Patient risk dynamic assessment method based on deep learning model and reinforcement learning algorithm

Abstract

The invention discloses a patient risk dynamic assessment method based on a deep learning model and a reinforcement learning algorithm. The method comprises the following steps: acquiring basic information, physiological parameters, laboratory detection data and medical record information of a patient to be assessed; preprocessing the acquired data to obtain preprocessed first data; inputting the first data into a trained risk assessment model for real-time risk assessment to obtain a first assessment result; obtaining detection data of re-admission of the to-be-evaluated patient and physiological parameters in the treatment process, and processing the data to obtain processed second data; inputting the second data into the trained risk assessment model to re-assess the risk to obtain a second assessment result; and generating a risk assessment report according to the basic information of the patient, the first assessment result and the second assessment result. According to the method, the risk of the patient can be accurately and dynamically evaluated in real time, the quality and efficiency of medical risk evaluation are improved, accurate suggestions are provided for doctors, and more effective medical guarantee is provided for the patient.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and particularly to a method, system, device and medium for dynamically evaluating patient risks based on a deep learning model and a reinforcement learning algorithm. Background Art

[0002] In the medical field, the existing technical solutions for patient risk assessment mainly include risk assessment methods such as rule-based, traditional statistical analysis, and simple machine learning models. Rule-based methods rely on doctors' experience and judge risks based on clinical manifestations and physiological parameter thresholds; traditional statistical analysis methods use statistical calculations of indicators and build models, such as logistic regression; simple machine learning model methods use models such as decision trees and naive Bayes for evaluation. These existing technologies have limited data processing capabilities, are difficult to integrate massive multi-source patient data, and cannot mine potential value. The model has poor adaptability and is difficult to cope with different patient groups and complex disease condition changes, and the adjustment cost is high. It is impossible to dynamically evaluate the changes in the disease condition in real time, and mostly provides static evaluation results, which are likely to delay treatment. Moreover, the accuracy and precision are insufficient. Due to the inability to handle complex data relationships, it is easy to lead to mistakes in treatment decisions, increase medical costs, and affect patient health. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, system, device and medium for dynamically evaluating patient risks based on a deep learning model and a reinforcement learning algorithm, so as to achieve a more accurate and real-time dynamic evaluation of patient risks and improve the quality and efficiency of medical risk assessment.

[0004] The present invention is realized by the following technical solutions:

[0005] In a first aspect, a method for dynamically evaluating patient risks based on a deep learning model and a reinforcement learning algorithm provided by an embodiment of the present invention includes:

[0006] Obtain the basic information, physiological parameters, laboratory test data and medical record information of the patient to be evaluated;

[0007] Preprocess the obtained data to obtain the first preprocessed data;

[0008] Input the first data into the trained risk assessment model for real-time risk assessment to obtain a first evaluation result;

[0009] Obtain the test data of the patient to be evaluated for readmission and the physiological parameters during the treatment process, and process the data to obtain the second processed data;

[0010] Input the second data into the trained risk assessment model to re-evaluate the risk and obtain a second evaluation result;

[0011] Generate a risk assessment report based on the basic information, the first assessment result, and the second assessment result of the patient.

[0012] Further, the risk assessment model adopts a long short-term memory network model, and the long short-term memory network model includes three hidden layers. Among them, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 64, and the number of nodes in the third hidden layer is 32.

[0013] Further, preprocess the obtained data to obtain the first preprocessed data, which specifically includes:

[0014] Clean the physiological parameters according to the threshold range set by medical statistics to obtain the cleaned physiological data;

[0015] Identify outliers for the laboratory test data based on the normal reference range, and use the median filling method to process the missing data to obtain the preprocessed test data;

[0016] For the medical record information, use natural language processing technology to perform semantic analysis and error correction on the text data in the electronic medical record to obtain the preprocessed medical record data;

[0017] Perform normalization and standardization processing on the cleaned physiological data, the preprocessed test data, and the preprocessed medical record data to obtain the preprocessed data;

[0018] Sort and align the preprocessed data according to the time series to make the data from different sources match in the time dimension, and obtain the first preprocessed data.

