Method for identifying unexpected outflow of anticoagulant for plasma platelet collection
By combining the combined prediction method of LSTM and Dense models, the anticoagulant dosage and chemical parameters during plasma platelet collection are analyzed, and the accuracy of the identification of unanticipated outflow of anticoagulant in the prior art is solved, and more accurate The rational prediction and timely warning of traffic improve the safety and efficiency of the collection process.
Patent Information
- Application Number
- CN202411042894.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-07-31
AI Technical Summary
The prior art is difficult to accurately identify and predict the unexpected outflow of anticoagulants during plasma platelet collection, especially in dynamically changing collection environments, and relying solely on physical measurements and chemical analysis is difficult to provide sufficient accuracy and timely response.
Using the combination of LSTM physical prediction model and Dense chemical prediction model, by constructing physical feature matrix and chemical feature matrix, using one-dimensional convolutional layer and fully connected layer to process time series data, constructing joint loss functions and training joints through gradient descent optimization algorithm Model to realize multi-dimensional analysis of anticoagulant dosage and chemical parameters, and timely identify unexpected outflows.
It improves the accuracy of predicting the flow rate rationality of the anticoagulant dosage, promptly detects signs of unexpected outflows and warns in advance, enhancing the safety and efficiency of the collection process.
Smart Images

Figure CN119069044A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of detection of unexpected outflow of anticoagulants, and in particular to a method for identifying unexpected outflow of anticoagulants for plasma and platelet collection. Background Art
[0002] In the field of plasma and platelet collection, it is crucial to accurately control the dosage of anticoagulants to ensure the safety and efficiency of the collection process. Existing technologies usually rely on electronic scales and mechanical rotation pulses of anticoagulant pumps to measure and control the dosage of anticoagulants. These devices can provide basic dosage data and assist decision-making through the analysis of chemical parameters. Electronic scales to measure the dosage of anticoagulants provide a direct way to monitor the accuracy of each collection. At the same time, the anticoagulant pump achieves precise control of the dosage by converting the number of rotation pulses to the unit pulse volume. In addition, real-time monitoring of chemical parameters provides important information for evaluating the status of the anticoagulant.
[0003] Although existing technologies can measure and control the dosage of anticoagulants to a certain extent, they are obviously insufficient in identifying and predicting unexpected outflow. Especially in a dynamically changing collection environment, it is difficult to accurately predict and respond to unexpected outflow events in a timely manner by relying solely on physical measurements and simple chemical analysis. In addition, existing technologies often lack in-depth analysis of historical data and time series changes, and cannot fully utilize this information to improve the accuracy and reliability of predictions. Summary of the invention
[0004] The purpose of the present invention is to provide a method for identifying unexpected outflow of anticoagulants for plasma and platelet collection. The method can quickly and accurately determine whether unexpected outflow of anticoagulants occurs in the plasma / platelet collection system during the plasma or platelet collection process, thereby helping the plasma / platelet collection system to identify unexpected outflow of anticoagulants and issue an alarm, thereby improving the health and safety of plasma workers.
[0005] The present invention is achieved through the following technical solutions:
[0006] A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection, comprising the following steps:
[0007] S1. Collect the amount of anticoagulant measured by the electronic scale at time t, the number of rotation pulses of the anticoagulant pump at time t, and the chemical parameters of the anticoagulant at time t, perform pulse amount conversion according to the number of rotation pulses of the anticoagulant pump at time t, calculate the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump, and calculate the deviation amount according to the amount of anticoagulant measured by the electronic scale at time t and the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump;
[0008] S2. extract the amount of anticoagulant and the deviation measured by the electronic scale at time t as physical features, construct a feature matrix of the physical model, and construct an LSTM physical prediction model based on the feature matrix of the physical model, wherein the LSTM physical prediction model performs flow rationality prediction;
[0009] S3. extracting chemical parameters of the anticoagulant at time t as chemical features, constructing a feature matrix of the chemical model, and constructing a Dense chemical prediction model according to the feature matrix of the chemical model, wherein the Dense chemical prediction model predicts the unexpected outflow prediction result;
[0010] S4. respectively calculate the losses of traffic rationality, unexpected outflow and cumulative unexpected outflow, and define a joint loss function, and construct and train a joint model through the defined joint loss function and gradient descent optimization algorithm;
[0011] S5. Input new data and predict the reasonableness and expected outflow prediction results of traffic through the joint model.
