AKI patient blood glucose dynamic monitoring system based on clustering processing

Abstract

The application discloses an AKI patient blood glucose dynamic monitoring system based on clustering processing, which comprises a data acquisition module, an initialization module, a blood glucose sample clustering processing module and a blood glucose dynamic monitoring module. The application belongs to the field of data processing and specifically refers to an AKI patient blood glucose dynamic monitoring system based on clustering processing. The scheme introduces a self-adaptive forgetting rate to construct a time decay weighted matrix, improves the sensitivity to sharp fluctuations, and avoids the response delay caused by excessive smoothing. The forgetting Mahalanobis distance fuses the time weight and the intra-cluster covariance information, reduces the noise interference, and further improves the blood glucose dynamic monitoring effect. Through non-correlation component filtering, the blood glucose is not dragged by the historical rigid components when the blood glucose suddenly drops or rises. Through elastic trust calibration, a secondary forgetting factor is introduced to exponentially attenuate the number of historical effective blood glucose samples, a stable detection factor is introduced to inhibit the small fluctuation error in the stable period, and the blood glucose mutation is quickly responded to, so that the blood glucose dynamic monitoring accuracy is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the field of data processing, in particular to an AKI patient blood glucose dynamic monitoring system based on clustering processing. BACKGROUND

[0002] The AKI patient blood glucose dynamic monitoring system is a real-time analysis and early warning system for blood glucose states of acute kidney injury patients, and the core goal is to quickly and accurately identify potential blood glucose abnormalities through dynamic modeling of the blood glucose changes of the patients. However, the general AKI patient blood glucose dynamic monitoring system lacks time sequence sensitivity to the dynamic changes of blood glucose, discriminates abnormal or marginal blood glucose samples, and thus leads to the problem of poor blood glucose dynamic monitoring effect. The general AKI patient blood glucose dynamic monitoring system cannot focus on effective information, is easily misled by outdated or noise centers, delays early warning, and the rare state is submerged, and thus the mutation sensitivity and stable suppression cannot be considered, leading to the problem of poor blood glucose dynamic monitoring accuracy. SUMMARY

[0003] In view of the above problems, in order to overcome the defects of the prior art, the AKI patient blood glucose dynamic monitoring system based on clustering processing is provided. In view of the problem that the general AKI patient blood glucose dynamic monitoring system lacks time sequence sensitivity to the dynamic changes of blood glucose, discriminates abnormal or marginal blood glucose samples, and thus leads to the problem of poor blood glucose dynamic monitoring effect, the adaptive forgetting rate is introduced to construct a time decay weighted matrix, the sensitivity to sharp fluctuations is improved, the latest blood glucose trend is captured and focused on in time, and the response delay caused by excessive smoothing is avoided. The forgetting Mahalanobis distance fuses the time weight and the intra-cluster covariance information, more accurately quantifies the probability that each new blood glucose sample belongs to each state cluster, and reduces the noise interference. The blood glucose dynamic monitoring effect is improved. In view of the problem that the general AKI patient blood glucose dynamic monitoring system cannot focus on effective information, is easily misled by outdated or noise centers, delays early warning, and the rare state is submerged, and thus the mutation sensitivity and stable suppression cannot be considered, leading to the problem of poor blood glucose dynamic monitoring accuracy, the non-correlated component filtering is performed, the system quickly completes the judgment when the blood glucose sharply drops or rises, and is not dragged by the historical rigid components. The elastic trustworthiness is calibrated, even if the rare severe hypoglycemia or acute hyperglycemia state, a minimum alarm probability can be maintained, the secondary forgetting factor is introduced to exponentially attenuate the number of historical effective blood glucose samples, the stable detection factor is introduced to suppress the false judgment of small fluctuations in the stable period, the blood glucose mutation is quickly responded, and the blood glucose dynamic monitoring accuracy is improved.

