A joint prediction method, system and medium for medical data

Abstract

The present invention provides a joint prediction method, system and medium for medical data, the method comprising: obtaining a data set of a medical data center; inputting the data set of the medical data center into a pre-trained data prediction model to obtain a local prediction result of the data set of the medical data center; using the quantum local data to train quantum node weights, and performing quantum joint reasoning calculations based on the weights and the local prediction results of the data set of the medical data center to obtain a joint prediction result of the data set of the medical data center. The present invention performs local prediction on the data set of the medical data center, avoiding the external transmission of quantum local data and parameters of the data prediction model during the training phase, and will not cause data privacy leakage, let alone face the risk of gradient attacks; no data transmission is performed, and only one quantum joint reasoning is performed during the joint reasoning phase, and there will be no performance loss on non-independent and identically distributed data sets.

Description

Technical Field

[0001] The present invention relates to the technical field of data privacy protection, and in particular to a joint prediction method, system and medium for medical data. Background Art

[0002] In the field of big data, fuzzy neural networks have attracted extensive attention and research due to their highly interpretable characteristics and reasonable accuracy in processing complex data sets. Their highly interpretable characteristics give fuzzy neural networks a unique advantage in understanding model decision-making processes and explaining complex data relationships. By introducing fuzzy reasoning mechanisms, neural networks are able to handle fuzziness and uncertainty, making modeling of complex data patterns more flexible. This flexibility is not only reflected in the performance of the model, but also provides interpretability in explaining the logic behind model behavior and decisions.

[0003] However, the high interpretability of fuzzy neural networks also brings about their high complexity. In the fuzzy layer of the fuzzy neural network, multiple membership function values ​​are calculated for each attribute of the feature, and the membership function values ​​of each attribute are combined through if-then rules. When the dimension of the data set increases, more membership functions need to be added to ensure the interpretability of the fuzzy neural network, which will bring an exponential increase in the complexity of the algorithm. In order to solve this problem, quantum neural networks and fuzzy neural networks can be combined. Quantum neural networks have the characteristics of parallelism, which makes quantum neural networks have significant advantages in processing high-dimensional data. This neural network that combines quantum neural networks with fuzzy neural networks is called a quantum fuzzy neural network. Quantum fuzzy neural networks have alleviated the high complexity problem of fuzzy neural networks to a certain extent.

[0004] In recent years, quantum fuzzy neural networks have also shown broad application prospects in the fields of medicine and social networking. However, with the accumulation of big data, quantum fuzzy neural networks inevitably need to solve the problem of data privacy protection. In recent years, the use of the Internet has gradually increased, and the number of IoT devices has increased significantly, and more sensitive data has been collected and analyzed. These sensitive and valuable data may be leaked and used by hackers for threats and blackmail. Encrypting private data is one of the solutions to protect privacy, but this method often sacrifices some features of the data, causing subsequent data analysis results to deviate. Another solution is to use a centralized distributed learning method to analyze data. However, this solution often requires frequent data exchange between the central node and other nodes. This will create the risk of data leakage and reduce the efficiency of data analysis. Therefore, a new privacy protection algorithm is needed to solve these problems.

[0005] Quantum federated learning enhances data privacy by mixing quantum classical machine learning models in a distributed environment. Studies have shown that it converges in fewer communication rounds than centralized models. So far, quantum federated learning has been widely used in data privacy protection. However, since these quantum federated learning algorithms need to share parameters, the overall learning efficiency of the algorithm will decrease. In addition, there is also the risk of gradient attacks. On the other hand, some quantum federated learning algorithms will also suffer performance losses on non-independent and identically distributed data sets. Therefore, the current quantum federated learning algorithms still need to be improved to alleviate the losses caused by these problems. Summary of the invention

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a joint prediction method, system and medium for medical data, which can overcome the problems of insufficient privacy protection of data in the medical data center during transmission, affecting data integrity and low efficiency, and does not require shared parameters, and has a high overall learning efficiency.

[0007] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0008] In one aspect, the present invention provides a joint prediction method for medical data, comprising:

[0009] Get the data set from the Medical Data Center;

[0010] Inputting the data set of the medical data center into a pre-trained data prediction model to obtain a local prediction result of the data set of the medical data center;

[0011] The training of the data prediction model includes:

[0012] Acquire quantum local data, use the quantum local data as a training sample of a data prediction model, perform fuzzification processing and intersection operation on the data, and obtain complete quantum local data; mine hidden features in the complete quantum local data through a quantum circuit to obtain a final state of the quantum local data; calculate an expected value of the quantum circuit using the final state of the quantum local data, and update the parameters of the data prediction model using the expected value to obtain a trained data prediction model;

[0013] The quantum local data is used to train quantum node weights, and quantum joint reasoning calculations are performed based on the quantum node weights and local prediction results of the data set of the medical data center to obtain the joint prediction results of the data set of the medical data center.

