screened medical data. The screened medical data is segmented to obtain segmented medical data, so as to facilitate the subsequent independent encryption of the segmented medical data, thereby improving the security of the medical data, regenerating the AES key corresponding to the segmented medical data, and encrypting the medical data segment according to the AES key to obtain data-encrypted medical data, thereby realizing data encryption of the medical data.

[0108] Specifically, in some embodiments, the step of performing key encryption on the encrypted medical data to obtain the secondary encrypted medical data includes:

[0109] Update the preset RSA public key;

[0110] Based on the preset RSA public key, the AES key corresponding to each medical data segment in the data-encrypted medical data is encrypted to obtain the encrypted AES key corresponding to each medical data segment;

[0111] The encrypted AES key corresponding to each medical data segment and each encrypted medical data segment constitute the secondary encrypted medical data.

[0112] Specifically, in some embodiments, the encryption method for encrypting the AES key corresponding to each medical data segment in the data-encrypted medical data is to use the RSA algorithm to achieve high-security encryption of small data volumes. The choice of encryption method here is only an exemplary explanation. In actual testing, technicians in this field can make a choice according to actual needs, which will not be repeated here.

[0113] Specifically, in some embodiments, the method for generating the original hash value corresponding to the medical data can be to calculate the medical data through the SHA-256 hash function, so as to verify the integrity of the decrypted data through the generated hash value, thereby ensuring that the data has not been tampered with during transmission. The selection of the method for generating the original hash value here is only an exemplary explanation. In actual testing, technicians in this field can make a choice according to actual needs, which will not be repeated here.

[0114] In the above embodiment, in the encryption processing of medical data by the encryption and decryption model, the AES key of each medical data segment is also encrypted by the latest preset RSA public key to obtain the encrypted AES key, thereby constituting the secondary encrypted medical data, thereby realizing the key encryption of the AES key of the medical data, thereby realizing the secondary encryption of the medical data, ensuring the security of the medical data, and then generating the original hash value for the medical data so that during the decryption operation, it is possible to determine whether the data has been tampered with based on the original hash value, thereby ensuring the integrity of the medical data.

[0115] Step S103: Substitute the processed medical data into an intelligent recognition model based on deep learning to obtain a recognition result that matches the medical data, wherein the recognition result includes abnormal markers in the medical data to achieve real-time detection and abnormality recognition of the medical data.

[0116] Specifically, in some embodiments, the medical data in an encrypted state includes an original hash value corresponding to the medical data, multiple encrypted medical data segments, and an encrypted AES key corresponding to each medical data segment. The encryption and decryption model also stores a preset user database, and the preset user database includes a periodically updated preset RSA public key, identity information of each user, device information of a collection device matched to each user, and a periodically updated RSA private key of each user. The medical data is decrypted to selectively obtain the decrypted medical data, including:

[0117] Based on the user corresponding to the medical data and a preset user database, obtaining the user's latest RSA private key;

[0118] Based on the user's latest RSA private key, decrypt each encrypted AES key in the medical data to obtain an AES key corresponding to each encrypted AES key;

[0119] Based on each AES key and the encrypted medical data segment corresponding to each AES key, decrypt each encrypted medical data segment to obtain each medical data segment, and sort each medical data segment to form decrypted medical data;

[0120] Based on the decrypted medical data, generating a decrypted hash value corresponding to the decrypted medical data;

[0121] Based on the decrypted hash value corresponding to the decrypted medical data and the original hash value corresponding to the medical data in the encrypted medical data, the decrypted medical data is selectively used to form decrypted medical data for output.

[0122] Specifically, in some embodiments, the decrypted hash value corresponding to the decrypted medical data can be generated by calculating the decrypted medical data through the SHA-256 hash function, so as to compare the original hash value with the decrypted hash value to determine the integrity of the decrypted data, and then verify whether the data has been tampered with during transmission. The choice of the method for generating the decrypted hash value here is only an exemplary explanation. In actual testing, technicians in this field can make a choice according to actual needs, which will not be repeated here.

[0123] Specifically, in some embodiments, the selectively outputting the decrypted medical data as the decrypted medical data based on the decrypted hash value corresponding to the decrypted medical data and the original hash value corresponding to the medical data in the encrypted medical data includes:

[0124] Compare the decrypted hash value corresponding to the decrypted medical data with the original hash value corresponding to the medical data in the encrypted medical data:

[0125] If the decrypted hash value corresponding to the decrypted medical data is consistent with the original hash value corresponding to the medical data in the encrypted medical data, the verification is determined to be successful, and the integrity of the decrypted data is determined to be that the data has not been tampered with, and the decrypted medical data constitutes the decrypted medical data for output;

[0126] Otherwise, it is determined that the verification has failed, and the integrity of the decrypted data is determined to be that the data has been tampered with, and an abnormal signal indicating that the data has been tampered with is output.

