An artificial intelligence medical auxiliary data processing method and system

Abstract

The present invention relates to an artificial intelligence medical auxiliary data processing method and system, the method comprising: constructing a medical data processing model, the medical data processing model comprising a K+1 channel feature fusion network model and a machine learning model; constructing a dynamic weight model; the input of the dynamic weight model is an L-element input vector, and the output is a K-element output vector; the K-element output vector is the input of the K+1 channel of the medical data processing model; obtaining medical data to be processed, constructing K modal medical data inputs based on the medical data to be processed; inputting them into the 1st to K channels of the medical data processing model respectively, and obtaining corresponding auxiliary data. The present invention realizes rapid training of medical data models under the support of less medical data and training complexity through the splitting of data dimensions, and improves the auxiliary efficiency of medical data.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to an artificial intelligence medical-assisted data processing method and system. Background Art

[0002] With the continuous advancement of technology, artificial intelligence technology is shifting from general applications to industry-specific applications, and is more deeply integrated into specific business scenarios in all walks of life, realizing applications that deeply integrate artificial intelligence technology from the bottom-level architecture. This trend indicates that artificial intelligence technology will play a more critical role in the future and promote the innovative development of social economy. The application of artificial intelligence technology in the medical industry includes auxiliary diagnosis, patient monitoring, drug research and development, etc. For example, Baidu's "Lingyi Big Model" is an AI big model for the medical industry that can provide services such as medical image analysis and disease prediction; among them, adaptive neural network is an advanced machine learning technology that simulates the connection between neurons in the human brain to build an adaptive learning neural network model. This model can automatically adjust the network structure and parameters according to the characteristics and rules of the input data, and assist doctors in early screening and diagnosis by analyzing a large amount of medical image data, improving the prediction accuracy and robustness of the model, and also improving the accuracy and efficiency of diagnosis.

[0003] The application of artificial intelligence in the medical field is developing rapidly, showing great potential and multiple advantages, but also facing some challenges and difficulties. In summary, there are mainly two aspects of the problem. First, the training cost is high, and a large amount of medical imaging data needs to be analyzed to assist doctors in early screening and diagnosis. In addition, the problems faced by the medical field itself are also very complex. It is difficult to train all medical problems one by one, and comprehensive training is obviously even more unattainable. Second, it is difficult to obtain medical data samples, data extraction is difficult, and a large amount of training is not supported, so it is difficult to obtain effective auxiliary data. Medical data samples require professional judgment. It is difficult to integrate various types of medical data, and it is even more difficult to professionally label them to obtain samples. In order to obtain accurate auxiliary data samples, doctors need to collect patient data for comprehensive judgment and labeling. Training relies on high-quality labeled data, and data quality issues may affect model performance. Therefore, medical data samples are very scarce and cannot support large-scale training of large-scale neural network models. Moreover, the privacy and security of medical data are major challenges in the application of artificial intelligence. It is necessary to ensure data protection and comply with regulations, which makes it more difficult to obtain medical data samples. Third, multimodal data fusion is imperative. Multimodal artificial intelligence technology is a deep learning model that comprehensively processes multiple types of data. It improves the performance and accuracy of tasks by fusing information from different modalities. Multimodal artificial intelligence aims to create models that can process multiple data types simultaneously to better understand and interpret complex real-world data. For medical assistance, obtaining relatively accurate auxiliary data requires the support of multimodal technology. How to introduce more modal information while minimizing the increase in training complexity is also urgent. Based on the above problems, the present invention achieves rapid training of medical data models with the support of less medical data and training complexity by splitting the data dimensions. This improves the efficiency of medical data assistance. Summary of the invention

[0004] In order to solve the above problems in the prior art, the present invention proposes an artificial intelligence medical auxiliary data processing method and system, the method comprising:

[0005] Step S1: construct a medical data processing model, the medical data processing model includes a K+1 channel feature fusion network model and a machine learning model; the K channels of the feature fusion network model accept K modal medical data inputs, the K+1th channel is used to accept the K-element output vector output by the dynamic weight model as the dynamic weight input, and feature extraction is performed on the K modal medical data inputs to obtain the corresponding features to be fused; in the first stage of medical data processing model training, feature fusion is performed based on the dynamic weight input of the K+1th channel; in the second stage of medical data processing model training, the dynamic weight is automatically adjusted based on the gradient descent method with the goal of minimizing the loss function of the medical data processing model to perform feature fusion; the fused features are input into the machine learning model and auxiliary data are obtained; the first stage is when the loss function of the dynamic weight model converges, then the first stage ends; the second stage ends when the loss function of the medical data processing model converges, then the second stage ends;