[0004] The technical scheme adopted by the application is as follows: The AKI patient blood glucose dynamic monitoring system based on clustering processing provided by the application comprises a data acquisition module, an initialization module, a blood glucose sample clustering processing module and a blood glucose dynamic monitoring module.

[0005] The data collection module collects a blood glucose measurement sequence of the AKI patient to form a blood glucose sample;

[0006] The initialization module trains a Gaussian mixture model offline based on historical continuous blood glucose data;

[0007] The blood glucose sample clustering processing module introduces a time-decay weighted Mahalanobis distance to calculate the trust degree, filters irrelevant components, performs elastic calibration, and combines an incremental parameter update with a forgetting factor to complete online clustering of a new blood glucose sample;

[0008] The blood glucose dynamic monitoring module performs blood glucose dynamic monitoring on the clustering result of the new blood glucose sample.

[0009] Further, the data collection module continuously collects a blood glucose measurement sequence of the AKI patient , and forms a blood glucose sample at the current time by taking the last L measurement values ; wherein, is the timestamp of the kth blood glucose measurement; T is the transpose operation; 、 and are the blood glucose values measured at the kth, i-L+1th, and ith times, respectively.

[0010] Further, the initialization module collects historical continuous blood glucose data of AKI patients and trains a Gaussian mixture model with C components offline; represented as: ; wherein, is the probability density function of the Gaussian mixture model for any blood glucose sample vector x; j is the component index; is the initial weight of the jth component; is the multivariate Gaussian distribution density function corresponding to the jth component; is the covariance matrix.

[0011] Further, the blood glucose sample clustering processing module applies a time-decay weighted Mahalanobis distance to calculate the trust degree for a new blood glucose sample, eliminates irrelevant components, and performs trust degree calibration, and then uses an incremental parameter update with a forgetting factor to cluster and assign the cluster to which the new blood glucose sample belongs; specifically including the following contents:

[0012] Calculate the original trust degree; for a new blood glucose sample , construct a time-sensitive weighted matrix , and introduce an adaptive forgetting rate, represented as: , ; wherein, 、 and are forgetting weights; is the adaptive forgetting rate; and is the adjustment parameter; calculate the blood glucose sample The forgotten Mahalanobis distance to each cluster center , expressed as: ;in, is the mean vector of the jth cluster at the tth iteration; is the covariance matrix of the jth cluster at the tth iteration; and then calculate the original trust , expressed as: ; ;in, is the mixing coefficient; and Blood glucose samples The trust strength of the j-th cluster and the k-th cluster;

[0013] Filtering of non-correlated components; calculating displacement ; Perform non-correlated component filtering, expressed as: ;in, is the mean vector of the jth cluster at the t-1th iteration; and They are respectively The forgotten Mahalanobis distance and displacement of each cluster; It is a collection after filtering out non-correlated components;

[0014] Flexible trust calibration: To ensure that the trust of each blood glucose sample and component has a lower bound, the minimum trust threshold is calculated using the following formula: ; ;in, is the minimum trust threshold; is the weight strength; the trust of non-correlated components is minimized, and the trust of the remaining clusters is uniformly amplified. The formula used is: ;in, is the calibrated trust; It is a blood sugar sample The original trust in the g-th cluster;

[0015] Incremental parameter update; introduce a forgetting factor to exponentially decay the number of historical valid blood glucose samples, and introduce a steady state detection factor when updating the mean vector , and finally perform incremental parameter update, the formula used is: ; ; ; ; ;in, and are the decayed weights of the jth cluster at the t-1th iteration and the tth iteration, respectively; is the number of samples in the jth cluster at the t-1th iteration; is the mixing coefficient of the jth cluster in the t+1th iteration; l is the cluster index; is the decayed weight of the lth cluster in the tth iteration; and are the mean vector and the covariance matrix of the jth cluster in the t+1th iteration, respectively; is the quadratic forgetting factor; and are the mean vectors of the jth cluster and the lth cluster in the t-1th iteration, respectively; is the exponential parameter

[0016] is assigned; according to the calibrated trustworthiness, the blood glucose is assigned to the cluster with the maximum trustworthiness; when the maximum number of iterations is reached, the iteration is completed; the clustering of the new blood glucose sample is completed.

