Hemodialysis sufficiency prediction method based on transfer learning

Abstract

The invention belongs to the technical field of medical data analysis, and discloses a hemodialysis sufficiency prediction method based on transfer learning, and the method comprises the steps: S1, collecting and processing dialysis data: taking time increment, conductivity and temperature as input variables, and carrying out the standardization processing; and S2, model training: carrying out regression prediction model training on the data processed in the step S1 by using an improved multi-source domain TrAdaboost algorithm. The algorithm comprises the following steps: firstly training a plurality of weak regression devices, gradually improving the performance of each weak regression device through multi-round iteration, and finally constructing a strong regression model as a final prediction model by integrating the plurality of weak regression devices; and 3, dialysis sufficiency prediction: inputting new patient data into the final regression model trained in the step 2, and performing dialysis sufficiency prediction. The dialysis sufficiency is effectively predicted through the conductivity and the temperature.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical data analysis, and in particular relates to a hemodialysis adequacy prediction method based on transfer learning. Background Art

[0002] In recent years, with the aging of the population, the incidence of chronic kidney disease has gradually increased, which will bring great pressure to hemodialysis work. The demand for dialysis equipment has also become increasingly greater. The dialysis adequacy index is a key parameter for evaluating the effect of hemodialysis treatment, which directly affects the patient's health prognosis and the adjustment of dialysis treatment. At present, high-end imported dialysis equipment can display the important indicator of dialysis adequacy (Kt / V) in real time, but a large number of mid- and low-end equipment generally lack this function, resulting in many hospitals being unable to continuously monitor and adjust the dialysis effect of patients.

[0003] There are two main methods for calculating dialysis adequacy. One is to collect blood samples before and after dialysis. By detecting the urea nitrogen concentration before and after dialysis, the dialysis result index can be calculated, but it can only reflect the physical condition at the time of blood drawing, and the cost is relatively high. Another way to calculate the dialysis adequacy index is to display it in real time through the equipment, mainly including ultraviolet absorption rate measurement method, conductivity detection method, and effective ion dialyzability (EID) method. The ultraviolet absorption rate measurement method realizes real-time monitoring of urea removal during dialysis. The device emits ultraviolet rays of a specific wavelength. When passing through the dialysis waste liquid, urea molecules absorb part of the ultraviolet rays, resulting in a decrease in the intensity of ultraviolet rays. This attenuation process is detected and recorded by the receiving head in the device, thereby reflecting the urea removal rate during the dialysis process, and then the dialysis adequacy index can be calculated. The effective ion dialyzability (EID) method evaluates the adequacy of hemodialysis in real time by measuring the diffusion of sodium ions in the dialysate. Its principle is based on the change of the conductivity of the dialysate during dialysis: when the conductivity of the dialysate (mainly determined by sodium ions) increases briefly, the sodium ions diffuse into the blood through the dialyzer membrane, which causes the conductivity of the dialysate at the dialyzer outlet to decrease. By analyzing the difference in the conductivity of the dialyzer inlet and outlet, the conductivity clearance (Kecn) can be derived, which is equivalent to the clearance of urea and is used to calculate the dialysis dose. Conductivity detection measures the conductivity change of the dialysate in real time by installing conductivity sensors at the inlet and outlet of the dialyzer, and automatically calculates the dialysis clearance in a non-invasive manner. The conductivity of the dialysate is proportional to the electrolyte concentration, so by monitoring the difference in the conductivity of the dialysate before and after entering the dialyzer, the dialysis rate of electrolytes (such as NaCl) can be inferred. Although the ultraviolet absorbance measurement method can monitor urea removal in real time, the equipment is expensive and has high requirements for the operating environment, and is greatly affected by interferences in the dialysis waste fluid. The effective ion dilatability (EID) method evaluates the dialysis effect by measuring the diffusion of sodium ions, but its results may be affected by conductivity fluctuations. When the electrolyte concentration changes dramatically, the calculation accuracy will decrease. The calculation of the conductivity detection method is relatively complex, and the data needs to be calculated through further formulas to derive the clearance rate and dialysis adequacy index.

