Medical information processing-based sepsis patient monitoring system

Abstract

The invention discloses a sepsis patient monitoring system based on medical information processing. The sepsis patient monitoring system comprises a data set construction module, a sepsis patient risk assessment model establishment module and a sepsis patient monitoring module. The invention belongs to the field of medical information processing, and particularly relates to a sepsis patient monitoring system based on medical information processing, according to the scheme, by defining mixed distribution and defining medical oriented divergence distribution, the early warning ability for extreme physiological states is enhanced; risk-sensitive weighting processing is introduced, so that excessive alarm is prevented; the sepsis patient monitoring accuracy is further improved; according to the method, a normal reference value is taken as a demarcation point, index saturation curves are respectively adopted for two sides with relatively high and relatively low indexes, asymmetric punishment of the risk degree is realized, and the actual risk condition when the physiological indexes of a sepsis patient deviate from a normal range can be reflected more accurately; for danger signals lower than a normal value, deterioration of the illness state of the patient is captured in time, and then the patient monitoring effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information processing, and specifically refers to a sepsis patient monitoring system based on medical information processing. Background Art

[0002] A sepsis patient monitoring system is a medical information system for real-time, dynamic monitoring and evaluation of sepsis patients, aiming to help medical staff timely grasp the changes in the patient's condition. However, the general sepsis patient monitoring system has problems such as weak risk capture ability, low early prediction accuracy, insufficient warning for extreme physiological states, and high monitoring false alarm rate affected by outlier monitoring values; the general sepsis patient monitoring system has problems such as being unable to effectively handle the asymmetric risk of index deviation from the normal range, insufficient monitoring of extreme situations, and thus poor patient monitoring effect. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides a sepsis patient monitoring system based on medical information processing. Aiming at the problems of the general sepsis patient monitoring system, such as weak risk capture ability, low early prediction accuracy, insufficient warning for extreme physiological states, and high monitoring false alarm rate affected by outlier monitoring values, this solution defines a mixed distribution, which can make full use of medical prior knowledge while learning historical monitoring data, and better adapt to the physiological changes of new patients; defines a medical-oriented divergence distribution to enhance the warning ability for extreme physiological states; introduces risk-sensitive weighted processing, automatically increasing the parameter when the current physiological state is not accurately grasped and decreasing the parameter for stable patients to prevent over-alarm; thereby improving the monitoring accuracy of sepsis patients; aiming at the problems of the general sepsis patient monitoring system, such as being unable to effectively handle the asymmetric risk of index deviation from the normal range, insufficient monitoring of extreme situations, and thus poor patient monitoring effect, this solution takes the normal reference value as the demarcation point, and uses exponential saturation curves on both sides of the index being too high and too low respectively, and uses different saturation level parameters to achieve asymmetric punishment of the risk degree, which can more accurately reflect the actual risk situation when the physiological indicators of sepsis patients deviate from the normal range; for the dangerous signal below the normal value, timely capture the deterioration of the patient's condition and improve the monitoring ability for extreme situations; thereby improving the patient monitoring effect.

[0004] The technical solution adopted by the present invention is as follows: The sepsis patient monitoring system based on medical information processing provided by the present invention includes a data set construction module, a sepsis patient risk assessment model establishment module, and a sepsis patient monitoring module;

[0005] The data set construction module collects historical monitoring data and risk labels, and performs feature engineering to generate a sample set;

[0006] The sepsis patient risk assessment model establishment module is based on BiLSTM and a mixture distribution; a risk-sensitive weighted loss function is introduced to establish a sepsis patient risk assessment model;

[0007] The sepsis patient monitoring module monitors patients based on the sepsis patient risk assessment model for real-time monitoring data.

[0008] Furthermore, the dataset construction module collects historical sepsis patient monitoring data and risk assessment levels; uses the risk assessment level as a label; and performs feature engineering on the sepsis patient monitoring data to obtain a patient sample set.

