Multimodal health data processing method, system, device and program product

Abstract

The application is suitable for the technical field of data processing, and provides a multi-modal health data processing method, system, device and program product. The method comprises: acquiring and preprocessing multi-modal health data to generate a unified to-be-operated data package; matching to a physically isolated layered coupling parameter memory container based on a data package label; generating a constraint parameter based on a historical layered average modal distortion degree stored in the container; performing pre-correction on a basic parameter based on the constraint parameter, and performing an upper limit constraint on a generated compensation value; performing joint solving operation to generate a contribution score and a current layered average modal distortion degree; and conducting the current layered average modal distortion degree to the container in a reverse direction to trigger updating of the constraint parameter. Through physical isolation, dynamic constraint and feedback regulation, the application improves processing efficiency, stability and calculation accuracy.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, and in particular relates to a multimodal health data processing method, system, device and program product. Background Technology

[0002] With the development of medical informatization, the demand for joint analysis of multimodal data such as imaging, genes, and physiological time series is increasing. Distributed parallel computing is a common technical architecture for processing such massive heterogeneous data. In existing technologies, to solve the resource contention problem when processing multiple tasks in parallel, some solutions adopt resource isolation strategies, such as allocating computing tasks with different characteristics to run in independent computing domains.

[0003] However, existing technologies still have the following drawbacks. Under a resource-isolated architecture, operational parameters (such as correction factors and compensation coefficients) within each computational domain are adjusted independently. The magnitude of parameter correction lacks constraints in massive, high-concurrency computation scenarios, leading to parameter values ​​exceeding reasonable ranges and generating invalid calculation results, requiring repeated recalculation and error correction. Error correction employs a post-processing approach, meaning that accumulated errors are uniformly compensated after all batches of data computation are completed. This method requires additional computing resources to perform correction calculations. Multimodal data computation involves multiple sources of bias, including information loss due to data anonymization, the scarcity of the data itself, and coupling errors arising from combinations of different modalities. Existing computational logic does not handle the cross-influence of these biases, causing computational errors to continuously increase batch by batch. Summary of the Invention

[0004] This application provides a multimodal health data processing method, system, device, and program product, which can solve the technical problems in the prior art when processing multimodal data, such as easy overflow of calculation parameters, large computational cost of error correction, and inability to continuously converge calculation accuracy due to the lack of effective dynamic control of parameters and error correction feedback mechanism.

[0005] In a first aspect, embodiments of this application provide a multimodal health data processing method, the method comprising: Acquire multimodal health data, perform preprocessing on the multimodal health data, and generate a unified data packet to be processed; Based on the tags carried by the unified data packet to be processed, perform hierarchical matching processing on the unified data packet to be processed to obtain the hierarchical coupling parameter memory container corresponding to the unified data packet to be processed in the distributed cluster; The basic parameters are loaded based on the hierarchical coupling parameter memory container, and constraint parameters are generated based on the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container. Based on the constraint parameters, the basic parameters are pre-corrected. According to the multimodal combination type corresponding to the unified data packet to be processed, the offset compensation data stored in the hierarchical coupling parameter memory container is called to generate a compensation value. The compensation value is then subject to an upper limit constraint based on the upper limit anchor point in the constraint parameters. Based on the unified data packet to be processed, the pre-corrected basic parameters, and the compensation value with upper limit constraint, a joint solution operation is performed to generate a contribution score and the current hierarchical average modal distortion corresponding to the hierarchical coupling parameter memory container; The current hierarchical average modal distortion is backpropagated to the hierarchical coupling parameter memory container to trigger an update of the constraint parameters.

[0006] In one possible implementation of the first aspect, the preprocessing of the multimodal health data to generate a unified data packet to be processed includes: Perform data access processing on the multimodal health data to extract metadata of the multimodal health data; the metadata includes the tags, lifecycle access frequency, and data scarcity level; The multimodal health data is anonymized, and a quantization calculation is performed based on the degree of effective feature attenuation caused by the anonymization process to generate a quantized value of anonymization loss. The processed multimodal health data, the desensitized loss quantification value, and the metadata are bound and encapsulated to generate the unified data package to be processed.

[0007] In one possible implementation of the first aspect, the tags include data reuse business scenario tags and data reuse depth tags.

[0008] In one possible implementation of the first aspect, the hierarchical coupling parameter memory container is a physically isolated independent memory space in a distributed cluster, and the hierarchical coupling parameter memory container corresponds to the layer indicated by the data reuse business scenario label and the data reuse depth label; the memory data, the basic parameters and the offset compensation data between each hierarchical coupling parameter memory container are independent of each other.

