A microcirculation disorder monitoring method and device based on multi-period time series analysis, an electronic device and a computer readable storage medium

Abstract

The application discloses a microcirculation disorder monitoring method and device based on multi-period time series analysis, electronic equipment and a computer readable storage medium. The method comprises the following steps: obtaining microcirculation data of a predetermined period; digitizing the microcirculation data according to a constructed microcirculation data index digital standard to obtain digitized microcirculation data; performing feature vectorization on the digitized microcirculation data to obtain a feature vector set; inputting the feature vector set in a time sequence into a multi-period time series analysis model to calculate a predicted two-dimensional feature vector array of the predetermined period, wherein the multi-period time series analysis model calculates the relationship between feature vectors in the feature vector set across periods and the relationship between feature vectors across dimensions according to a second-order attention mechanism, and then calculates the predicted two-dimensional feature vector array; and analyzing the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain predicted microcirculation data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of microcirculation disorder monitoring, and in particular to a microcirculation disorder monitoring method and device based on multi-period time series analysis, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Microcirculation disorder is related to the occurrence, development and outcome of common diseases of various human systems such as hypertension, diabetes, rheumatic immune diseases and cardiovascular and cerebrovascular diseases. Therefore, regular monitoring of microcirculation can timely assess the situation of microcirculation disorder that cannot be perceived.

[0003] T2DM is a lifelong metabolic disease, and the duration of the patient's course of disease varies from 2 to 60 years. Therefore, the diagnosis and treatment of diabetes is a long-term process. The nature, severity and duration of microcirculation disorder have a great relationship with the prognosis of patients. In clinical practice, continuous monitoring of 16 microcirculation index data of patients can predict the change trend of each observation index of patient microcirculation and the overall microcirculation disorder condition of patients, but there is usually a deviation between the prediction result and the actual result. The present application extracts not only the autocorrelation of single feature sequences such as clarity or number of loop turns in a quarter cycle, but also the interaction and cross-dimensional dependence of multiple features through multi-period time series analysis modeling, so as to improve the accuracy of the prediction result of the microcirculation index of T2DM patients. SUMMARY

[0004] The present application provides a microcirculation disorder monitoring method based on multi-period time series analysis.

[0005] According to a first aspect of the present application, a microcirculation disorder monitoring method based on multi-period time series analysis is provided, which comprises:

[0006] acquiring microcirculation data of a predetermined period;

[0007] digitizing the microcirculation data according to the constructed digital standard of microcirculation data index to obtain digital microcirculation data;

[0008] vectorizing the digital microcirculation data to obtain a feature vector set;

[0009] inputting the feature vector set in time series into a multi-period time series analysis model to calculate a predicted two-dimensional feature vector array of a predetermined period, wherein the multi-period time series analysis model calculates the relationship between the feature vectors in different periods and the relationship between the feature vectors in different dimensions in the feature vector set according to a second-order attention mechanism, and calculates the predicted two-dimensional feature vector array based on the relationship between the feature vectors in different periods and the relationship between the feature vectors in different dimensions.

[0010] According to the microcirculation data index digital standard, the predicted two-dimensional feature vector array is analyzed to obtain predicted microcirculation data.

[0011] In some embodiments, the microcirculation data includes: clarity, number of loops, tube diameter, tube length, loop shape, flow rate, vascular motion type, red blood cell aggregation, white blood cell count, white microthrombus, blood color, exudation, hemorrhage, subpapillary venous plexus, nipple and sweat duct.

[0012] In some embodiments, the training of the multi-period time series analysis model includes the following steps:

[0013] Obtaining a feature vector set input in time series T ∈R T*D ; wherein, x is a feature vector, D is the number of vector elements contained in each feature vector in the feature vector set, R is a real number;

[0014] Dividing the feature vector set into a training set and a validation set based on time periods; let the time period length be L seg Then including The feature vector set of the booking number of time periods is selected as the training set and the validation set, respectively;

[0015] Extending the time period length feature and the time period feature to each feature vector in the training set and the validation set to obtain a two-dimensional feature vector array training set and a two-dimensional feature vector array validation set;

[0016] Inputting the two-dimensional feature vector array training set into a second-order attention mechanism to calculate the relationship between feature vectors across periods and the relationship between feature vectors across dimensions to obtain a joint feature vector;

[0017] According to the joint feature vector, a predicted two-dimensional feature vector array is calculated using linear transformation;

[0018] Inputting the two-dimensional feature vector array validation set to verify the accuracy of the predicted two-dimensional feature vector array.