[0012] Furthermore, in step S1, the pulse dosage conversion is performed according to the rotation pulse number of the anticoagulant pump at time t, and the anticoagulant dosage corresponding to the rotation pulse number of the anticoagulant pump is calculated, and the specific calculation method is as follows:
[0013] F pump (t) = P pump (t)×V pulse ;
[0014] Among them, the F pump (t) represents the amount of anticoagulant, the P pump (t) represents the number of rotation pulses of the anticoagulant pump at time t, and V pulse Indicates unit pulse volume.
[0015] Furthermore, in step S1, the deviation amount is calculated based on the anticoagulant dosage measured by the electronic scale at time t and the anticoagulant dosage corresponding to the number of rotation pulses of the anticoagulant pump, and the specific calculation method is as follows:
[0016] ΔF(t)=|F scale (t)-F pump (t)|;
[0017] Wherein, ΔF(t) represents the deviation, and F scale (t) represents the amount of anticoagulant measured by an electronic scale at time t, and the F pump (t) represents the amount of anticoagulant used.
[0018] Furthermore, the step S2 specifically includes the following sub-steps:
[0019] S201. Construct a physical feature matrix, including the amount of anticoagulant and the deviation measured by the electronic scale:
[0020]
[0021] Among them, the X phys represents the physical characteristic matrix, the F scale (·) represents the amount of anticoagulant measured by an electronic scale at a certain moment, the ΔF(.) represents the deviation per unit time, and T represents the time point, that is, the total number of time points in the time series data;
[0022] S202. Process the feature matrix through a one-dimensional convolutional layer to obtain the convolutional layer output: Among them, the c phys (t) represents the convolutional layer output at time point t, and b conv Represents the bias term of the convolutional layer, the W conv,k Indicates that K represents the width of the convolution kernel, represents the physical characteristic input at time point tk, the W conv,k Represents the weight in the convolution kernel;
[0023] S203. Process the convolution layer output through the LSTM layer to obtain the LSTM layer output;
[0024] S204. Use the fully connected layer to process the output of the LSTM layer and predict the rationality of the traffic.
[0025] Furthermore, the step S204 is specifically as follows: Among them, the represents the predicted value of the rationality of the flow at time point t, the W phys represents the weight matrix of the fully connected layer, the b phys (t) represents the LSTM layer output at time point t, and b phys Represents the bias term of the fully connected layer.
[0026] Furthermore, the step S3 specifically includes the following sub-steps:
[0027] S301. Construct a chemical feature matrix including anticoagulant concentration and viscosity:
[0028]
[0029] Among them, the X chem represents the chemical feature matrix, the C conc (·) represents the anticoagulant concentration at a certain moment, and μ vise (·) represents the viscosity of the anticoagulant at a certain moment, and T represents the time point;
[0030] S302. Process the chemical feature matrix through a fully connected layer to obtain a chemical feature processing output;
[0031] S303. Output unexpected outflow risk prediction through the fully connected layer.
[0032] Furthermore, the step S303 is specifically as follows: Among them, the represents the unexpected outflow prediction result at time point t, the W leak represents the fully connected layer weight matrix for unexpected outflow prediction, the h chem (t) represents the output of the chemical feature at time point t after the ReLU activation function, and b leak Represents the fully connected layer bias term for unexpected outflow prediction.
[0033] Furthermore, in step S4, the loss calculation of the flow rationality, unexpected outflow and cumulative unexpected outflow specifically includes:
[0034] Calculate the loss of traffic rationality: Among them, the L flow-MSE The mean square error loss represents the rationality of the flow, represents the rationality of the predicted flow at time point t, the V valid (t) represents the rationality of the actual flow at time point t;
[0035] Calculate the loss for unexpected outflow: Among them, the L leakage-MSE represents the mean square error loss of unexpected outflow, represents the model prediction of unexpected outflow at time point t, and L actual-leakage (t) represents the actual unexpected outflow at time point t;
[0036] Calculate the loss of the accumulated unexpected outflow: Among them, the L cum-leakage represents the mean square error loss of the accumulated unexpected outflow, the L actual-leakage (t) represents the actual unexpected outflow at time point t, and η represents the weight parameter of the regularization term.
