Physiological parameter dynamic early warning method and device
Through artificial intelligence processing of physiological parameter information and using deep learning models to conduct early warning, the dynamic warning system of sepsis patients neglects the impact of single and multiple injuries, and improves early warning capabilities and medical resource utilization efficiency.
Patent Information
- Application Number
- CN202411054754.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-11-19
AI Technical Summary
The existing dynamic warning system for adverse prognosis of sepsis patients fails to effectively consider the impact of single and multiple injuries on this patient population, resulting in insufficient dynamic warning capabilities, affecting the patient's survival rate and medical resource utilization efficiency.
Using artificial intelligence methods, we use deep learning models to preprocess and train them by obtaining and processing physiological parameter information, generating dynamic early warning information, and updating the model through incremental training when the environment changes to adapt to different environments.
Dynamic warnings for septic patients are achieved, the survival rate and quality of life of patients are improved, the consumption of medical resources is reduced, and the dynamic warning capabilities with higher computational complexity and automation are provided.
Smart Images

Figure CN118983097A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a physiological parameter dynamic early warning method and device. Background Art
[0002] Physiological parameter information can reflect the physical condition of the human body, and evaluating physiological parameter information can realize dynamic early warning of the physical state. For example, during the high temperature period in summer and the cold period in winter, collecting the physiological parameter information of the staff is conducive to effectively protecting the life, health and safety of the personnel.
[0003] In the medical field, sepsis is a systemic inflammatory response syndrome caused by infection that may lead to organ dysfunction and death. The incidence of sepsis is higher in patients with single and multiple injuries, who are often at higher risk due to the complexity of their injuries and the difficulty of treatment. Therefore, it is particularly important to provide dynamic early warning of adverse prognosis for these patients.
[0004] In recent years, machine learning has attracted increasing attention in the field of sepsis, and has made some breakthroughs in the clinical diagnosis, precision treatment, and prognosis assessment of sepsis. However, most studies on the adverse prognosis of sepsis patients have not taken into account the important impact of single and multiple injuries on this patient group.
[0005] In patients with single and multiple injuries and sepsis, dynamic early warning systems can help the medical team identify patients at risk of developing severe sepsis and other complications early. Timely intervention can reduce the mortality rate and the occurrence of adverse prognosis of patients. Studies have shown that high ISS (injury severity score), AHR (airway hyperresponsiveness) level 1 day after injury, and PCT (procalcitonin) level 3 days after injury are independent risk factors for sepsis in patients with severe multiple injuries.
[0006] For patients with sepsis caused by single or multiple injuries, establishing an effective dynamic early warning system is crucial to improving patient prognosis. This can not only improve the survival rate of patients, but also improve their quality of life and reduce the consumption of medical resources. Therefore, a dynamic early warning model for adverse prognosis of patients with sepsis caused by single or multiple injuries is constructed to improve the quality and efficiency of patient care. Summary of the invention
[0007] The technical problem to be solved by the present invention is to provide a physiological parameter dynamic early warning method and device, which uses artificial intelligence methods to collect physiological parameter information of personnel and process it to achieve dynamic early warning.
[0008] In order to solve the above technical problems, the first aspect of the embodiment of the present invention discloses a physiological parameter dynamic early warning method, the method comprising:
[0009] S1, obtaining physiological parameter information; the physiological parameter information includes first physiological parameter information, second physiological parameter information and third physiological parameter information;
[0010] S2, using the physiological parameter information to train a preset deep learning model to obtain an optimized deep learning model;
[0011] S3, using the optimized deep learning model to process the physiological parameter information to be processed to obtain dynamic warning information.
[0012] As an optional implementation, in the first aspect of the embodiment of the present invention, the method of using the physiological parameter information to train a preset deep learning model to obtain an optimized deep learning model includes:
[0013] S21, preprocessing the physiological parameter information to obtain preprocessed physiological parameter information;
[0014] The preprocessed physiological parameter information includes first preprocessed physiological parameter information, second preprocessed physiological parameter information and third preprocessed physiological parameter information;
[0015] S22, processing the pre-processed physiological parameter information to obtain physiological parameter evaluation information;
[0016] S23, using the physiological parameter evaluation information to train a preset deep learning model to obtain an optimized deep learning model.
[0017] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the preprocessing of the physiological parameter information to obtain the preprocessed physiological parameter information includes:
[0018] S211, normalizing the physiological parameter information to obtain normalized physiological parameter information;
[0019] S212, performing missing value processing on the normalized physiological parameter information to obtain preprocessed physiological parameter information.
[0020] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the processing of the pre-processed physiological parameter information to obtain physiological parameter evaluation information includes:
[0021] S221, processing the first pre-processed physiological parameter information to obtain a first test index judgment matrix A;
[0022] Using a first geometric mean model, the first test indicator judgment matrix A is processed to obtain a geometric mean of all elements in each row of the first test indicator judgment matrix A;
[0023] The first geometric mean model is:
[0024]
[0025] Among them, the geometric mean of all elements in each row of the judgment matrix A is a ij is the element in the i-th row and j-th column of the judgment matrix A, and n is the number of rows and columns of the judgment matrix A;
[0026] The geometric mean is Perform the first normalization to obtain the first normalization parameter information w a ;
[0027] The first normalization is:
[0028]
[0029] Among them, the first normalized parameter information w a =(w a1 ,w a2 ,...w an );
[0030] S222, processing the second pre-processed physiological parameter information to obtain a second test index judgment matrix B;
[0031] The judgment matrix B is processed using the second geometric mean model to obtain the second normalized parameter information w b ;
[0032] S223, processing the third pre-processed physiological parameter information to obtain a third test index judgment matrix C;
[0033] The judgment matrix C is processed using the third geometric mean model to obtain the third normalized parameter information w c ;
[0034] S224: Fusing the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information.