[0019] In some embodiments, the time period length feature and the time period feature are extended to each feature vector in the training set and the validation set to obtain a two-dimensional feature vector array training set and a two-dimensional feature vector array validation set: first, input ∈R T*D , which is denoted as: wherein,​​ is the d th time period in the i th vector element, then, the features of each time series are embedded into a feature vector using linear projection and positional embedding: + ， where ∈R, E is a learnable mapping matrix, ∈R is (i, d) is a learnable positional embedding of the position, and finally, a two-dimensional feature vector array is obtained: .

[0020] In some embodiments, the two-dimensional feature vector array training set is input into a second-order attention mechanism to calculate the relationship between the feature vectors across periods and the relationship between the feature vectors across dimensions to obtain a joint feature vector. Specifically, it includes:

[0021] Let L= , then the two-dimensional feature vector array is H ；

[0022] The relationship between the feature vectors across periods is calculated, and the operation formula of the dth feature vector in the period is:

[0023] = LayerNorm( , , ））

[0024] = LayerNorm( + MLP( ））

[0025] wherein, is the feature vector across periods, 1 d D , all dimensions share multi-head self-attention, is the dth feature vector in the period, d The first input of , LayerNorm represents layer normalization calculation,​MSA represents a matrix , , ） Computing the self-attention mechanism;

[0026] The relationship between the feature vectors across dimensions is calculated, and the operation formula of all features in the i-th time period is:

[0027] = , , ），1 L

[0028] ），1 L

[0029] = LayerNorm( + ）

[0030] = LayerNorm( + ））

[0031] wherein, is a cross-dimensional feature vector, is an intermediate operation vector feature, is all cross-period feature vectors in the i-th period;

[0032] Joint feature vector = ⊕ wherein, , l is the number of model layers.

[0033] In some embodiments, the predicted two-dimensional feature vector array calculated according to the joint feature vector with a linear transformation specifically includes the following formula:

[0034] = *

[0035] { |1 i ,1 }

[0036] =

[0037] wherein, is a learnable matrix vector, is a predicted feature vector of the dth feature in the ith cycle, is a length of a cycle to be predicted, is a predicted two-dimensional feature vector array.

[0038] In some embodiments, the predicted microcirculation data is calculated according to the Tian Niu Jia plica microcirculation weighted integral method, and a microcirculation state detection evaluation result is output.

[0039] According to a second aspect of the present application, a microcirculation disorder monitoring device based on multi-cycle time series is provided, comprising:

[0040] a data acquisition module for acquiring microcirculation data of a predetermined cycle;

[0041] a data digitization module for digitizing the microcirculation data according to a constructed microcirculation data index digital standard to obtain digitized microcirculation data;

[0042] a data feature vectorization module for feature vectorizing the digitized microcirculation data to obtain a feature vector set;

[0043] a multi-cycle time series analysis module for inputting the feature vector set into a multi-cycle time series analysis model according to a time sequence to calculate a predicted two-dimensional feature vector array of the predetermined cycle;

[0044] a resolution module for resolving the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain predicted microcirculation data.

[0045] According to a third aspect of the present application, an electronic device is provided, comprising a processor and a memory coupled to the processor; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the following method:

[0046] acquiring microcirculation data of a predetermined cycle;

[0047] digitizing the microcirculation data according to a constructed microcirculation data index digital standard to obtain digitized microcirculation data;

[0048] feature vectorizing the digitized microcirculation data to obtain a feature vector set;

[0049] inputting the feature vector set into a multi-period time series analysis model in time sequence to obtain a predicted two-dimensional feature vector array of the predetermined period, wherein the multi-period time series analysis model calculates the relationship between the feature vectors in the feature vector set across periods and the relationship between the feature vectors across dimensions according to a second-order attention mechanism, and obtains the predicted two-dimensional feature vector array based on the relationship between the feature vectors across periods and the relationship between the feature vectors across dimensions;

[0050] analyzing the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain predicted microcirculation data.