[0019] Further, the normalization and standardization processing of the cleaned physiological data specifically includes:

[0020] Use the min-max normalization method to map the cleaned physiological data to the 0-1 interval.

[0021] Further, the risk assessment report includes one or more risk characteristics of the patient to be evaluated, the risk characteristic scores of each risk characteristic, and the explanations corresponding to the risk characteristic scores.

[0022] In a second aspect, a patient risk dynamic assessment system based on a deep learning model and a reinforcement learning algorithm provided by an embodiment of the present invention includes: a first acquisition module, a preprocessing module, a first assessment module, a processing module, a second assessment module, and a report generation module.

[0023] The first acquisition module is used to acquire the basic information, physiological parameters, laboratory test data, and medical record information of the patient to be evaluated;

[0024] The preprocessing module preprocesses the acquired data to obtain the first preprocessed data;

[0025] The first evaluation module is used to input the first data into the trained risk assessment model for real-time risk assessment to obtain the first evaluation result;

[0026] The processing module is used to obtain the detection data of the patient's readmission and the physiological parameters during the treatment process, and process the data to obtain the second processed data;

[0027] The second evaluation module is used to input the second data into the trained risk assessment model to re-evaluate the risk and obtain the second evaluation result;

[0028] The report generation module is used to generate a risk assessment report based on the patient's basic information, the first evaluation result, and the second evaluation result.

[0029] Further, the risk assessment model adopts a long short-term memory network model, and the long short-term memory network model includes three hidden layers. Among them, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 64, and the number of nodes in the third hidden layer is 32.

[0030] Further, the preprocessing module includes a cleaning unit, an outlier processing unit, a semantic analysis unit, a normalization processing unit, and a time alignment unit;

[0031] The cleaning unit is used to clean the physiological parameters according to the threshold range set by medical statistics to obtain the cleaned physiological data;

[0032] The outlier processing unit is used to identify outliers in the laboratory test data based on the normal reference range and use the median filling method to process the missing data to obtain the preprocessed test data;

[0033] The semantic analysis unit is used to perform semantic analysis and error correction on the text data in the electronic medical record of the medical record information by using natural language processing technology to obtain the preprocessed medical record data;

[0034] The normalization processing unit performs normalization and standardization processing on the cleaned physiological data, the preprocessed test data, and the preprocessed medical record data to obtain the preprocessed data;

[0035] The time alignment unit is used to organize and align the preprocessed data according to the time series to make the data from different sources match in the time dimension to obtain the first preprocessed data.

[0036] In a third aspect, an electronic device provided by an embodiment of the present invention includes a processor, an input device, an output device, and a memory. The processor is respectively connected to the input device, the output device, and the memory. The memory is used to store a computer program, and the computer program includes program instructions. It is characterized in that the processor is configured to call the program instructions to execute the method described in the first embodiment.

[0037] In a fourth aspect, a computer-readable storage medium provided by an embodiment of the present invention stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, the processor is caused to execute the method described in the first embodiment.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] A method, system, device, and medium for dynamically evaluating patient risk based on a deep learning model and a reinforcement learning algorithm provided by an embodiment of the present invention use a risk assessment model constructed by a long short-term memory network model to mine hidden patterns and complex relationships in patient data, learn long-term dependence features in time series data, and the reinforcement learning algorithm optimizes the evaluation strategy according to the patient's real-time feedback and dynamically adjusts the model parameters. By combining the long short-term memory network model with the reinforcement learning algorithm, the risk assessment model has a powerful adaptive ability, can achieve more accurate and real-time dynamic evaluation of patient risk according to different patient groups and complex and changeable conditions, improve the quality and efficiency of medical risk assessment, provide accurate auxiliary suggestions for doctors, and provide more effective medical protection for patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts. In the drawings:

[0041] Figure 1 is a flowchart of a method for dynamically evaluating patient risk based on a deep learning model and a reinforcement learning algorithm provided by the first embodiment of the present invention;