[0035] As an optional implementation, in the first aspect of the embodiment of the present invention, the judgment matrix B is processed using the second geometric mean model to obtain the second normalized parameter information w b ,include:
[0036] Using a second geometric mean model, the judgment matrix B is processed to obtain a geometric mean of all elements in each row of the judgment matrix B;
[0037] The second geometric mean model is:
[0038]
[0039] Among them, the geometric mean of all elements in each row of the judgment matrix B is a pq is the element in the pth row and qth column of the judgment matrix B, and m is the number of rows and columns of the judgment matrix B;
[0040] The geometric mean is Perform the second normalization to obtain the second normalization parameter information w b ;
[0041] The second normalization is:
[0042]
[0043] Among them, the second normalized parameter information w b =(w b1 ,w b2 ,...w bm ).
[0044] As an optional implementation, in the first aspect of the embodiment of the present invention, the judgment matrix C is processed using the third geometric mean model to obtain the third normalized parameter information w c ,include:
[0045] Using a third geometric mean model, the judgment matrix C is processed to obtain a geometric mean of all elements in each row of the judgment matrix C;
[0046] The third geometric mean model is:
[0047]
[0048] Among them, the geometric mean of all elements in each row of the judgment matrix C is a i′j′ is the element in the i′th row and j′th column of the judgment matrix C, and n′ is the number of rows and columns of the judgment matrix C;
[0049] The geometric mean Perform the third normalization to obtain the third normalization parameter information w c ;
[0050] The third normalization is:
[0051]
[0052] Among them, the third normalized parameter information wc =(w c1 ,w c2 ,...w cn′ ).
[0053] As an optional implementation, in the first aspect of the embodiment of the present invention, the fusing of the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information includes:
[0054] S2241, using a data centralization model, processing the first normalized parameter information, the second normalized parameter information, and the third normalized parameter information to obtain centralization parameter information;
[0055] The model expression in the data center is:
[0056]
[0057] The centralization parameter information is X=[w a -u,w b -u,w c -u], w a =(w a1 ,w a2 ,...w an ), w b =(w b1 ,w b2 ,...w bm ), w c =(w c1 ,w c2 ,...w cn′ ) in the l3th element;
[0058] S2242, processing the centralization parameter information to obtain a covariance matrix;
[0059] The covariance matrix expression is:
[0060] ∑=XX T
[0061] Where ∑ is the covariance matrix and T is the transpose;
[0062] S2243, performing eigenvalue decomposition on the covariance matrix to obtain m eigenvalues;
[0063] S2244, sorting the m eigenvalues, and extracting eigenvectors corresponding to the first k eigenvalues;
[0064] S2245, using the eigenvector reconstruction model, processing the eigenvectors corresponding to the first k eigenvalues to obtain physiological parameter evaluation information;
[0065] The feature vector reconstruction model expression is:
[0066] Y=[u1X,u2X,…,u k X]
[0067] Among them, Y is the physiological parameter evaluation information, [u1,u2,…,u k ] is the eigenvector corresponding to the first k eigenvalues.
[0068] As an optional implementation, in the first aspect of the embodiment of the present invention, after using the optimized deep learning model to process the physiological parameter information to be processed and obtaining dynamic warning information, the method further includes:
[0069] When the environment changes, the optimized deep learning model parameters are updated through incremental training using a preset deep learning model update subsystem, so that the optimized deep learning model adapts to the current environment;
[0070] The incremental training method is: randomly selecting small-scale data in the current environment to add to the original training set, retraining the optimized deep learning model, and obtaining updated model parameters.
[0071] A second aspect of an embodiment of the present invention discloses a physiological parameter dynamic warning device, the device comprising:
[0072] A parameter acquisition module, used to acquire physiological parameter information; the physiological parameter information includes first physiological parameter information, second physiological parameter information and third physiological parameter information;
[0073] A model training module, used to train a preset deep learning model using the physiological parameter information to obtain an optimized deep learning model;
[0074] The dynamic warning module is used to process the physiological parameter information to be processed using the optimized deep learning model to obtain dynamic warning information.
[0075] As an optional implementation, in the second aspect of the embodiment of the present invention, the using of the physiological parameter information to train a preset deep learning model to obtain an optimized deep learning model includes:
[0076] S21, preprocessing the physiological parameter information to obtain preprocessed physiological parameter information;
[0077] The preprocessed physiological parameter information includes first preprocessed physiological parameter information, second preprocessed physiological parameter information and third preprocessed physiological parameter information;
[0078] S22, processing the pre-processed physiological parameter information to obtain physiological parameter evaluation information;
[0079] S23, using the physiological parameter evaluation information to train a preset deep learning model to obtain an optimized deep learning model.
[0080] As an optional implementation manner, in the second aspect of the embodiment of the present invention, the preprocessing of the physiological parameter information to obtain the preprocessed physiological parameter information includes:
[0081] S211, normalizing the physiological parameter information to obtain normalized physiological parameter information;
[0082] S212, performing missing value processing on the normalized physiological parameter information to obtain preprocessed physiological parameter information.