[0051] According to a fourth aspect of the present application, a computer readable storage medium is provided, which stores a computer program, wherein the computer program, when executed by a processor, implements the following method:

[0052] obtaining microcirculation data of a predetermined period;

[0053] digitizing the microcirculation data according to the constructed microcirculation data index digital standard to obtain digital microcirculation data;

[0054] performing feature vectorization on the digital microcirculation data to obtain a feature vector set;

[0055] inputting the feature vector set into a multi-period time series analysis model in time sequence to obtain a predicted two-dimensional feature vector array of the predetermined period, wherein the multi-period time series analysis model calculates the relationship between the feature vectors in the feature vector set across periods and the relationship between the feature vectors across dimensions according to a second-order attention mechanism, and obtains the predicted two-dimensional feature vector array based on the relationship between the feature vectors across periods and the relationship between the feature vectors across dimensions;

[0056] analyzing the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain predicted microcirculation data.

[0057] Compared with the prior art, the present application has the following beneficial effects:

[0058] The application can automatically and accurately identify the relationship of each microcirculation index within the period and between the periods by using the natural multi-periodicity of time series, innovatively proposes an algorithm for multi-dimensional time prediction, expands the characteristics of time into microcirculation data to form a two-dimensional array, captures the dependence relationship across time and dimensions through a second-order attention mechanism, thereby more efficiently monitoring these indicators and timely discovering abnormal points deviating from periodicity, achieving the purpose of accurately predicting the trend of each time of each observation index of the patient's microcirculation and the overall microcirculation disorder condition of the patient, and the purpose of evaluation, and the monitoring and evaluation of microcirculation not only can improve the patient's cognition and management ability of the microcirculation, but also can provide doctors with more real-time and accurate data reference, and better realize personalized treatment. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 Flow chart of the microcirculation disorder monitoring method based on multi-period time series of the application;

[0060] Figure 2 Digital standard diagram of microcirculation data index of the application;

[0061] Figure 3 Structure schematic diagram of the microcirculation disorder monitoring device based on multi-period time series of the application;

[0062] Figure 4 Structure schematic diagram of an electronic device of the application;

[0063] Figure 5 Structure schematic diagram of a computer readable storage medium of the application. DETAILED DESCRIPTION

[0064] The preferred embodiments of the application will be described in more detail with reference to the accompanying drawings. Although the preferred embodiments of the application are shown in the drawings, it should be understood that the application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the application more thorough and complete, and to fully convey the scope of the application to those skilled in the art.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs. The terminology used in the specification of the application herein is only for the purpose of describing specific embodiments and is not intended to limit the application. In the present application and the appended claims, the singular forms "a", "an" and "the" are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein means and includes any or all possible combinations of one or more associated listed items.

[0066] It should be understood that although the terms "first", "second", "third" and the like can be employed in this disclosure to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another type of information.

[0067] The methods and devices employed in the present application are conventional reagents, methods and devices in the art, unless otherwise specified.

[0068] The present application will be further described below with reference to the accompanying drawings. Figure 1 The present application will be further described below with reference to the accompanying drawings.

[0069] According to a first aspect of the present application, a microcirculation disorder monitoring method based on multi-period time series analysis is provided, comprising:

[0070] S100, obtaining microcirculation data of a predetermined period;

[0071] S200, digitizing the microcirculation data according to the constructed microcirculation data index digital standard to obtain digitized microcirculation data;

[0072] S300, performing feature vectorization on the digitized microcirculation data to obtain a feature vector set;

[0073] S400, inputting the feature vector set into a multi-period time series analysis model according to a time sequence to calculate a predicted two-dimensional feature vector array of the predetermined period, wherein the multi-period time series analysis model calculates the relationship between the feature vectors across periods and the relationship between the feature vectors across dimensions in the feature vector set according to a second-order attention mechanism, and calculates the predicted two-dimensional feature vector array based on the relationship between the feature vectors across periods and the relationship between the feature vectors across dimensions;

[0074] S500, analyzing the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain predicted microcirculation data;

[0075] In the specific embodiment, the following steps are further included

[0076] S600, calculating the predicted microcirculation data according to the weighted integral method of bovine auricle flap microcirculation, and outputting a microcirculation state detection evaluation result.

[0077] Hereinafter, each step will be described in detail in combination with specific embodiments:

[0078] S100, obtaining microcirculation data of a predetermined period;

[0079] Specifically, in the embodiment, the microcirculation data is from the nailfold microcirculation, and the microcirculation data includes: clarity, loop number, tube diameter, tube length, loop morphology, flow rate, vascular motility, red blood cell aggregation, white blood cell number, white microthrombus, blood color, exudation, hemorrhage, subpapillary venous plexus, nipple, and sweat gland duct. Of course, other microcirculation data related to the disease can be used as samples for training and prediction. In a specific implementation, when the model is trained, the microcirculation data of one or more patients for more than 360 days can be collected as basic training data.