[0004] With the rapid development of deep learning technology, if data-driven deep learning methods can be combined with medical devices, by collecting a large amount of patient data to train models, the dialysis adequacy indicator can be predicted in a simple and effective way, which will help solve the problem that many low-end and mid-range devices cannot display this important function in real time. Through effective preprocessing of the collected data, the dialysis adequacy indicator can be accurately predicted based on the conductivity and temperature of the dialysate after dialysis, providing new ideas for data-driven methods to predict dialysis adequacy (Kt / V). Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for predicting hemodialysis adequacy based on transfer learning, which uses the conductivity and temperature of the dialysate after dialysis to perform deep learning modeling to predict the adequacy of hemodialysis; and uses transfer learning to effectively predict the dialysis adequacy of different patients.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for predicting hemodialysis adequacy based on transfer learning comprises the following steps: Step S1, collecting and processing dialysis data: Collect data during the dialysis process as basic input variables, including conductivity, temperature, and dialysis adequacy value; introduce time increment as an additional input variable; take time increment, conductivity, and temperature as input variables and perform standardization; Step S2: Model training: The data processed in step S1 is trained with the improved multi-source domain TrAdaboost algorithm for prediction model training; the algorithm first trains multiple weak regressors, and after multiple rounds of iterations, gradually improves the performance of each weak regressor, and finally integrates multiple weak regressors to construct a strong regression model as the final prediction model; Step 3: Prediction of dialysis adequacy: The new patient data is input into the final prediction model trained in step 2 to predict dialysis adequacy.

[0007] Furthermore, step S1 is specifically implemented as follows: Step S11: Data collection: The data of the dialysis patients during the dialysis process is collected through the dialysis equipment. Specifically, the conductivity and temperature of the dialysate after dialysis are monitored and collected in real time by sensors, and the dialysis adequacy value displayed by the equipment is manually recorded; the conductivity, temperature and dialysis adequacy value collected during the dialysis process are used as the basic data source; Step S12: Data processing: The collected conductivity and temperature data are preprocessed, that is, the conductivity and temperature are averaged by minute to ensure that the conductivity and temperature data are consistent with the sampling frequency of the dialysis adequacy value; at the same time, the time increment is introduced, and the time increment, conductivity and temperature are used as input variables of the prediction model together, and the input variables are standardized; The prediction model is a model for predicting the adequacy of hemodialysis, which is trained using an improved multi-source domain TrAdaboost algorithm and takes conductivity, temperature and time increment as input variables and a dialysis adequacy index as a label value.

[0008] Further, step S12 is specifically: Manually recorded dialysis adequacy values ​​at each moment A time increment is calculated accordingly. The time increment is the accumulated time based on the start time of dialysis. Its initial value is Set to 0, the time increment is calculated as follows: (1); in, For the The time corresponding to the dialysis adequacy value manually collected at each time point. At each sampling moment, the time increment, conductivity and temperature are used as the input variables of the model. , specifically expressed as: (2); in, For at time point The measured conductivity, For at time point The measured temperature, is the time increment calculated from the start of dialysis; All input variables are standardized, and the standardization formula is as follows: (3); in is the mean vector of the three input variables, is the standard deviation vector of the three input variables; After standardization, the data of each dialysis patient is expressed as , where n is the number of samples of all patients, The input variables were standardized for each patient. Manually annotated dialysis adequacy values ​​were collected for each patient.