[0127] like Figure 2 As shown, specifically, in some embodiments, taking the encryption processing or decryption processing in the encryption and decryption model as an example, the edge node, the AES algorithm, the RSA algorithm, and the SHA-256 hash function are respectively used for encryption processing or decryption processing, the processing steps of the model architecture of the encryption processing or decryption processing of the encryption and decryption model include:

[0128] Encryption Processing:

[0129] Pre-process the medical data through edge nodes to obtain filtered medical data;

[0130] Segmenting the screened medical data to obtain multiple medical data segments;

[0131] Generate an AES key corresponding to each medical data segment through the AES algorithm, and encrypt each medical data segment to obtain data-encrypted medical data;

[0132] Encrypt the AES key corresponding to each medical data segment through the RSA algorithm to obtain the corresponding encrypted AES key to form the secondary encrypted medical data;

[0133] The original hash value of the filtered medical data is calculated by the SHA-256 hash function.

[0134] Decryption process:

[0135] Determine the user's latest RSA private key based on the user corresponding to the medical data and the preset user database;

[0136] Decrypt each encrypted AES key using the RSA algorithm to obtain the corresponding AES key;

[0137] Decrypting the encrypted medical data segments corresponding to each AES key by using an AES algorithm to obtain each medical data segment, and sorting each medical data segment to form decrypted medical data;

[0138] The decrypted hash value of the decrypted medical data is calculated using the SHA-256 hash function, and the decrypted hash value and the original hash value are used to verify whether the data has been tampered with during transmission.

[0139] In the above embodiment, in the decryption processing of medical data by the encryption and decryption model, the user's latest RSA private key is obtained through the user corresponding to the medical data and the preset user database, and the encrypted AES key corresponding to the user is decrypted according to the user's latest RSA private key to obtain each AES key, thereby realizing one-time decryption of the encrypted medical data, and then the encrypted medical data segment corresponding to each AES key is decrypted according to the AES key to obtain each medical data segment and constitute the decrypted medical data, thereby realizing secondary decryption of the encrypted medical data, and then the corresponding decrypted hash value is generated according to the decrypted medical data, and whether the medical data has been tampered with is determined according to the decrypted hash value and the original hash value, and then the decrypted medical data is selectively outputted as decrypted medical data, thereby ensuring the integrity and security of the medical data.

[0140] Furthermore, in some embodiments, before substituting the processed medical data into the intelligent recognition model based on deep learning to obtain a recognition result matching the medical data, the method further trains the intelligent recognition model through the following steps:

[0141] Using multiple groups of processed medical data as multiple groups of training sample sets;

[0142] Preprocess each group of training sample sets to obtain each group of filtered training sample sets;

[0143] Perform feature extraction on each group of training sample sets after screening to obtain multiple groups of feature values ​​corresponding to each group of training sample sets;

[0144] Identify and classify multiple sets of feature values ​​corresponding to each set of training samples, and selectively mark each set of training samples as abnormal, so as to obtain recognition results that match each set of training samples;

[0145] Based on the abnormal labeling of each set of training samples, the training results of the intelligent recognition model are determined.

[0146] Deep learning algorithms can model and analyze data based on historical data and existing medical knowledge to identify abnormal patterns or outliers in the data. Machine learning (ML) networks preprocess data by performing preprocessing operations such as cleaning, denoising, and standardization on the collected medical data to improve data quality. Model performance is evaluated through methods such as cross-validation, loss function adjustment, and model fusion, and model parameters are continuously optimized to improve recognition accuracy.

[0147] Specifically, in some embodiments, during the model training process of the intelligent recognition model, the following steps are adopted:

[0148] Preprocessing each group of training sample sets to obtain each group of filtered training sample sets, wherein the multiple medical data in each group of filtered training sample sets are RGB images with a shape of (None, 64, 64, 3), that is, a 64x64 RGB image sequence;

[0149] The feature extraction module and the self-attention mechanism module are used to extract features from each set of training samples, and multiple sets of feature values ​​corresponding to each set of training samples are obtained;

[0150] The multiple sets of feature values ​​corresponding to each set of training samples are identified and classified through the identification and classification module, and each set of training samples is selectively marked as abnormal, so as to obtain the identification results matching each set of training samples;

[0151] Based on the abnormal labeling of each set of training samples, the training results of the intelligent recognition model are determined.