[0009] Furthermore, the sepsis patient risk assessment model establishment module divides the patient sample set into a training set and a test set to establish a sepsis patient risk assessment model; specifically including the following:

[0010] Network architecture design unit; the front end extracts temporal dependencies by multiple layers of bidirectional LSTM, expressed as: , the last layer outputs , and then takes the attention aggregation as a fixed-length representation z; where, and are the bidirectional hidden states of the l-th layer at time t and time t-1 respectively; is the bidirectional LSTM of the l-th layer; is the input vector of the l-th layer at time t; H is the matrix formed by concatenating the hidden states of all time moments of the top layer L over the entire window in time; and are the bidirectional hidden states of the L-th layer at time t-T and time t respectively; the backend uses a fully connected layer to output the sepsis risk probability, expressed as: ; where, is the predicted probability of the model for the sepsis risk of the patient within the current window; is the activation function; and are the weights and biases of the fully connected layer respectively;

[0011] Mixture distribution definition unit; the mixture distribution is expressed as: ; where, is the mixture distribution; is the variational approximation distribution, representing the distribution of the model parameters under the condition of given parameters ; is the prior distribution; is the mixture weight; reflects the learning of the network for historical monitoring data;

[0012] Medical-oriented divergence distribution definition unit; the medical-oriented divergence distribution is expressed as: ; where is the medical-oriented divergence distribution; is the Rényi-β divergence; is divergence;

[0013] Loss function design unit; introducing risk-sensitive weighting, the final loss function is designed as: ; ; where L is the final loss function; is the loss weight; N is the total number of samples; i is the sample index; is the input feature of the known i-th sample and the model parameters , the sample label is probability; is the basic weight constant; is the sensitivity coefficient; is the variance of the overall prediction distribution;

[0014] Activation function design unit; taking the normal reference value s as the demarcation point, when the output at x = s is , there is no risk increment; for the two sides with high and low indicators respectively, an exponential saturation curve is adopted to obtain the activation function as: ; where a and c are saturation level parameters; b is the slope scale;

[0015] Model establishment; setting the maximum number of training times, training convergence threshold and prediction threshold; when the risk assessment of sepsis patients is based on the convergence of the training set loss or reaches the maximum number of training times, but the prediction for the test set does not meet the standard, parameter tuning is performed based on the particle swarm search algorithm and retraining is carried out; otherwise, the risk assessment model for sepsis patients is established.

[0016] Furthermore, the sepsis patient monitoring module is based on the established sepsis patient risk assessment model, and performs risk assessment on the real-time collected sepsis patient monitoring data, thereby realizing sepsis patient monitoring; when the output label is medium risk, the sampling frequency of the patient is increased, and when the output label is high risk, a warning is given.

[0017] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0018] (1) Aiming at the problems of the general sepsis patient monitoring system, such as weak risk capture ability, low early prediction accuracy, insufficient warning for extreme physiological states, and high monitoring false alarm rate caused by outlier monitoring values, this solution can better adapt to the physiological changes of new patients by defining a mixed distribution and making full use of medical prior knowledge while learning historical monitoring data; defining a medical-oriented divergence distribution to enhance the warning ability for extreme physiological states; introducing risk-sensitive weighted processing, automatically increasing parameters when the current physiological state is not well grasped and decreasing parameters for stable patients to prevent over-alarm; thereby improving the monitoring accuracy of sepsis patients.

[0019] (2) Aiming at the problem that the general sepsis patient monitoring system cannot effectively handle the asymmetric risk of indicators deviating from the normal range, insufficient monitoring of extreme situations, and thus poor patient monitoring effect, this solution takes the normal reference value as the demarcation point, and uses exponential saturation curves on both sides of the indicators being too high and too low respectively, with different saturation level parameters, to achieve asymmetric punishment of the risk level, which can more accurately reflect the actual risk situation when the physiological indicators of sepsis patients deviate from the normal range; for the dangerous signals below the normal value, timely capture the deterioration of the patient's condition and improve the monitoring ability for extreme situations; thereby improving the patient monitoring effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic flowchart of the sepsis patient monitoring system based on medical information processing provided by the present invention;

[0021] Figure 2 It is a schematic flowchart of the sepsis patient risk assessment model establishment module.

