CN-122000045-A - Long-term blood pressure early warning method and system based on multi-decision hybrid model

CN122000045ACN 122000045 ACN122000045 ACN 122000045ACN-122000045-A

Abstract

The invention belongs to the technical field of blood pressure monitoring, and provides a long-term blood pressure early warning method and a system based on a multi-decision mixed model, wherein the method can collect and process physiological signals of a user to extract characteristics, and the characteristics are extracted through the multi-decision mixed model comprising a gate control network and a plurality of groups of expert networks, the gate control network intelligently distributes input data to the network of a plurality of expert network groups for processing, comprehensively judges and dynamically calculates blood pressure integral according to the results generated by each group, and accordingly early warning is carried out based on the blood pressure integral. The scheme adopts a multi-task expert network architecture, can cooperatively analyze blood pressure from different dimensions, greatly improves the accuracy and reliability of early warning, can dynamically allocate computing resources according to real-time data characteristics of users to realize personalized analysis, can capture abnormal precursors of blood pressure earlier and more stably by fusing multi-expert decisions and integrating quantitative risks, realizes real long-term risk early warning, and realizes personalized and prospective blood pressure health management.

Inventors

CHEN HANJIE
ZHANG YUANTING
ZENG ZEZHEN
He Pinyuan

Assignees

联感医疗科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251229

Claims (10)