[0152] Specifically, in some embodiments, the feature extraction module can adopt a convolutional neural network, wherein the convolutional neural network includes a Conv2D convolutional layer and a MaxPooling2D maximum pooling layer, so as to perform a convolution operation on each group of training sample sets through the Conv2D convolutional layer to extract multiple feature vectors from each group of training sample sets, wherein the convolution kernel size is 3x3, the activation function is ReLU, and the filling method is 'same', and then each group of training sample sets after convolution is downsampled through the MaxPooling2D maximum pooling layer to obtain multiple groups of feature values ​​after nonlinear transformation, wherein the pooling window The size is 2x2 to reduce the spatial dimension of the feature vector, and then the Conv2D convolution layer is used to perform a convolution operation on the multiple groups of eigenvalues ​​after the nonlinear transformation to obtain convolution feature data, so as to extract the local features of the data, and extract the local features in the multiple groups of eigenvalues ​​of each training sample set to achieve temporary splitting of the multiple groups of eigenvalues, so as to facilitate the subsequent separate processing of each eigenvector in each group of eigenvalues, thereby achieving feature extraction for each group of training sample sets. The setting of the feature extraction module here is only an exemplary description. In actual tests, those skilled in the art can set it according to actual needs, which will not be repeated here.

[0153] Specifically, in some embodiments, the recognition and classification module can sequentially adopt an attention mechanism, a deep residual convolutional neural network, a bidirectional long short-term memory network, a global average pooling layer, a fully connected layer, and regularization to realize the recognition and classification of multiple groups of feature values ​​corresponding to each group of training sample sets. The setting of the recognition and classification module here is only an exemplary description. In actual testing, those skilled in the art can set it according to actual needs, which will not be repeated here.

[0154] Specifically, in some embodiments, the self-attention mechanism module may adopt a multi-head attention mechanism to perform self-attention calculations on the features of each group of training sample sets, and obtain the attention weights of each feature vector to determine the more important feature data in the features corresponding to each group of training sample sets, thereby achieving selective elimination of the features of each group of training sample sets. The setting of the self-attention mechanism module here is only an exemplary description. In actual testing, those skilled in the art can set it according to actual needs, which will not be repeated here.

[0155] Specifically, in some embodiments, the deep residual convolutional neural network may adopt a residual block, wherein the residual block includes two convolutional layers, wherein the convolution kernel size is 3x3, the activation function is ReLU, and the filling method is 'same'. Through the residual block, the jump connection between the feature vectors is added to realize the feature extraction capability of the model. The setting of the deep residual convolutional neural network here is only an exemplary description. In actual tests, those skilled in the art can set it according to actual needs, which will not be repeated here.

[0156] Specifically, in some embodiments, the bidirectional long short-term memory network can be composed of a time distribution flattening layer and a bidirectional LSTM layer. The bidirectional long short-term memory network is used to add a time series to each feature vector to obtain sequence data, and then the bidirectional information in the sequence data is extracted to obtain sequence data represented by bidirectional encoding, so as to improve the data processing capability of the model. The setting of the bidirectional long short-term memory network here is only an exemplary description. In actual tests, those skilled in the art can set it according to actual needs, which will not be repeated here.

[0157] Specifically, in some embodiments, the sequence data represented by the bidirectional encoding is reduced in dimension through a global average pooling layer to obtain a one-dimensional vector of each eigenvalue to achieve screening of the eigenvalues. In actual tests, those skilled in the art can set the global average pooling layer according to actual needs. As long as it is possible to reduce the dimension of the sequence data represented by the bidirectional encoding through the global average pooling layer to obtain a one-dimensional vector of each eigenvalue, it will not be repeated here.

[0158] Specifically, in some embodiments, the fully connected layer may include a first fully connected layer and a second fully connected layer, and the one-dimensional vector of the eigenvalue is linearly transformed and combined through the first fully connected layer to facilitate the subsequent classification of the eigenvalue, and the data output by the regularization is classified through the second fully connected layer, wherein the first fully connected layer contains 256 units, and the activation function is ReLU, and the second fully connected layer contains 10 units, and the activation function is Softmax, which is used for classification tasks. The setting of the fully connected layer here is only an exemplary description. In actual tests, those skilled in the art can set it according to actual needs, which will not be repeated here.

[0159] Specifically, in some embodiments, the regularization may be a Dropout layer, through which the one-dimensional data and its classification output by the first fully connected layer are screened to improve the accuracy of model classification and reduce the error rate of model classification, wherein the Dropout layer uses a dropout rate of 0.5 to prevent overfitting. The regularization setting here is only an exemplary description, and those skilled in the art can set it according to actual needs in actual testing, which will not be repeated here.