[0083] As an optional implementation, in the second aspect of the embodiment of the present invention, the processing of the pre-processed physiological parameter information to obtain physiological parameter evaluation information includes:
[0084] S221, processing the first pre-processed physiological parameter information to obtain a first test index judgment matrix A;
[0085] Using a first geometric mean model, the first test indicator judgment matrix A is processed to obtain a geometric mean of all elements in each row of the first test indicator judgment matrix A;
[0086] The first geometric mean model is:
[0087]
[0088] Among them, the geometric mean of all elements in each row of the judgment matrix A is a ij is the element in the i-th row and j-th column of the judgment matrix A, and n is the number of rows and columns of the judgment matrix A;
[0089] The geometric mean is Perform the first normalization to obtain the first normalization parameter information w a ;
[0090] The first normalization is:
[0091]
[0092] Among them, the first normalized parameter information wa =(w a1 ,w a2 ,...w an );
[0093] S222, processing the second pre-processed physiological parameter information to obtain a second test index judgment matrix B;
[0094] The judgment matrix B is processed using the second geometric mean model to obtain the second normalized parameter information w b ;
[0095] S223, processing the third pre-processed physiological parameter information to obtain a third test index judgment matrix C;
[0096] The judgment matrix C is processed using the third geometric mean model to obtain the third normalized parameter information w c ;
[0097] S224: Fusing the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information.
[0098] As an optional implementation, in the second aspect of the embodiment of the present invention, the judgment matrix B is processed using the second geometric mean model to obtain the second normalized parameter information w b ,include:
[0099] Using a second geometric mean model, the judgment matrix B is processed to obtain a geometric mean of all elements in each row of the judgment matrix B;
[0100] The second geometric mean model is:
[0101]
[0102] Among them, the geometric mean of all elements in each row of the judgment matrix B is a pq is the element in the pth row and qth column of the judgment matrix B, and m is the number of rows and columns of the judgment matrix B;
[0103] The geometric mean is Perform the second normalization to obtain the second normalization parameter information w b ;
[0104] The second normalization is:
[0105]
[0106] Among them, the second normalized parameter information wb =(w b1 ,w b2 ,...w bm ).
[0107] As an optional implementation, in the second aspect of the embodiment of the present invention, the judgment matrix C is processed using the third geometric mean model to obtain the third normalized parameter information w c ,include:
[0108] Using a third geometric mean model, the judgment matrix C is processed to obtain a geometric mean of all elements in each row of the judgment matrix C;
[0109] The third geometric mean model is:
[0110]
[0111] Among them, the geometric mean of all elements in each row of the judgment matrix C is a i′j′ is the element in the i′th row and j′th column of the judgment matrix C, and n′ is the number of rows and columns of the judgment matrix C;
[0112] The geometric mean Perform the third normalization to obtain the third normalization parameter information w c ;
[0113] The third normalization is:
[0114]
[0115] Among them, the third normalized parameter information w c =(w c1 ,w c2 ,...w cn′ ).
[0116] As an optional implementation, in the second aspect of the embodiment of the present invention, the fusing of the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information includes:
[0117] S2241, using a data centralization model, processing the first normalized parameter information, the second normalized parameter information, and the third normalized parameter information to obtain centralization parameter information;
[0118] The model expression in the data center is:
[0119]
[0120] The centralization parameter information is X=[w a -u,w b -u,w c -u], w a =(w a1 ,w a2 ,...w an ), w b =(w b1 ,w b2 ,...w bm ), w c =(w c1 ,w c2 ,...w cn′ ) in the l3th element;
[0121] S2242, processing the centralization parameter information to obtain a covariance matrix;
[0122] The covariance matrix expression is:
[0123] ∑=XX T
[0124] Where ∑ is the covariance matrix and T is the transpose;
[0125] S2243, performing eigenvalue decomposition on the covariance matrix to obtain m eigenvalues;
[0126] S2244, sorting the m eigenvalues, and extracting eigenvectors corresponding to the first k eigenvalues;
[0127] S2245, using the eigenvector reconstruction model, processing the eigenvectors corresponding to the first k eigenvalues to obtain physiological parameter evaluation information;
[0128] The feature vector reconstruction model expression is:
[0129] Y=[u1X,u2X,…,u k X]
[0130] Among them, Y is the physiological parameter evaluation information, [u1,u2,…,u k ] is the eigenvector corresponding to the first k eigenvalues.
[0131] As an optional implementation, in the second aspect of the embodiment of the present invention, after using the optimized deep learning model to process the physiological parameter information to be processed and obtaining dynamic warning information, the method further includes:
[0132] When the environment changes, the optimized deep learning model parameters are updated through incremental training using a preset deep learning model update subsystem, so that the optimized deep learning model adapts to the current environment;
[0133] The incremental training method is: randomly selecting small-scale data in the current environment to add to the original training set, retraining the optimized deep learning model, and obtaining updated model parameters.
[0134] The third aspect of the present invention discloses another physiological parameter dynamic warning device, the device comprising:
[0135] A memory storing executable program code;
[0136] a processor coupled to the memory;
[0137] The processor calls the executable program code stored in the memory to execute part or all of the steps in the physiological parameter dynamic early warning method disclosed in the first aspect of the embodiment of the present invention.