[0080] S200, digitizing the microcirculation data according to the constructed microcirculation data index digital standard to obtain digital microcirculation data;

[0081] Specifically, in combination with Figure 2 , the microcirculation data index digital standard is constructed, according to the collected values of each microcirculation data index, the normal is level 1, the attribute value is set to 0, the abnormal change is divided into levels 2, 3, 4, 5, and 6 according to different degrees, the attribute value of level 2 is set to a smaller value, the attribute value of level 6 is set to a larger value, and the values from 1 to 5 are sequentially digitized. The collected 16 microcirculation data indexes are digitized. Taking a T2DM patient as an example, 16 microcirculation index data of the patient are collected once a day within one year. The feature values of the 16 microcirculation index data of the patient on the first day are as follows:

[0082] Clarity, loop number reduction of 35%, tube diameter increase of 15%, tube length increase of 70%, loop morphology crossing of 50%, flow rate of granular slow flow, vascular motility of 3 times / min, red blood cell aggregation, white blood cell number of 20 / 15s, white microthrombus of 2 / min, blood color of dark red, exudation of ++, hemorrhage of 1 loop, subpapillary venous plexus of 1 row, nipple flat, and sweat gland duct of 2 / 1 nail fold.

[0083] The 16 microcirculation data of the patient are digitally represented according to the "microcirculation data index digital standard" table.

[0084] S300, digitizing the microcirculation data according to the constructed microcirculation data index digital standard to obtain digital microcirculation data;

[0085] According to the 16 microcirculation index data of the patient on the first day, the feature vector is obtained as follows: P1=[0, 1, 0, 2, 1, 3, 1, 0, 0, 1, 2, 2, 1, 1, 2, 0], and the set of feature vectors of the patient for 365 days is the feature vector set.

[0086] S400, inputting the feature vector set in time sequence into a multi-period time sequence analysis model to obtain a predicted two-dimensional feature vector array of a predetermined period, wherein the multi-period time sequence analysis model calculates the relationship between feature vectors across periods and the relationship between feature vectors across dimensions in the feature vector set according to a second-order attention mechanism, and obtains the predicted two-dimensional feature vector array based on the relationship between feature vectors across periods and the relationship between feature vectors across dimensions.

[0087] Specifically, the training of the multi-period time sequence analysis model includes the following steps:

[0088] obtaining a feature vector set input in time sequence T ∈R T*D ; wherein, x is a feature vector, D is the number of vector elements contained in each feature vector in the feature vector set, R is a real number;

[0089] dividing the feature vector set into a training set and a validation set based on time periods; let the time period length be L seg , including time periods, and selecting a predetermined number of feature vector sets of time periods as the training set and the validation set, respectively;

[0090] extending the time period length feature and the time period feature to each feature vector in the training set and the validation set to obtain a two-dimensional feature vector array training set and a two-dimensional feature vector array validation set;

[0091] inputting the two-dimensional feature vector array training set into a second-order attention mechanism to calculate the relationship between feature vectors across periods and the relationship between feature vectors across dimensions to obtain a joint feature vector;

[0092] calculating a predicted two-dimensional feature vector array from the joint feature vector using linear transformation;

[0093] inputting the two-dimensional feature vector array validation set to verify the accuracy of the predicted two-dimensional feature vector array;

[0094] Specifically, the time period length feature and the time period feature are extended to each feature vector in the training set and the validation set to obtain a two-dimensional feature vector array training set and a two-dimensional feature vector array validation set: first, input ∈R T*D , let be denoted as: wherein, is the d ​The first i vector set of the feature vectors in the first + In this embodiment, the fourth feature "tube length" of the second quarter in the first three quarters is represented as ∈R, E is a learnable mapping matrix, ∈R is (i, d) The learnable position embedding of the position, and finally, obtain a two-dimensional feature vector array: ;

[0095] Specifically, the two-dimensional feature vector array training set is input into the second-order attention mechanism to calculate the relationship between the cross-period feature vectors and the relationship between the cross-dimensional feature vectors to obtain the joint feature vector, which specifically includes:

[0096] Let L= The two-dimensional feature vector array is H ；

[0097] Calculate the relationship between the cross-period feature vectors, and the operation formula of the dth feature in the cycle is:

[0098] = LayerNorm( , , ））

[0099] = LayerNorm( + MLP( ））

[0100] wherein, is the cross-period feature vector, 1 d D All dimensions share multi-head self-attention, is the dth d feature in the cycle, The first input of , LayerNorm represents layer normalization calculation,​MSA represents a matrix , , ） performing the calculation of the self-attention mechanism;

[0101] The relationship between the feature vectors across dimensions is calculated, and the operation formula for all features in the i-th time period is:

[0102] = , , ），1 L

[0103] ），1 L

[0104] = LayerNorm( + ）

[0105] = LayerNorm( + ））

[0106] wherein, is a cross-dimensional feature vector, is an intermediate operation vector feature, is all cross-period feature vectors in the i-th period;

[0107] Joint feature vector = ⊕ wherein, , l is the number of model layers, in the present embodiment l is 12, represents the dimension of the hidden layer of the model, which in the present case is specifically 512 。

[0108] Specifically, the predicted two-dimensional feature vector array is calculated according to the joint feature vector using linear transformation, and specifically includes the following formula:

[0109] = *

[0110] ={ |1 i ,1 }

[0111] =

[0112] wherein, is a learnable matrix vector, is a predicted feature vector of the dth feature in the ith cycle, is a length of a cycle to be predicted, is a predicted two-dimensional feature vector array.

[0113] S500, according to the microcirculation data index digital standard, the predicted two-dimensional feature vector array is parsed to obtain predicted microcirculation data. It should be noted that the two-dimensional feature vector array includes a plurality of feature vectors arranged according to time series, according to Figure 2 , the vector element value of each feature vector in the two-dimensional feature vector array is obtained by corresponding to the microcirculation data index digital standard to obtain the predicted microcirculation data.

[0114] S600, according to the Tian Niu nail fold microcirculation weighted integral method, the predicted microcirculation data is calculated, and the microcirculation state detection evaluation result is output.

[0115] In the specific embodiment, the nail fold microcirculation data of the patient for 360 days is collected or input, that is, T=360, D=16, the time cycle length L seg =90, then including 4 time cycles, in training the model, the two-dimensional feature vector set of the previous three quarters ∈R 270*16 is the training set, and the two-dimensional feature vector set of the time sequence ∈R 90*16 is the validation set, the model is constructed based on the training set and the validation set, and when the model is used, the data of one quarter of the patient is input ∈ R 90*16 The predicted two-dimensional feature vector array of the future one quarter is obtained by the multi-cycle time series analysis model ∈ R 90*16 , according to Figure 2The 16 microcirculation data of the patient in the future quarter are obtained by analyzing and interpreting the microcirculation data index digital standard, the overall microcirculation characteristic integral prediction of the patient in the future period of time is obtained by calculating through the "Tianniujia fold microcirculation weighted integral method", and the prediction judgment of "normal, approximately normal, mild abnormality, moderate abnormality and severe abnormality" is output to monitor and evaluate the overall microcirculation state. Finally, the microcirculation observation indexes and the overall microcirculation disorder state of the T2DM patient are predicted, of course, the length of the prediction is related to the training amount of the model and the set parameters, which will not be described here. It should be noted that the cycle length L seg is 90 days because the microcirculation of the human body does not change much in a short period of time, and the long-term change trend needs to be observed, for example, the change of the microcirculation data in a quarter is meaningful, because the microcirculation data in a quarter can fully reflect the change trend of the microcirculation data over time, so that the future microcirculation data can be correctly predicted. The change difference of the existing short-period microcirculation data, such as 7-day microcirculation data, is very small, and it is impossible to make accurate prediction based on the short-period microcirculation data as the basis for predicting future microcirculation data.

[0116] The present application can automatically and accurately identify the relationship of each microcirculation index within the period and between the periods by using the natural multi-periodicity of time series, innovatively proposes an algorithm for multi-element time prediction across dimensions, expands the characteristics of time into microcirculation data to form a two-dimensional array, captures the dependence relationship across time and across dimensions through a second-order attention mechanism, thereby more efficiently monitoring these indicators and timely discovering abnormal points deviating from periodicity, achieving the purpose of predicting and evaluating the trend of each time of the future microcirculation observation indexes and the overall microcirculation disorder state of the patient. The monitoring and evaluation of microcirculation not only improve the patient's cognition and management ability of the microcirculation, but also provide doctors with more real-time and accurate data reference, and better realize personalized treatment.