[0160] Specifically, in some embodiments, during the model training process of the intelligent recognition model, iterative training of the model can be achieved through adaptive parameter adjustment to accelerate the convergence speed and improve the training effect of the model.

[0161] Specifically, in some embodiments, after each model training of the intelligent recognition model, the training results are predicted by prediction technology in the prior art to obtain prediction results.

[0162] Specifically, in some embodiments, a cross entropy loss function is used in the process of identifying and classifying multiple groups of feature values ​​corresponding to each group of training sample sets, so as to focus on difficult samples by reducing the weights of samples that are easy to predict, wherein the cross entropy loss function formula used in the present invention is:

[0163] FL(p t )=-αy(1-p) γ log(p)-(1-α)(1-y)p γ log(1-p)+δ

[0164] Among them, FL(p t ) represents Focal Loss, which is a loss function used to deal with imbalance problems in classification tasks. y is the true label, p is the probability of predicting the positive class, δ is the cost factor that changes adaptively through forward feedback, α = 0.25, γ = 2;

[0165] Among them, the parameter α in the formula is an important parameter for balancing positive and negative samples. When the positive and negative samples in the data set are unbalanced, by adjusting the value of α, the penalty for fewer categories (such as positive samples) can be increased and the penalty for more categories (such as negative samples) can be reduced. The value of α is usually between [0,1], and the common setting is α=0.25. γ is an important parameter for adjusting the loss between easy-to-classify samples and difficult-to-classify samples. By adjusting the value of γ, the loss of easy-to-classify samples can be reduced and the loss of difficult-to-classify samples can be increased, so as to better learn difficult-to-classify samples. The value of γ is usually a non-negative number, generally ranging from 0 to 5, and the common setting is γ=2.

[0166] According to the above embodiments, the present invention targets unbalanced samples by using AdaptiveFocal Loss in deep learning, which is a cross-entropy loss function optimized on category-imbalanced data. It focuses on difficult samples by reducing the weights of easily predicted samples; through cost-sensitive learning, that is, explicitly adding cost factors to the loss function, different penalties are imposed on different categories of erroneous predictions, thereby achieving dynamic adjustment of loss weights and improving model performance.

[0167] Specifically, in some embodiments, determining the training result of the intelligent recognition model based on the abnormal labeling of each group of training sample sets includes:

[0168] Based on the abnormal marking of each group of training sample sets, determining the recognition result of each group of training sample sets;

[0169] Based on the recognition results of each set of training samples, the training results of the intelligent recognition model are determined:

[0170] If the training result of the intelligent recognition model is determined to be a training failure, the cost factor is adjusted, and based on the adjusted cost factor, the training of the intelligent recognition model is re-executed.

[0171] Specifically, in some embodiments, the model training process of the intelligent recognition model further includes the following steps:

[0172] Based on a preset cycle time, the intelligent recognition model is periodically retrained through cross-validation to improve the stability and generalization ability of the model.

[0173] Specifically, in some embodiments, the cross-validation may use the Stratified K-Fold method to divide the training sample set into multiple subsets, and perform multiple training and validation based on the multiple subsets. The selection of the cross-validation method here is only an exemplary description. In actual testing, those skilled in the art may make a selection based on actual needs, which will not be elaborated here.

[0174] like Figure 3 As shown, specifically, in some embodiments, taking the convolutional neural network, multi-head attention mechanism, deep residual convolutional neural network, bidirectional long short-term memory network, global average pooling layer, fully connected layer and regularization, adaptive parameter adjustment and loss function and Stratified K-Fold method as examples in the process of recognition, classification and verification of the training sample set by the intelligent recognition model, the processing steps of the model architecture of the recognition, classification and verification of the training sample set by the intelligent recognition model include:

[0175] Perform feature extraction on each set of training samples through a convolutional neural network to obtain multiple sets of feature values ​​in each set of training samples, wherein each set of feature values ​​includes a local feature vector and a general feature vector;

[0176] The attention mechanism is used to perform self-attention calculation on each group of feature values ​​to obtain the attention weight of each feature vector;

[0177] Through the deep residual convolutional neural network, jump connections are made between each feature vector;

[0178] A bidirectional long short-term memory network is used to add a time series to each feature vector and extract bidirectional information to obtain sequence data represented by bidirectional coding.