[0138] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:
[0139] The present invention realizes a physiological parameter dynamic early warning method, mines and analyzes multiple physiological indicators, and realizes dynamic early warning of physiological parameters. Compared with the traditional linear addition regression model, the method of the present invention has a higher degree of automation and process, has a higher computational complexity, is more suitable for common nonlinear problems in real problems, and can provide better dynamic early warning capabilities. The method of the present invention can automatically update the model weight according to environmental changes to adapt to different environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0140] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0141] Figure 1 It is a flow chart of a physiological parameter dynamic early warning method disclosed in an embodiment of the present invention;
[0142] Figure 2 It is a flow chart of another physiological parameter dynamic early warning method disclosed in an embodiment of the present invention;
[0143] Figure 3 It is a structural schematic diagram of a physiological parameter dynamic early warning device disclosed in an embodiment of the present invention;
[0144] Figure 4 It is a structural schematic diagram of another physiological parameter dynamic early warning device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION
[0145] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0146] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or equipment.
[0147] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0148] The present invention discloses a physiological parameter dynamic early warning method and device, the method comprising: obtaining physiological parameter information; the physiological parameter information comprises first physiological parameter information, second physiological parameter information and third physiological parameter information; using the physiological parameter information, training a preset deep learning model to obtain an optimized deep learning model; using the optimized deep learning model, processing the physiological parameter information to be processed to obtain dynamic early warning information. The method of the present invention has a higher degree of automation and process, has a higher computational complexity, is more suitable for common nonlinear problems in real problems, and can provide better dynamic early warning capabilities. The method of the present invention can automatically update the model weights according to environmental changes to adapt to different environments. The following are detailed descriptions.
[0149] Embodiment 1
[0150] See also Figure 1 , Figure 11 is a flow chart of a physiological parameter dynamic early warning method disclosed in an embodiment of the present invention. Figure 1 The physiological parameter dynamic warning method described is applied to the field of artificial intelligence technology, and the physiological parameter information to be processed is processed to obtain dynamic warning information. The present invention is not limited to this embodiment. Figure 1 As shown, the physiological parameter dynamic early warning method may include the following operations:
[0151] S1, obtaining physiological parameter information; the physiological parameter information includes first physiological parameter information, second physiological parameter information and third physiological parameter information;
[0152] Optionally, the first physiological parameter information includes age, gender, body mass index, heart rate, respiratory rate, body temperature, pulse oxygen saturation, non-invasive systolic blood pressure and non-invasive diastolic blood pressure;
[0153] Optionally, the second physiological parameter information includes Glasgow Coma Scale, Glasgow Coma Scale-Verbal Response, Glasgow Coma Scale-Motor Response and Glasgow Coma Scale-Eye Response;
[0154] Optionally, the third physiological parameter information includes mechanical ventilation status, inspired oxygen partial pressure, positive end-expiratory pressure, end-tidal carbon dioxide partial pressure, tidal volume and urine volume;
[0155] The physiological parameter information is all non-invasive parameters, which are obtained by connecting to a local hospital database or manually inputting data by staff.
[0156] S2, using the physiological parameter information to train a preset deep learning model to obtain an optimized deep learning model;
[0157] The optional, preset deep learning model is a BERT-based model. After training, four models are fine-tuned to ultimately output the mortality rate within the next 7 days, the mortality rate within 30 days, the estimated ICU length of stay, and the estimated total length of stay.
[0158] Optionally, the preset deep learning model is the CTPN model. After training, four models are fine-tuned to ultimately output the mortality rate within the next 7 days, the mortality rate within 30 days, the estimated ICU length of stay, and the estimated total length of stay.
[0159] The CTPN model includes a VGG16 convolutional model, a bidirectional LSTM model, and a fully connected layer;
[0160] Inputting the physiological parameter information into the VGG16 convolutional model; the VGG16 network includes 13 convolutional layers, 5 maximum pooling layers and 3 fully connected layers;
[0161] In the convolution layer, a two-dimensional convolution kernel w∈R is used. 3×3Extract the physiological parameter information to obtain the feature matrix C n ;
[0162]
[0163] Among them, n represents the number of convolution operations, m represents the number of convolution kernels, and p i represents the i-th feature matrix obtained, f represents the nonlinear activation function, · represents the corresponding operation of the shared weight of the convolution kernel and the feature matrix, w represents the weight of the convolution kernel, b represents the bias value, R 3×3 represents a 3×3 real matrix;
[0164] The maximum pooling method is used in the pooling layer, and the feature extraction formula is expressed as follows:
[0165] p u =Max 2×2 [C n ]
[0166] Among them, u represents the number of pooling times, Max 2×2 represents the maximum pooling operation method of a 2×2 matrix, p u is the extracted feature; after multiple convolution and pooling operations, the Reshape processed data stream is input into the bidirectional LSTM model to obtain a feature vector with time series attributes, which is expressed as follows:
[0167] o t =g(V st +V s ' T′ )
[0168] s t =f(U Xt +W St-1 )
[0169]
[0170] Among them, s t represents the output of the forward sequence at time t, s t ' represents the output of the reverse sequence at time t, U Xt Represents the initial input of the forward sequence, U′ Xt Represents the initial input of the reverse sequence, W St-1 Represents the input at the previous moment in the forward sequence. The input of the next moment in reverse time sequence, o tRepresents the output at time t. After feature extraction by the time series model, the spatial + sequence feature vector is input into the RPN network and passes through two feature extraction networks at the same time. One of them obtains positive feedback and negative feedback classification through the softmax classification feature vector set, and the other is used to calculate the boundary value regression offset for the feature vector set to obtain accurate recognition results.
[0171] S3, using the optimized deep learning model to process the physiological parameter information to be processed to obtain dynamic warning information.
[0172] The dynamic warning information includes the mortality rate within the next 7 days, the mortality rate within 30 days, the expected ICU stay and the expected total hospital stay.