[0009] Further, step S2 is implemented by: Step S21, initializing weights of source domain dataset: Data from multiple dialysis patients on the same device are collected, and the training data is used as the source domain dataset, and the test set is used as the target domain dataset; the source domain dataset is defined as ,in They represent different source domain datasets respectively; A patient’s dialysis data as the target domain ; For each source domain dataset , assign initial weights, and the initial weights of samples in the source domain dataset are uniformly distributed: (4); in Indicates the number of iterations At that time, The initial weight of the source domain data; Step S22, start iterative training: Use samples from all source datasets for training and to update the parameters of the weak regressor; train the weak regressor using A gated recurrent unit GRU is trained for each source dataset; During training, the mean square error (MSE) is used as the loss function, and the maximum mean difference (MMD) loss is additionally introduced to calculate the difference between the source domain data and the target domain data, and added to the total loss function. By minimizing the combined loss, the model gradually learns the migration law from the source data set to the target data set. The specific formula is expressed as follows: (5); in It is the kernel function that maps the data to the high-dimensional reproducing kernel Hilbert space RKHS, and H represents the norm of the difference in the reproducing kernel Hilbert space. is the number of samples in the target domain dataset, is the number of samples in the source domain dataset; Step S23: Update source domain data weight: Adjust the weight parameters of the source domain data It is updated based on the mean square error and maximum mean difference MMD on the source domain data, and its weight update formula is as follows: (6); in is the mean square error on the source domain data set, reflecting the prediction ability of the current weak regressor on the source domain data; is the adjustment coefficient used to balance the influence of error and MMD. MMD is the maximum mean difference, which reflects the difference in data distribution between the source domain and the target domain. After each iteration, the sample weights are adjusted according to the prediction error of the weak regressor; for samples in the source dataset , the weight update formula of source domain data is: (7); in The source domain samples are The weight after round iteration, The source domain samples are The weight of the round iteration; Step S24: Weight normalization: After each round of iteration, the weight of the source data is normalized. The specific formula is: (8); Step S25: Output the final regression model: Experienced The final model is obtained by iterating The weak regressors are weighted averagely to obtain the final strong regression model .

[0010] Furthermore, in step 3, when a new patient undergoes dialysis, the sensor collects the conductivity and temperature data during the dialysis process in real time. Consistent with the processing method in step 1, the collected data will introduce time increment as an additional input variable; then the conductivity, temperature and time increment are standardized; the processed data are input into the final regression model trained in step 2, so as to predict the dialysis adequacy index of the new patient.

[0011] Compared with the prior art, the present invention has the following advantages: Although the existing blood sample detection method before and after dialysis can directly and accurately reflect the changes in urea nitrogen concentration before and after dialysis, this method can only reflect the patient's physical condition at the time of blood sampling and cannot monitor the dynamic changes during dialysis in real time. In addition, frequent blood sampling is invasive to patients, so it is not suitable for frequent use in routine dialysis. Although the ultraviolet absorbance measurement method can monitor the clearance of urea in real time, it has high requirements for the operating environment, and the equipment cost is expensive and the maintenance cost is high. The effective ion dialyzability (EID) method evaluates the dialysis effect by the diffusion of sodium ions. However, this method depends on the sodium ion concentration in the dialysate, which does not always directly reflect the urea clearance rate, so its accuracy may be affected. Although the conductivity detection method can measure the conductivity change of the dialysate in real time in a non-invasive manner and calculate the dialysis clearance rate by installing conductivity sensors at the inlet and outlet of the dialyzer, its clearance calculation process is complicated and has limited accuracy.

[0012] The present invention introduces a prediction model based on data-driven methods (such as transfer learning) to predict the dialysis adequacy (Kt / V) through two easily accessible parameters: conductivity and temperature of the dialysate after dialysis. Compared with traditional methods, the present invention can effectively predict the dialysis adequacy (Kt / V) through conductivity and temperature. Through large-scale data training, the data-driven method can effectively capture the complex relationship between conductivity, temperature and dialysis adequacy, thereby improving the accuracy of the prediction. At the same time, this method also reduces the need for expensive and high-maintenance equipment, and has wider application potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0014] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0016] like Figure 1 As shown, this embodiment provides a method for predicting hemodialysis adequacy based on transfer learning, comprising the following steps: Step S1, collecting and processing dialysis data: Data during the dialysis process were collected as basic input variables, including conductivity, temperature, and dialysis adequacy value; time increment was introduced as an additional input variable; time increment, conductivity, and temperature were taken as input variables together and standardized.