[0014] Optionally, using the quantum local data to train quantum node weights includes:

[0015] A density estimator of a quantum node is trained by using the quantum local data, a density distribution value of the quantum local data is calculated by using the density estimator, and a weight of the quantum node is calculated according to the density distribution value.

[0016] Optionally, the calculation formula of the quantum node weight is:

[0017] ;

[0018] In the formula, For the The weight of a quantum node; For the The density weight of quantum nodes; For the The density weight of quantum nodes; To traverse the sequence; is the number of quantum nodes; For the A density estimator for quantum nodes; For the A density estimator for quantum nodes; It is the data to be inferred in the joint reasoning stage.

[0019] Optionally, the calculation formula for performing quantum joint reasoning calculation is:

[0020] ;

[0021] In the formula, Joint prediction results for the dataset of the medical data center; For quantum nodes; For the Local prediction results for datasets from medical data centers.

[0022] Optionally, the fuzzy processing and intersection operation of the data include:

[0023] Using a fuzzy module, the quantum local data is fuzzified to obtain fuzzified quantum local data;

[0024] The fuzzified quantum local data is subjected to an intersection operation using a quantum T-norm layer to obtain complete quantum local data.

[0025] Optionally, the calculation formula for the final state of the quantum local data is:

[0026] ;

[0027] In the formula, For the The final state of the quantum local data of samples; For the The entire data prediction model of quantum nodes; For the The parameter set of the revolving door in the data prediction model of the quantum nodes; For the The initial state of the quantum local data of the samples.

[0028] Optionally, the expected value of the quantum circuit is as follows:

[0029] ;

[0030] ;

[0031] In the formula, For the The expected value of the quantum circuit of samples; For the The conjugate transpose of the final state of the quantum local data of samples; For the The final state of the quantum local data of samples; is the measurement operator.

[0032] Optionally, the calculation formula for updating the parameters of the data prediction model by the expected value is:

[0033] ;

[0034] In the formula, In order to seek partial guidance; For the The loss function of a quantum node; For the The parameters of the fuzzy processing of quantum nodes; For the Parameters of the data prediction model for quantum nodes; For the The quantum local data of each quantum node.

[0035] In a second aspect, the present invention provides a computer system, comprising:

[0036] Memory, for storing computer instructions;

[0037] A processor is used to execute the computer instructions to implement the steps of the joint prediction method of medical data described in the first aspect.

[0038] In a third aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method for joint prediction of medical data described in the first aspect.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention performs local prediction on the data set of the medical data center through the trained data prediction model, avoiding the external transmission of quantum local data and parameters of the data prediction model in the training phase, and can overcome the insufficient privacy protection of the data in the medical data center during the transmission process. There is no need to share parameters, the overall learning efficiency is high, data privacy leakage will not occur, and there is no risk of gradient attack. No data transmission is performed, and only one quantum joint reasoning is performed in the joint reasoning phase, which avoids the problems of affecting data integrity and low efficiency, and has a high degree of expression ability for complex data, and there will be no performance loss on non-independent and identically distributed data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 FIG. 1 is a schematic diagram of a process of a joint prediction method for medical data of the present invention in one embodiment;

[0042] Figure 2 FIG. 1 is a flow chart of another embodiment of the combined prediction method for medical data of the present invention;

[0043] Figure 3 Shown is a schematic diagram of the structure of a data prediction model of the present invention in one embodiment;

[0044] Figure 4 Shown is a schematic diagram of the structure of a quantum circuit in a data prediction model of the present invention in one embodiment. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.

[0046] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.

[0047] Example 1

[0048] like Figure 1 As shown, this embodiment introduces a joint prediction method for medical data, which specifically includes:

[0049] Get the Medical Data Center dataset.

[0050] The data set of the medical data center is input into a pre-trained data prediction model to obtain a local prediction result of the data set of the medical data center.

[0051] The training of the data prediction model includes:

[0052] The method comprises obtaining quantum local data, taking the quantum local data as a training sample of a data prediction model, performing fuzzification processing and intersection operation on the data, and obtaining complete quantum local data; mining hidden features in the complete quantum local data through a quantum circuit, and obtaining a final state of the quantum local data; calculating an expected value of the quantum circuit using the final state of the quantum local data, and updating parameters of the data prediction model using the expected value, and obtaining a trained data prediction model.