[0037] Furthermore, in step S4, the joint loss function is defined as a weighted average of the flow rationality loss, the unexpected outflow loss and the cumulative unexpected outflow loss: L total =λL flow-MSE +(1-λ)(L leakage-MSE +γL cum-leakage ), wherein the Ltotal represents the joint loss function, λ represents the weight for balancing the traffic rationality loss, and γ represents the weight for balancing the cumulative unexpected outflow loss.
[0038] Furthermore, in step S4, the gradient descent optimization algorithm specifically updates the model parameters by gradient descent:
[0039]
[0040]
[0041] Wherein, W represents the weight matrix in the model, b represents the bias term in the model, α represents the learning rate, Denotes the loss function L total The partial derivative of the weight is Denotes the loss function L total Partial derivative with respect to the bias.
[0042] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0043] The present invention uses the LSTM model to process time series data characteristics, captures the changing trend of anticoagulant dosage and chemical parameters over time, thereby providing a more accurate prediction of flow rationality, and analyzes chemical characteristics through the Dense model to identify unexpected outflow risks. The present invention combines the above two models to analyze data from multiple dimensions, timely discover signs of unexpected outflow, and provide early warning. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:
[0045] Figure 1 The present invention is a flow chart of a method for identifying unexpected outflow of anticoagulants for plasma and platelet collection. DETAILED DESCRIPTION
[0046] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following.
[0047] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.
[0048] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative work are within the scope of protection of the present invention. It should be noted that relational terms such as the terms "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 such actual relationship or order between these entities or operations.
[0049] Moreover, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a list of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus. In the absence of more restrictions, an element defined by the phrase "comprises a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.
[0050] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.
[0051] A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection, such as Figure 1 , including the following steps:
[0052] S1. Collect the amount of anticoagulant measured by the electronic scale at time t, the number of rotation pulses of the anticoagulant pump at time t, and the chemical parameters of the anticoagulant at time t, perform pulse amount conversion according to the number of rotation pulses of the anticoagulant pump at time t, calculate the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump, and calculate the deviation amount according to the amount of anticoagulant measured by the electronic scale at time t and the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump;
[0053] S2. extract the amount of anticoagulant and the deviation measured by the electronic scale at time t as physical features, construct a feature matrix of the physical model, and construct an LSTM physical prediction model based on the feature matrix of the physical model, wherein the LSTM physical prediction model performs flow rationality prediction;
[0054] S3. extracting chemical parameters of the anticoagulant at time t as chemical features, constructing a feature matrix of the chemical model, and constructing a Dense chemical prediction model according to the feature matrix of the chemical model, wherein the Dense chemical prediction model predicts the unexpected outflow prediction result;
[0055] S4. respectively calculate the losses of traffic rationality, unexpected outflow and cumulative unexpected outflow, and define a joint loss function, and construct and train a joint model through the defined joint loss function and gradient descent optimization algorithm;
[0056] S5. Input new data and predict the reasonableness and expected outflow prediction results of traffic through the joint model.
[0057] Specifically, at each time point t, the system collects the amount of anticoagulant measured by the electronic scale and the number of rotation pulses of the anticoagulant pump, and records the chemical parameters of the anticoagulant. The number of rotation pulses is converted into the amount of anticoagulant through the known unit pulse volume, and the deviation between the electronic scale measurement and the pump calculated amount is calculated. The deviation is used to evaluate the accuracy of the anticoagulant pump. Secondly, the system extracts physical features, including the amount of anticoagulant measured by the electronic scale and the calculated deviation, and constructs a feature matrix. The feature matrix is processed using a one-dimensional convolutional layer to extract local features from the time series data; the output of the convolutional layer is processed by an LSTM network to capture long-term dependencies in the data and predict the rationality of the flow; further, the system extracts chemical features. For example, this embodiment selects the concentration and viscosity of the anticoagulant, constructs a chemical feature matrix, processes the chemical feature matrix through a fully connected layer, learns the relationship between chemical properties and unexpected outflow, and outputs unexpected outflow through another fully connected layer. Risk prediction; the system calculates the mean square error loss of traffic rationality, the mean square error loss of unexpected outflow and the cumulative unexpected outflow loss respectively. The above loss functions evaluate the difference between the model prediction value and the actual value, define a joint loss function, balance the impact on model training through the weighted average loss value, and then use the gradient descent optimization algorithm to update the weights and biases according to the gradient of the loss function with respect to the model parameters to minimize the joint loss function; when the model training is completed and its accuracy is verified, the new data is input into the trained joint model, and the traffic rationality and expected outflow prediction results are predicted through the joint model.