[0179] The sequence data represented by the bidirectional encoding is reduced in dimension through the global average pooling layer to obtain a one-dimensional vector for each eigenvalue;

[0180] Linearly transforming and combining the one-dimensional vectors of the eigenvalues ​​through the first fully connected layer;

[0181] The one-dimensional data and its classification output by the first fully connected layer are filtered through the Dropout layer;

[0182] The data output by regularization is classified through the second fully connected layer to obtain the classification result of each feature;

[0183] Adaptively adjust the model parameters through adaptive parameter adjustment to improve the training effect of the model, and predict the training results of the model to obtain the prediction results;

[0184] The prediction results are tested by the cross entropy loss function, and the adaptive change component factors are selectively adjusted according to the test results;

[0185] The training sample set is trained and validated multiple times through cross-validation to improve the stability and generalization ability of the model.

[0186] This system uses deep learning algorithms to intelligently identify medical data. The system uses the self-developed deep learning network UDC-DNN (Unbalanced Dataset Cross-fusion Deep Neural Network) based on unbalanced medical samples and an intelligent recognition algorithm combined with big data analysis technology. It uses data-driven feature extraction and pattern recognition to perform real-time anomaly detection on the massive data collected by medical sensors. The UDC-DNN network structure is as follows: Figure 4 As shown, the model contains the following main parts:

[0187] 1. Input layer:

[0188] -The input is medical data with a shape of (None, 64, 64, 3), representing multiple 64x64 RGB image sequences.

[0189] 2. Feature extraction module:

[0190] - Convolutional layer: Use Conv2D layer for convolution operation, with kernel size of 3x3, activation function of ReLU, and padding of 'same'.

[0191] -Maximum pooling layer: Use the MaxPooling2D layer for downsampling with a pooling window size of 2x2 to reduce the spatial dimension of the feature map.

[0192] 3. Self-attention mechanism module:

[0193] -Self-attention mechanism module (self_attention_block): Use the Multi-Head Attention mechanism to perform self-attention calculations on features and extract more important features.

[0194] 4. Deep Residual Convolutional Neural Network (Deep Residual CNN):

[0195] -Residual_block: Contains two convolutional layers, with a kernel size of 3x3, an activation function of ReLU, and padding of 'same'. By adding skip connections, the input information is retained and activated.

[0196] 5. Bidirectional Long Short-Term Memory Network (BiLSTM):

[0197] - Temporal flattening layer: flattens each frame of the input sequence.

[0198] - Bidirectional LSTM layer: uses 128 units, processes sequence data bidirectionally, and returns the final state of the sequence.

[0199] 6. Global average pooling layer (GlobalAveragePooling2D):

[0200] -Perform global average pooling on each feature map to reduce feature dimensions and preserve global information.

[0201] 7. Fully connected layer and regularization:

[0202] -Fully connected layer: contains 256 units and the activation function is ReLU.

[0203] -Dropout layer: Use a dropout rate of 0.5 to prevent overfitting.

[0204] -Fully connected layer: The last layer contains 10 units and the activation function is Softmax, which is used for classification tasks.

[0205] 8. Adaptive parameter adjustment and loss function:

[0206] -Adjust parameters based on judgment result feedback, use RLA (Reinforcement Learning Adaptive) mechanism and Adaptive Focal Loss function to dynamically adjust loss weights and improve model performance.

[0207] 9. Model verification and evaluation:

[0208] -Cross-validation (Stratified K-Fold cross-validation): Use the Stratified K-Fold method to divide the data into multiple subsets and perform multiple training and validation to ensure the stability and generalization ability of the model.

[0209] The adaptive adjustment mechanism of the model can dynamically optimize parameters according to feedback, improving the accuracy and robustness of the model in practical applications. Model verification and evaluation ensure the stability and generalization ability of the model on different data sets through cross-validation methods, so that the model has high accuracy and reliability when processing medical data. The model adopts a cross-fusion structure, which can effectively integrate data from different sources and properties, such as different medical images, physiological signals and text data. This cross-fusion helps to improve the generalization ability of the model for multi-source data, so as to better deal with the imbalance problem. By introducing the reinforcement learning agent RLA (Reinforcement Learning Agent) to dynamically adjust the network parameters, positive feedback is introduced to achieve the adjustment of the adaptive cost factor of the penalty function. The algorithm can automatically learn the characteristic patterns in the data, and continuously optimize the model through the reinforcement learning mechanism to improve the accuracy and real-time performance of anomaly detection. Through a large amount of data training and optimization, the model can gradually adapt to various complex scenarios and improve the accuracy and efficiency of recognition. This combination not only improves the intelligent level of medical data recognition, but also provides more possibilities for future medical research and applications. The deep reinforcement learning algorithm is applied to the anomaly detection of medical sensor data, which improves the accuracy and real-time performance of detection.