[0017] Further, the blood glucose dynamic monitoring module, after the calculation of the blood glucose sample clustering processing module, obtains the blood glucose state based on the cluster to which the new blood glucose sample belongs; and performs blood glucose dynamic monitoring based on the current blood glucose state of the blood glucose sample.

[0018] The above-mentioned scheme has the following beneficial effects:

[0019] (1) For the problem that the general AKI patient blood glucose dynamic monitoring system lacks time sequence sensitivity to the dynamic change of blood glucose, discriminates abnormal or marginal blood glucose samples, and thus leads to poor blood glucose dynamic monitoring effect, the scheme introduces an adaptive forgetting rate to construct a time decay weighting matrix, improves the sensitivity to sharp fluctuations, timely captures and focuses on the latest blood glucose trend, and avoids the response delay caused by excessive smoothing; the forgetting Mahalanobis distance fuses the time weight and the intra-cluster covariance information, more accurately quantifies the probability of each new blood glucose sample belonging to each state cluster, and reduces noise interference; and thus improves the blood glucose dynamic monitoring effect.

[0020] (2) For the problem that the general AKI patient blood glucose dynamic monitoring system cannot focus on effective information, is easily misled by outdated or noise centers, and delays early warning; rare states are submerged, and the mutation sensitivity and stable suppression cannot be considered, thus leading to poor blood glucose dynamic monitoring accuracy, the scheme filters the non-correlated components, quickly completes the judgment when the blood glucose sharply drops or rises, and is not dragged by historical rigid components; through elastic trustworthiness calibration, even in the case of rare severe hypoglycemia or acute hyperglycemia, a minimum alarm probability can be maintained, a quadratic forgetting factor is introduced to exponentially decay the number of historical effective blood glucose samples, to suppress the false judgment of small fluctuations in the stable period, a stable detection factor is introduced to quickly respond to blood glucose mutations, and the blood glucose dynamic monitoring accuracy is improved. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1A flowchart of the AKI patient blood glucose dynamic monitoring system based on clustering processing provided by the present application is shown in the figure.

[0022] Figure 2 A flowchart of the blood glucose sample clustering processing module is shown in the figure.

[0023] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with embodiments of the present application, and do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0025] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0026] Embodiment one, refer to Figure 1 The AKI patient blood glucose dynamic monitoring system based on clustering processing provided by the present application comprises a data acquisition module, an initialization module, a blood glucose sample clustering processing module and a blood glucose dynamic monitoring module.

[0027] The data acquisition module acquires a blood glucose measurement sequence of an AKI patient to form a blood glucose sample, and sends data to the initialization module.

[0028] The initialization module trains a Gaussian mixture model based on historical continuous blood glucose data offline, and sends data to the blood glucose sample clustering processing module.

[0029] The blood glucose sample clustering processing module introduces a time-decay weighted Mahalanobis distance to calculate a trust degree, filters non-related components, performs elastic calibration, and combines an incremental parameter update with a forgetting factor to complete online clustering of a new blood glucose sample, and sends data to the blood glucose dynamic monitoring module.

[0030] The blood glucose dynamic monitoring module realizes blood glucose dynamic monitoring on the clustering result of the new blood glucose sample.

[0031] Embodiment two, refer to Figure 1The embodiment is based on the above embodiment, the data acquisition module continuously acquires the blood glucose measurement sequence of the AKI patient At the current time, the last L measurement values constitute a blood glucose sample ; wherein, is the timestamp of the kth blood glucose measurement; T is the transpose operation; , and are the blood glucose values of the kth, i-L+1th and ith measurements, respectively.