[0117] According to a second aspect of the present application, referring to Figure 3 , a microcirculation disorder monitoring device based on multi-period time series is provided, comprising:

[0118] a data acquisition module for acquiring microcirculation data of a predetermined period;

[0119] a data digitization module for digitizing the microcirculation data to obtain digitized microcirculation data by constructing a microcirculation data index digital standard;

[0120] a data feature vectorization module for feature vectorizing the digitized microcirculation data to obtain a feature vector set;

[0121] a multi-period time series analysis module, configured to input the feature vector set into a multi-period time series analysis model in a time sequence to calculate a predicted two-dimensional feature vector array of a predetermined period;

[0122] a parsing module, configured to parse the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain predicted microcirculation data.

[0123] According to a third aspect of the present application, referring to Figure 4 , an electronic device 110 is provided, which comprises a processor 111 and a memory 112 connected to the processor; the memory is configured to store a computer program, and the processor is configured to execute the computer program to implement the following method:

[0124] obtaining microcirculation data of a predetermined period;

[0125] digitizing the microcirculation data according to the constructed microcirculation data index digital standard to obtain digitized microcirculation data;

[0126] performing feature vectorization on the digitized microcirculation data to obtain a feature vector set;

[0127] inputting the feature vector set into a multi-period time series analysis model in a time sequence to calculate a predicted two-dimensional feature vector array of a predetermined period, wherein the multi-period time series analysis model calculates the relationship between feature vectors across periods and the relationship between feature vectors across dimensions in the feature vector set according to a second-order attention mechanism, and calculates the predicted two-dimensional feature vector array based on the relationship between feature vectors across periods and the relationship between feature vectors across dimensions;

[0128] parsing the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain predicted microcirculation data.

[0129] According to a fourth aspect of the present application, referring to Figure 5 , a computer readable storage medium 140 is provided, which stores a computer program 141, and the computer program 141, when executed by a processor, implements the following method:

[0130] obtaining microcirculation data of a predetermined period;

[0131] digitizing the microcirculation data according to the constructed microcirculation data index digital standard to obtain digitized microcirculation data;

[0132] performing feature vectorization on the digitized microcirculation data to obtain a feature vector set;

[0133] The feature vector set is input into a multi-period time series analysis model in time sequence to obtain a predicted two-dimensional feature vector array of a predetermined period, wherein the multi-period time series analysis model calculates the relationship between feature vectors in the feature vector set across periods and the relationship between feature vectors across dimensions according to a second-order attention mechanism, and obtains the predicted two-dimensional feature vector array based on the relationship between feature vectors across periods and the relationship between feature vectors across dimensions.

[0134] The predicted two-dimensional feature vector array is analyzed according to a microcirculation data index digital standard to obtain predicted microcirculation data.

[0135] In summary, the embodiments of the present application improve the microcirculation disorder monitoring method, device, electronic equipment and readable storage medium based on the multi-period time series.

[0136] In the embodiments of the present application, the types of the devices are not limited unless otherwise specified, and any device that can complete the above functions can be used.

[0137] The above only describes some embodiments of the present application. For those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A method for monitoring microcirculatory disturbances based on multi-period time series analysis, characterized in that, It includes: Acquire microcirculation data for a predetermined period; The microcirculation data is digitized according to the constructed microcirculation data index digital standard to obtain digital microcirculation data; The digitized microcirculation data is vectorized to obtain a feature vector set; The feature vector set is input into a multi-period time series analysis model according to the time series to calculate a predicted two-dimensional feature vector array for a predetermined period. The multi-period time series analysis model calculates the relationship between feature vectors across periods and the relationship between feature vectors across dimensions in the feature vector set according to the second-order attention mechanism, and calculates the predicted two-dimensional feature vector array based on the relationship between feature vectors across periods and the relationship between feature vectors across dimensions. The predicted microcirculation data is obtained by parsing the predicted two-dimensional feature vector array based on the microcirculation data index digital standard.

2. The microcirculation disorder monitoring method based on multi-period time series analysis according to claim 1, characterized in that, The microcirculation data includes: clarity, number of vascular loops, vascular diameter, vascular length, vascular loop morphology, flow velocity, vasomotor pattern, erythrocyte aggregation, white blood cell count, white microthrombi, blood color, exudation, hemorrhage, subpapillary venous plexus, nipple and sweat gland ducts.