[0210] The experimental parameters are shown in Table 1, including the following: the datasets are the MIT-BIH arrhythmia database (for heartbeat type detection) and a synthetic body temperature dataset; the sensor types include ECG sensor, body temperature sensor, and pulse oximeter; the sampling frequencies are 360 ​​Hz (ECG) and 1 Hz (body temperature, blood oxygen); the labels cover five heartbeat types (normal heartbeat, supraventricular premature beat, premature heartbeat, fused ventricle, unclassifiable heartbeat), body temperature (fever / non-fever), heart rate, and blood oxygen level; the training data ratio is 80% training, 10% validation, and 10% testing; the learning rate is set to 0.001, the optimization algorithm is the Adam algorithm, the batch size is 32, the loss function is Focal Loss, and the evaluation indicators include accuracy, recall, F1 score, and AUC.

[0211] Table 1 Experimental parameters

[0212]

[0213]

[0214] The experimental environment is shown in Table 2, including the following: the programming language is Python 3.8, the deep learning frameworks are TensorFlow 2.x and Keras, the hardware uses NVIDIA GPU (Tesla V100), the operating system is Ubuntu 20.04 LTS, the dependent libraries include numpy, pandas, scikit-learn, matplotlib, seaborn, tensorflow and keras, and the IDE is Jupyter Notebook or PyCharm.

[0215] Table 2 Experimental environment

[0216]

[0217]

[0218] The training and optimization process is shown in Table 3, which includes the following: data preprocessing normalizes and denoises ECG, body temperature, and blood oxygen data (using wavelet transform and filtering technology); model training uses the above network structure to train the data, using the Adam optimizer and Adaptive Focal Loss loss function; model evaluation uses cross-validation, confusion matrix, ROC curve and other methods to evaluate model performance; model fusion uses ensemble learning methods to fuse the prediction results of multiple models to improve prediction accuracy; reinforcement learning is trained through historical data, continuously optimizes model parameters, and improves the model's ability to detect abnormal patterns.

[0219] Table 3 Training and optimization process

[0220]

[0221] According to the embodiments of the present disclosure, the following technical effects are achieved:

[0222] By using the medical data received in real time, the data type of the medical data is determined, and then the medical data of different data types are substituted into the encryption and decryption model, so that the medical data can be selectively encrypted or decrypted to obtain the processed medical data, thereby ensuring that the medical data matches the user and that the data security and integrity between the system and each acquisition device are ensured. The processed medical data is then substituted into an intelligent recognition model based on deep learning to obtain recognition results that match the medical data, so as to realize real-time detection and abnormality recognition of medical data, thereby avoiding the technical problems in the prior art that the medical health monitoring system, when simultaneously processing the remote health monitoring data of multiple users, often has the problem of easy confusion of multi-user data and misidentification of data, resulting in low data processing efficiency and untimely feedback on monitoring problems, and even delays in the user's condition.

[0223] Furthermore, the present disclosure also provides a medical data intelligent recognition system based on deep learning.

[0224] See attached Figure 5 , Figure 5 FIG. 1 is a block diagram of a medical data intelligent recognition system based on deep learning according to an embodiment of the present disclosure. Figure 5 As shown, the medical data intelligent recognition system 200 in the embodiment of the present disclosure interacts with multiple acquisition devices. The system includes an encryption and decryption model 202, an intelligent recognition model 203, and a control device 201. The encryption and decryption model 202 stores a preset user database 2021. The control device 201 includes a processor 2011 and a memory 2012. The memory 2012 can be configured to store a program code 2013 for executing the medical data intelligent recognition method based on deep learning of the above method embodiment. The processor 2011 can be configured to execute the program code 2013 in the memory 2012, and the program code 2013 includes but is not limited to executing the program code 2013 for the medical data intelligent recognition method based on deep learning of the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present disclosure. The control device 201 can be a control device formed by various electronic devices.

[0225] Specifically, the acquisition devices are in direct or indirect contact with the body of a user wearing the acquisition devices, and are used for real-time monitoring and collection of medical data.

[0226] Specifically, in some embodiments, the user database 2021 includes a periodically updated preset RSA public key, identity information of each user, device information of a collection device matched to each user, and a periodically updated RSA private key of each user.

[0227] In one implementation, the description of the specific implementation functions may refer to steps S101 to S103.