[0009] In one possible implementation of the first aspect, the basic parameters include a modal basic contribution correction factor, a lifecycle scarcity offset, and a desensitization loss adaptation coefficient; the modal basic contribution correction factor is a single-modal basic contribution benchmark correction coefficient adapted to the current data reuse business scenario and data reuse depth; the lifecycle scarcity offset is a value offset correction amount generated based on the lifecycle access frequency and the data scarcity level; and the desensitization loss adaptation coefficient is a weighted adjustment factor used to quantify the attenuation of the desensitization loss quantization value on the modal basic contribution. The modal base contribution correction factor, the lifetime scarcity offset, and the desensitization loss adaptation coefficient are locked together through a synchronization lock mechanism. The base parameters are synchronized, including synchronization update, synchronization verification, and synchronization adjustment.

[0010] In one possible implementation of the first aspect, the constraint parameters include an upper limit anchor point and a lower limit anchor point; The step of performing pre-correction on the basic parameters based on the constraint parameters, generating compensation values ​​by calling the offset compensation data stored in the hierarchical coupling parameter memory container according to the multimodal combination type corresponding to the unified data packet to be processed, and applying upper limit constraints on the compensation values ​​based on the upper limit anchor points in the constraint parameters, includes: Based on the upper limit anchor point and the lower limit anchor point, the basic parameters are pre-corrected, and parameter values ​​that exceed the range defined by the upper limit anchor point and the lower limit anchor point are truncated. Based on the multimodal combination type corresponding to the unified data packet to be processed, the offset compensation data stored in the hierarchical coupling parameter memory container is called, and multi-level superposition operation is performed according to the hierarchical step-by-step transmission rule to generate compensation value; An upper limit constraint is applied to the total amount of the compensation value based on the upper limit anchor point.

[0011] In one possible implementation of the first aspect, the contribution score is calculated using the following formula: H = (MD × L) + CT; where H is the contribution score, M is the modal basis contribution correction factor, D is the desensitization loss adaptation coefficient, L is the lifetime scarcity offset, C is the compensation value, and T is the single-sample modal distortion.

[0012] Secondly, embodiments of this application provide a multimodal health data processing system, the system comprising: The data preprocessing module is used to acquire multimodal health data, perform preprocessing on the multimodal health data, and generate a unified data packet to be processed; The hierarchical matching module is used to perform hierarchical matching processing on the unified data packet to be processed based on the label carried by the unified data packet to be processed, so as to obtain the hierarchical coupling parameter memory container in the distributed cluster corresponding to the unified data packet to be processed; The parameter loading and constraint generation module is set in the hierarchical coupling parameter memory container. It is used to load basic parameters and generate constraint parameters based on the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container. The parameter pre-calibration and compensation module is set in the hierarchical coupling parameter memory container. It is used to perform pre-calibration on the basic parameters based on the constraint parameters, generate compensation values ​​by calling the offset compensation data stored in the hierarchical coupling parameter memory container according to the multimodal combination type corresponding to the unified data packet to be calculated, and perform upper limit constraints on the compensation values ​​based on the upper limit anchor point in the constraint parameters. The joint solution module is used to perform joint solution operations based on the unified data packet to be calculated, the pre-corrected basic parameters, and the compensation value constrained by the upper limit, to generate contribution scores and the current hierarchical average modal distortion corresponding to the hierarchical coupling parameter memory container; The feedback update module is used to backpropagate the current hierarchical average modal distortion to the hierarchical coupling parameter memory container to trigger the update of the constraint parameters.

[0013] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multimodal health data processing method described in any one of the first aspects above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the multimodal health data processing method described in any one of the first aspects.

[0015] Fifthly, embodiments of this application provide a computer program product that, when run on a computer device, causes the computer device to execute the multimodal health data processing method described in any one of the first aspects.

[0016] In this embodiment, by setting physically isolated hierarchical coupling parameter memory containers, computational tasks and their required parameters at different levels are isolated, avoiding read / write conflicts and memory blocking issues caused by sharing global parameters, thus improving the system's parallel processing capabilities. By dynamically generating constraint parameters (upper and lower anchor points) based on historical distortion, and using these to pre-correct and truncate adjustments to basic parameters, dynamic boundary constraints are provided for all parameter corrections, preventing parameter value overflow and dirty data generation. By generating compensation values ​​before core operations to offset foreseeable coupling errors, replacing the traditional post-error correction mode, the computational steps required for the correction process are reduced. By backpropagating the current hierarchical average modal distortion generated in this round of operations to trigger the update of constraint parameters in the next round, a feedback control mechanism is formed, enabling the system to adjust constraint boundaries according to the operational state and suppress the continuous increase of multimodal operational errors.

[0017] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart of a multimodal health data processing method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the system architecture of a multimodal health data processing system provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0022] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0024] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0025] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0026] Figure 1 A schematic flowchart of a multimodal health data processing method provided in an embodiment of this application is shown.

[0027] S101, acquire multimodal health data, perform preprocessing on the multimodal health data, and generate a unified data packet to be processed.

[0028] Among them, the unified data packet to be processed refers to the data unit with a fixed structure generated after the original multimodal health data has been standardized and preprocessed. It contains the processed health data ontology, the desensitization loss quantification value, and metadata.