[0017] Step S11: Data collection: The data of dialysis patients during the dialysis process are collected through dialysis equipment. Specifically, sensors are used to monitor and collect the conductivity and temperature of the dialysate after dialysis in real time, and the dialysis adequacy (Kt / V) value displayed by the equipment is manually recorded.

[0018] For each patient, the conductivity, temperature, and dialysis adequacy values ​​collected during dialysis were used as the basic data source.

[0019] Step S12: Data processing: The collected conductivity and temperature data were preprocessed. Since the conductivity and temperature data collected during the dialysis process were collected by sensors at a high frequency of 5 (seconds / time), and the dialysis adequacy value was usually recorded manually for a longer period of time (minutes), the collection interval was long and the intervals were different (in minutes). In order to ensure that the collected data were reasonable and consistent, the conductivity and temperature were averaged by minute to ensure that the conductivity and temperature data were consistent with the sampling frequency of the dialysis adequacy (Kt / V) value, so that these input variables could effectively correspond to the label values.

[0020] Since dialysis adequacy (Kt / V) is a cumulative variable and the time intervals of manual recording are different, directly using conductivity and temperature as input variables often does not work well. Therefore, the present invention introduces time increment, uses time increment together with conductivity and temperature as input variables of the prediction model, and standardizes the input variables. The model here refers to a model for predicting the adequacy of hemodialysis (Kt / V) using conductivity, temperature and time increment as input variables, dialysis adequacy index as label value, and uses the improved multi-source domain TrAdaboost algorithm to train a model for predicting the adequacy of hemodialysis (Kt / V).

[0021] Specifically, step S12: Manually recorded dialysis adequacy (Kt / V) values ​​at each moment (in minutes) corresponds to a time increment, which is the cumulative time based on the start of dialysis. Its initial value is Set to 0, the time increment is calculated as follows: (1); in, For the The time corresponding to the dialysis adequacy (Kt / V) value manually collected at each time point.

[0022] At each sampling moment, the time increment, conductivity and temperature are used as input variables of the model , specifically expressed as: (2); in, For at time point The measured conductivity, For at time point The measured temperature, is the time increment calculated from the start of dialysis; this input feature not only includes physical data (conductivity, temperature), but also reflects the change of sampling time through the time increment, solving the problem of inconsistent time intervals.

[0023] At each sampling moment, the manually annotated dialysis adequacy (Kt / V) values ​​correspond to the input features By incorporating time increments as input variables, the model is able to more accurately capture the temporal dependencies among variables during the dialysis process, especially when sampling intervals are irregular.

[0024] In order to further optimize the input data, all input variables are standardized. The standardization formula is as follows: (3); in is the mean vector of the three input variables, is the vector of standard deviations of the three input variables.

[0025] After standardization, the data of each dialysis patient is expressed as , where n is the number of samples of all patients, The input variables were standardized for each patient. Manually annotated dialysis adequacy values ​​were collected for each patient.

[0026] In summary, by introducing time increment as input feature and standardizing the collected dialysis data of each patient, the deep learning model can be effectively used to accurately predict dialysis adequacy (Kt / V). This method not only solves the problem caused by inconsistent data collection time intervals, but also enhances the model's learning ability for time dependence by introducing cumulative time, thereby improving the model's training effect and prediction performance, and solving the problem of dialysis adequacy (Kt / V) of cumulative values ​​of conductivity and temperature prediction that are not time-series.