3. The microcirculation disorder monitoring method based on multi-period time series analysis according to claim 1, characterized in that, The training of the multi-period time series analysis model includes the following steps: Get the time series input T The feature vector set of the day ∈R T*D ;in, x For feature vectors, D This represents the number of vector elements contained in each feature vector in the feature vector set. R It is a real number; The feature vector set is divided into a training set and a validation set based on a time period; let the time period length be L. seg Then it includes For each time period, feature vector sets from the pre-ordered number of time periods are selected as the training set and the validation set, respectively. The time period length feature and the time period feature are extended to each feature vector in the training set and the validation set to obtain a two-dimensional feature vector array training set and a two-dimensional feature vector array validation set. The training set of two-dimensional feature vector arrays is input into a second-order attention mechanism to calculate the relationship between feature vectors across periods and the relationship between feature vectors across dimensions to obtain joint feature vectors. The predicted two-dimensional feature vector array is obtained by linear transformation based on the joint feature vector; Input a two-dimensional feature vector array validation set to verify the accuracy of the predicted two-dimensional feature vector array.

4. The microcirculation disorder monitoring method based on multi-period time series analysis according to claim 3, characterized in that, Extending the time period length feature and time period feature to each feature vector in the training and validation sets yields two-dimensional feature vector arrays: a training set and a validation set. First, input... ∈R T*D ,Will Recorded as: ,in, For the first d The first vector element i The feature vector set for each time period is then used to embed the features of each time series into a feature vector using linear projection and positional embedding: + ,in ∈R,E A learnable mapping matrix, ∈R for (i,d) The learnable location embeddings are then used to obtain a two-dimensional feature vector array. .

5. The microcirculation disorder monitoring method based on multi-period time series analysis according to claim 4, characterized in that, The training set of two-dimensional feature vector arrays is input into a second-order attention mechanism to calculate the relationship between feature vectors across periods and the relationship between feature vectors across dimensions, resulting in a joint feature vector. Specifically, this includes: remember L= Then the two-dimensional feature vector array is H ； The formula for calculating the relationship between eigenvectors across periods, and for all eigenvectors of the d-th eigenvalue within a period, is as follows: =LayerNorm ( , , ）） =LayerNorm ( +MLP ( ）） in, For cross-period eigenvectors, 1 d D All dimensions share multi-head self-attention. For the first d The vector representation of each feature within the period. The first input is , LayerNorm Representation layer normalization calculation, MSA Represents a matrix , , ） Perform calculations on the self-attention mechanism; The formula for calculating the relationship between feature vectors across dimensions, for all features in the i-th time period, is: = , , ），1 L ），1 L =LayerNorm ( + ） =LayerNorm ( + ）） in, For cross-dimensional feature vectors, Features of intermediate operation vectors These are all cross-period eigenvectors within the i-th period; Joint eigenvectors = ⊕ ,in, , l For the number of model layers, This represents the dimension of the hidden layer of the model.

6. The microcirculation disorder monitoring method based on multi-period time series analysis according to claim 5, characterized in that, The predicted two-dimensional feature vector array is calculated using a linear transformation based on the joint feature vector, specifically including the following formula: = * ={ |1 i ,1 } = in, For learnable matrix vectors, Let be the predicted feature vector of the d-th feature in the i-th period. The length of the period to be predicted. This is the predicted two-dimensional feature vector array.

7. The microcirculation disorder monitoring method based on multi-period time series analysis according to claim 1, characterized in that, The method further includes: The predicted microcirculation data is calculated using the weighted integral method of microcirculation in the armor fold of the bullhead, and the microcirculation status detection and evaluation results are output.

8. A microcirculation disorder monitoring device based on multi-period time series, characterized in that, It includes: The data acquisition module is used to acquire microcirculation data within a predetermined period. The data digitization module is used to digitize the microcirculation data to obtain digitized microcirculation data by constructing a digital standard for microcirculation data indicators. Data feature vectorization module: used to vectorize the digitized microcirculation data to obtain a feature vector set; The multi-period time series analysis module is used to input the feature vector set into a multi-period time series analysis model according to the time series to calculate a two-dimensional feature vector array for a predetermined period. The parsing module is used to parse the predicted two-dimensional feature vector array according to the microcirculation data index digital standard to obtain the predicted microcirculation data.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory coupled to the processor; the memory is used to store a computer program, and the processor is used to execute the computer program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images