[0032] Embodiment three, see Figure 1 The embodiment is based on the above embodiment, the initialization module collects continuous blood glucose data of historical AKI patients, and trains a C-component Gaussian mixture model offline; represented as: ; wherein, is the probability density function of the Gaussian mixture model for any blood glucose sample vector x; j is the component index; is the initial weight of the jth component; is the multivariate Gaussian distribution density function corresponding to the jth component; is the covariance matrix; the number of clinical clusters C=4, corresponding to severe hypoglycemia, mild hypoglycemia, normal and hyperglycemia; the blood glucose clinical limits are used to initialize ; and the overall blood glucose sample covariance is used to initialize ; after training, the four components correspond to different Gaussian distributions, and each distribution represents a specific blood glucose state cluster.

[0033] Embodiment four, see Figure 1 and Figure 2 The embodiment is based on the above embodiment, the blood glucose sample clustering processing module includes calculating the original trust degree; for a newly arrived blood glucose sample , a time-sensitive weighted matrix is constructed, and an adaptive forgetting rate is introduced to adaptively adjust according to the speed of blood glucose change, which is more sensitive to rapid blood glucose change, avoids response delay caused by excessive smoothing, and is represented as: , ; the recent data is strengthened, so that the distance can better reflect the true difference between the current blood glucose trend and the state center, wherein, , and are the forgetting weights; is the adaptive forgetting rate; and are adjustment parameters; the blood glucose sample and the forgetting Mahalanobis distance of each cluster center is calculated, represented as: ; wherein, is the mean vector of the jth cluster at the tth iteration; is the covariance matrix of the jth cluster at the tth iteration; and then calculate the original trust , expressed as: ; ; Makes recent blood sugar fluctuations contribute more to the judgment of current status; Among them, is the mixing coefficient; and Blood glucose samples The confidence strength of the j-th cluster and the k-th cluster.

[0034] By performing the above operations, this solution addresses the problem that the general AKI patient blood glucose dynamic monitoring system lacks temporal sensitivity to dynamic blood glucose changes and lacks the ability to identify abnormal or marginal blood glucose samples, which leads to poor blood glucose dynamic monitoring results. This solution introduces an adaptive forgetting rate to construct a time-attenuated weighted matrix to improve sensitivity to drastic fluctuations, timely capture and focus on the latest blood glucose trends, and avoid slow response caused by over-smoothing. The forgotten Mahalanobis distance is used to fuse time weights with intra-cluster covariance information to more accurately quantify the probability that each new blood glucose sample belongs to each state cluster, reducing noise interference and thereby improving the effect of blood glucose dynamic monitoring.

[0035] Example 5, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The blood glucose sample clustering processing module applies the time-decayed weighted Mahalanobis distance to calculate the confidence of new blood glucose samples, removes irrelevant components, and performs confidence calibration. Then, an incremental parameter update with a forgetting factor is used to cluster the new blood glucose samples and assign them to clusters. The module also includes the following:

[0036] Filtering of non-correlated components; calculating displacement Filter out irrelevant components. If a component is always farther away from the current nearest component of the blood glucose sample, it can be considered irrelevant. The trust has little impact on this update, which can be expressed as: ;in, is the mean vector of the jth cluster at the t-1th iteration; and They are respectively The forgotten Mahalanobis distance and displacement of each cluster; It is a collection of non-correlated components after filtering; for continuous blood glucose monitoring, most irrelevant centers are eliminated, significantly reducing the amount of subsequent calculations; component displacement is used to avoid misjudgment caused by center updates;

[0037] Flexible trust calibration: To ensure that the trust of each blood glucose sample and component has a lower bound, the minimum trust threshold is calculated using the following formula: ; ;in, is the minimum trust threshold; is the weight strength; the trust of non-correlated components is minimized, and the trust of the remaining clusters is uniformly amplified. The formula used is: ;in, is the calibrated trust; It is a blood sugar sample The original trust in the g-th cluster; retain the minimum responsiveness;