[0027] Step S2: Model training: The data processed in step S1 is trained with an improved multi-source domain TrAdaboost algorithm (i.e., Multi SourceTrAdaboost algorithm) for a prediction model. The algorithm first trains multiple weak regressors, and after multiple rounds of iterations, gradually improves the performance of each weak regressor, and finally integrates multiple weak regressors to construct a strong regression model as the final prediction model. Multi Source TrAdaboost is a transfer learning algorithm in machine learning. In Chinese, it can usually be translated as "Multi-source domain TrAdaboost algorithm. TrAdaboost is the abbreviation of Transfer AdaBoost, which means an AdaBoost algorithm based on transfer learning. It can be directly transliterated as "TrAdaboost" in Chinese.

[0028] Step S2 is implemented by: Step S21, initializing weights of source domain dataset: The present invention collects data from multiple dialysis patients on the same device, uses the training data as the source domain dataset, and the test set as the target domain dataset. The source domain dataset is defined as ,in They represent different source domain datasets respectively.

[0029] A patient’s dialysis data as the target domain ; For each source domain dataset , assign initial weights, and the initial weights of samples in the source domain dataset are uniformly distributed: (4); in Indicates the number of iterations At that time, The initial weight of the source domain data.

[0030] Step S22, start iterative training: Use samples from all source datasets for training and to update the parameters of the weak regressor; train the weak regressor using A gated recurrent unit GRU is used to train each source data set; GRU is a recurrent neural network, where k is 3, and each source data set is trained.

[0031] During training, the mean square error (MSE) is used as the loss function, and the maximum mean difference (MMD) loss is additionally introduced to calculate the difference between the source domain data and the target domain data, and added to the total loss function. By minimizing the combined loss, the model gradually learns the migration law from the source data set to the target data set. The specific formula is expressed as follows: (5); in It is the kernel function that maps the data to the high-dimensional reproducing kernel Hilbert space RKHS, and H represents the norm of the difference in the reproducing kernel Hilbert space. is the number of samples in the target domain dataset, is the number of samples in the source domain dataset.

[0032] Step S23: Update source domain data weight: Adjust the weight parameters of the source domain data It is updated based on the mean square error and maximum mean difference MMD on the source domain data, and its weight update formula is as follows: (6); in is the mean square error on the source domain data set, reflecting the prediction ability of the current weak regressor on the source domain data; is an adjustment coefficient used to balance the impact of error and MMD. MMD is the maximum mean difference, which reflects the difference in data distribution between the source domain and the target domain.

[0033] After each iteration, the sample weights are adjusted according to the prediction error of the weak regressor; for samples in the source dataset , if the prediction error of the source data is large, the weight of the source data will be reduced to avoid the negative impact of the source data on the model; the update formula is: (7); in The source domain samples are The weight after round iteration, The source domain samples are The weight of the round iteration; if Smaller, As it approaches 1, the weight update amplitude is smaller, so the weight remains relatively large. A negative exponent means that the larger the prediction error of the source domain data, the smaller its weight is, thereby reducing its impact on subsequent model training.

[0034] Step S24: Weight normalization: After each round of iteration, in order to prevent the sample weight from increasing or decreasing infinitely, the weight of the source data must be normalized. The specific formula is: (8); Formula (8) is the weight normalization processing for the source domain. Weight normalization ensures that the sum of the sample weights remains 1 in each iteration, avoiding excessive weights from affecting the training process.

[0035] Step S25: Output the final regression model: Experienced The final model is obtained by iterating The weak regressors are weighted averagely to obtain the final strong regression model .

[0036] Step 3: Prediction of dialysis adequacy: The new patient data is input into the final prediction model trained in step 2 to predict dialysis adequacy.

[0037] When a new patient undergoes dialysis, the sensor collects the conductivity and temperature data during the dialysis process in real time. Consistent with the processing method in step 1, the collected data will introduce time increment as an additional input variable; then the conductivity, temperature and time increment are standardized; the processed data are input into the final regression model trained in step 2 to predict the dialysis adequacy index of the new patient.