[0022] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present invention.

[0024] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0025] Example 1. Refer to Figure 1 , the sepsis patient monitoring system based on medical information processing provided by the present invention includes a data set construction module, a sepsis patient risk assessment model establishment module, and a sepsis patient monitoring module;

[0026] The data set construction module collects historical monitoring data and risk tags, and performs feature engineering to generate a sample set; and sends the data to the sepsis patient risk assessment model establishment module;

[0027] The sepsis patient risk assessment model establishment module is based on BiLSTM and a mixed distribution; introduces a risk-sensitive weighted loss function to establish a sepsis patient risk assessment model; and sends the data to the sepsis patient monitoring module;

[0028] The sepsis patient monitoring module monitors patients based on the sepsis patient risk assessment model for real-time monitoring data.

[0029] Example 2. Refer to Figure 1 , based on the above example, the data set construction module collects historical sepsis patient monitoring data and risk assessment levels; uses the risk assessment level as a tag; performs feature engineering processing on the sepsis patient monitoring data to obtain a patient sample set; the historical sepsis patient monitoring data includes blood pressure, heart rate, respiratory rate, blood oxygen saturation, body temperature, white blood cell count, lactate, creatinine, bilirubin, platelet count, and C-reactive protein; the risk assessment levels include no risk, low risk, medium risk, and high risk; a time window length T is taken, and sampling is performed once per hour; the feature engineering processing includes data conversion, data cleaning, and standardization processing.

[0030] Example 3. Refer to Figure 1 and Figure 2 , based on the above example, the sepsis patient risk assessment model establishment module divides the patient sample set into a training set and a test set, and establishes a sepsis patient risk assessment model; specifically includes the following content:

[0031] Network architecture design unit; the front end extracts temporal dependencies by multiple layers of bidirectional LSTM, expressed as: , the last layer outputs , then take the attention aggregation as the fixed-length representation z; where, and are the bidirectional hidden states of the l-th layer at time t and time t-1 respectively; is the bidirectional LSTM of the l-th layer; is the input vector of the l-th layer at time t; H is the matrix formed by concatenating the hidden states of the top layer L at all times in the entire window along the time dimension; and are the bidirectional hidden states of the L-th layer at time t-T and time t respectively; the backend uses a fully connected layer to output the sepsis risk probability, denoted as: ; where, is the predicted probability of the model for the sepsis risk of the patient within the current window; is the activation function; and are the weight and bias of the fully connected layer respectively;

[0032] Mixture distribution definition unit; the mixture distribution is denoted as: ; where, is the mixture distribution; is the variational approximation distribution, representing the distribution of the model parameters under the condition of given parameters ; is the prior distribution; is the mixture weight, with a value range of [0.1, 0.9]; reflects the learning of the network from historical monitoring data; through regulation, it can not only retain the medical prior, but also allow the model to flexibly adapt to the physiological changes of new patients;

[0033] Medical-oriented divergence distribution definition unit; the medical-oriented divergence distribution is denoted as: ; where, is the medical-oriented divergence distribution; is the Rényi-β divergence, and β is used to adjust the sensitivity to the tail probability, with a value range of [0.5, 2.0]; is divergence, which emphasizes the matching of the prior and the mixture distribution in the low probability of the tail; through the Rényi-β divergence for tail risk capture, it is extremely sensitive to the deviation of q in the low probability area of the tail, and can quickly amplify the warning signal in the extreme physiological states of sharp blood pressure drop and cytokine storm; while divergence raises the penalty intensity of the denominator when is very small, which can ensure that when the patient's indicators show abnormal values, the posterior will not be pushed to the dangerous area deviating from the medical prior due to data noise;

[0034] Loss function design unit; introducing risk-sensitive Weighted, the final loss function is designed as: ; ; where L is the final loss function; is the loss weight; N is the total number of samples; i is the sample index; is the input feature of the known i-th sample and the model parameters , and the sample label is probability; is the basic weight constant, and its value range is [0.1, 1.0]; is the sensitivity coefficient, and its value range is [0.1, 1.0]; is the variance of the overall prediction distribution.