[0173] Optionally, dynamic warning information such as patient injury severity score, mortality rate within the next 7 days, mortality rate within 30 days, expected ICU hospital stay and expected total hospital stay needs to be obtained by processing physiological parameter information and historical data stored in the database, and the obtained result is used as the target, and then the model is trained. This embodiment is not limited.
[0174] As an optional implementation, in the first aspect of the embodiment of the present invention, the method of using the physiological parameter information to train a preset deep learning model to obtain an optimized deep learning model includes:
[0175] S21, preprocessing the physiological parameter information to obtain preprocessed physiological parameter information;
[0176] The preprocessed physiological parameter information includes first preprocessed physiological parameter information, second preprocessed physiological parameter information and third preprocessed physiological parameter information;
[0177] The first pre-processed physiological parameter information is a result obtained by processing the first physiological parameter information;
[0178] The second pre-processed physiological parameter information is a result obtained by processing the second physiological parameter information;
[0179] The third pre-processed physiological parameter information is a result obtained by processing the third physiological parameter information;
[0180] S22, processing the pre-processed physiological parameter information to obtain physiological parameter evaluation information;
[0181] S23, using the physiological parameter evaluation information to train a preset deep learning model to obtain an optimized deep learning model.
[0182] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the preprocessing of the physiological parameter information to obtain the preprocessed physiological parameter information includes:
[0183] S211, normalizing the physiological parameter information to obtain normalized physiological parameter information;
[0184] Age, sex, body mass index, heart rate, respiratory rate, body temperature, pulse oximetry, noninvasive systolic blood pressure and noninvasive diastolic blood pressure, Glasgow Coma Score, Glasgow Coma Score-verbal response, Glasgow Coma Score-motor response and Glasgow Coma Score-eye response, mechanical ventilation status, inspired oxygen partial pressure, positive end-expiratory pressure, end-tidal carbon dioxide partial pressure, tidal volume and urine output were normalized to values between 0 and 1;
[0185] S212, performing missing value processing on the normalized physiological parameter information to obtain preprocessed physiological parameter information.
[0186] Normalization and missing value processing are conventional processing methods in the art, and the present invention does not limit them.
[0187] Optionally, the processing the pre-processed physiological parameter information to obtain physiological parameter evaluation information includes:
[0188] S221, processing the first pre-processed physiological parameter information to obtain a first test index judgment matrix A;
[0189] The age, gender, body mass index, heart rate, respiratory rate, body temperature, pulse oxygen saturation, non-invasive systolic pressure and non-invasive diastolic pressure in the first preprocessed physiological parameter information normalized to between 0 and 1 form a 3×3 first test index judgment matrix A, where age, gender, and body mass index are the first row of the matrix, heart rate, respiratory rate, and body temperature are the second row of the matrix, and pulse oxygen saturation, non-invasive systolic pressure and non-invasive diastolic pressure are the third row of the matrix;
[0190] Using a first geometric mean model, the first test indicator judgment matrix A is processed to obtain a geometric mean of all elements in each row of the first test indicator judgment matrix A;
[0191] The first geometric mean model is:
[0192]
[0193] Among them, the geometric mean of all elements in each row of the judgment matrix A is a ij is the element in the i-th row and j-th column of the judgment matrix A, and n is the number of rows and columns of the judgment matrix A;
[0194] The geometric mean is Perform the first normalization to obtain the first normalization parameter information w a ;
[0195] The first normalization is:
[0196]
[0197] Among them, the first normalized parameter information w a =(w a1 ,w a2 ,...w an );
[0198] S222, processing the second pre-processed physiological parameter information to obtain a second test index judgment matrix B;
[0199] The second preprocessed physiological parameter information is the Glasgow Coma Scale, Glasgow Coma Scale-verbal response, Glasgow Coma Scale-motor response and Glasgow Coma Scale-eye response normalized between 0 and 1, forming a 2×2 second test index judgment matrix B, where the Glasgow Coma Scale and Glasgow Coma Scale-verbal response are the first row of the matrix B, and the Glasgow Coma Scale-motor response and Glasgow Coma Scale-eye response are the second row of the matrix B;
[0200] The judgment matrix B is processed using the second geometric mean model to obtain the second normalized parameter information w b ;
[0201] S223, processing the third pre-processed physiological parameter information to obtain a third test index judgment matrix C;
[0202] The third preprocessing physiological parameter information includes mechanical ventilation status, inspired oxygen partial pressure, positive end-expiratory pressure, end-tidal carbon dioxide partial pressure, tidal volume and urine volume normalized to between 0 and 1, forming a 3×2 third test index judgment matrix C, where the mechanical ventilation status and inspired oxygen partial pressure are the first row of the matrix C, the positive end-expiratory pressure and end-tidal carbon dioxide partial pressure are the second row of the matrix C, and the tidal volume and urine volume are the third row of the matrix C;
[0203] The judgment matrix C is processed using the third geometric mean model to obtain the third normalized parameter information w c ;
[0204] S224: Fusing the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information.
[0205] Optionally, the judgment matrix B is processed using a second geometric mean model to obtain second normalized parameter information w b ,include:
[0206] Using a second geometric mean model, the judgment matrix B is processed to obtain a geometric mean of all elements in each row of the judgment matrix B;
[0207] The second geometric mean model is:
[0208]
[0209] Among them, the geometric mean of all elements in each row of the judgment matrix B is a pq is the element in the pth row and qth column of the judgment matrix B, and m is the number of rows and columns of the judgment matrix B;
[0210] The geometric mean is Perform the second normalization to obtain the second normalization parameter information w b ;
[0211] The second normalization is:
[0212]
[0213] Among them, the second normalized parameter information w b =(w b1 ,w b2 ,...w bm ).