[0042] Figure 2 is a structural block diagram of a system for dynamically evaluating patient risk based on a deep learning model and a reinforcement learning algorithm provided by another embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the embodiments and the accompanying drawings. The illustrative embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0044] Embodiment 1

[0045] As Figure 1 shown, a method for dynamically evaluating the risk of patients based on a deep learning model and a reinforcement learning algorithm provided by the first embodiment of the present invention is applicable to the system for dynamically evaluating the risk of patients based on a deep learning model and a reinforcement learning algorithm provided by the embodiments of the present invention. The method includes the following steps:

[0046] Obtain the basic information, physiological parameters, laboratory test data, and medical record information of the patient to be evaluated;

[0047] Preprocess the obtained data to obtain the first preprocessed data;

[0048] Input the first data into the trained risk assessment model for real-time risk assessment to obtain the first assessment result;

[0049] Obtain the test data for the readmission of the patient to be evaluated and the physiological parameters during the treatment process, and process the data to obtain the second processed data;

[0050] Input the second data into the trained risk assessment model to re-evaluate the risk and obtain the second assessment result;

[0051] Generate a risk assessment report based on the basic information, the first assessment result, and the second assessment result of the patient.

[0052] The method for dynamically evaluating the risk of patients based on a deep learning model and a reinforcement learning algorithm provided by the embodiments of the present invention needs to analyze and evaluate the data of multiple aspects of the patient. The data of the patient to be evaluated is comprehensively collected from a variety of medical devices and information systems. For example: The key physiological parameters such as the patient's heart rate, blood pressure (systolic blood pressure, diastolic blood pressure), blood oxygen saturation, and respiratory rate are obtained from the bedside detector in real time at a frequency of 1 to 5 times per second to ensure the timeliness and continuity of the data. The latest blood test results are obtained from the laboratory test equipment every 1 to 24 hours (set according to the urgency of the test item and the clinical needs), including the hemoglobin concentration accurate to 0.1 g / dL, blood glucose accurate to 0.1 mmol / L, electrolyte levels (such as the concentrations of potassium, sodium, chloride ions, etc. accurate to 0.1 mmol / L), etc., and the latest results are automatically obtained at the set time. For emergency test items such as myocardial enzymes, the data will be transmitted to the system server within 5 minutes after the test is completed.

[0053] Periodically and accurately extract the detailed medical history of patients from the electronic medical record system using Structured Query Language (SQL), covering past diseases (disease names, diagnosis time, treatment process, etc.), surgical records (surgical names, surgery time, surgical results, etc.), and allergy history (names of allergic substances, severity of allergic reactions, etc.). Obtain treatment process data from treatment equipment every 5 to 60 minutes (depending on the working mode of the treatment equipment and the data update frequency), such as the medication dose of the infusion pump accurate to 0.01 ml / h, the respiratory parameters of the ventilator (tidal volume, respiratory rate, inspiratory pressure, etc.), the dialysis time of the dialysis equipment accurate to 1 minute, and the dialysis fluid flow accurate to 1 ml / min, etc. Patient movement data, sleep data, etc. can also be obtained through the intelligent wearable devices worn by patients to enrich the patient data dimension.

[0054] After obtaining data from multiple devices, preprocess the data. For physiological parameters, set a reasonable threshold range based on medical statistics. For example, a heart rate below 40 beats per minute or above 150 beats per minute, a systolic blood pressure below 90 mmHg or above 180 mmHg, a diastolic blood pressure below 60 mmHg or above 110 mmHg, a blood oxygen saturation below 90% etc. are regarded as outliers, and methods such as replacing with the mean of adjacent data or linear interpolation are used for correction. For laboratory test data, identify and process outliers according to the normal reference range of the test items (such as the normal range of hemoglobin for men is 120 - 160 g / dL, for women is 110 - 150 g / dL, etc.), and the median filling method can be used to handle missing data. For the text data in the electronic medical record, use natural language processing technology for semantic analysis and error correction to ensure the accuracy of the information. Then, perform normalization and standardization processing on the data.