[0038] Incremental parameter update; using the adjusted trust to more accurately reflect the trust of the current blood glucose sample in each state, especially when the blood glucose rises / drops suddenly, it can quickly push the component to the real gathering area, introduce the forgetting factor, and perform exponential decay on the number of historical valid blood glucose samples to avoid the hysteresis caused by excessive historical accumulation. It is sensitive to blood glucose mutations in AKI patients and has a smoothing effect on the stable stage; introduce a stable state detection factor when updating the mean vector , to avoid the fluctuation in the stable stage being misjudged as a sudden change state; finally, the incremental parameter update is performed, and the formula used is: ; ; ; ; ;in, and are the decayed weights of the jth cluster at the t-1th iteration and the tth iteration, respectively; is the number of samples in the jth cluster at the t-1th iteration; is the mixing coefficient of the jth cluster at the t+1th iteration; l is the cluster index; is the attenuated weight of the lth cluster at the tth iteration; and are the mean vector and covariance matrix of the jth cluster at the t+1th iteration; is the secondary forgetting factor; and are the mean vectors of the jth cluster and the lth cluster at the t-1th iteration respectively; It is an exponential parameter; it can capture new blood glucose dynamic patterns more quickly and reduce the risk of misjudging a sudden change as normal.

[0039] Assignment; According to the calibrated confidence, the blood glucose is assigned to the cluster with the largest confidence; When the maximum number of iterations is reached, the iteration is completed; The clustering of the new blood glucose sample is completed.

[0040] By performing the above operation, the blood glucose dynamic monitoring system cannot focus on effective information for general AKI patients, and is easily misled by outdated or noise center, delaying early warning; Rare state is submerged, and mutation sensitivity and stable suppression cannot be considered, which leads to the problem of poor blood glucose dynamic monitoring accuracy. The scheme filters non-related components, quickly completes judgment when blood glucose drops or rises, and is not hindered by historical rigid components; Through elastic trust calibration, even in rare cases of severe hypoglycemia or acute hyperglycemia, a minimum probability can be maintained, a secondary forgetting factor is introduced to exponentially attenuate the number of historical effective blood glucose samples, and a stable detection factor is introduced to suppress small fluctuations in the stable period, quickly respond to blood glucose mutations, and improve blood glucose dynamic monitoring accuracy.

[0041] Embodiment six, refer to Figure 1 This embodiment is based on the above-mentioned embodiments. The blood glucose dynamic monitoring module is calculated by the blood glucose sample clustering processing module. For the newly arrived blood glucose sample, the blood glucose state is obtained based on the cluster; the blood glucose dynamic monitoring is carried out based on the current blood glucose state of the blood glucose sample; for the severe hypoglycemia state, the high-priority alarm system is triggered to inform the medical staff; for the mild hypoglycemia state, the relatively low-level alarm is issued to remind the patient and the medical staff to pay attention to the low blood glucose condition, and the monitoring frequency is increased; for the normal state, the blood glucose monitoring is carried out at the normal monitoring frequency; for the hyperglycemia state, the alarm is issued to inform the patient and the medical staff of the high blood glucose condition.

[0042] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application.

[0043] The above describes the present application and its embodiments, which is not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited. In general, if those skilled in the art are inspired, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution can be designed, which should belong to the protection scope of the present application.