[0053] The quantum local data is used to train quantum node weights, and quantum joint reasoning calculations are performed based on the quantum node weights and local prediction results of the data set of the medical data center to obtain the joint prediction results of the data set of the medical data center.

[0054] When this embodiment is applied, the trained data prediction model is used to perform local prediction on the data set of the medical data center, thereby avoiding the external transmission of quantum local data and parameters of the data prediction model during the training phase, and can overcome the insufficient privacy protection of the data in the medical data center during transmission. There is no need to share parameters, the overall learning efficiency is high, data privacy leakage will not occur, and there is no risk of gradient attack. No data transmission is performed, and only one quantum joint reasoning is performed in the joint reasoning phase, which avoids problems affecting data integrity and inefficiency, and has a high degree of expression ability for complex data, and there will be no performance loss on non-independent and identically distributed data sets.

[0055] Example 2

[0056] like Figure 2 As shown, based on Example 1, this embodiment further performs the following designs, specifically including:

[0057] Get the Medical Data Center dataset.

[0058] The data set of the medical data center is input into a pre-trained data prediction model to obtain a local prediction result of the data set of the medical data center.

[0059] The training of the data prediction model includes:

[0060] Obtain quantum local data, use the quantum local data as training samples for the data prediction model, perform fuzzy processing and intersection operations on the data, and obtain complete quantum local data. Specifically:

[0061] like Figure 3 As shown, the fuzzy module is used to perform a fuzzification operation on the quantum local data to obtain the fuzzified quantum local data. The fuzzy module can be a fuzzy layer of a fuzzy neural network, specifically including:

[0062] In the In the data prediction model of quantum nodes, the quantum local data set to be trained is , is the size of the data set. The data samples are , is the size of each data feature. Assume that the membership function set used is , among which, The membership function group is , for The number of membership functions in . The Attributes correspond membership functions. Then there are fuzzy sets:

[0063] ;

[0064] ;

[0065]

[0066] ;

[0067]

[0068] ;

[0069] Through the above formula, we get of fuzzy sets, a total of fuzzy sets. From its Select one fuzzy set from the fuzzy sets and form a sequence of all the selected fuzzy sets . is the number of the fuzzy set sequence, , , is the number of quantum bits; for Selected fuzzy sets.

[0070] Next, The membership function values ​​of all fuzzy sets in are calculated. Let The value of the membership function is . The calculated values ​​of all membership functions form a new feature , which is the fuzzified quantum local data. All the membership functions in the fuzzy set sequence are calculated, and we get .

[0071] Using the quantum T-norm layer, the fuzzified quantum local data is subjected to an intersection operation to obtain complete quantum local data. Specifically, it includes:

[0072] Obtained the blurred quantum local data Afterwards, these data features need to be intersected. The T norm is generally used to refer to the intersection operation in the fuzzy system. The specific implementation is shown in the formula:

[0073] ;

[0074] This is the general formula for the T-norm. Let's choose the Algebraic product as the function of the T-norm. Then we have:

[0075] ;

[0076] This formula gives the exact calculation formula of the T norm. In order to use quantum bits to transmit data later, the present invention will implement a quantum T norm method with the same effect as this formula in the quantum circuit.

[0077] First, The membership function values ​​in are encoded into the quantum circuit. Gates are used to perform encoding operations. In quantum circuits, The matrix representation of the gate is:

[0078] ;

[0079] in, is a variable parameter. For the base state For a single quantum bit, the main steps of encoding are as follows:

[0080] ;

[0081] ;

[0082] It should be noted that, since the value of the membership function is always greater than or equal to 0 in fuzzy logic, is established. To encode the state of the quantum bit, Bring in of In, we get From the formula we can see The value is encoded in of amplitude.

[0083] The next step is to implement quantum T-norm operations. This can be achieved by using the Toffoli gate in the quantum circuit. The Toffoli gate requires two qubits as control bits and one qubit as the controlled bit. When the control bits are all 1, the controlled bit will flip. For example, Under the action of Toffoli gate, it will become , There will be no change under the action of the Toffoli gate. For any two single-qubit states and The quantum T-norm operation is , use the Toffoli gate to achieve, such as the formula:

[0084] ;

[0085] set up , ,get:

[0086] ;

[0087] In the results obtained, the third quantum bit The probability of Then, for the encoded Perform quantum T-norm operation to obtain complete quantum local data:

[0088] ;

[0089] It can be found that The last qubit middle The probability is:

[0090] ;

[0091] This shows that quantum T-norm can also realize T-norm operation on fuzzy sets. Put it into the same line, and we get .