[0055] Specifically include: For physiological parameters, use the min - max normalization method to map them to the [0, 1] interval. The formula is:

[0056]

[0057] where x is the original data, x min is the minimum value in the dataset, x max is the maximum value in the dataset, and x norm is the normalized data. Using the min - max normalization method to map the data to the [0, 1] interval retains the relative distribution of the physiological data but is sensitive to outliers.

[0058] For laboratory test data, according to the data distribution characteristics, if it is approximately normally distributed, use the Z - score standardization method. The formula is:

[0059]

[0060] Among them, y is the original data point, μ is the mean of the dataset, and σ is the standard deviation of the dataset. Using the Z-score normalization method, the data is mapped to the standard normal distribution (mean 0, standard deviation 1). The distribution of the normalized data is consistent with the distribution form of the original data, facilitating model training and feature comparison.

[0061] For categorical data, such as gender (male, female), disease type (such as heart disease, diabetes, etc.), one-hot encoding is used to convert it into a numerical vector. Finally, all the processed data is sorted and aligned according to the time series to ensure that data from different sources matches in the time dimension, facilitating subsequent model analysis.

[0062] This embodiment can collect data from multiple sources such as bedside monitors, laboratory testing equipment, electronic medical record systems, and treatment equipment, set thresholds using medical statistical principles, clean the data by combining natural language processing techniques, and perform preprocessing on different types of data using specific normalization, standardization, and encoding methods, realizing the effective integration and in-depth mining of massive multi-source patient data, improving data quality, and laying a foundation for subsequent accurate evaluation.

[0063] The deep learning model can adopt gated recurrent unit (GRU), long short-term memory network (LSTM), or use the time series version in the Transformer architecture. In this embodiment, LSTM with strong time series data processing ability is adopted as the basic model architecture. The long short-term memory network model structure includes 3 hidden layers. The number of nodes in the first hidden layer is set to 128, the second layer is 64, and the third layer is 32. The input layer determines the dimension according to the collected physiological parameters and medical history data. For example, if time series data of 8 data types such as heart rate, blood pressure, blood oxygen saturation, blood sugar, hemoglobin concentration, and age, gender (after one-hot encoding) are input at the same time, considering the data one by one within the past 24 hours, with one hour as a time step and a total of 24 time steps, the input dimension is 8×24 = 192. The output layer adopts a classification output form, dividing the patient risk into three categories: low risk (0 - 0.3), medium risk (0.3 - 0.7), and high risk (0.7 - 1). Through the softmax activation function for probability conversion, the obtained risk level represents the risk of the patient re-entering the acute care facility from the post-acute care facility.

[0064] After constructing the long short-term memory network model, the long short-term memory network model is trained. A number of historical patient data are obtained from at least one medical record database, and the historical patient data is preprocessed to obtain the preprocessed patient data. The preprocessed patient data is divided into a training set, a validation set, and a test set according to the ratio of 70%, 15%, and 15%. Define the cross-entropy loss function as the loss measurement index for model training. The formula is:

[0065]

[0066] Among them, y i is the true label, and is the model prediction probability.

[0067] The Adam optimization algorithm is selected, the initial learning rate is set to 0.001, the learning rate decay strategy adopts exponential decay with a decay rate of 0.95, and decay is performed every 10 training epochs. At the same time, to prevent gradient explosion, the gradient clipping threshold is set to 1.0. The number of iterations (epochs) of model training is set to 200 times. During the training process, after each epoch ends, appropriate evaluation metrics are used to evaluate the performance of the model. For classification problems, common evaluation metrics include accuracy, precision, recall, F1-score, and confusion matrix, etc. In this example, accuracy is used as the evaluation metric.

[0068] The formula for calculating the accuracy ACC is:

[0069]

[0070] In the formula, TP is the number of true positives, that is, the number of samples that the model predicts as positive samples and are actually positive samples; TN is the number of true negatives, that is, the number of samples that the model predicts as negative samples and are actually negative samples; FP is the number of false positives, that is, the number of samples that the model predicts as positive samples but are actually negative samples; FN is the number of false negatives, that is, the number of samples that the model predicts as negative samples but are actually positive samples.