[0029] In this embodiment of the application, multimodal health data from different sources and in different formats are transformed into a unified data packet to be processed, providing standardized input for hierarchical matching, parameter loading, and joint solving.

[0030] S102, perform hierarchical matching processing on the unified data packet to be processed based on the label carried by the unified data packet to be processed, and obtain the hierarchical coupling parameter memory container in the distributed cluster corresponding to the unified data packet to be processed.

[0031] Among them, the tag refers to the attribute information used to identify the processing layer to which the data packet belongs, including data reuse business scenario tags and data reuse depth tags.

[0032] Layered coupled parameter memory containers refer to physically isolated independent memory spaces in a distributed cluster. Each container corresponds to a specific layer, and the memory data, basic parameters, and offset compensation data stored inside are independent of other containers.

[0033] In this embodiment of the application, memory read / write conflicts between different layered tasks are avoided at the architectural level by routing data packets of different layers to physically isolated memory containers.

[0034] S103, load basic parameters based on the hierarchical coupling parameter memory container, and generate constraint parameters based on the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container.

[0035] Among them, the basic parameters refer to a set of core input parameters required to perform joint solution operations, including modal basic contribution correction factor, lifetime scarcity offset and desensitization loss adaptation coefficient.

[0036] Historical stratified average modal distortion refers to the arithmetic mean of the modal distortion of a single sample generated when performing joint solution operations on one or more batches of data in a specific stratum.

[0037] Constraint parameters are a set of parameters used to set dynamic boundaries for parameter adjustments in this round of calculations, including upper limit anchor points and lower limit anchor points.

[0038] In this embodiment of the application, by using the distortion of history as the constraint basis for the current operation, the system can dynamically adjust the parameter boundaries based on historical performance.

[0039] S104, perform pre-correction on the basic parameters based on the constraint parameters, generate compensation values ​​by calling the offset compensation data stored in the hierarchical coupling parameter memory container according to the multimodal combination type corresponding to the unified data packet to be processed, and perform upper limit constraints on the compensation values ​​based on the upper limit anchor point in the constraint parameters.

[0040] The compensation value refers to the amount of compensation generated after querying the offset compensation data based on the multimodal combination type of the data packet, which is used to offset the multimodal superposition error in advance. The value range is positive floating-point compensation value.

[0041] In this embodiment of the application, by performing boundary correction on the basic parameters and generating constrained compensation values ​​before the calculation, the risk of parameter out-of-bounds errors and systematic errors are eliminated before the calculation begins.

[0042] S105, perform joint solution operation based on the unified data packet to be calculated, the pre-corrected basic parameters, and the compensation value with upper limit constraint to generate contribution score and current hierarchical average modal distortion corresponding to the hierarchical coupling parameter memory container.

[0043] The contribution score refers to the quantitative value obtained after a single piece of multimodal health data has undergone a complete processing flow, which represents the degree of effective contribution to a specific business scenario. The value range is a non-negative floating-point number.

[0044] The current stratified average modal distortion refers to the arithmetic mean of the modal distortion of all single samples in this round of batch computation within a specific stratum.

[0045] In this embodiment of the application, by simultaneously generating contribution scores and distortion indices in a single joint solution operation, the business result output and accuracy evaluation are completed synchronously.

[0046] S106, the current hierarchical average modal distortion is propagated in reverse to the hierarchical coupling parameter memory container to trigger the update of the constraint parameters.

[0047] In this embodiment, by transmitting the current distortion degree back to the memory container as the historical distortion degree, a feedback control mechanism of operation, feedback, and update is formed, so that the constraint parameters can be dynamically adjusted according to the operation state.

[0048] In an optional embodiment, step S101, which involves preprocessing the multimodal health data to generate a unified data packet to be processed, includes: Step a1: Perform data access processing on the multimodal health data and extract the metadata of the multimodal health data; the metadata includes the tag, lifecycle access frequency and data scarcity level.

[0049] Metadata refers to structured data used to describe the attribute information of multimodal health data. Tags are used for subsequent hierarchical matching processing. Lifecycle access frequency records the number of times the data has been accessed in historical operations. Data scarcity level measures how rare the data is within the overall dataset.

[0050] In this embodiment of the application, metadata is extracted to provide a basis for fine-grained scheduling and value assessment.

[0051] Step a2: Perform desensitization processing on the multimodal health data, and perform quantization calculation based on the degree of effective feature attenuation caused by the desensitization processing to generate a quantized value of desensitization loss.

[0052] Among them, the desensitization loss quantification value refers to the quantified value used to characterize the degree of attenuation caused by the desensitization operation to the effective features of the data.

[0053] In the embodiments of this application, by quantifying the attenuation of effective features caused by desensitization into specific values, the negative impact of desensitization can be traced and corrected.

[0054] Step a3: Bind and encapsulate the processed multimodal health data, the desensitized loss quantification value, and the metadata to generate the unified data packet to be processed.