[0092] The final state of the quantum local data is obtained by mining the hidden features in the complete quantum local data through quantum circuits. The circuit diagram structure of the quantum circuit is as follows: Figure 4 shown.

[0093] For the fuzzified quantum local data, further analysis is required according to the type of task. The data prediction model can be a quantum neural network, which can discover the hidden features in the data by parameterizing quantum circuits. Depending on the type of task, the structure of the quantum neural network will also be different.

[0094] The quantum neural network structure in this embodiment is designed based on the classification task. The quantum neural network is parameterized as a quantum circuit with a total of L layers. The structure of each layer is as follows: Figure 4 Each layer contains RX gates and RZ gates, and a set of controlled NOT gates on adjacent qubits and on the last and first qubits. Indicates The entire data prediction model of quantum nodes, is the parameter set of the revolving door in the data prediction model. Then:

[0095] ;

[0096] For the The final state of the quantum local data of samples; For the The initial state of the quantum local data of the samples.

[0097] The expected value of the quantum circuit is calculated using the final state of the quantum local data, and the parameters of the data prediction model are updated by the expected value to obtain a trained data prediction model. Specifically, it includes:

[0098] use As a measurement operator, the expected value of the entire quantum circuit is measured for:

[0099] ;

[0100] in:

[0101] ;

[0102] In the formula, For the The conjugate transpose of the final state of the quantum local data of samples. and is the Dirac symbol, which consists of two parts: the left arrow (bra) and the right arrow (ket), respectively expressed in angle brackets. and The left vector is used to represent the conjugate transpose of a state, while the right vector is used to represent a specific state vector.

[0103] Will Substitute it into the activation function to get the predicted result. Calculate the loss value through the loss function and update the parameters of the data prediction model. It is worth noting that The gradient of needs to be solved by the quantum shift rule. The specific formula is as follows:

[0104] ;

[0105] In the formula, In order to seek partial guidance; For the The loss function of a quantum node; For the The parameters of the fuzzy processing of quantum nodes; For the Parameters of the data prediction model for quantum nodes; For the The quantum local data of each quantum node.

[0106] Finally, the data prediction model of each quantum node is trained. During the entire local data prediction model training process, no data is shared between quantum nodes. Therefore, at this stage, the privacy of the data prediction model in the local quantum node is absolutely safe.

[0107] The quantum local data is used to train quantum node weights, and quantum joint reasoning calculations are performed based on the quantum node weights and local prediction results of the data set of the medical data center to obtain the joint prediction results of the data set of the medical data center.

[0108] The weights are trained using the training samples of the data prediction model, including:

[0109] The density estimator of the quantum node is trained by the quantum local data, the density distribution value of the quantum local data is calculated by the density estimator, and the weight of the quantum node is calculated according to the density distribution value. Specifically, it includes:

[0110] In this embodiment, each quantum node first trains a density estimator with quantum local data to obtain the density distribution of quantum local data. The density estimator uses a kernel density estimator with a Gaussian function as the kernel function. The trained density estimator is used to estimate the density distribution of new data, and the calculated value is used to determine the weight of each quantum node.

[0111] In the In quantum nodes, the quantum local data set to be trained is , is the size of the data set; let There is a trained density estimator in each node ;set up , , is the number of quantum nodes. The weight of a quantum node The calculation of is as follows:

[0112] ;

[0113] In the formula, For the The density weight of quantum nodes; For the The density weight of quantum nodes; To traverse the sequence; is the number of quantum nodes; For the A density estimator for quantum nodes; For the A density estimator for quantum nodes; It is the data to be inferred in the joint reasoning stage.

[0114] Different from the traditional federated learning algorithm, in this embodiment, no data transmission is performed between different nodes, and only quantum joint reasoning is performed once in the reasoning stage.

[0115] The data to be inferred in the joint inference phase Use density matrix representation, denoted as pure state . Assume that in each quantum node For the The local prediction results of the dataset of a medical data center. , . Then we have:

[0116] ;

[0117] In the formula, Joint prediction results for the dataset of the medical data center.

[0118] In a specific embodiment, the calculation process of quantum joint reasoning is given and its rationality is proved. First, the quantum nodes in the training phase are expressed in the form of density matrix, which can facilitate the subsequent description of the algorithm. The density matrix of the quantum local data set to be trained by quantum nodes is expressed as .

[0119] With a set of orthogonal primitives Represents each quantum node, then the joint density matrix of the quantum local data set and the quantum node can be expressed as .