[0214] Optionally, the judgment matrix C is processed using a third geometric mean model to obtain third normalized parameter information w c ,include:
[0215] Using a third geometric mean model, the judgment matrix C is processed to obtain a geometric mean of all elements in each row of the judgment matrix C;
[0216] The third geometric mean model is:
[0217]
[0218] Among them, the geometric mean of all elements in each row of the judgment matrix C is a i′j′ is the element in the i′th row and j′th column of the judgment matrix C, and n′ is the number of rows and columns of the judgment matrix C;
[0219] The geometric mean Perform the third normalization to obtain the third normalization parameter information w c ;
[0220] The third normalization is:
[0221]
[0222] Among them, the third normalized parameter information w c =(w c1 ,w c2 ,...w cn′ ).
[0223] Optionally, the fusing the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information includes:
[0224] S2241, using a data centralization model, processing the first normalized parameter information, the second normalized parameter information, and the third normalized parameter information to obtain centralization parameter information;
[0225] The model expression in the data center is:
[0226]
[0227] The centralization parameter information is X=[w a -u,w b -u,w c -u], w a =(w a1 ,w a2 ,...w an ), w b =(w b1 ,w b2 ,...w bm ), w c =(w c1 ,w c2 ,...w cn′ ) in the l3th element;
[0228] S2242, processing the centralization parameter information to obtain a covariance matrix;
[0229] The covariance matrix expression is:
[0230] ∑=XX T
[0231] Where ∑ is the covariance matrix and T is the transpose;
[0232] S2243, performing eigenvalue decomposition on the covariance matrix to obtain m eigenvalues;
[0233] S2244, sorting the m eigenvalues, and extracting eigenvectors corresponding to the first k eigenvalues;
[0234] S2245, using the eigenvector reconstruction model, processing the eigenvectors corresponding to the first k eigenvalues to obtain physiological parameter evaluation information;
[0235] The feature vector reconstruction model expression is:
[0236] Y=[u1X,u2X,…,u k X]
[0237] Among them, Y is the physiological parameter evaluation information, [u1,u2,…,u k ] is the eigenvector corresponding to the first k eigenvalues.
[0238] Optionally, after using the optimized deep learning model to process the physiological parameter information to be processed and obtaining dynamic warning information, the method further includes:
[0239] When the environment changes, the optimized deep learning model parameters are updated through incremental training using a preset deep learning model update subsystem, so that the optimized deep learning model adapts to the current environment;
[0240] The incremental training method is: randomly selecting small-scale data in the current environment to add to the original training set, retraining the optimized deep learning model, and obtaining updated model parameters.
[0241] It can be seen that the present invention realizes a dynamic early warning method for physiological parameters, mines and analyzes multiple physiological indicators, and realizes dynamic early warning of physiological parameters. Compared with the traditional linear addition regression model, the method of the present invention has a higher degree of automation and process, has a higher computational complexity, is more suitable for common nonlinear problems in real problems, and can provide better dynamic early warning capabilities. The method of the present invention can automatically update the model weights according to environmental changes to adapt to different environments.
[0242] Embodiment 2
[0243] See also Figure 2 , Figure 2 is a flow chart of another physiological parameter dynamic early warning method disclosed in an embodiment of the present invention. Figure 2 The physiological parameter dynamic warning method described is applied to the field of artificial intelligence technology, and the physiological parameter information to be processed is processed to obtain dynamic warning information. The present invention is not limited to this embodiment. Figure 2 As shown, the physiological parameter dynamic early warning method may include the following operations:
[0244] The purpose of this embodiment is to propose a method of using physiological parameters collected by medical equipment to construct a dynamic warning model for adverse prognosis of patients with single or multiple injuries through artificial intelligence algorithms. It is mainly aimed at filling the gap in the analysis of adverse prognosis of patients with sepsis with single or multiple injuries in the current analysis of adverse prognosis of patients with sepsis.
[0245] The dynamic early warning system for adverse prognosis of patients with single or multiple injuries and sepsis includes: a physiological parameter sensing subsystem, a dynamic early warning subsystem for adverse prognosis of patients with single or multiple injuries and sepsis, a dynamic early warning subsystem for adverse prognosis of patients with multiple injuries and sepsis, and a deep learning model updating subsystem.
[0246] The system first monitors the patient's routine physiological parameters in real time through the physiological ventilation parameter sensing subsystem. Then, every new data is input into the corresponding adverse prognosis dynamic warning subsystem through the physiological parameter sensing subsystem, where the patient's single injury or multiple injuries are judged based on the ISS score (Injury Severity Score). The corresponding adverse prognosis dynamic warning subsystem outputs the patient's mortality rate within the next 7 days, mortality rate within 30 days, expected ICU hospital stay, and expected total hospital stay based on the patient's physiological parameter time series data within one hour. If the medical environment changes, the deep learning model update subsystem will use the incremental learning method to update the model weights based on the patient's physiological parameters and prognosis information collected in the current medical environment, so that the model warning results can adapt to the current medical environment.
[0247] The physiological ventilation parameter sensing subsystem has the function of real-time monitoring of non-invasive physiological parameters of patients, and can realize real-time monitoring of parameters such as age, gender, body mass index, mechanical ventilation status, Glasgow Coma Scale, Glasgow Coma Scale-verbal response, Glasgow Coma Scale-motor response, Glasgow Coma Scale-eye response, inspired oxygen partial pressure, positive end-expiratory pressure, end-expiratory carbon dioxide partial pressure, tidal volume, urine volume, heart rate, respiratory rate, body temperature, pulse oxygen saturation, non-invasive systolic blood pressure, non-invasive diastolic blood pressure, etc. The physiological ventilation parameter sensing subsystem is connected to the local hospital database or the medical staff manually inputs the data.