[0058] Furthermore, in step S1, the pulse dosage conversion is performed according to the rotation pulse number of the anticoagulant pump at time t, and the anticoagulant dosage corresponding to the rotation pulse number of the anticoagulant pump is calculated, and the specific calculation method is as follows:
[0059] F pump (t) = P pump (t)×V pulse ;
[0060] Among them, the F pump (t) represents the amount of anticoagulant, the P pump (t) represents the number of rotation pulses of the anticoagulant pump at time t, and V pulse Indicates unit pulse volume.
[0061] Furthermore, in step S1, the deviation amount is calculated based on the anticoagulant dosage measured by the electronic scale at time t and the anticoagulant dosage corresponding to the number of rotation pulses of the anticoagulant pump, and the specific calculation method is as follows:
[0062] ΔF(t)=|F scale (t)-F pump (t)|;
[0063] Wherein, ΔF(t) represents the deviation, and F scale (t) represents the amount of anticoagulant measured by an electronic scale at time t, and the F pump (t) represents the amount of anticoagulant used.
[0064] Furthermore, the step S2 specifically includes the following sub-steps:
[0065] S201. Construct a physical feature matrix, including the amount of anticoagulant and the deviation measured by the electronic scale:
[0066]
[0067] Among them, the X phys represents the physical characteristic matrix, the F scale (·) represents the amount of anticoagulant measured by an electronic scale at a certain moment, the ΔF(.) represents the deviation per unit time, and T represents the time point, that is, the total number of time points in the time series data;
[0068] S202. Process the feature matrix through a one-dimensional convolutional layer to obtain the convolutional layer output: Among them, the c phys (t) represents the convolutional layer output at time point t, and b conv Represents the bias term of the convolutional layer, the W conv,k Indicates that K represents the width of the convolution kernel, represents the physical characteristic input at time point tk, the W conv,k Represents the weight in the convolution kernel;
[0069] S203. Process the convolution layer output through the LSTM layer to obtain the LSTM layer output;
[0070] S204. Use the fully connected layer to process the output of the LSTM layer and predict the rationality of the traffic.
[0071] Specifically, a physical feature matrix is constructed, which integrates the anticoagulant dosage measured by the electronic scale at each time point and the calculated deviation. The physical feature matrix is a two-dimensional array, in which each row represents a feature vector at a time point, including dosage and deviation, and the columns correspond to different features. The physical feature matrix is processed by a one-dimensional convolutional layer to extract local features in the time series data. The convolution operation is filtered through a sliding window feature matrix; the LSTM layer receives the output of the convolution layer and processes it to capture the long-term dependencies in the time series data. The LSTM layer output is processed using a fully connected layer to predict the rationality of the flow, where the fully connected layer is a set of linear equations that convert the high-dimensional output of the LSTM layer into the final prediction result.
[0072] Furthermore, the step S204 is specifically as follows: Among them, the represents the predicted value of the rationality of the flow at time point t, the W phys represents the weight matrix of the fully connected layer, the h phys (t) represents the LSTM layer output at time point t, and b phys Represents the bias term of the fully connected layer.
[0073] Furthermore, the step S3 specifically includes the following sub-steps:
[0074] S301. Construct a chemical feature matrix including anticoagulant concentration and viscosity:
[0075]
[0076] Among them, the X chem represents the chemical feature matrix, the C conc (·) represents the anticoagulant concentration at a certain moment, and μ vise (·) represents the viscosity of the anticoagulant at a certain moment, and T represents the time point;
[0077] S302. Process the chemical feature matrix through a fully connected layer to obtain a chemical feature processing output;
[0078] S303. Output unexpected outflow risk prediction through the fully connected layer.