[0055] In this embodiment of the application, the data ontology, the desensitization loss quantification value, and metadata are integrated into an indivisible unified data packet by binding and encapsulation, ensuring that the information is complete and consistent when the data flows through the processing link.

[0056] In one optional embodiment, the tags include data reuse business scenario tags and data reuse depth tags.

[0057] Among them, the data reuse business scenario label refers to the label information used to identify the computing scenario in which multimodal health data will be used. The data reuse depth label refers to the label information used to identify the processing level of multimodal health data.

[0058] In this embodiment of the application, by setting dual-dimensional labels, it is possible to distinguish the usage scenarios and processing levels of different data, so that hierarchical matching processing can route data with different characteristics to different memory containers and achieve isolation of computing tasks.

[0059] In an optional embodiment, step S102, which involves performing hierarchical matching processing on the unified data packet to be processed based on the label carried by the unified data packet to be processed, to obtain a hierarchical coupling parameter memory container in the distributed cluster corresponding to the unified data packet to be processed, includes: Step b1: Based on the tags carried by the unified data packet to be processed, determine the processing layer corresponding to the data packet.

[0060] The tags include data reuse business scenario tags and data reuse depth tags, which together determine a unique processing layer.

[0061] In this embodiment of the application, the layer to which the data packet belongs is determined by parsing the two-dimensional label, providing a basis for routing.

[0062] Step b2: Based on the determined processing layer, match the memory container with the layer coupling parameter corresponding to the processing layer in the distributed cluster.

[0063] In this embodiment of the application, a corresponding hierarchical coupling parameter memory container is pre-created in the cluster for each tag combination. By querying the preset correspondence, a unique corresponding container can be matched to realize the routing of data packets to the container.

[0064] Step b3: The unified data packet to be processed is routed to the memory container of the matched hierarchical coupling parameters.

[0065] In this embodiment of the application, the routing of data packets to the physical isolation container is completed, providing an isolated environment for independent computing.

[0066] In one optional embodiment, the hierarchical coupling parameter memory container is a physically isolated independent memory space in a distributed cluster. The hierarchical coupling parameter memory container corresponds to the hierarchical layer indicated by the data reuse business scenario label and the data reuse depth label. The memory data, the basic parameters, and the offset compensation data between each hierarchical coupling parameter memory container are independent of each other.

[0067] In this context, a distributed cluster refers to a computer cluster system composed of multiple computing nodes used for parallel processing of large-scale data. The hierarchical coupling parameter memory container's correspondence with the data reuse business scenario label and the data reuse depth label indicates that each container uniquely corresponds to a processing layer determined by both the data reuse business scenario label and the data reuse depth label; that is, data packets with the same dual-dimensional label combination are routed to the same container. Memory data refers to all intermediate computational data and temporary variables related to that layer, stored in each hierarchical coupling parameter memory container.

[0068] In this embodiment of the application, through the design of physically isolated memory space, different layers of computing tasks run in their own independent containers, and there is no shared memory area between containers, thereby eliminating memory read and write conflicts between parallel tasks from the hardware architecture level.

[0069] In one optional embodiment, the basic parameters include a modal basis contribution correction factor, a lifetime scarcity offset, and a desensitization loss adaptation coefficient.

[0070] Among them, the basic parameters refer to a set of core input parameters required to perform joint solution operations. The modal basic contribution correction factor is a single-modal basic contribution benchmark correction coefficient adapted to the current data reuse business scenario and data reuse depth, with a value range of a preset positive range of floating-point values. The lifecycle scarcity offset is a value offset correction amount generated based on the lifecycle access frequency and the data scarcity level, with a value range of controllable positive and negative floating-point values; high-frequency, low-scarcity data has a negative offset, and low-frequency, high-scarcity data has a positive offset. The desensitization loss adaptation coefficient is a decay adaptation weight used to quantify the desensitization loss quantization value on the modal basic contribution, with a value range of a preset positive weight range of floating-point values. The modal basic contribution correction factor, the lifecycle scarcity offset, and the desensitization loss adaptation coefficient are linked and locked through a synchronization lock mechanism. The basic parameters undergo synchronization operations, including synchronization update, synchronization verification, and synchronization adjustment.

[0071] The synchronization lock mechanism refers to a constraint rule that forces multiple parameters to be changed as a whole. Synchronization operation means that multiple parameters perform the same type of operation simultaneously, including synchronous update, synchronous verification, and synchronous adjustment, prohibiting any single parameter from being modified or changed independently.

[0072] In this embodiment, a synchronization lock mechanism is used to lock and update the three sets of parameters in a coordinated manner, preventing independent modification of a single parameter from disrupting the balance between parameters and ensuring the logical consistency of the calculation model.

[0073] In an optional embodiment, constraint parameters are generated based on the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container, including: Step c1: Read the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container.

[0074] The initial value of the historical hierarchical average modal distortion is a system preset value, which is subsequently updated by the current hierarchical average modal distortion fed back from the previous round of calculation.