1. A long-term blood pressure early warning method based on a multi-decision hybrid model, the method comprising: acquiring personal information and physiological signals of a user, and preprocessing the physiological signals to extract signal characteristics; The personal information and the signal characteristics are used as input data and are input into a multi-decision mixed model comprising a gate control network and a plurality of expert networks, wherein the expert networks are mutually independent and are divided into at least three expert task groups in advance, the expert networks in the expert task groups are preconfigured to execute blood pressure analysis tasks of the same task type, and the task types executed by different expert task groups are different from each other; Dynamically distributing the input data to at least one expert network of each expert task group through the gate control network, and enabling each distributed expert network to independently process the input data to generate an intermediate result; fusing all intermediate results belonging to the same expert task group to obtain comprehensive processing results of the expert task groups; And dynamically adjusting the blood pressure integral based on each comprehensive processing result according to a preset scoring rule, and sending out early warning information when the blood pressure integral reaches a preset warning threshold value.
2. The multi-decision hybrid model based long-term blood pressure warning method of claim 1, wherein the preprocessing the physiological signal to extract signal features comprises: Performing periodic segmentation on the physiological signal to obtain a plurality of signal segments, performing quality evaluation on each signal segment, and further selecting the signal segments meeting a preset quality standard; Extracting waveform characteristic parameters used for representing cardiovascular states from the selected signal fragments as the signal characteristics, wherein the waveform characteristic parameters cover at least one of a time domain analysis domain, an amplitude analysis domain, a phase space analysis domain and a frequency domain analysis domain.
3. The method for long-term blood pressure warning based on a multi-decision hybrid model according to claim 2, wherein the quality evaluation of each of the signal segments comprises: And comprehensively calculating waveform bias consistency, sequence similarity during the period and correlation coefficient consistency of the period of each signal segment by using a multi-index fusion model optimized by a genetic algorithm, and applying nonlinear penalty by combining the number of outlier periods so as to evaluate the quality of each signal segment.
4. A method of long-term blood pressure warning based on a multi-decision hybrid model according to any one of claims 1-3, characterized in that the multi-decision hybrid model further comprises a generic task group consisting of several generic networks for processing input data not assigned to an expert task group by the gating network or for processing input data exceeding a preset capacity of the expert task group that has been assigned to any one of the expert task groups, the method further comprising: determining the blood pressure analysis task type corresponding to each intermediate result generated by the general task group; And adding each intermediate result into an intermediate result set of the expert task group for executing the corresponding blood pressure analysis task respectively, and participating in the step of fusing all intermediate results belonging to the same expert task group.
5. The long-term blood pressure pre-warning method based on a multi-decision hybrid model according to claim 4, wherein the multi-decision hybrid model comprises a first expert task group for performing a blood pressure numerical regression prediction task, a second expert task group for performing a blood pressure state classification task, and a third expert task group for performing a blood pressure classification task; The intermediate results generated by the expert network in the first expert task group comprise predicted values of systolic pressure and/or diastolic pressure, the intermediate results generated by the expert network in the second expert task group comprise classification labels used for indicating normal blood pressure states or abnormal blood pressure states, and the intermediate results generated by the expert network in the third expert task group comprise multi-classification labels used for indicating blood pressure levels.
6. The multi-decision hybrid model based long-term blood pressure warning method of claim 5, wherein the method of training the multi-decision hybrid model comprises: setting a preset capacity value for each expert network in the first expert task group, the second expert task group and the third expert task group respectively; dynamically inputting training data into the expert networks of each expert task group through the gate control network to train, and changing the training data exceeding the preset capacity corresponding to each expert network into the universal network in the universal task group to train; Determining a mean square error between a predicted value of the private network in the first expert task group and a real blood pressure value as a first loss, determining a cross entropy between a classification result of the private network in the second expert task group and a real label as a second loss, determining a cross entropy between a classification result of the private network in the third expert task group and the real label as a third loss, and selecting the mean square error or the cross entropy as a general loss according to an original expert task group type corresponding to training data processed by a general network in the general task group; and calculating the total loss of the multi-decision hybrid model based on the first loss, the second loss, the third loss and the general loss, and performing joint optimization on parameters of the gating network, each expert task group and the general task group according to the total loss through a back propagation algorithm.
7. The method for long-term blood pressure early warning based on a multi-decision hybrid model according to claim 6, wherein the multi-decision hybrid model further comprises a balance network, and the method for obtaining the comprehensive processing result of a group of expert task groups comprises: distributing a first weight to intermediate results generated by each expert network in the expert task group based on current input data through the gating network, and carrying out weighted fusion to obtain expert fusion results; And dynamically generating balance parameters for the expert task groups based on the input data through the balance network, balancing the expert fusion result and the intermediate result added by the general task groups based on the balance parameters to obtain the comprehensive processing result of the expert task groups, wherein the balance network is mutually independent of each balance parameter generated for each expert task group.
8. The long-term blood pressure early warning method based on a multi-decision hybrid model according to claim 5, wherein the blood pressure integral is preset with a lower threshold, and the dynamically adjusting the blood pressure integral based on each comprehensive processing result according to a preset scoring rule comprises: based on the comprehensive processing results of the first expert task group, the second expert task group and the third expert task group, tendency judgment is carried out on the current blood pressure state of the user; If the judging result is prone to a normal state, correspondingly reducing the blood pressure integral based on the consistency degree between the comprehensive processing results until the blood pressure integral reaches the lower limit threshold; If the determination result is prone to an abnormal state, determining an adjustment strategy for the blood pressure integral based on the consistency degree between the comprehensive processing results, and correspondingly increasing the blood pressure integral based on the comprehensive processing results of the first expert task group and the third expert task group when the adjustment strategy comprises increasing the blood pressure integral.
9. A long-term blood pressure early warning system based on a multi-decision hybrid model, the system comprising: The acquisition module is used for acquiring personal information and physiological signals of a user and preprocessing the physiological signals to extract signal characteristics; The system comprises an input module, a multi-decision mixing model, a data processing module and a data processing module, wherein the input module is used for inputting the personal information and the signal characteristics as input data to the multi-decision mixing model comprising a gate control network and a plurality of expert networks; The processing module is used for dynamically distributing the input data to at least one expert network of each expert task group through the gate control network, and enabling each distributed expert network to independently process the input data to generate an intermediate result; The output module is used for fusing all intermediate results belonging to the same expert task group to obtain the comprehensive processing result of each expert task group; the prompting module is used for dynamically adjusting the blood pressure integral based on each comprehensive processing result according to a preset scoring rule, and sending out early warning information when the blood pressure integral reaches a preset warning threshold value.
10. A computer readable storage medium, characterized in that it has stored thereon a computer program for implementing the long-term blood pressure pre-warning method based on a multi-decision hybrid model according to any one of claims 1-8 when executed by a processor.