[0038] In this embodiment, dialysis data of 6 patients were collected using a dialysis device. The average number of data samples for each patient was 40, and the sample time intervals were different. Table 1 shows the collected data variables. In order to ensure the effectiveness of model training, this embodiment preprocessed the data of each patient, especially introduced the key variable of time increment to reflect the dynamic changes in the time series. The preprocessed data is shown in Table 2.

[0039] In order to comprehensively evaluate the generalization ability of the model, this embodiment adopts a 6-fold cross-validation method. Specifically, in each experiment, the data of one patient is selected as the test set (Test), and the data of the remaining 5 patients are used as the training set for model training. In this way, the performance and generalization ability of the model on different patients can be fully evaluated. It is worth noting that all variables in Tables 1 and 2 are in minutes, where the manually collected dialysis adequacy (Kt / V) index is based on data sampled at different time intervals, while the data collection frequency of conductivity and temperature is once every 5 seconds. Therefore, when processing data, the average values ​​of conductivity and temperature within each minute are calculated to match the corresponding dialysis adequacy indicators.

[0040] In the process of model training, in order to avoid information leakage, this embodiment does not use the labeled test set patient data when using the maximum mean difference (MMD) loss. In addition, since the dialysis equipment does not immediately display the dialysis adequacy (Kt / V) indicator at the beginning of dialysis, this embodiment uses the patient's conductivity and temperature data in the first 15 minutes as the basic data for MMD loss calculation to migrate the target domain. This approach avoids information leakage when migrating between the source domain and the target domain.

[0041] Table 1 Data collected manually Table 2 Preprocessed data in Indicates Conductivity of the dialysate at each time point after dialysis. Indicates The temperature of the dialysate after dialysis at a certain time point. Indicates The cumulative time increment between the time point and the start of dialysis. Indicates The dialysis adequacy values ​​(Kt / V) were manually recorded at each time point.

[0042] In the model training process, this embodiment uses time increment, conductivity and temperature as input variables, dialysis adequacy (Kt / V) as a label, and applies the improved Multi Source TrAdaboost model for training. This model successfully addresses the imbalance problem between multi-source domain data by dynamically adjusting the weights of source domain and target domain data, thereby effectively improving the prediction performance in the target domain.

[0043] In order to further reduce the difference in data distribution between the source domain and the target domain, the model introduces the maximum mean difference (MMD) loss. MMD loss helps the model better adapt to different data distributions, thereby improving the accuracy of prediction. In each iteration, the Multi Source TrAdaboost model optimizes the contribution of source domain data by adjusting the weights of samples, gradually reducing the prediction error. As the iterations proceed, multiple weak models are gradually integrated to form a strong prediction model. Through this weighted integration strategy, the model can better adapt to the target domain data, thereby significantly improving the accuracy of dialysis adequacy prediction.

[0044] Performance evaluation and comparison: The prediction of dialysis adequacy is a regression task, so in order to evaluate the performance of the present invention in the prediction of dialysis adequacy (Kt / V), this embodiment uses a variety of regression evaluation indicators, including root mean square error (RMSE), mean absolute error (MAE), and determination coefficient (R²). At the same time, the consistency evaluation Bland-Altman method is used as an additional evaluation indicator to calculate the prediction value. With the true value These indicators comprehensively evaluate the prediction accuracy, stability and consistency of the model from multiple perspectives.

[0045] Specifically, the root mean square error (RMSE) is used to measure the average deviation between the predicted value and the actual value. The specific formula is: (9); The mean absolute error is the average of the absolute errors between the predicted value and the actual value. The specific formula is: (10); The coefficient of determination (R²) indicates the model's ability to explain the changes in actual data. The specific formula is: (11).