[0075] In this embodiment of the application, the historical distortion is read as the basis for generating constraint parameters, so that the parameter boundaries can change dynamically with the computation history.

[0076] Step c2: Perform gradient boundary calculation based on the historical hierarchical average modal distortion to generate constraint parameters, which include upper limit anchor points and lower limit anchor points.

[0077] Among them, gradient boundary calculation refers to the operation of multiplying the historical average modal distortion as the central reference value by a preset upper and lower floating coefficient to generate upper and lower limit anchor points.

[0078] In this embodiment of the application, the historical distortion is mapped to a specific numerical boundary through gradient boundary calculation, so that the constraint boundary has a quantitative mathematical basis.

[0079] In an optional embodiment, the constraint parameters include an upper limit anchor point and a lower limit anchor point. S104, performing pre-correction on the basic parameters based on the constraint parameters, generating a compensation value by calling the offset compensation data stored in the hierarchical coupling parameter memory container according to the multimodal combination type corresponding to the unified data packet to be processed, and applying an upper limit constraint to the compensation value based on the upper limit anchor point in the constraint parameters, includes: Step d1: Based on the upper limit anchor point and the lower limit anchor point, perform pre-correction on the basic parameters, and perform truncation on parameter values ​​that exceed the range defined by the upper limit anchor point and the lower limit anchor point.

[0080] Here, the interval refers to the range of values ​​determined by the upper limit anchor point and the lower limit anchor point.

[0081] Truncation refers to the operation of forcibly limiting parameter values ​​that are outside the range to the nearest boundary value.

[0082] In this embodiment of the application, the basic parameters are limited within the dynamic boundary by truncation processing to prevent the parameter values ​​from overflowing out of the boundary.

[0083] Step d2: Based on the multimodal combination type corresponding to the unified data packet to be processed, call the offset compensation data stored in the hierarchical coupling parameter memory container, perform multi-level superposition operation according to the hierarchical step-by-step transmission rule, and generate compensation value.

[0084] Among them, multimodal combination type refers to the combination of modal data types contained in a unified data packet to be processed.

[0085] The hierarchical transmission rule refers to the operational rule of superimposing sub-compensation values ​​at each level according to the multimodal combination type.

[0086] Multi-level superposition operation refers to the mathematical operation of weighted summation of each level of sub-compensation value according to preset weights.

[0087] Offset compensation data refers to a data structure that is pre-set in a hierarchical coupling parameter memory container and contains multi-level compensation coefficients corresponding to different multimodal combination types.

[0088] In the embodiments of this application, compensation values ​​are generated through multi-level superposition operations to simulate the error propagation process generated when different modal data are combined, thereby achieving more sufficient multimodal coupling error cancellation than single-level compensation.

[0089] Step d3: Apply an upper limit constraint to the total amount of the compensation value based on the upper limit anchor point.

[0090] The upper limit constraint refers to using the upper limit anchor point as the maximum allowable threshold for the total amount of compensation value. When the total amount of compensation value exceeds the threshold, it is compressed to the threshold.

[0091] In this embodiment of the application, an upper limit constraint is imposed on the total amount of compensation value to avoid excessive error compensation.

[0092] In an optional embodiment, step S105, which involves performing a joint solution operation based on the unified data packet to be processed, the pre-corrected basic parameters, and the compensation value subject to upper limit constraints, to generate a contribution score and the current hierarchical average modal distortion corresponding to the hierarchical coupling parameter memory container, includes: Step e1: Extract the desensitization loss quantization value carried in the unified data packet to be processed, as well as the pre-corrected basic parameters and the compensation value constrained by the upper limit, as inputs for the joint solution operation.

[0093] In this embodiment of the application, all parameters that have been preprocessed are extracted to prepare complete input for the joint solution operation.

[0094] Step e2, the contribution score is calculated using the following formula: H = (MD × L) + CT; Wherein, H is the contribution score, M is the modal basis contribution correction factor, D is the desensitization loss adaptation coefficient, L is the lifetime scarcity offset, C is the compensation value, and T is the single-sample modal distortion.

[0095] Here, the single-sample modal distortion T refers to the quantified deviation value generated by a single data point in this round of calculation. It is obtained by comparing the difference between the actual calculation result and the theoretical benchmark calculation result, and its value range is a non-negative floating-point value. The theoretical benchmark calculation result refers to the ideal contribution value without considering any attenuation and error.

[0096] In this embodiment of the application, the above formula is used to cross-balance and solve the deviations of multiple dimensions such as basic contribution, desensitization loss, lifetime scarcity, multimodal superposition error and random distortion in a unified framework.

[0097] In one specific embodiment of this application, assuming M=1.2, D=0.8, L=0.15, C=0.05, T=0.02, then H=(1.2-0.8×0.15)+0.05-0.02=1.11.