[0228] The medical data intelligent recognition system 200 is used to perform Figure 1 The embodiment of the medical data intelligent recognition method based on deep learning shown in the figure has similar technical principles, technical problems solved and technical effects produced. Technicians in this technical field can clearly understand that for the convenience and conciseness of description, the specific working process and related instructions of the medical data intelligent recognition system 200 can refer to the contents described in the embodiment of the medical data intelligent recognition method based on deep learning, and will not be repeated here.

[0229] It is understood by those skilled in the art that all or part of the processes in the method for implementing the above-mentioned embodiment of the present disclosure can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor 2011, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electric carrier signal and telecommunication signal.

[0230] Furthermore, the system of the present disclosure also includes a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present disclosure, the computer-readable storage medium can be configured to store a program code 2013 for executing the medical data intelligent recognition method based on deep learning of the above-mentioned method embodiment, and the program code 2013 can be loaded and run by the processor 2011 to implement the above-mentioned medical data intelligent recognition method based on deep learning. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present disclosure. The computer-readable storage medium can be a memory 2012 device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present disclosure is a non-temporary computer-readable storage medium.

[0231] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present disclosure, the physical devices corresponding to these modules may be the processor 2011 itself, or a part of the software, a part of the hardware, or a part of the combination of software and hardware in the processor 2011. Therefore, the number of each module in the figure is only schematic.

[0232] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor 2011, which can be a dedicated or general programmable processor 2011, which can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0233] The program code 2013 for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes 2013 can be provided to a processor 2011 or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code 2013, when executed by the processor 2011 or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code 2013 can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0234] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0235] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0236] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0237] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0238] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A medical data intelligent recognition method based on deep learning, characterized in that: The method is applied to a medical data intelligent recognition system, wherein the system interacts with multiple acquisition devices, and the method includes: In response to the medical data received in real time, determining a data type of the medical data; Substituting medical data of different data types into the encryption and decryption model, so as to determine at least one user corresponding to the medical data and the current state of the medical data according to the medical data, and selectively encrypting or decrypting the medical data to obtain processed medical data; Substituting the processed medical data into an intelligent recognition model based on deep learning to obtain a recognition result that matches the medical data, wherein the recognition result includes abnormal markers in the medical data; In response to the medical data received in real time, determining the data type of the medical data includes: In response to the medical data received in real time, determining device information of at least one acquisition device corresponding to the medical data; determining a data type of the medical data based on device information of at least one acquisition device corresponding to the medical data and the medical data; Substituting medical data of different data types into the encryption and decryption model, determining at least one user corresponding to the medical data and the current state of the medical data according to the medical data, and selectively encrypting or decrypting the medical data to obtain the processed medical data includes: Determining at least one user corresponding to the medical data based on the medical data of different data types and device information of at least one acquisition device corresponding to the medical data; Based on the medical data and at least one user corresponding to the medical data, determine a current state of the medical data corresponding to each user, wherein the current state is an encrypted state or an original state; Based on the current state of the medical data corresponding to each user, each medical data is selectively encrypted or decrypted to obtain processed medical data, including: If the current state of the medical data corresponding to the user is the original state, encrypting the medical data to obtain encrypted medical data; or, If the current state of the medical data corresponding to the user is an encrypted state, decrypting the medical data to selectively obtain the decrypted medical data; The encrypted medical data includes the original hash value corresponding to the medical data, multiple encrypted medical data segments, and the encrypted AES key corresponding to each medical data segment. The encryption and decryption model also stores a preset user database, which includes a periodically updated preset RSA public key, identity information of each user, device information of a collection device matched to each user, and a periodically updated RSA private key of each user. Based on the decrypted hash value corresponding to the decrypted medical data and the original hash value corresponding to the medical data in the encrypted medical data, the decrypted medical data is selectively used to form decrypted medical data for output.

2. The method according to claim 1, characterized in that The encrypting the medical data to obtain the encrypted medical data comprises: Preprocessing the medical data to obtain screened medical data; The screened medical data is encrypted in sections to obtain data-encrypted medical data; Performing key encryption on the encrypted medical data to obtain secondary encrypted medical data; Based on the medical data, generating an original hash value corresponding to the medical data; The encrypted medical data is composed of the twice-encrypted medical data and the original hash value corresponding to the medical data.

3. The method according to claim 2, characterized in that The medical data after screening is encrypted in sections to obtain encrypted medical data including: Segmenting the screened medical data to obtain segmented medical data, wherein the segmented medical data includes a plurality of medical data segments; Based on the multiple medical data segments in the segmented medical data, respectively generate an AES key corresponding to each medical data segment; Based on the AES key corresponding to each medical data segment, each medical data segment is encrypted respectively to obtain data-encrypted medical data.