[0248] The two adverse prognosis dynamic warning subsystems implement the following operations: the patient's routine non-invasive physiological parameters within one hour obtained by the physiological ventilation parameter sensing subsystem are input into the trained deep learning model. The deep learning model is based on BERT. After pre-training, four models are fine-tuned to finally output the mortality rate within the next 7 days, the mortality rate within 30 days, the expected ICU stay and the expected total hospital stay.
[0249] The deep learning model update subsystem implements the following operations: When the clinical medical environment changes, the model parameters are updated through incremental training to adapt the model to the current medical environment. The specific method of incremental training is to randomly select small-scale data in the current clinical medical environment and add it to the original training set, retrain the model, and update the model parameters.
[0250] Example 1: After ICU patients are connected to the adverse prognosis dynamic warning system for single-injury and multiple-injury sepsis patients. The physiological ventilation parameter sensing subsystem monitors the patient's physiological parameters in real time, and transmits the patient's parameters within one hour to the corresponding adverse prognosis dynamic warning subsystem in real time. Taking hours as the minimum time unit, if there are multiple valid sampling points within a one-hour span in the original data, the intelligent dynamic warning subsystem calculates the total urine volume of the patient within the hour for the urine volume, and calculates the median for other parameters, and then interpolates the missing data values. According to the ISS score, it is determined whether the patient is a single-injury or multiple-injury sepsis patient, and then the interpolated physiological parameters are input into the corresponding adverse prognosis dynamic warning model. The model outputs the patient's corresponding mortality rate within the next 7 days, mortality rate within 30 days, expected ICU hospitalization time, and expected total hospitalization time.
[0251] Example 2: Changes in the clinical medical environment. After ICU patients are connected to the adverse prognosis dynamic warning system for single-injury and multiple-injury sepsis patients. The physiological ventilation parameter sensing subsystem monitors the patient's physiological parameters in real time. Taking hours as the minimum time unit, if there are multiple valid sampling points within a one-hour span in the original data, the corresponding adverse prognosis dynamic warning subsystem calculates the total urine volume of the patient within the hour for the urine volume, and calculates the median for other parameters, and then interpolates the missing data values. The deep learning model update subsystem uses the input physiological parameters and prognosis information, combined with the incremental training method, that is, updates the model parameters without changing the model structure, and transmits the patient's parameters within one hour in real time to the adverse prognosis dynamic warning subsystem after the model parameters are updated.
[0252] Embodiment 3
[0253] See also Figure 3 , Figure 3 : is a schematic diagram of the structure of a physiological parameter dynamic warning device disclosed in an embodiment of the present invention. Figure 3 The physiological parameter dynamic warning device described is applied to the field of artificial intelligence technology, and processes the physiological parameter information to be processed to obtain dynamic warning information. The present invention is not limited to this embodiment. Figure 3 As shown, the physiological parameter dynamic early warning device may include the following operations:
[0254] S301, a parameter acquisition module, used to acquire physiological parameter information; the physiological parameter information includes first physiological parameter information, second physiological parameter information and third physiological parameter information;
[0255] S302, a model training module, used to train a preset deep learning model using the physiological parameter information to obtain an optimized deep learning model;
[0256] S303, a dynamic warning module, is used to use the optimized deep learning model to process the physiological parameter information to be processed to obtain dynamic warning information.
[0257] Embodiment 4
[0258] See also Figure 4 , Figure 4 is a schematic diagram of the structure of another physiological parameter dynamic warning device disclosed in an embodiment of the present invention. Figure 4 The physiological parameter dynamic warning device described is applied to the field of artificial intelligence technology, and processes the physiological parameter information to be processed to obtain dynamic warning information. The present invention is not limited to this embodiment. Figure 4 As shown, the physiological parameter dynamic early warning device may include the following operations:
[0259] A memory 401 storing executable program codes;
[0260] a processor 402 coupled to the memory 401;
[0261] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the physiological parameter dynamic early warning method described in the first and second embodiments.
[0262] Embodiment 5
[0263] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the physiological parameter dynamic early warning method described in Embodiment 1 and Embodiment 2.
[0264] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, i.e., they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.
[0265] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution can be essentially or partly contributed to the prior art in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.
[0266] Finally, it should be noted that the physiological parameter dynamic warning method and device disclosed in the embodiment of the present invention discloses only the preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, ordinary technicians in this field should understand that it is still possible to modify the technical solutions recorded in the aforementioned embodiments, or to replace some of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A physiological parameter dynamic early warning method, characterized in that: The method comprises: S1, obtaining physiological parameter information; the physiological parameter information includes first physiological parameter information, second physiological parameter information and third physiological parameter information; S2, using the physiological parameter information to train a preset deep learning model to obtain an optimized deep learning model; S3, using the optimized deep learning model to process the physiological parameter information to be processed to obtain dynamic warning information.
2. The physiological parameter dynamic early warning method according to claim 1, characterized in that: The method of using the physiological parameter information to train a preset deep learning model to obtain an optimized deep learning model includes: S21, preprocessing the physiological parameter information to obtain preprocessed physiological parameter information; The preprocessed physiological parameter information includes first preprocessed physiological parameter information, second preprocessed physiological parameter information and third preprocessed physiological parameter information; S22, processing the pre-processed physiological parameter information to obtain physiological parameter evaluation information; S23, using the physiological parameter evaluation information to train a preset deep learning model to obtain an optimized deep learning model.