[0098] Step e3: Calculate the single-sample modal distortion of all data packets within this layering round, and then use the following formula to calculate the average modal distortion of the current layer: T bar =ΣT / N; Among them, T bar The current stratum average modal distortion is T, where T is the single-sample modal distortion, and N is the total number of valid data packets in this round of batch computation for this stratum, with values ​​ranging from positive integers.

[0099] In one specific embodiment of this application, assuming that a total of 100 data packets are processed in this round of layering, and the sum of the single-sample modal distortion of all data packets ΣT=5.0, then T bar =5.0 / 100=0.05.

[0100] In this embodiment of the application, the hierarchical average modal distortion is obtained by statistics, providing a quantitative basis for updating the constraint parameters.

[0101] In an optional embodiment, step S106, which involves backpropagating the current hierarchical average modal distortion to the hierarchical coupling parameter memory container to trigger an update of the constraint parameters, includes: Step f1: Write the current hierarchical average modal distortion into the hierarchical coupling parameter memory container, overwriting the original historical hierarchical average modal distortion.

[0102] In this embodiment, the historical distortion is always kept up-to-date by overwrite updates, ensuring that the constraint parameters are generated based on the most recent computational state.

[0103] Step f2: When the next batch of unified data packets to be processed enters the hierarchical coupling parameter memory container, the gradient boundary calculation is re-executed based on the updated historical hierarchical average modal distortion to generate new upper limit anchor points and lower limit anchor points.

[0104] In this embodiment of the application, the constraint boundary is dynamically and adaptively adjusted by regenerating the constraint parameters based on the updated historical distortion.

[0105] In one optional embodiment, the label includes a data reuse business scenario label and a data reuse depth label. Alternatively, the label may be a combination of a data source institution label and a data timeliness level label. Alternatively, the label may be a combination of a data source credibility label and a processing timeliness requirement label.

[0106] Among them, the data source organization label refers to the label information used to identify the data source organization.

[0107] Data timeliness rating labels are labels used to identify the freshness or oldness of data collection time.

[0108] Data source credibility labels are labels used to identify the credibility of data source organizations.

[0109] The processing timeliness requirement label refers to label information used to identify the urgency of data processing.

[0110] In this application embodiment, by changing the label dimension, different business classification strategies can be flexibly adapted. The same set of hierarchical isolation and feedback control framework can be used for different label systems without modification, indicating that the architecture of this application has good scalability and versatility.

[0111] In an optional embodiment, when the multimodal health data to be processed is a standardized public dataset that has been uniformly standardized during collection, has not undergone de-identification, and is considered to have equal scarcity of all data, the de-identification loss quantification value can be regarded as 0, and the de-identification loss adaptation coefficient D and the life cycle scarcity offset L can also be regarded as 0.

[0112] Accordingly, the formula for calculating the contribution score is simplified as follows: H = M + CT; Wherein, H is the contribution score, M is the modal baseline contribution correction factor, C is the compensation value, and T is the single-sample modal distortion.

[0113] In the embodiments of this application, even if the core solution formula is simplified, the hierarchical isolation, constraint compensation, joint solution, and feedback update linkage control framework constructed in this application still operates completely: the system still avoids resource conflicts through physically isolated memory containers, still dynamically generates constraint parameters based on historical distortion and constrains the adjustment boundary of the basic parameter M, still handles the inherent error C generated by multimodal combinations through a pre-compensation mechanism, and still calculates and back-propagates the current hierarchical average modal distortion to drive the dynamic refresh of the constraint system. This shows that the core of the linkage control framework of this application lies in the framework itself, rather than relying on a specific formula form.

[0114] In an optional embodiment, the contribution score can be calculated using the following extended formula: H=(Mw d ×D×L)+w c ×Cw t ×T; Wherein, H is the contribution score, M is the modal basis contribution correction factor, D is the desensitization loss adaptation coefficient, L is the lifetime scarcity offset, C is the compensation value, T is the single-sample modal distortion, and w d w is the deviation weighting coefficient. c To compensate for the weighting coefficient, w t This is the distortion weighting coefficient.

[0115] Among them, w d w c w t It is a weight coefficient preset in the layered coupling parameter memory container and associated with the label of that layer, and its value range is a preset positive weight interval floating-point value. d This is used to adjust the correction strength of desensitization loss and lifetime scarcity on the contribution score, w c Used to adjust the contribution of the compensation value to the contribution score, w t Used to adjust the deduction of contribution score by single-sample modal distortion.

[0116] In this embodiment, by introducing configurable weight coefficients, the core algorithm can treat each deviation item differently according to different business scenarios, thereby achieving a trade-off between accuracy and efficiency.

[0117] In one optional embodiment, the mechanism for generating the compensation value can be further refined by introducing a long-cycle learning optimization mechanism.

[0118] Among them, the long-cycle learning optimization mechanism refers to adding a step of asynchronously updating the offset compensation data on the basis of real-time feedback control.

[0119] Specifically, it includes: In step g1, the system runs an independent statistical analysis module in the background. This module periodically collects the single-sample modal distortion T and the corresponding multimodal combination type generated in all strata.