[0071] The early stopping method is adopted to prevent overfitting. When the loss function value on the validation set does not decrease for 5 consecutive epochs or the evaluation metric no longer improves, the training is stopped. The model parameters with the best performance during the training process are saved as the final risk assessment model. Through such a training process, the model can learn the complex patterns and relationships in the patient data, so as to accurately classify and evaluate the patient risk.

[0072] Taking patient Wang as an example for evaluation, the basic information, physiological parameters, laboratory test data, and medical record information of Wang are obtained. The acquired data is preprocessed to obtain the first preprocessed data. The first data is input into the trained risk assessment model for real-time risk assessment, and the assessment result is medium risk. During the treatment process, Wang's heart rate suddenly increased and blood pressure decreased. The system quickly captured these data changes, processed the newly acquired data to obtain the second data, input the second data into the trained risk assessment model to re-evaluate the risk, and the assessment result was high risk. A risk assessment report is generated based on Wang's basic information, the first assessment result, and the second assessment result. The risk assessment report not only includes the risk level but also lists the key risk characteristics that led to the risk change, such as abnormal changes in heart rate and blood pressure, and the contribution degree of these risk characteristics to the risk score (e.g., abnormal heart rate contributed 30%, and abnormal blood pressure contributed 25%). At the same time, the report also gives explanations for the risk characteristics, such as too high heart rate may increase the heart burden, and too low blood pressure may lead to insufficient organ blood supply, etc. The risk assessment report is sent to a remote device for display, and medical staff can view the report according to the remote device. The risk assessment report can be in the form of a spreadsheet or a document. The risk assessment report includes one or more note input fields configured to receive input from the caregiver's device. The patient data is updated after receiving the input in the note field from the caregiver. The risk assessment report also includes one or more notification input fields that accept user customization. After receiving the notification input field sent by the user and the user-defined notification request, a notification time is generated according to the user-defined notification request, and a notification is sent when a notification event occurs. The notification event includes determining whether the subsequent determined risk score of the corresponding patient is higher than the user-defined risk threshold, the user-specified time, and triggers sent at one or more patient times.

[0073] After receiving the risk assessment report, medical staff adjusted the treatment plan in a timely manner according to the assessment result, increased the dosage of vasoactive drugs, and strengthened the monitoring frequency of Wang's vital signs. As the treatment progressed, Wang's various physiological indicators gradually returned to normal. The risk assessment model performed a dynamic re-evaluation based on real-time data, and the risk level evaluated by the model dropped to medium risk. Medical staff adjusted the treatment plan accordingly, reduced the drug dosage, and continued to closely observe. Through this real-time dynamic assessment, it provides medical staff with timely and accurate decision-making basis, effectively ensuring the treatment effect and safety of patients.

[0074] A patient risk dynamic assessment method using a deep learning model and a reinforcement learning algorithm provided by an embodiment of the present invention can collect data from multiple sources such as bedside monitors, laboratory testing equipment, electronic medical record systems, and treatment equipment, and set thresholds using medical statistical principles and clean the data in combination with natural language processing technology. Specific normalization, standardization, and encoding methods are used for preprocessing different types of data to achieve effective integration and in-depth mining of massive multi-source patient data and improve data quality. A risk assessment model constructed using a long short-term memory network model mines hidden patterns and complex relationships in patient data, learns long-term dependence features in time series data, and a reinforcement learning algorithm optimizes the assessment strategy according to the patient's real-time feedback and dynamically adjusts the model parameters. Combining the long short-term memory network model with the reinforcement learning algorithm enables the risk assessment model to have a strong adaptive ability, and can achieve more accurate and real-time dynamic assessment of patient risks according to different patient groups and complex and changeable conditions, improve the quality and efficiency of medical risk assessment, provide accurate auxiliary suggestions for doctors, and provide more effective medical protection for patients.