4. The method according to claim 3, characterized in that The encrypted medical data is encrypted with a key to obtain the secondary encrypted medical data, including: Update the preset RSA public key; Based on the preset RSA public key, the AES key corresponding to each medical data segment in the data-encrypted medical data is encrypted to obtain the encrypted AES key corresponding to each medical data segment; The encrypted AES key corresponding to each medical data segment and each encrypted medical data segment constitute the secondary encrypted medical data.

5. The method according to claim 4, characterized in that The medical data in an encrypted state includes an original hash value corresponding to the medical data, a plurality of encrypted medical data segments, and an encrypted AES key corresponding to each medical data segment. The encryption and decryption model also stores a preset user database, and the preset user database includes a preset RSA public key that is periodically updated, identity information of each user, device information of a collection device matched to each user, and an RSA private key of each user that is periodically updated; The step of performing decryption processing on the medical data to selectively obtain the decrypted medical data comprises: Based on the user corresponding to the medical data and a preset user database, obtaining the user's latest RSA private key; Based on the user's latest RSA private key, decrypt each encrypted AES key in the medical data to obtain an AES key corresponding to each encrypted AES key; Based on each AES key and the encrypted medical data segment corresponding to each AES key, decrypt each encrypted medical data segment to obtain each medical data segment, and sort each medical data segment to form decrypted medical data; Based on the decrypted medical data, generating a decrypted hash value corresponding to the decrypted medical data; Based on the decrypted hash value corresponding to the decrypted medical data and the original hash value corresponding to the medical data in the encrypted medical data, the decrypted medical data is selectively used to form decrypted medical data for output.

6. The method according to claim 1, characterized in that in, The network structure of the intelligent recognition model based on deep learning is a UDC-DNN network structure, which specifically includes: an input layer, a feature extraction module, a self-attention mechanism module, a deep residual convolutional neural network, a bidirectional long short-term memory network, a global average pooling layer, a fully connected layer and a regularization module, and an output layer; wherein, The input layer is used to receive medical data, and its shape is (None, 64, 64, 3), which represents multiple 64x64 RGB image sequences; The feature extraction module includes: a convolution layer and a maximum pooling layer; the convolution layer uses a Conv2D layer for convolution operation, the convolution kernel size is 3x3, the activation function is ReLU, and the filling method is 'same'; the maximum pooling layer uses a MaxPooling2D layer for downsampling, and the pooling window size is 2x2; The self-attention mechanism module uses a multi-head attention mechanism to perform self-attention calculations on features and extract more important features; The deep residual convolutional neural network includes a residual block; the residual block contains two convolutional layers, the convolution kernel size is 3x3, the activation function is ReLU, the padding mode is 'same', and the input information is retained and activated by adding a skip connection; The bidirectional long short-term memory network includes: a time distribution flattening layer and a bidirectional LSTM layer; the time distribution flattening layer is used to flatten each frame of the input sequence; the bidirectional LSTM layer uses 128 units, processes the sequence data bidirectionally, and returns the final state of the sequence; The fully connected layer and regularization module includes a first fully connected layer, a Dropout layer, and a second fully connected layer; the first fully connected layer contains 256 units, and the activation function is ReLU; the Dropout layer uses a dropout rate of 0.5; the last layer of the second fully connected layer contains 10 units, and the activation function is Softmax, which is used for classification tasks; The output layer is used to output prediction results.

7. The method according to claim 1 or 6, characterized in that: Before substituting the processed medical data into the intelligent recognition model based on deep learning to obtain the recognition result matching the medical data, the method further trains the intelligent recognition model through the following steps: Using multiple groups of processed medical data as multiple groups of training sample sets; Preprocess each group of training sample sets to obtain each group of filtered training sample sets; Perform feature extraction on each group of training sample sets after screening to obtain multiple groups of feature values ​​corresponding to each group of training sample sets; Identify and classify multiple sets of feature values ​​corresponding to each set of training samples, and selectively mark each set of training samples as abnormal, so as to obtain recognition results that match each set of training samples; Based on the abnormal labeling of each set of training samples, the training results of the intelligent recognition model are determined.

8. A medical data intelligent recognition system based on deep learning, characterized in that: The system interacts with multiple acquisition devices, and includes an encryption and decryption model, an intelligent recognition model, and a control device. The encryption and decryption model also stores a preset user database. The control device includes a processor and a memory. The memory is suitable for storing multiple program codes, and the program code is suitable for being loaded and run by the processor to execute the medical data intelligent recognition method based on deep learning as described in any one of claims 1 to 7.

Images