Claims

1. An AKI patient blood glucose dynamic monitoring system based on clustering processing, characterized in that: The system comprises a data acquisition module, an initialization module, a blood glucose sample clustering processing module and a blood glucose dynamic monitoring module. The data acquisition module acquires a blood glucose measurement sequence of an AKI patient to form a new blood glucose sample. The initialization module trains a Gaussian mixture model offline based on historical continuous blood glucose data. The blood glucose sample clustering processing module introduces a time-decay weighted Mahalanobis distance to calculate a trust degree, filters non-relevant components, calibrates the elasticity, and combines an incremental parameter update with a forgetting factor to complete online clustering of the new blood glucose sample. The blood glucose dynamic monitoring module realizes blood glucose dynamic monitoring based on the clustering result of the new blood glucose sample. The initialization module collects historical AKI patient continuous glucose data, and trains a C-component Gaussian mixture model offline; represented as: ; wherein, is the probability density function of the Gaussian mixture model for any glucose sample vector x; j is the component index; is the initial weight of the jth component; is the multivariate Gaussian distribution density function corresponding to the jth component; is the covariance matrix; The blood glucose sample clustering processing module specifically comprises the following contents: calculating the original trust degree; calculating the new blood glucose sample , constructing a time-sensitive weighted matrix , and introducing an adaptive forgetting rate, denoted as: , ; wherein, , and are forgetting weights; the blood glucose value of the ith measurement; is the adaptive forgetting rate; and are adjustment parameters; calculating the new blood glucose sample the forgetting Mahalanobis distance from each cluster center , denoted as: ; wherein, is the mean vector of the jth cluster at the tth iteration; is the covariance matrix of the jth cluster at the tth iteration; further calculating the original trust degree , denoted as: ; ; wherein, is a mixing coefficient; and are the trust strengths of the new blood glucose sample to the jth cluster and the kth cluster, respectively; Non-relevant component filtering; Elasticity trust degree calibration; Incremental parameter update; Assignment; according to the calibrated trust degree, the blood glucose is assigned to the cluster with the maximum trust degree; when the maximum iteration number is reached, the iteration is completed; the clustering of the new blood glucose sample is completed.

2. The AKI patient blood glucose dynamics monitoring system based on clustering processing according to claim 1, characterized in that: The non-relevant component filtering is calculating a displacement amount ; performing non-relevant component filtering, denoted as: ; wherein, is a mean vector of the jth cluster at the t-1th iteration; and are a forgetting Mahalanobis distance and a displacement amount, respectively, to the nth cluster; ; and is a set after non-relevant component filtering.

3. The AKI patient blood glucose dynamics monitoring system based on clustering processing according to claim 2, characterized in that: The elastic trust degree calibration is to ensure that each blood glucose sample has a lower bound with the component trust degree, and the minimum trust degree threshold is calculated by the formula: ; ; wherein, is the minimum trust degree threshold; is the weight intensity; the non-related component trust degree is pressed to the minimum, and the remaining cluster trust degree is uniformly amplified, and the formula used is: ; wherein, is the calibrated trust degree; is the blood glucose sample original trust degree of the gth cluster.

4. The AKI patient blood glucose dynamics monitoring system based on clustering processing according to claim 3, characterized in that: The incremental parameter update is to introduce a forgetting factor, exponentially decay the number of historical effective blood glucose samples, and introduce a stationary state detection factor in the mean vector update Finally, incremental parameter update is performed, and the formula used is: ; wherein, and are the decayed weights of the jth cluster in the t-1th iteration and the tth iteration, respectively; is the number of samples of the jth cluster in the t-1th iteration; is the mixing coefficient of the jth cluster in the t+1th iteration; l is the cluster index; is the decayed weight of the lth cluster in the tth iteration; and are the mean vector and the covariance matrix of the jth cluster in the t+1th iteration, respectively; is a quadratic forgetting factor; is an exponential parameter; and are the mean vectors of the jth cluster and the lth cluster in the t-1th iteration, respectively.​​​​ 5. The AKI patient blood glucose dynamics monitoring system based on clustering processing according to claim 4, characterized in that: The data acquisition module continuously acquires a sequence of blood glucose measurements of the AKI patient At the current time, the last L measurements are taken to form a new blood glucose sample ; wherein, is the timestamp of the kth blood glucose measurement; T is the transpose operation; and are the kth and i-L+1th blood glucose values, respectively.

6. The AKI patient blood glucose dynamics monitoring system based on clustering processing according to claim 5, characterized in that: The blood glucose dynamic monitoring module, after calculation by the blood glucose sample clustering processing module, obtains the blood glucose state based on the cluster to which the new blood glucose sample belongs, and performs blood glucose dynamic monitoring based on the current blood glucose state of the blood glucose sample.

Images