[0075] Embodiment 2

[0076] As Figure 2 As shown in the figure, a patient risk dynamic assessment system based on a deep learning model and a reinforcement learning algorithm provided by the second embodiment of the present invention includes: a first acquisition module, a preprocessing module, a first assessment module, a processing module, a second assessment module, and a report generation module. The first acquisition module is used to acquire the basic information, physiological parameters, laboratory test data, and medical record information of the patient to be evaluated; the preprocessing module preprocesses the acquired data to obtain the first preprocessed data; the first assessment module is used to input the first data into the trained risk assessment model for real-time risk assessment to obtain a first assessment result; the processing module is used to acquire the test data of the patient to be evaluated for readmission and the physiological parameters during the treatment process, and process the data to obtain the second processed data; the second assessment module is used to input the second data into the trained risk assessment model to re-evaluate the risk to obtain a second assessment result; the report generation module is used to generate a risk assessment report according to the basic information, the first assessment result, and the second assessment result of the patient.

[0077] Among them, the risk assessment model uses a long short-term memory network model, and the long short-term memory network model includes three hidden layers. Among them, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 64, and the number of nodes in the third hidden layer is 32.

[0078] The preprocessing module includes a cleaning unit, an outlier processing unit, a semantic analysis unit, a normalization processing unit, and a time alignment unit; the cleaning unit is used to clean the physiological parameters according to the threshold range set by medical statistics to obtain the cleaned physiological data;

[0079] The outlier processing unit is used to identify outliers in laboratory test data based on the normal reference range, and use the median filling method to process missing data to obtain preprocessed test data;

[0080] The semantic analysis unit is used to perform semantic analysis and error correction on the text data in the electronic medical record by using natural language processing technology on the medical record information to obtain preprocessed medical record data;

[0081] The normalization processing unit performs normalization and standardization processing on the cleaned physiological data, preprocessed test data, and preprocessed medical record data to obtain preprocessed data;

[0082] The time alignment unit is used to organize and align the preprocessed data according to the time series, so that the data from different sources match in the time dimension, and obtain the first preprocessed data.

[0083] Among them, the execution processes of each module and unit can be executed according to the process steps of a patient risk dynamic assessment method using a deep learning model and a reinforcement learning algorithm in Embodiment 1, and will not be elaborated one by one in this embodiment.

[0084] Embodiment 3

[0085] An electronic device provided in the third embodiment of the present invention includes a processor, an input device, an output device, and a memory. The processor is respectively connected to the input device, the output device, and the memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is configured to call the program instructions to execute the method described in the first embodiment above.

[0086] It should be understood that in the embodiments of the present invention, the so-called processor may be a central processing unit (CPU), and this processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or this processor may also be any conventional processor, etc.

[0087] The input device may include a touchpad, a fingerprint acquisition sensor (for acquiring the fingerprint information and the orientation information of the fingerprint of the user), a microphone, etc., and the output device may include a display (such as an LCD), a speaker, etc.

[0088] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0089] In a specific implementation, the processor, the input device, and the output device described in the embodiments of the present invention may implement the implementation manners described in the method embodiments provided by the present invention, or may also implement the implementation manners of the system embodiments described in the embodiments of the present invention, which will not be elaborated herein.

[0090] Embodiment 4

[0091] In the fourth embodiment of the present invention, there is also provided an embodiment of a computer-readable storage medium. The computer-readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by the processor, the processor is caused to execute the method described in the first embodiment above.

[0092] The computer-readable storage medium may be an internal storage unit of the terminal described in the foregoing embodiments, such as the hard disk or memory of the terminal. The computer-readable storage medium may also be an external storage device of the terminal, such as a plug-in hard disk equipped on the terminal, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the computer-readable storage medium may also include both the internal storage unit of the terminal and the external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the terminal. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0093] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. A professional technician may use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0094] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described terminals and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0095] In several embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices or units, and can also be electrical, mechanical or other forms of connection.