[0079] Specifically, the construction of the chemical feature matrix is similar to that of the physical feature matrix, but the focus is on chemical parameters, such as the concentration and viscosity of the anticoagulant, which affect the flow characteristics and reactivity of the anticoagulant. The fully connected layer processes the chemical feature matrix and learns the impact of chemical parameters on unexpected outflow. The ReLU activation function is used to introduce nonlinearity so that the model can learn and simulate more complex chemical behaviors. The second fully connected layer converts the processed output of the chemical features into a prediction of unexpected outflow, and predicts potential unexpected events by learning the relationship between chemical parameters and outflow risk.
[0080] Furthermore, the step S303 is specifically as follows: Among them, the represents the unexpected result at time point t, the W leak represents the fully connected layer weight matrix for unexpected outflow prediction, the h chem (t) represents the output of the chemical feature at time point t after the ReLU activation function, and b leak Represents the fully connected layer bias term for unexpected outflow prediction.
[0081] Furthermore, in step S4, the loss calculation of the flow rationality, unexpected outflow and cumulative unexpected outflow specifically includes:
[0082] Calculate the loss of traffic rationality: Among them, the L flow-MSE The mean square error loss represents the rationality of the flow, represents the rationality of the predicted flow at time point t, the V valid (t) represents the rationality of the actual flow at time point t;
[0083] Calculate the loss for unexpected outflow: Among them, the L leakage-MSE represents the mean square error loss of unexpected outflow, represents the model prediction of unexpected outflow at time point t, and L actual-leakage (t) represents the actual unexpected outflow at time point t;
[0084] Calculate the loss of the accumulated unexpected outflow: Among them, the L cum-leakage represents the mean square error loss of the accumulated unexpected outflow, the L actual-leakage (t) represents the actual unexpected outflow at time point t, and η represents the weight parameter of the regularization term.
[0085] Specifically, the mean square error loss L flow-MSE and L leakage-MSEThe difference between the reasonableness of the flow predicted by the quantified model and the actual measured value, and the cumulative unexpected outflow loss L cum-leakage To evaluate the long-term accuracy of the model prediction, it should be noted that in plasma platelet collection, even if a small amount of unexpected outflow may have little impact in the short term, the long-term accumulation may have a significant impact on the safety and efficiency of the entire collection process. The regularization term of the cumulative unexpected outflow loss function is used to impose additional penalties on the situation where the predicted non-zero outflow amount is predicted. In addition, for the rationality of the flow rate, it means the rationality of the flow rate under specific conditions (such as a specific anticoagulant dosage, time point, etc.). This can be a continuous value, indicating the rationality of the flow rate, or a binary value, indicating whether the flow rate is reasonable. For the above-mentioned actual flow rationality and actual unexpected outflow values, the flow rationality is obtained by cross-validation by comparing the anticoagulant dosage measured by the electronic scale and the output of the anticoagulant pump, and the actual unexpected outflow value is obtained by the unexpected outflow data collected by the system, which is mainly determined based on the deviation value between the expected anticoagulant dosage and the actual anticoagulant dosage. For example, when the deviation value is in the set threshold interval, it is determined as expected outflow, and when the deviation value is not in the set threshold interval, it is determined as unexpected outflow.
[0086] Furthermore, in step S4, the joint loss function is defined as a weighted average of the flow rationality loss, the unexpected outflow loss and the cumulative unexpected outflow loss: L total =λL flow-MSE +(1-λ)(L leakage-MSE +γL cum-leakage ), wherein the L total represents the joint loss function, λ represents the weight for balancing the traffic rationality loss, and γ represents the weight for balancing the cumulative unexpected outflow loss. Specifically, the purpose of the joint loss function is to provide a unified optimization goal so that the model can simultaneously learn features related to traffic rationality and unexpected outflow risk prediction during training, coordinate the optimization process of the two subtasks, balance the impact on model parameter updates, and improve the overall performance and generalization ability of the model.
[0087] Furthermore, in step S4, the gradient descent optimization algorithm specifically updates the model parameters by gradient descent:
[0088]
[0089]
[0090] Wherein, W represents the weight matrix in the model, b represents the bias term in the model, α represents the learning rate, Denotes the loss function L total The partial derivative of the weight is Denotes the loss function L total Partial derivative of the bias. Specifically, the parameters updated by the gradient descent optimization algorithm are used for the output of the integration layer of the joint model, that is:
[0091] Furthermore, as a preferred implementation of the above embodiment, a specific implementation scheme of the nonlinear integration of the above joint model is proposed, and its process is: for the output unexpected outflow risk results and flow rationality prediction results, nonlinear integration is performed using a nonlinear activation function: Among them, the represents the joint model output, σ represents the activation function, and w 1 and w 2 The determination is made through gradient descent optimization, based on the unexpected outflow risk results and the flow rationality prediction model. Indicates the unexpected outflow risk result of the output. Indicates the output traffic rationality prediction result, the b combined Represents the bias term of the joint model. The joint model is trained based on the outputs of the two subtasks as features. During training, the model of each subtask is trained independently.