[0120] In this embodiment of the application, historical distortion data is collected periodically to provide statistical samples for the optimization of offset compensation data.

[0121] Step g2: The statistical analysis module performs statistical regression analysis on the collected historical data to identify systematic deviation patterns under each multimodal combination type.

[0122] In this embodiment of the application, statistical regression analysis is used to identify whether the average deviation under a specific combination of modes is higher or lower than expected in the long term, providing a basis for adjusting the compensation mapping relationship.

[0123] Step g3: When it is identified that the average deviation of a certain multimodal combination type deviates from the expected value for a long period of time, the statistical analysis module automatically fine-tunes the compensation mapping relationship in the corresponding offset compensation data in the hierarchical coupling parameter memory container to make it closer to the actual error situation.

[0124] The compensation mapping relationship refers to the correspondence between the multimodal combination type and the sub-compensation values ​​at each level in the offset compensation data.

[0125] In the embodiments of this application, by enabling the pre-error compensation to have parameter adjustment capabilities, the system can predict and offset systematic errors, thereby improving the long-term stability and final accuracy of the calculation.

[0126] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0127] Corresponding to the multimodal health data processing method described in the above embodiments, this application also provides a multimodal health data processing system. Figure 2This is a schematic diagram of the system architecture of a multimodal health data processing system provided in an embodiment of this application. The system includes: The data preprocessing module is used to acquire multimodal health data, perform preprocessing on the multimodal health data, and generate a unified data packet to be processed; The hierarchical matching module is used to perform hierarchical matching processing on the unified data packet to be processed based on the label carried by the unified data packet to be processed, so as to obtain the hierarchical coupling parameter memory container in the distributed cluster corresponding to the unified data packet to be processed; The parameter loading and constraint generation module is set in the hierarchical coupling parameter memory container. It is used to load basic parameters and generate constraint parameters based on the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container. The parameter pre-calibration and compensation module is set in the hierarchical coupling parameter memory container. It is used to perform pre-calibration on the basic parameters based on the constraint parameters, generate compensation values ​​by calling the offset compensation data stored in the hierarchical coupling parameter memory container according to the multimodal combination type corresponding to the unified data packet to be calculated, and perform upper limit constraints on the compensation values ​​based on the upper limit anchor point in the constraint parameters. The joint solution module is used to perform joint solution operations based on the unified data packet to be calculated, the pre-corrected basic parameters, and the compensation value constrained by the upper limit, to generate contribution scores and the current hierarchical average modal distortion corresponding to the hierarchical coupling parameter memory container; The feedback update module is used to backpropagate the current hierarchical average modal distortion to the hierarchical coupling parameter memory container to trigger the update of the constraint parameters.

[0128] It should be noted that the information interaction and execution process between the above-mentioned components are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0129] This application also provides a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0130] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0131] This application provides a computer program product that, when run on a computer device, enables the computer device to execute the steps described in the various method embodiments above.

[0132] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 3 As shown, the computer device of this embodiment includes: at least one processor 30 ( Figure 3 (Only one is shown in the diagram), memory 31, and computer program 32 stored in the memory 31 and executable on the at least one processor 30, wherein the processor 30 executes the computer program 32 to implement the steps in any of the above embodiments of the multimodal health data processing methods.

[0133] The computer device may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will understand that... Figure 3 The examples of computer devices are merely examples and do not constitute a limitation on computer devices. They may include more or fewer components than shown in the illustration, or combinations of certain components, or different components. For example, they may also include input / output devices, network access devices, etc.

[0134] The processor 30 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0135] In some embodiments, the memory 31 may be an internal storage unit of the computer device, such as a hard drive or memory. In other embodiments, the memory 31 may be an external storage device of the computer device, such as a plug-in hard drive, smart media card (SMC), secure digital card (SD), flash card, etc. Furthermore, the memory 31 may include both internal and external storage units of the computer device. The memory 31 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0136] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a device / computer equipment, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0137] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0138] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0139] In the embodiments provided in this application, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0140] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0141] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for processing multimodal health data, characterized in that, The method includes: Acquire multimodal health data, perform preprocessing on the multimodal health data, and generate a unified data packet to be processed; Based on the tags carried by the unified data packet to be processed, perform hierarchical matching processing on the unified data packet to be processed to obtain the hierarchical coupling parameter memory container corresponding to the unified data packet to be processed in the distributed cluster; The basic parameters are loaded based on the hierarchical coupling parameter memory container, and constraint parameters are generated based on the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container. Based on the constraint parameters, the basic parameters are pre-corrected. According to the multimodal combination type corresponding to the unified data packet to be processed, the offset compensation data stored in the hierarchical coupling parameter memory container is called to generate a compensation value. The compensation value is then subject to an upper limit constraint based on the upper limit anchor point in the constraint parameters. Based on the unified data packet to be processed, the pre-corrected basic parameters, and the compensation value with upper limit constraint, a joint solution operation is performed to generate a contribution score and the current hierarchical average modal distortion corresponding to the hierarchical coupling parameter memory container; The current hierarchical average modal distortion is backpropagated to the hierarchical coupling parameter memory container to trigger an update of the constraint parameters.