[0120] First Finding the deviation can get the average density matrix of the data :

[0121] ;

[0122] The conditional density matrix can describe the data at a given quantum node. For the The projection measurement operator of quantum nodes satisfies . Then the conditional density matrix is:

[0123] ;

[0124] because is a set of orthogonal primitives, so ; hour, ; hour, .because ,therefore ;right Finding the deviation trace can obtain the quantum local data set: .

[0125] Next, let the projection operator , the density estimator is expressed as .Will Expand it and get:

[0126] .

[0127] Therefore, in this embodiment, multiple pre-trained data prediction models can correctly analyze and predict the data through quantum joint reasoning, which proves that it is feasible to only train data prediction models locally to perform data reasoning and prediction. This method greatly enhances the protection of privacy.

[0128] Example 3

[0129] This embodiment introduces a computer system, including:

[0130] Memory, for storing computer instructions;

[0131] A processor is used to execute the computer instructions to implement the steps of the joint prediction method for medical data described in Example 1 or 2.

[0132] Example 4

[0133] This embodiment introduces a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the steps of the joint prediction method for medical data described in Embodiment 1 or 2 are implemented.

[0134] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0135] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0136] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0137] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0138] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, which all fall within the protection of the present invention.

Claims

1. A joint prediction method for medical data, characterized in that: include: Get the data set from the Medical Data Center; Inputting the data set of the medical data center into a pre-trained data prediction model to obtain a local prediction result of the data set of the medical data center; The training of the data prediction model includes: Acquire quantum local data, use the quantum local data as a training sample of a data prediction model, perform fuzzification processing and intersection operation on the data, and obtain complete quantum local data; mine hidden features in the complete quantum local data through a quantum circuit to obtain a final state of the quantum local data; calculate an expected value of the quantum circuit using the final state of the quantum local data, and update the parameters of the data prediction model using the expected value to obtain a trained data prediction model; Using the quantum local data to train quantum node weights, performing quantum joint reasoning calculations based on the quantum node weights and local prediction results of the data set of the medical data center, and obtaining joint prediction results of the data set of the medical data center; Using the quantum local data to train quantum node weights includes: Training a density estimator of a quantum node through the quantum local data, calculating a density distribution value of the quantum local data using the density estimator, and calculating a weight of the quantum node according to the density distribution value; The calculation formula of the quantum node weight is: In the formula, λ h is the weight of the hth quantum node; μ h is the density weight of the hth quantum node; μ t is the density weight of the tth quantum node; t is the traversal sequence; N is the number of quantum nodes; E h (X als ) is the density estimator of the hth quantum node; E t (X als ) is the density estimator of the tth quantum node; X als The data to be inferred in the joint reasoning phase; The calculation formula for performing quantum joint reasoning calculation is: In the formula, P(X als ) is the joint prediction result of the data set of the medical data center; h is the quantum node; F h (X als ) is the local prediction result of the data set of the h-th medical data center; The calculation formula for updating the parameters of the data prediction model by the expected value is: In the formula, To seek partial derivative; Loss h (·) is the loss function of the hth quantum node; ω h * is the parameter of fuzzy processing of the hth quantum node; θ h * is the parameter of the data prediction model of the hth quantum node; X h is the quantum local data of the hth quantum node.

2. The joint prediction method of medical data according to claim 1, characterized in that: The fuzzy processing and intersection operation of the data include: Using a fuzzy module, the quantum local data is fuzzified to obtain fuzzified quantum local data; The fuzzified quantum local data is subjected to an intersection operation using a quantum T-norm layer to obtain complete quantum local data.

3. The joint prediction method of medical data according to claim 1, characterized in that: The calculation formula for the final state of the quantum local data is: In the formula, is the final state of the quantum local data of the i-th sample; U(θ h * ) is the entire data prediction model of the hth quantum node; θ h * is the parameter set of the revolving gate in the data prediction model of the hth quantum node; is the initial state of the quantum local data of the i-th sample.

4. The joint prediction method for medical data according to claim 1, characterized in that: The expected value of the quantum circuit is as follows: In the formula, r i is the expected value of the quantum circuit of the i-th sample; is the conjugate transpose of the final state of the quantum local data of the i-th sample; is the final state of the quantum local data of the i-th sample; is the measurement operator.

5. A computer system, characterized in that: include: Memory, for storing computer instructions; A processor, configured to execute the computer instructions to implement the steps of the method for joint prediction of medical data according to any one of claims 1 to 4.

6. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the steps of the joint prediction method for medical data described in any one of claims 1 to 4 are implemented.

Images