[0092] Further, as a preferred implementation scheme of the above embodiment of this embodiment, a method for outputting results through decision rules is proposed, which is exemplary:
[0093]
[0094] Here, ∈ represents a set threshold value, which is used to determine the size of the difference between the two outputs, so as to determine which output is selected as the final output.
[0095] Furthermore, through the joint model proposed in the above embodiment, the final output of the model can be set according to the threshold and decision rules. According to the changes in the threshold and decision rules, the output priority of the prediction results output by the two tasks can be set. Specifically, as shown in the above example, when the difference between the prediction result values of the two tasks is too small, the output priority can be set first. Right now has a higher priority.
[0096] Further, as a preferred implementation of this embodiment, a system for identifying unexpected outflow of anticoagulants for plasma and platelet collection is proposed, which can be applied to a method for identifying unexpected outflow of anticoagulants for plasma and platelet collection, including: a data acquisition and calculation module, which is used to collect the amount of anticoagulant measured by an electronic scale at time t, the number of rotation pulses of an anticoagulant pump at time t, and the chemical parameters of the anticoagulant at time t, perform pulse dosage conversion according to the number of rotation pulses of the anticoagulant pump at time t, calculate the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump, and calculate the deviation amount according to the amount of anticoagulant measured by the electronic scale at time t and the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump;
[0097] A physical model building module is used to extract the amount of anticoagulant and the deviation measured by the electronic scale at time t as physical features, build a feature matrix of the physical model, and build an LSTM physical prediction model according to the feature matrix of the physical model. The LSTM physical prediction model performs flow rationality prediction;
[0098] A chemical model building module is used to extract chemical parameters of the anticoagulant at time t as chemical features, build a feature matrix of the chemical model, and build a dense chemical prediction model according to the feature matrix of the chemical model, wherein the dense chemical prediction model predicts the unexpected outflow prediction result;
[0099] The model training and optimization module is used to calculate the losses of traffic rationality, unexpected outflow and cumulative unexpected outflow respectively, and define the joint loss function. The joint model is constructed and trained through the defined joint loss function and gradient descent optimization algorithm. The data prediction module is used to input new data and predict the prediction results of traffic rationality and unexpected outflow through the joint model.
[0100] Furthermore, the specific implementation principle process of the above embodiment is as follows:
[0101] Step 1: Perform a self-check of the electronic scale before sampling, check the current working status of the electronic scale and obtain the zero value of the electronic scale, and store it in the system;
[0102] Step 2: Self-check the anticoagulant pump before sampling to check the working status of the anticoagulant pump;
[0103] Step 3: Pre-fill the pipeline with anticoagulant to obtain the amount of anticoagulant used on the electronic scale and the number of rotation pulses of the anticoagulant pump;
[0104] Step 4: Calculate the anticoagulant pump flow rate. Use the electronic scale to divide the anticoagulant dosage by the number of rotation pulses of the anticoagulant pump to calculate the anticoagulant pump flow rate.
[0105] Step 5: Check the rationality of the anticoagulant pump flow rate. After calculating the current anticoagulant pump flow rate data, check the threshold interval to [M0, M1]. If the check is qualified, continue the process. If the check is unqualified, alarm.
[0106] Step 6: Collect data for preprocessing, record the current weighing data of the electronic scale, and clear the rotation pulse of the anticoagulant pump;
[0107] Step 7: During the collection process, the electronic scale calculates the amount of anticoagulant, and the amount of anticoagulant W0 is calculated according to the weighing reading of the electronic scale; Step 8: During the collection process, the anticoagulant pump calculates the amount of anticoagulant, and the amount of anticoagulant W1 is calculated according to the number of rotation pulses of the anticoagulant pump multiplied by the flow rate of the anticoagulant pump;
[0108] Step 9: Determination of unexpected outflow of anticoagulant: the amount of anticoagulant calculated by the electronic scale W0 minus the amount of anticoagulant calculated by the anticoagulant pump W1 is the unexpected outflow value, and the deviation amount is determined to be [-B0, B0].