[0035] By performing the above operations, for the problems of weak risk capture ability, low early prediction accuracy, insufficient warning of extreme physiological states, and high monitoring false alarm rate caused by outlier monitoring values in the general sepsis patient monitoring system, this solution can fully utilize medical prior knowledge while learning historical monitoring data by defining a mixed distribution, and better adapt to the physiological changes of new patients; define a medical-oriented divergence distribution to enhance the warning ability for extreme physiological states; introduce risk-sensitive weighted processing, automatically increase the parameters when the current physiological state is not accurately grasped, and decrease the parameters for stable patients to prevent over-alarm; thereby improving the monitoring accuracy of sepsis patients.

[0036] Example 4, refer to Figure 1 and Figure 2 , based on the above example, the sepsis patient risk assessment model establishment module further includes:

[0037] Activation function design unit; taking the normal reference value s as the demarcation point, when x = s, the output is , without risk increment; for both sides of the index being high and low, an exponential saturation curve is adopted respectively, and the activation function obtained is: ; where a and c are saturation level parameters, and their value range is [0.5, 3.0], realizing asymmetric punishment of the risk level; b is the slope scale, and its value range is [0.1, 10.0], determining how much deviation is regarded as significantly abnormal; gradient vanishing mitigation: compared with the traditional ReLU activation function, the negative interval is not always 0, but has a negative output. When the patient's index drops to an extremely low value in a short time during the shock stage, an effective gradient can still be obtained instead of completely losing voice, without missing dangerous signals below the normal value, which helps to quickly converge to the correct high-risk prediction; the translation parameter s aligns the maximum slope with the true clinical threshold, making the inflection point of the activation function closer to the key feature interval, so as to retain stronger gradient sensitivity at the disease grading boundary, facilitating the capture of minority sample features in unbalanced scenarios;

[0038] Model establishment; set the maximum number of training times, with a value range of [50, 200], the training convergence threshold, with a value range of [10 -4 , 10 -2 , and the prediction threshold, with a value range of [0.8, 0.95]; when the risk assessment of sepsis patients converges based on the training set loss or reaches the maximum number of training times, but the prediction for the test set does not meet the standard, parameter tuning is performed based on the particle swarm search algorithm and retraining is carried out; otherwise, the establishment of the sepsis patient risk assessment model is completed.

[0039] By performing the above operations, for the problem that the general sepsis patient monitoring system has the asymmetry risk of being unable to effectively handle the deviation of indicators from the normal range, insufficient monitoring of extreme cases, and thus poor patient monitoring effect, this solution uses the normal reference value as the demarcation point, and respectively uses exponential saturation curves on both sides of the higher and lower indicators, and uses different saturation level parameters to achieve asymmetric punishment of the risk level, which can more accurately reflect the actual risk situation when the physiological indicators of sepsis patients deviate from the normal range; for the dangerous signals below the normal value, the deterioration of the patient's condition is promptly captured, improving the monitoring ability for extreme cases; and thus improving the patient monitoring effect.

[0040] Example 6, refer to Figure 1 , based on the above example, the sepsis patient monitoring module is based on the established sepsis patient risk assessment model to perform risk assessment on the real-time collected sepsis patient monitoring data, and thus achieve sepsis patient monitoring; when the output label is medium risk, increase the sampling frequency of the patient, and when the output label is high risk, give an alarm.

[0041] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0042] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0043] The above describes the present invention and its embodiments. Such description is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and, without departing from the gist of the present invention, creatively design structural modes and embodiments similar to the technical solution, they shall fall within the protection scope of the present invention.