[0046] In clinical medicine, especially in the research of medical devices, Bland-Altman consistency evaluation is widely used because of its simple understanding, intuitive analysis, and reliable results. In order to ensure the accuracy of the results, it is also necessary to check whether the randomness, homoscedasticity, and normality of the comparison data difference are met before using this calculation indicator. The main use of Bland-Altman's inconsistent interval ratio is to calculate the mean bias and the limits of agreement. The Lower Limit is the lower limit of consistency, calculated as the mean deviation minus 1.96 times the standard deviation, and the Upper Limit is the upper limit of consistency, calculated as the mean deviation plus 1.96 times the standard deviation. The inconsistent interval ratio refers to the proportion of samples that fall outside the consistency limit, reflecting the degree of inconsistency between the predicted value and the actual value. A lower inconsistent interval ratio indicates that the model prediction results have good consistency with the actual value, which is an important indicator for evaluating the reliability of the model.

[0047] This example analyzes the data of 6 patients and evaluates the prediction performance of the model through 6-fold cross validation, where each fold represents the data of one patient as the test set. In each experiment, the data of one patient is selected as the test set, and the data of the remaining 5 patients are used as the training set. A total of 6 cross validations are performed. Table 3 shows the specific evaluation results.

[0048] The fluctuation of the evaluation index between each fold is within a certain range, indicating that the performance of the model under different test sets is relatively stable. In particular, the coefficient of determination (R²) of Fold 43 is 0.845, which is the lowest among all folds, but still at a high level, indicating that even in the most undesirable conditions, the model still has strong predictive ability and the results are relatively reliable. This also further proves that through effective data preprocessing, the model can effectively perform data-driven modeling under different conditions. Overall, the predictive performance of the model is stable and reliable, and can provide strong support for the prediction of dialysis adequacy.

[0049] Table 3 Prediction results of transfer learning model Before the Bland-Altman method analysis, the prerequisite verification needs to be performed. The difference randomness, homoscedasticity, and normality analysis were performed for the predicted value and the true value of each fold of dialysis adequacy. The homoscedasticity assumption and the normality assumption were all met, and the difference randomness assumption was basically not met, which may be due to the insufficiency of the model in some aspects or the incomplete capture of data characteristics. The results of the Bland-Altman consistency evaluation are shown in Table 4. Except for Fold 2, which is slightly higher than 5%, the other inconsistent interval ratios are all within 5%. The low inconsistent ratio indicates that most of the differences between the predicted value and the actual value fall within the consistency limit, indicating that the effectiveness of this method is feasible.

[0050] Table 4 Bland-Altman consistency analysis results Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Any changes, modifications, additions or substitutions made by ordinary technicians in this technical field within the essential scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A method for predicting hemodialysis adequacy based on transfer learning, characterized in that: The following steps are involved: Step S1, collecting and processing dialysis data: Collect data during the dialysis process as basic input variables, including conductivity, temperature, and dialysis adequacy value; introduce time increment as an additional input variable; take time increment, conductivity, and temperature as input variables and perform standardization; Step S2: Model training: The data processed in step S1 is trained with the improved multi-source domain TrAdaboost algorithm for prediction model training; the algorithm first trains multiple weak regressors, and after multiple rounds of iterations, gradually improves the performance of each weak regressor, and finally integrates multiple weak regressors to construct a strong regression model as the final prediction model; Step 3: Prediction of dialysis adequacy: The new patient data is input into the final prediction model trained in step 2 to predict dialysis adequacy.

2. A hemodialysis adequacy prediction method based on transfer learning according to claim 1, characterized in that: Step S1 is specifically implemented as follows: Step S11: Data collection: The data of the dialysis patients during the dialysis process is collected through the dialysis equipment. Specifically, the conductivity and temperature of the dialysate after dialysis are monitored and collected in real time by sensors, and the dialysis adequacy value displayed by the equipment is manually recorded; the conductivity, temperature and dialysis adequacy value collected during the dialysis process are used as the basic data source; Step S12: Data processing: The collected conductivity and temperature data are preprocessed, that is, the conductivity and temperature are averaged by minute to ensure that the conductivity and temperature data are consistent with the sampling frequency of the dialysis adequacy value; at the same time, the time increment is introduced, and the time increment, conductivity and temperature are used as input variables of the prediction model together, and the input variables are standardized; The prediction model is a model for predicting the adequacy of hemodialysis, which is trained using an improved multi-source domain TrAdaboost algorithm and takes conductivity, temperature and time increment as input variables and a dialysis adequacy index as a label value.