[0006] Step S2: construct a dynamic weight model; the input of the dynamic weight model is an L-element input vector, and the output is a K-element output vector; wherein: each element in the L-element input vector is used to indicate L types of external attribute information; the external attribute information is user medical data other than the image information of the internal structure of the user's body; the k-th element in the K-element output vector is used to indicate the dynamic weight corresponding to the k-th modal medical data; the dynamic weight model uses the external attribute information of the multimodal medical data used in the medical data processing model based on the current medical data set, and the channel correlation between the auxiliary data and the auxiliary data samples obtained based on the k-th modal medical data for training; then for the next medical data set, obtain an L-element input vector based on the next medical data set; input the L-element input vector into the dynamic weight model to obtain a K-element output vector corresponding to the L-element input vector; use the K-element output vector to set the K+1-th channel input of the medical data processing model for the next medical data set in the first stage of training;

[0007] The K-modality medical data is image information from the internal structure of the user's body, including X-ray, CT, MRI and / or PET; the external attribute information includes multiple of user attribute information, electronic health records, genome data, physiological signal data, and biological sample detection data;

[0008] Step S3: After the second training phase is completed, the medical data to be processed is obtained, and K modal medical data inputs are constructed based on the medical data to be processed; they are respectively input into the 1st to Kth channels of the medical data processing model, and corresponding auxiliary data are obtained.

[0009] Furthermore, the loss function includes: one or more of mean square error, mean absolute error, cross entropy loss, logarithmic loss, and Hinge loss.

[0010] Furthermore, the feature fusion is performed based on the dynamic weight input of the K+1th channel, specifically: accepting the features to be fused of K modalities, and performing forward propagation to perform feature fusion based on the dynamic weight input.

[0011] Furthermore, K modal features to be fused are accepted, and based on the dynamic weight input, K+1 modal input features are spliced ​​based on dynamic weights and / or linearly spliced ​​to generate fused features.

[0012] Furthermore, user attribute information includes one or more of height, weight and / or gender; electronic health records include one or more of the user's medical history, diagnosis results and / or treatment records; genomic data includes one or more of the user's gene sequence, mutation and / or expression data; physiological signal data includes the user's electrocardiogram and / or electroencephalogram; laboratory test results include blood and / or urine.

[0013] Furthermore, the auxiliary data is presented to the client in an additional manner.

[0014] An artificial intelligence medical-assisted data processing system is used to implement the above-mentioned artificial intelligence medical-assisted data processing method.

[0015] An artificial intelligence medical-assisted data processing device, wherein the artificial intelligence medical-assisted data processing device is used to implement the above-mentioned artificial intelligence medical-assisted data processing method.

[0016] An artificial intelligence medical-assisted data processing module, wherein the artificial intelligence medical-assisted data processing module is used to implement the above-mentioned artificial intelligence medical-assisted data processing method.

[0017] An artificial intelligence medical-assisted data processing server, characterized in that the artificial intelligence medical-assisted data processing server is used to implement the above-mentioned artificial intelligence medical-assisted data processing method.

[0018] The beneficial effects of the present invention include:

[0019] (1) The dynamic weight model is based on the external attribute information of the multimodal medical data used in the medical data processing model, which is different from the same medical data set. It obtains the correlation between the auxiliary data and the reference auxiliary data based on the k-th modal medical data and the correlation between the medical data in different dimensions from different dimensions. Since the input and output of the dynamic weight model are relatively simple, the dynamic weight model will reach a convergence state first and its training complexity is also low, so that relatively accurate dynamic weights can be quickly obtained, so that the dynamic weights of the medical data model can be quickly adjusted, thereby guiding the rapid descent iteration of the medical data processing model. Therefore, the rapid training of the medical data model can be completed with the support of less medical data, thereby improving the efficiency of medical data assistance.

[0020] (2) When training the medical data processing model, different feature extraction methods are set based on the complexity, credibility or information entropy content of K-modal medical data, and feature extraction is performed differentially. Dynamic weights are used for feature fusion, which improves the efficiency of the use of multimodal medical data.