[0096] The above-described specific embodiments have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A dynamic patient risk assessment method based on a deep learning model and a reinforcement learning algorithm, characterized in that, Including: Obtain the basic information, physiological parameters, laboratory test data, and medical record information of the patient to be evaluated; Preprocess the obtained data to obtain the first preprocessed data; Input the first data into the trained risk assessment model for real-time risk assessment to obtain the first assessment result; Obtain the test data for the readmission of the patient to be evaluated and the physiological parameters during the treatment process, and process the data to obtain the second processed data; Input the second data into the trained risk assessment model to re-evaluate the risk and obtain the second assessment result; Generate a risk assessment report based on the basic information of the patient, the first assessment result, and the second assessment result.

2. The method according to claim 1, wherein The risk assessment model uses a long short-term memory network model, and the long short-term memory network model includes three hidden layers. Among them, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 64, and the number of nodes in the third hidden layer is 32.

3. The method according to claim 1, wherein The preprocessing of the obtained data to obtain the first preprocessed data specifically includes: Clean the physiological parameters according to the threshold range set by medical statistics to obtain the cleaned physiological data; Identify outliers for the laboratory test data based on the normal reference range, and use the median filling method to process the missing data to obtain the preprocessed test data; For the medical record information, use natural language processing technology to perform semantic analysis and error correction on the text data in the electronic medical record to obtain the preprocessed medical record data; Perform normalization and standardization processing on the cleaned physiological data, the preprocessed test data, and the preprocessed medical record data to obtain the preprocessed data; Organize and align the preprocessed data according to the time series to make the data from different sources match in the time dimension to obtain the first preprocessed data.

4. The method according to claim 3, characterized in that, The normalization and standardization processing of the cleaned physiological data specifically includes: Use the min-max normalization method to map the cleaned physiological data to the 0-1 interval.

5. The method according to claim 1, characterized in that The risk assessment report includes one or more risk characteristics of the patient to be evaluated, the risk characteristic scores of each risk characteristic, and the explanations corresponding to the risk characteristic scores.

6. A patient risk dynamic assessment system based on a deep learning model and a reinforcement learning algorithm, characterized in that, Including: The first acquisition module, the preprocessing module, the first evaluation module, the processing module, the second evaluation module, and the report generation module, The first acquisition module is used to obtain the basic information, physiological parameters, laboratory test data, and medical record information of the patient to be evaluated; The preprocessing module preprocesses the obtained data to obtain the first preprocessed data; The first evaluation module is used to input the first data into the trained risk assessment model for real-time risk assessment to obtain the first assessment result; The processing module is used to obtain the test data for the readmission of the patient to be evaluated and the physiological parameters during the treatment process, and process the data to obtain the second processed data; The second evaluation module is used to input the second data into the trained risk assessment model to re-evaluate the risk and obtain the second assessment result; The report generation module is used to generate a risk assessment report based on the basic information of the patient, the first assessment result, and the second assessment result.

7. The system according to claim 6, wherein The risk assessment model uses a long short-term memory network model, and the long short-term memory network model includes three hidden layers. Among them, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 64, and the number of nodes in the third hidden layer is 32.

8. The system according to claim 6, wherein The preprocessing module includes a cleaning unit, an outlier processing unit, a semantic analysis unit, a normalization processing unit, and a time alignment unit; The cleaning unit is used to clean the physiological parameters according to the threshold range set by medical statistics to obtain the cleaned physiological data; The outlier processing unit is used to identify outliers in the laboratory test data based on the normal reference range and process the missing data using the median filling method to obtain the preprocessed test data; The semantic analysis unit is used to perform semantic analysis and error correction on the text data in the electronic medical record of the medical record information using natural language processing technology to obtain the preprocessed medical record data; The normalization processing unit performs normalization and standardization processing on the cleaned physiological data, the preprocessed test data, and the preprocessed medical record data to obtain the preprocessed data; The time alignment unit is used to organize and align the preprocessed data according to the time series, so that the data from different sources are matched in the time dimension to obtain the preprocessed first data.

9. An electronic device, comprising a processor, an input device, an output device, and a memory, the processor is respectively connected to the input device, the output device, and the memory, the memory is used for storing a computer program, the computer program includes program instructions, and is characterized in that, The processor is configured to call the program instructions to execute the method according to any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by the processor, the processor is caused to execute the method according to any one of claims 1-5.

Images