3. The physiological parameter dynamic early warning method according to claim 2, characterized in that: The preprocessing of the physiological parameter information to obtain preprocessed physiological parameter information includes: S211, normalizing the physiological parameter information to obtain normalized physiological parameter information; S212, performing missing value processing on the normalized physiological parameter information to obtain preprocessed physiological parameter information.
4. The physiological parameter dynamic early warning method according to claim 2, characterized in that: The processing of the pre-processed physiological parameter information to obtain physiological parameter evaluation information includes: S221, processing the first pre-processed physiological parameter information to obtain a first test index judgment matrix A; Using a first geometric mean model, the first test indicator judgment matrix A is processed to obtain a geometric mean of all elements in each row of the first test indicator judgment matrix A; The first geometric mean model is: Among them, the geometric mean of all elements in each row of the judgment matrix A is a ij is the element in the i-th row and j-th column of the judgment matrix A, and n is the number of rows and columns of the judgment matrix A; The geometric mean is Perform the first normalization to obtain the first normalization parameter information w a ; The first normalization is: Among them, the first normalized parameter information w a =(w a1 ,w a2 ,...w an ); S222, processing the second pre-processed physiological parameter information to obtain a second test index judgment matrix B; The judgment matrix B is processed using the second geometric mean model to obtain the second normalized parameter information w b ; S223, processing the third pre-processed physiological parameter information to obtain a third test index judgment matrix C; The judgment matrix C is processed using the third geometric mean model to obtain the third normalized parameter information w c ; S224: Fusing the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information.
5. The physiological parameter dynamic early warning method according to claim 4, characterized in that: The second geometric mean model is used to process the judgment matrix B to obtain the second normalized parameter information w b ,include: Using a second geometric mean model, the judgment matrix B is processed to obtain a geometric mean of all elements in each row of the judgment matrix B; The second geometric mean model is: Among them, the geometric mean of all elements in each row of the judgment matrix B is a pq is the element in the pth row and qth column of the judgment matrix B, and m is the number of rows and columns of the judgment matrix B; The geometric mean is Perform the second normalization to obtain the second normalization parameter information w b ; The second normalization is: Among them, the second normalized parameter information w b =(w b1 ,w b2 ,...w bm ).
6. The physiological parameter dynamic early warning method according to claim 4, characterized in that: The third geometric mean model is used to process the judgment matrix C to obtain the third normalized parameter information w c ,include: Using a third geometric mean model, the judgment matrix C is processed to obtain a geometric mean of all elements in each row of the judgment matrix C; The third geometric mean model is: Among them, the geometric mean of all elements in each row of the judgment matrix C is a i′j′ is the element in the i′th row and j′th column of the judgment matrix C, and n′ is the number of rows and columns of the judgment matrix C; The geometric mean Perform the third normalization to obtain the third normalization parameter information w c ; The third normalization is: Among them, the third normalized parameter information w c =(w c1 ,w c2 ,...w cn′ ).
7. The physiological parameter dynamic early warning method according to claim 4, characterized in that: The fusing the first normalized parameter information, the second normalized parameter information and the third normalized parameter information to obtain physiological parameter evaluation information includes: S2241, using a data centralization model, processing the first normalized parameter information, the second normalized parameter information, and the third normalized parameter information to obtain centralization parameter information; The model expression in the data center is: The centralization parameter information is X=[w a -u,w b -u,w c -u], w a =(w a1 ,w a2 ,...w an ), w b =(w b1 ,w b2 ,...w bm ), w c =(w c1 ,w c2 ,...w cn′ ) in the l3th element; S2242, processing the centralization parameter information to obtain a covariance matrix; The covariance matrix expression is: ∑=XX T Where ∑ is the covariance matrix and T is the transpose; S2243, performing eigenvalue decomposition on the covariance matrix to obtain m eigenvalues; S2244, sorting the m eigenvalues, and extracting eigenvectors corresponding to the first k eigenvalues; S2245, using the eigenvector reconstruction model, processing the eigenvectors corresponding to the first k eigenvalues to obtain physiological parameter evaluation information; The feature vector reconstruction model expression is: Y=[u1X,u2X,…,u k X] Among them, Y is the physiological parameter evaluation information, [u1,u2,…,u k ] is the eigenvector corresponding to the first k eigenvalues.
8. The physiological parameter dynamic early warning method according to claim 1, characterized in that: After the physiological parameter information to be processed is processed by using the optimized deep learning model to obtain dynamic warning information, the method further includes: When the environment changes, the optimized deep learning model parameters are updated through incremental training using a preset deep learning model update subsystem, so that the optimized deep learning model adapts to the current environment; The incremental training method is: randomly selecting small-scale data in the current environment to add to the original training set, retraining the optimized deep learning model, and obtaining updated model parameters.
9. A physiological parameter dynamic early warning device, characterized in that: The device comprises: A parameter acquisition module, used to acquire physiological parameter information; the physiological parameter information includes first physiological parameter information, second physiological parameter information and third physiological parameter information; A model training module, used to train a preset deep learning model using the physiological parameter information to obtain an optimized deep learning model; The dynamic warning module is used to process the physiological parameter information to be processed using the optimized deep learning model to obtain dynamic warning information.
10. A physiological parameter dynamic early warning device, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the physiological parameter dynamic early warning method as described in any one of claims 1-8.