[0021] The intrinsic relationship between the external attribute information discovered by setting the channel correlation and the multimodal medical data is quantified to support the repeated iteration of the medical data processing model and the dynamic weight model on the same medical data set, so that the dynamic weight model with faster iteration speed can promote the rapid training of the medical data processing model. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application, but do not constitute an improper limitation of the present invention. In the drawings:

[0023] Figure 1 Schematic diagram of the artificial intelligence medical-assisted data processing method provided by the present invention. DETAILED DESCRIPTION

[0024] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, wherein the illustrative embodiments and descriptions are only used to explain the present invention but are not intended to limit the present invention.

[0025] The present invention proposes an artificial intelligence medical auxiliary data processing method and system, as shown in the attached Figure 1 As shown, the method comprises the following steps:

[0026] Step S1: construct a medical data processing model, the medical data processing model includes a K+1 channel feature fusion network model and a machine learning model; the K channels of the feature fusion network model accept K modal medical data inputs, the K+1th channel is used to accept the K-element output vector output by the dynamic weight model as the dynamic weight input, and feature extraction is performed on the K modal medical data inputs to obtain the corresponding features to be fused; in the first stage of the medical data processing model training, feature fusion is performed based on the dynamic weight input of the K+1th channel; in the second stage of the medical data processing model training, the dynamic weight input is automatically adjusted based on the gradient descent method with the goal of minimizing the loss function to perform feature fusion; the fused features are input into the machine learning model to obtain auxiliary data;

[0027] Preferably, the loss function includes one or more of: mean square error, mean absolute error, cross entropy loss, logarithmic loss, and Hinge loss;

[0028] The feature fusion based on the dynamic weight input of the K+1th channel is specifically: accepting the features to be fused of K modalities, performing forward propagation to perform feature fusion based on the dynamic weight input; for example: creating a fusion_net instance of the FeatureFusionNet class, and using the fusion_net instance to perform forward propagation on the features to be fused of all K modalities to perform feature fusion; of course, it also depends on the specific encoder model adopted by the feature fusion network model;

[0029] Alternatively: accept K modal features to be fused, and perform dynamic weight-based splicing and / or linear (or nonlinear transformation) splicing on K+1 modal input features based on dynamic weight input to generate fused features;

[0030] Preferably: the K-modality medical data is image information from the internal structure of the user's body; including X-ray, CT (computed tomography), MRI (magnetic resonance imaging), PET (positron emission tomography);

[0031] Preferably, the initial value of the dynamic weight input is set before the first stage; for example, the dynamic weight values ​​of all modalities are set to be equal; or they are set according to experience; the first stage ends when the loss function of the dynamic weight model converges; the second stage ends when the loss function of the medical data processing model converges;

[0032] Preferably: the auxiliary data is medical auxiliary information such as annotation information about the image of the internal structure of the user's body;

[0033] Preferably: the medical data processing model is a multimodal artificial intelligence model, a multimodal neural network, or an adaptive neural network model;

[0034] Preferably: the medical data processing model adopts different feature extraction methods for K-channel multimodal medical data; specifically: different feature extraction methods are set based on the complexity (credibility or information entropy content, etc.) of K-modal medical data; the higher the complexity, the larger the number of the feature extraction method, and vice versa, the lower the complexity, the smaller the number of the feature extraction method; wherein: method 1: perform maximum pooling on the image information to extract local features and / or average pooling to extract global features; method 2: first use multiple groups of 1×1 convolutions to reduce the data dimension without losing features, and then use N*N convolutions to extract features for each group of 1×1 convolution results and then splice them; wherein: N=3, 5, 7, etc.; method 3: use a group of 1×1 convolutions to extract all image features and retain all original data information;

[0035] Furthermore: different feature extraction methods are set based on the credibility or information entropy content of K-modal medical data; the higher the credibility or information entropy content, the larger the number of the feature extraction method, and vice versa, the lower the credibility or information entropy content, the smaller the number of the feature extraction method; specifically: the maximum pooling is performed on the X-ray image to extract local features, and the average pooling is performed on the CT image to extract global features; the method 2 is used for feature extraction of MRI images, and the method 3 is used for feature extraction of PET images;