2. The method according to claim 1, characterized in that, The preprocessing of the multimodal health data to generate a unified data packet to be processed includes: Perform data access processing on the multimodal health data to extract metadata of the multimodal health data; the metadata includes the tags, lifecycle access frequency, and data scarcity level; The multimodal health data is anonymized, and a quantization calculation is performed based on the degree of effective feature attenuation caused by the anonymization process to generate a quantized value of anonymization loss. The processed multimodal health data, the desensitized loss quantification value, and the metadata are bound and encapsulated to generate the unified data package to be processed.

3. The method according to claim 2, characterized in that, The tags include data reuse business scenario tags and data reuse depth tags.

4. The method according to claim 3, characterized in that, The hierarchical coupling parameter memory container is a physically isolated independent memory space in the distributed cluster. The hierarchical coupling parameter memory container corresponds to the hierarchical layer indicated by the data reuse business scenario label and the data reuse depth label. The memory data, the basic parameters, and the offset compensation data between the memory containers of each layered coupling parameter are independent of each other.

5. The method according to claim 4, characterized in that, The basic parameters include a modal basic contribution correction factor, a lifecycle scarcity offset, and a desensitization loss adaptation coefficient. The modal basic contribution correction factor is a single-modal basic contribution benchmark correction coefficient adapted to the current data reuse business scenario and data reuse depth. The lifecycle scarcity offset is a value offset correction amount generated based on the lifecycle access frequency and the data scarcity level. The desensitization loss adaptation coefficient is a weighted adjustment factor used to quantify the attenuation of the desensitization loss quantization value on the modal basic contribution. The modal base contribution correction factor, the lifetime scarcity offset, and the desensitization loss adaptation coefficient are locked together through a synchronization lock mechanism. The base parameters are synchronized, including synchronization update, synchronization verification, and synchronization adjustment.

6. The method according to claim 1, characterized in that, The constraint parameters include upper limit anchor points and lower limit anchor points; The step of performing pre-correction on the basic parameters based on the constraint parameters, generating compensation values ​​by calling the offset compensation data stored in the hierarchical coupling parameter memory container according to the multimodal combination type corresponding to the unified data packet to be processed, and applying upper limit constraints on the compensation values ​​based on the upper limit anchor points in the constraint parameters, includes: Based on the upper limit anchor point and the lower limit anchor point, the basic parameters are pre-corrected, and parameter values ​​that exceed the range defined by the upper limit anchor point and the lower limit anchor point are truncated. Based on the multimodal combination type corresponding to the unified data packet to be processed, the offset compensation data stored in the hierarchical coupling parameter memory container is called, and multi-level superposition operation is performed according to the hierarchical step-by-step transmission rule to generate compensation value; An upper limit constraint is applied to the total amount of the compensation value based on the upper limit anchor point.

7. The method according to claim 5, characterized in that, The contribution score is calculated using the following formula: H = (MD × L) + CT; where H is the contribution score, M is the modal basis contribution correction factor, D is the desensitization loss adaptation coefficient, L is the lifetime scarcity offset, C is the compensation value, and T is the single-sample modal distortion.

8. A multimodal health data processing system, characterized in that, The system includes: The data preprocessing module is used to acquire multimodal health data, perform preprocessing on the multimodal health data, and generate a unified data packet to be processed; The hierarchical matching module is used to perform hierarchical matching processing on the unified data packet to be processed based on the label carried by the unified data packet to be processed, so as to obtain the hierarchical coupling parameter memory container in the distributed cluster corresponding to the unified data packet to be processed; The parameter loading and constraint generation module is set in the hierarchical coupling parameter memory container. It is used to load basic parameters and generate constraint parameters based on the historical hierarchical average modal distortion stored in the hierarchical coupling parameter memory container. The parameter pre-calibration and compensation module is set in the hierarchical coupling parameter memory container. It is used to perform pre-calibration on the basic parameters based on the constraint parameters, generate compensation values ​​by calling the offset compensation data stored in the hierarchical coupling parameter memory container according to the multimodal combination type corresponding to the unified data packet to be calculated, and perform upper limit constraints on the compensation values ​​based on the upper limit anchor point in the constraint parameters. The joint solution module is used to perform joint solution operations based on the unified data packet to be calculated, the pre-corrected basic parameters, and the compensation value constrained by the upper limit, to generate contribution scores and the current hierarchical average modal distortion corresponding to the hierarchical coupling parameter memory container; The feedback update module is used to backpropagate the current hierarchical average modal distortion to the hierarchical coupling parameter memory container to trigger the update of the constraint parameters.

9. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1 to 7.

10. A computer program product, characterized in that, Includes a computer program that, when run, implements the method as described in any one of claims 1 to 7.

Images