Claims

1. A sepsis patient monitoring system based on medical information processing, characterized in that: The system includes a dataset construction module, a sepsis patient risk assessment model establishment module, and a sepsis patient monitoring module; The dataset construction module collects historical monitoring data and risk tags, and performs feature engineering to generate a sample set; The sepsis patient risk assessment model establishment module is based on BiLSTM and a mixed distribution; introduces a risk-sensitive weighted loss function to establish a sepsis patient risk assessment model; The sepsis patient monitoring module monitors patients based on the sepsis patient risk assessment model for real-time monitoring data.

2. The sepsis patient monitoring system based on medical information processing according to claim 1, wherein: The sepsis patient risk assessment model establishment module divides the patient sample set into a training set and a test set, and establishes a sepsis patient risk assessment model; specifically including the following: Network architecture design unit; The front end extracts temporal dependencies through multiple layers of bidirectional LSTM, expressed as: , the output of the last layer , and then take the attention aggregation as the fixed-length representation z; where, and are the bidirectional hidden states of the l-th layer at time t and time t-1 respectively; is the bidirectional LSTM of the l-th layer; is the input vector of the l-th layer at time t; H is the matrix formed by concatenating the hidden states of all time moments of the top layer L over the entire window in time; and are the bidirectional hidden states of the L-th layer at time t-T and time t respectively; The back end uses a fully connected layer to output the sepsis risk probability, expressed as: ; where, is the predicted probability of the model for the sepsis risk of the patient within the current window; is the activation function; and are the weights and biases of the fully connected layer respectively; Mixed distribution definition unit; Medical-oriented divergence distribution definition unit; Loss function design unit; Activation function design unit; Model establishment; set the maximum number of training times, training convergence threshold, and prediction threshold; when the sepsis patient risk assessment converges based on the training set loss or reaches the maximum number of training times, but the prediction for the test set does not meet the standard, perform parameter tuning based on the particle swarm search algorithm and retrain; otherwise, the sepsis patient risk assessment model establishment is completed.

3. The sepsis patient monitoring system based on medical information processing according to claim 2, characterized in that: In the mixture distribution definition unit, the mixture distribution is expressed as: ; where is the mixture distribution; is the variational approximation distribution, representing the distribution of the model parameter under the condition of the given parameter ; is the prior distribution; is the mixture weight; reflects the learning of the network from historical monitoring data.

4. The sepsis patient monitoring system based on medical information processing according to claim 3, wherein: In the medical-oriented divergence distribution definition unit, the medical-oriented divergence distribution is expressed as: ; where is the medical-oriented divergence distribution; is the Rényi-β divergence; is divergence.

5. The sepsis patient monitoring system based on medical information processing according to claim 4, characterized in that: The loss function design unit introduces risk-sensitive weighting, and the final loss function is designed as: ; ; where L is the final loss function; is the loss weight; N is the total number of samples; i is the sample index; is the input feature of the known i-th sample and the model parameters , and the probability that the sample label is ; is the base weight constant; is the sensitivity coefficient; is the variance of the overall prediction distribution.

6. The sepsis patient monitoring system based on medical information processing according to claim 5, characterized in that: The activation function design unit uses the normal reference value s as the demarcation point, and outputs at x = s , with no risk increment; for the two sides with high and low indicators respectively, an exponential saturation curve is adopted to obtain the activation function as: ; where a and c are saturation level parameters; b is the slope scale.

7. The sepsis patient monitoring system based on medical information processing according to claim 6, characterized in that: The dataset construction module collects historical sepsis patient monitoring data and risk assessment levels; uses the risk assessment level as a tag; performs feature engineering processing on the sepsis patient monitoring data to obtain a patient sample set.

8. The sepsis patient monitoring system based on medical information processing according to claim 7, characterized in that: The sepsis patient monitoring module performs risk assessment on the real-time collected sepsis patient monitoring data based on the established sepsis patient risk assessment model, and then realizes sepsis patient monitoring; when the output tag is medium risk, increase the sampling frequency of the patient, and when the output tag is high risk, give an alarm.

Images