[0036] Step S2: construct a dynamic weight model; the input of the dynamic weight model is an L-element input vector, and the output is a K-element output vector; wherein: each element in the L-element input vector is used to indicate L types of external attribute information; the external attribute information is user medical data other than the image information of the internal structure of the user's body; the k-th element in the K-element output vector is used to indicate the dynamic weight corresponding to the k-th modal medical data; that is, the correlation between the k-th modal medical data and the obtained auxiliary data; the dynamic weight model is based on the external attribute information of the same medical data set that is different from the multimodal medical data used in the medical data processing model, and obtains the correlation between the degree of correlation between the auxiliary data obtained based on the k-th modal medical data and the reference auxiliary data and the correlation between the medical data of different dimensions from different dimensions, so as to quickly adjust the dynamic weight, thereby guiding the rapid descent iteration of the medical data processing model;

[0037] Preferably, the L value and K value are both preset values; for example, L=5-10, K=4;

[0038] Preferably, the external attribute information includes one or more of user attribute information, electronic health records, genome data, physiological signal data, biological sample test data, etc.; specifically, user attribute information includes height, weight, gender, etc.; electronic health records include the user's medical history, diagnosis results, treatment records, etc.; genome data includes the user's gene sequence, mutation and expression data, etc.; physiological signal data includes the user's electrocardiogram, electroencephalogram, etc.; laboratory test results include blood, urine, etc.;

[0039] Step S3: In the first stage of training, construct an e-th fused sample set based on the e-th medical data set; wherein: the e-th fused sample set contains I fused samples, and the i-th fused sample contains a K-modal medical data input sample and its auxiliary data sample; use each i-th fused sample in the e-th fused sample set to train the medical data processing model; wherein: ; e is the number of the medical data set; when entering for the first time, set the initial weight for each channel; set the initial value of e to 1;

[0040] Preferably: set initial weights for each channel, and the weight values ​​of each channel in the initial weights are the same; for example: use the nn.Parameter parameter in the PyTorch library to initialize the weight input; in this way, we believe that the contribution of each modality is equal;

[0041] Preferably: the auxiliary data sample is auxiliary data obtained through manual annotation based on the e-th medical data set;

[0042] Preferably: the medical data is divided into a plurality of e-th medical data sets, and used in batches in the first stage; for example: each e-th medical data set contains 10-50 pieces of medical data; each piece of medical data is for the same user, and different pieces of medical data are for the same or different users;

[0043] Step S4: construct an e-th channel sample set based on the e-th fusion sample set; each e-th channel sample set contains K*I channel samples; for each i-th fusion sample, for each k-th modality medical data thereof, in turn, set the non-k-th modality medical data part to the default value (0) and retain only the k-th modality medical data, and split an i-th fusion sample into K channel samples; input each (k, i)-th channel sample in the K*I channel samples into the medical data processing model in turn to obtain the corresponding (k, i)-th auxiliary data; calculate the (k, i)-th channel correlation between the (k, i)-th auxiliary data and the auxiliary data sample of the i-th fusion sample; construct an e-th weight training sample set for training the dynamic weight model based on the channel correlation and the e-th medical data set; and use the e-th weight training sample set to train the dynamic weight model;

[0044] The calculation of the (k, i)th channel correlation between the (k, i)th auxiliary data and the auxiliary data sample ; Specifically, it includes the following steps:

[0045] Step S4A1: Construct decision matrix ;in: is the normalized value of the auxiliary data of the i-th fused sample; is the normalized value of the (k, i)th auxiliary data;

[0046] Step S4A2: Calculate the correlation coefficient component ; ;in: and are the minimum difference between the two levels and the maximum difference between the two levels corresponding to the i-th column in the decision matrix, is the resolution coefficient. The smaller the resolution coefficient, the larger the resolution granularity of the decision matrix among the k modes and the stronger the adjustment ability.