3. A method for predicting hemodialysis adequacy based on transfer learning according to claim 2, characterized in that: Specifically, step S12: Manually recorded dialysis adequacy values ​​at each moment A time increment is calculated accordingly. The time increment is the accumulated time based on the start time of dialysis. Its initial value is Set to 0, the time increment is calculated as follows: （1）； in, For the The time corresponding to the dialysis adequacy value manually collected at each time point. At each sampling moment, the time increment, conductivity and temperature are used as the input variables of the model. , specifically expressed as: （2）； in, For at time point The measured conductivity, For at time point The measured temperature, is the time increment calculated from the start of dialysis; All input variables are standardized, and the standardization formula is as follows: （3）； in is the mean vector of the three input variables, is the standard deviation vector of the three input variables; After standardization, the data of each dialysis patient is expressed as , where n is the number of samples of all patients, The input variables were standardized for each patient. Manually annotated dialysis adequacy values ​​were collected for each patient.

4. A method for predicting hemodialysis adequacy based on transfer learning according to claim 2, characterized in that: Step S2 is implemented by: Step S21, initializing weights of source domain dataset: Data from multiple dialysis patients on the same device are collected, and the training data is used as the source domain dataset, and the test set is used as the target domain dataset; the source domain dataset is defined as ,in They represent different source domain datasets respectively; A patient’s dialysis data as the target domain ; For each source domain dataset , assign initial weights, and the initial weights of samples in the source domain dataset are uniformly distributed: （4）； in Indicates the number of iterations At that time, The initial weight of the source domain data; Step S22, start iterative training: Use samples from all source datasets for training and to update the parameters of the weak regressor; Train a weak regressor using A gated recurrent unit GRU is trained for each source dataset; During training, the mean square error (MSE) is used as the loss function, and the maximum mean difference (MMD) loss is additionally introduced to calculate the difference between the source domain data and the target domain data, and added to the total loss function. By minimizing the combined loss, the model gradually learns the migration law from the source data set to the target data set. The specific formula is expressed as follows: （5）； in It is the kernel function that maps the data to the high-dimensional reproducing kernel Hilbert space RKHS, and H represents the norm of the difference in the reproducing kernel Hilbert space. is the number of samples in the target domain dataset, is the number of samples in the source domain dataset; Step S23: Update source domain data weight: Adjust the weight parameters of the source domain data It is updated based on the mean square error and maximum mean difference MMD on the source domain data, and its weight update formula is as follows: （6）； in is the mean square error on the source domain data set, reflecting the prediction ability of the current weak regressor on the source domain data; is the adjustment coefficient used to balance the influence of error and MMD. MMD is the maximum mean difference, which reflects the difference in data distribution between the source domain and the target domain. After each iteration, the sample weights are adjusted according to the prediction error of the weak regressor; for samples in the source dataset , the weight update formula of source domain data is: （7）； in The source domain samples are The weight after round iteration, The source domain samples are The weight of the round iteration; Step S24: Weight normalization: After each round of iteration, the weight of the source data is normalized. The specific formula is: （8）； Step S25: Output the final regression model: Experienced The final model is obtained by iterating The weak regressors are weighted averagely to obtain the final strong regression model .

5. The method for predicting hemodialysis adequacy based on transfer learning according to claim 1, characterized in that: Step 3, when a new patient undergoes dialysis, the sensor collects the conductivity and temperature data during the dialysis process in real time. Consistent with the processing method in step 1, the collected data will introduce time increment as an additional input variable; then the conductivity, temperature and time increment are standardized; the processed data are input into the final prediction model trained in step 2, so as to predict the dialysis adequacy index of the new patient.

Images