[0109] On the basis of the above steps, this embodiment also adds deep learning technology and performs joint prediction on chemical features to make the prediction results more accurate. Specifically:
[0110] The physical model building module specifically includes:
[0111] The physical feature construction unit is used to construct the physical feature matrix, including the amount of anticoagulant and the deviation measured by the electronic scale:
[0112]
[0113] Among them, the X phys represents the physical characteristic matrix, the F scale (·) represents the amount of anticoagulant measured by an electronic scale at a certain moment, the ΔF(.) represents the deviation per unit time, and T represents the time point, that is, the total number of time points in the time series data; the model convolution unit is used to process the feature matrix through a one-dimensional convolution layer to obtain the convolution layer output: Among them, the c phys (t) represents the convolutional layer output at time point t, and b conv Represents the bias term of the convolutional layer, the W conv,k Indicates that K represents the width of the convolution kernel, represents the physical characteristic input at time point tk, the W conv,k Represents the weight in the convolution kernel;
[0114] LSTM output unit, used to process the convolution layer output through the LSTM layer to obtain the LSTM layer output;
[0115] The first prediction output unit uses a fully connected layer to process the output of the LSTM layer and predict the rationality of the traffic.
[0116] Furthermore, the chemical model building module specifically includes:
[0117] Chemical feature building block for building a chemical feature matrix including anticoagulant concentration and viscosity:
[0118]
[0119] Among them, the X chem represents the chemical feature matrix, the C conc (·) represents the anticoagulant concentration at a certain moment, and μ vise (·) represents the viscosity of the anticoagulant at a certain moment, and T represents the time point;
[0120] A model processing unit, used for processing the chemical feature matrix through a fully connected layer to obtain a chemical feature processing output;
[0121] The second prediction output unit is used to output the unexpected outflow risk prediction through the fully connected layer.
[0122] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for identifying unexpected outflow of anticoagulants for plasma and platelet collection, characterized in that: The following steps are involved: S1. Collect the amount of anticoagulant measured by the electronic scale at time t, the number of rotation pulses of the anticoagulant pump at time t, and the chemical parameters of the anticoagulant at time t, perform pulse amount conversion according to the number of rotation pulses of the anticoagulant pump at time t, calculate the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump, and calculate the deviation amount according to the amount of anticoagulant measured by the electronic scale at time t and the amount of anticoagulant corresponding to the number of rotation pulses of the anticoagulant pump; S2. extract the amount of anticoagulant and the deviation measured by the electronic scale at time t as physical features, construct a feature matrix of the physical model, and construct an LSTM physical prediction model based on the feature matrix of the physical model, wherein the LSTM physical prediction model performs flow rationality prediction; S3. extracting chemical parameters of the anticoagulant at time t as chemical features, constructing a feature matrix of the chemical model, and constructing a Dense chemical prediction model according to the feature matrix of the chemical model, wherein the Dense chemical prediction model predicts the unexpected outflow prediction result; S4. respectively calculate the losses of traffic rationality, unexpected outflow and cumulative unexpected outflow, and define a joint loss function, and construct and train a joint model through the defined joint loss function and gradient descent optimization algorithm; S5. Input new data and predict the reasonableness and expected outflow prediction results of traffic through the joint model.
2. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 1, characterized in that: In step S1, pulse dosage conversion is performed according to the number of rotation pulses of the anticoagulant pump at time t, and the anticoagulant dosage corresponding to the number of rotation pulses of the anticoagulant pump is calculated. The specific calculation method is as follows: F pump (t)=P pump (t)×V pulse ; Among them, the F pump (t) represents the amount of anticoagulant, the P pump (t) represents the number of rotation pulses of the anticoagulant pump at time t, and V pulse Indicates unit pulse volume.
3. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 1, characterized in that: In step S1, the deviation is calculated based on the anticoagulant dosage measured by the electronic scale at time t and the anticoagulant dosage corresponding to the number of rotation pulses of the anticoagulant pump. The specific calculation method is as follows: ΔF(t)=|F scal e(t)-F pump (t)|; Wherein, ΔF(t) represents the deviation, and F scale (t) represents the amount of anticoagulant measured by an electronic scale at time t, and the F pump (t) represents the amount of anticoagulant used.
4. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 1, characterized in that: The step S2 specifically includes the following sub-steps: S201. Construct a physical feature matrix, including the amount of anticoagulant and the deviation measured by the electronic scale: Among them, the X phys represents the physical characteristic matrix, the F scale (·) represents the amount of anticoagulant measured by an electronic scale at a certain moment, the ΔF(.) represents the deviation at a certain moment, and the T represents the time point, that is, the total number of time points in the time series data; S202. Process the feature matrix through a one-dimensional convolutional layer to obtain the convolutional layer output: Among them, the c phys (t) represents the convolutional layer output at time point t, and b conv represents the bias term of the convolution layer, K represents the width of the convolution kernel, represents the physical characteristic input at time point tk, the W conv,k Represents the weight in the convolution kernel; S203. Process the convolution layer output through the LSTM layer to obtain the LSTM layer output; S204. Use the fully connected layer to process the output of the LSTM layer and predict the rationality of the traffic.
5. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 4, characterized in that: The step S204 is specifically as follows: Among them, the represents the predicted value of the rationality of the flow at time point t, the W phys represents the weight matrix of the fully connected layer, the h phys (t) represents the LSTM layer output at time point t, and b phys Represents the bias term of the fully connected layer.
6. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 1, characterized in that: The step S3 specifically includes the following sub-steps: S301. Construct a chemical feature matrix including anticoagulant concentration and viscosity: Among them, the X chem represents the chemical feature matrix, the C conc (·) represents the anticoagulant concentration at a certain moment, and μ vise (·) represents the viscosity of the anticoagulant at a certain moment, and T represents the time point; S302. Process the chemical feature matrix through a fully connected layer to obtain a chemical feature processing output; S303. Output unexpected outflow risk prediction through the fully connected layer.
7. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 6, characterized in that: The step S303 is specifically as follows: Among them, the represents the unexpected outflow prediction result at time point t, the W leak represents the fully connected layer weight matrix for unexpected outflow prediction, the h chem (t) represents the output of the chemical feature at time point t after the ReLU activation function, and b leak Represents the fully connected layer bias term for unexpected outflow prediction.
8. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 1, characterized in that: In step S4, the loss calculation of flow rationality, unexpected outflow and cumulative unexpected outflow specifically includes: Calculate the loss of traffic rationality: Among them, the L flow-MSE The mean square error loss represents the rationality of the flow, represents the rationality of the predicted flow at time point t, the V valid (t) represents the rationality of the actual flow at time point t; Calculate the loss for unexpected outflow: Among them, the L leakage-MSE represents the mean square error loss of unexpected outflow, represents the model prediction of unexpected outflow at time point t, and L actual-leakage (t) represents the actual unexpected outflow at time point t; Calculate the loss of the accumulated unexpected outflow: Among them, the L cum-leakage represents the mean square error loss of the accumulated unexpected outflow, the L actual-leakage (t) represents the actual unexpected outflow at time point t, and η represents the weight parameter of the regularization term.
9. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection as claimed in claim 8, characterized in that: In step S4, the joint loss function is defined as a weighted average of the flow rationality loss, the unexpected outflow loss and the cumulative unexpected outflow loss: L total =λL flow-MSE +(1-λ)(L leakage-MSE +γL cum-leakage ), wherein the L total represents the joint loss function, λ represents the weight for balancing the traffic rationality loss, and γ represents the weight for balancing the cumulative unexpected outflow loss.
10. A method for identifying unexpected outflow of anticoagulant for plasma and platelet collection according to claim 1, characterized in that: In step S4, the gradient descent optimization algorithm specifically updates the model parameters by gradient descent: Wherein, W represents the weight matrix in the model, b represents the bias term in the model, α represents the learning rate, Denotes the loss function L total The partial derivative of the weight is Denotes the loss function L total Partial derivative with respect to the bias.
Citation Information
Patent Citations
Methods, devices, and systems for measuring physical properties of fluid
CN109520889A
Recommendation method and system for therapeutic scheme for renal dialysis
CN110473629A
Non-invasive blood glucose prediction method and system for smart watch
CN118197629A