[0047] Preferred: Setting The initial value of is equal to 0.5; as the training phase progresses, this value can be gradually reduced. value;

[0048] Step S4A3: Normalize the correlation coefficient component of the kth mode and set the correlation coefficient of the (k, i)th channel ;

[0049] Constructing an e-th weight training sample set for training a dynamic weight model based on the channel correlation and the e-th medical data set; specifically comprising the following steps:

[0050] Step S4B1: Based on the L types of external attribute information of each i-th medical data in the e-th medical data set, construct an L-element input vector of the i-th weighted training sample in the e-th weighted training sample set; based on the channel correlation Construct K-tuple output vector< >, The K-tuple output vector of the dynamic weight model in the i-th weight training sample;

[0051] Step S4B2: for each i-th medical data, use the L-element input vector and the K-element output vector to construct each i-th weighted training sample in the e-th weighted training sample set based on the e-th medical data set; and use the e-th weighted training sample set to train a dynamic weight model;

[0052] Step S5: taking the next medical data set as the e-th medical data set (that is, the current medical data set); constructing an L-element input vector in the e-th weight training sample set based on the e-th medical data set; inputting the L-element input vector into the dynamic weight model to obtain a K-element output vector corresponding to the L-element input vector; using the K-element output vector to set the K+1-th channel input of the medical data processing model for the next medical data set in the first stage of training; since the input and output of the dynamic weight model are relatively simple, the dynamic weight model will reach a convergence state first and its training complexity is also relatively low;

[0053] The use of the K-element output vector to set the K+1th channel input of the medical data processing model for the e-th medical data set (the next medical data set) in the first stage of training is specifically: when the medical data processing model is trained using the e-th medical data set (the next medical data set), each i-th fused sample in the e-th fused sample set is used to train the medical data processing model, and the K-element output vector obtained in the dynamic weight model for the i-th medical data corresponding to the i-th fused sample is used as the K+1th channel input of the medical data processing model;

[0054] Step S6: Determine whether the first training stage is completed, if not, return to step S3; if yes, enter the second training stage; use the gradient descent algorithm or other optimization algorithms to minimize the loss function; in the forward propagation process, automatically adjust the weight of each branch according to the gradient information of the loss function; enable the model to learn the contribution of each feature channel to the final auxiliary data result; that is, when entering the second training stage, no longer change or set the weight input through external interference, and turn off the external setting of the K+1th channel; when this step returns to S3, since the next medical data set has been taken as the e-th medical data set in step S5, it is still in the first training stage at this time, and the new e-th medical data set will be continuously trained. The weight setting at this time still adopts the channel setting method;

[0055] The determining whether the first training stage is completed is specifically: when the number of training times in the first stage reaches a preset value, determining that the first training stage is completed;

[0056] Alternatively: the determining whether the first training stage is completed is specifically: when the loss function of the dynamic weight model stops decreasing or decreases very slightly, it is determined that the first training stage is completed;

[0057] Alternatively, the determination of whether the first training phase is completed is specifically as follows: when the similarity between the K-ary output vectors recently output by the dynamic weight model exceeds a similarity threshold, the first training phase is determined to be completed; recently can be quantified by the most recent N times or the like;

[0058] Step S7: After the second training stage is finished, the medical data to be processed is obtained, and K modal medical data inputs are constructed based on the medical data to be processed; they are respectively input into the 1st to Kth channels of the medical data processing model, and corresponding auxiliary data are obtained; the end of the second training stage is the achievement of the training target of the medical data processing model itself for its loss function, and of course, corresponding verification and testing are also required;

[0059] Preferably: when the medical data processing model itself reaches the training target of its loss function, the second training stage ends;

[0060] Preferably: present auxiliary data to the client in an additional manner; design interactive methods, such as click, hover, search, etc., so that users can obtain auxiliary information as needed when processing medical data;

[0061] A specific implementation method of the second stage is to not perform K+1 channel input, treat the 1st to Kth modal input as the 1st to Kth features, use the forward method in the PyTorch library, which accepts the 1st to Kth features, and then create an instance fusion_net of the FeatureFusionNet class, use fusion_net to perform forward propagation on features_1 to features_K, and obtain the fused feature fused_output; input the feature fused_output into the machine learning model, and finally obtain the input result, that is, the auxiliary data, through CreatePredictionEngine.Predict();

[0062] A computer program (also referred to as a program, software, software application, script, or code) can be written in any form of a programming language, including an assembled or interpreted language, a declarative or procedural language, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program, or in multiple collaborative files (e.g., files storing one or more modules, subroutines, or code portions). A computer program can be deployed to execute on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communication network.

[0063] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take 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 code.

[0064] The present invention 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 block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks 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.

[0065] 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.