Description

Long-term blood pressure early warning method and system based on multi-decision hybrid model Technical Field The invention belongs to the technical field of blood pressure monitoring, and particularly relates to a long-term blood pressure early warning method and system based on a multi-decision hybrid model. Background Hypertension is one of the most common risk factors for cardiovascular diseases worldwide, and long-term poor control is a major cause of serious complications such as heart disease, cerebral apoplexy, kidney disease and the like. Therefore, effective and continuous monitoring and early warning of blood pressure are important. At present, the mainstream blood pressure monitoring technology mainly depends on even blood pressure measurement and dynamic blood pressure monitoring (ABPM), wherein the even blood pressure measurement mode is single, the frequency is low, the daily fluctuation law of blood pressure is difficult to capture, the ABPM can provide 24-hour data, but the equipment is heavy and uncomfortable to wear, and is mainly used for short-term diagnosis, and is not suitable for being used as a long-term and daily early warning means. With the popularization of wearable equipment, noninvasive continuous blood pressure monitoring technologies based on physiological signals such as photoplethysmography (PPG), electrocardiogram (ECG) and the like become research hotspots, however, these technical schemes often lack comprehensive judgment capability on multidimensional blood pressure states, have extremely limited generalization capability when facing actual scenes with large individual differences and large signal quality fluctuation, and cannot realize long-term, accurate and personalized monitoring and early risk prompting of blood pressure. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides a long-term blood pressure early warning method based on a multi-decision hybrid model, which comprises the following steps: acquiring personal information and physiological signals of a user, and preprocessing the physiological signals to extract signal characteristics; The personal information and the signal characteristics are used as input data and are input into a multi-decision mixed model comprising a gate control network and a plurality of expert networks, wherein the expert networks are mutually independent and are divided into at least three expert task groups in advance, the expert networks in the expert task groups are preconfigured to execute blood pressure analysis tasks of the same task type, and the task types executed by different expert task groups are different from each other; Dynamically distributing the input data to at least one expert network of each expert task group through the gate control network, and enabling each distributed expert network to independently process the input data to generate an intermediate result; fusing all intermediate results belonging to the same expert task group to obtain comprehensive processing results of the expert task groups; And dynamically adjusting the blood pressure integral based on each comprehensive processing result according to a preset scoring rule, and sending out early warning information when the blood pressure integral reaches a preset warning threshold value. Specifically, the preprocessing the physiological signal to extract signal features includes: Performing periodic segmentation on the physiological signal to obtain a plurality of signal segments, performing quality evaluation on each signal segment, and further selecting the signal segments meeting a preset quality standard; Extracting waveform characteristic parameters used for representing cardiovascular states from the selected signal fragments as the signal characteristics, wherein the waveform characteristic parameters cover at least one of a time domain analysis domain, an amplitude analysis domain, a phase space analysis domain and a frequency domain analysis domain. Preferably, the quality evaluation of each signal segment includes: And comprehensively calculating waveform bias consistency, sequence similarity during the period and correlation coefficient consistency of the period of each signal segment by using a multi-index fusion model optimized by a genetic algorithm, and applying nonlinear penalty by combining the number of outlier periods so as to evaluate the quality of each signal segment. Further, the multi-decision hybrid model further comprises a universal task group consisting of a plurality of universal networks, wherein the universal task group is used for processing input data which is not distributed to an expert task group by the gating network or processing input data which is distributed to any expert task group and exceeds the preset capacity of the expert task group, and the method further comprises: determining the blood pressure analysis task type corresponding to each intermediate result gen