[0066] 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 in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An artificial intelligence medical assisted data processing method, characterized in that: The method comprises: Step S1: constructing a medical data processing model, the medical data processing model comprising a K+1 channel feature fusion network model and a machine learning model; the K channels of the feature fusion network model accept K modal medical data inputs, the K+1th channel is used to accept the K-element output vector output by the dynamic weight model as the dynamic weight input, and feature extraction is performed on the K modal medical data inputs to obtain the corresponding features to be fused; In the first stage of medical data processing model training, feature fusion is performed based on the dynamic weight input of the K+1th channel; In the second stage of the medical data processing model training, the dynamic weights are automatically adjusted based on the gradient descent method to perform feature fusion with the goal of minimizing the loss function of the medical data processing model; the fused features are input into the machine learning model and auxiliary data is obtained; the first stage ends when the loss function of the dynamic weight model converges; the second stage ends when the loss function of the medical data processing model converges; Step S2: construct a dynamic weight model; the input of the dynamic weight model is an L-element input vector, and the output is a K-element output vector; wherein: each element in the L-element input vector is used to indicate L types of external attribute information; the external attribute information is user medical data other than the image information of the internal structure of the user's body; Step S3: In the first stage of training, construct an e-th fused sample set based on the e-th medical data set; wherein: the e-th fused sample set includes I fused samples, and the i-th fused sample includes a K-modal medical data input sample and its auxiliary data sample; Step S4: construct an e-th channel sample set based on the e-th fusion sample set; each e-th channel sample set contains K*I channel samples; for each i-th fusion sample, for each k-th modality medical data thereof, in turn, the non-k-th modality medical data part is set to a default value (0) and only the k-th modality medical data is retained, so that an i-th fusion sample is split into K channel samples; each (k, i)-th channel sample in the K*I channel samples is input into the medical data processing model in turn to obtain the corresponding (k, i)-th auxiliary data; Calculating the (k, i)th channel correlation between the (k, i)th auxiliary data and the auxiliary data sample of the i-th fusion sample; constructing an e-th weight training sample set for training a dynamic weight model based on the channel correlation and the e-th medical data set; and training the dynamic weight model using the e-th weight training sample set; Step S5: After the second training phase is completed, the medical data to be processed is obtained, and K modal medical data inputs are constructed based on the medical data to be processed; they are respectively input into the 1st to Kth channels of the medical data processing model, and corresponding auxiliary data are obtained.

2. The artificial intelligence medical assisted data processing method according to claim 1, characterized in that: The loss function includes one or more of mean square error, mean absolute error, cross entropy loss, logarithmic loss, and Hinge loss.

3. The artificial intelligence medical-assisted data processing method according to claim 2, characterized in that: The feature fusion based on the dynamic weight input of the K+1th channel is specifically: accepting the features to be fused of K modalities, and performing forward propagation to perform feature fusion based on the dynamic weight input.

4. The artificial intelligence medical-assisted data processing method according to claim 3 is characterized in that: Accept K modal features to be fused, and generate fused features by performing dynamic weight-based splicing and / or linear splicing on K+1 modal input features based on dynamic weight input.

5. The artificial intelligence medical-assisted data processing method according to claim 4 is characterized in that: User attribute information includes one or more of height, weight and / or gender; electronic health records include one or more of the user's medical history, diagnosis results and / or treatment records; genomic data includes one or more of the user's gene sequence, mutation and / or expression data; physiological signal data includes the user's electrocardiogram and / or electroencephalogram; laboratory test results include blood and / or urine.

6. The artificial intelligence medical-assisted data processing method according to claim 5, characterized in that: The auxiliary data is presented to the client in an additional manner.

7. An artificial intelligence medical auxiliary data processing system, characterized in that: The artificial intelligence medical-assisted data processing system is used to implement the artificial intelligence medical-assisted data processing method described in any one of claims 1 to 6.

8. An artificial intelligence medical auxiliary data processing device, characterized in that: The artificial intelligence medical-assisted data processing device is used to implement the artificial intelligence medical-assisted data processing method described in any one of claims 1 to 6.

9. An artificial intelligence medical auxiliary data processing module, characterized in that: The artificial intelligence medical-assisted data processing module is used to implement the artificial intelligence medical-assisted data processing method described in any one of claims 1 to 6.

10. An artificial intelligence medical auxiliary data processing server, characterized in that: The artificial intelligence medical-assisted data processing server is used to implement the artificial intelligence medical-assisted data processing method described in any one of claims 1 to 6.

Images