Method and system for regulating and controlling rotating speed of ventricular assist device based on deep learning

Abstract

The invention belongs to the technical field of medical instruments, and particularly relates to a ventricular assist device rotating speed regulation and control method and system based on deep learning. The ventricular assist device rotating speed regulation and control method based on deep learning comprises the steps of collecting current pressure information and current pump speed information of a left ventricle, sending the current pressure information and the current pump speed information into a trained long-short-term memory network model, obtaining to-be-regulated and controlled pump speed information through the long-short-term memory network model, and obtaining to-be-regulated and controlled pump speed information through the long-short-term memory network model; and sending the to-be-regulated pump speed information to the ventricle auxiliary device. According to the invention, the monitoring module of the human body LVAD can automatically adjust the control target of the rotating speed of the pump according to the hemodynamic change, and the LVAD can realize adaptive control for individual patients.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical devices, and particularly relates to a method and system for regulating the rotation speed of a ventricular assist device based on deep learning. Background Art

[0002] Cardiovascular diseases have always been the number one threat to human health and the primary factor leading to mortality. Although heart transplantation remains the most effective treatment for heart failure, the shortage of donors and clinical complications (including heart failure caused by advanced age and chemotherapy) severely limit its application. Left Ventricular Assistant Devices (LVADs), a type of mechanical pump implanted in patients with heart failure, can be used as mechanical circulatory support to assist the failing left ventricle for bridge-to-transplantation, bridge-to-recovery, and destination therapy. The emergence of ventricular assist devices can overcome the problem of insufficient transplant donors and partially or fully replace the function and role of the natural heart, which is one of the main measures for treating end-stage heart failure.

[0003] With the successful clinical application of LVADs, how to make LVADs better assist the heart in blood supply has become an issue of more concern to researchers. Currently, from the LVAD care experience shared clinically, all LVADs are controlled at a constant speed without relying on the physiological state of the human body. Running the LVAD at a constant speed may lead to ventricular suction (ventricular collapse due to too low pressure in the ventricle) or pulmonary congestion (a condition caused by too much fluid in the lungs due to ventricular hypertension). Ventricular suction may cause hemolysis, damage to heart tissue near the pump inlet, right ventricular dysfunction, or the release of ventricular thrombus and subsequent stroke, while pulmonary congestion may cause congestion of the lungs (pulmonary edema) and shortness of breath.

[0004] At present, the constant-speed continuous operation mode of the left ventricular assist device lacks a physiological feedback mechanism and requires clinicians to monitor regularly to ensure that the assisted blood pressure and flow are suitable for the patient's physiological state. The human body is a complex time-varying system, and the ventricular contractility, heart rate, arterial pressure, and venous return related to ventricular load also have time-varying properties. However, the response ability of the constant-speed continuous flow operation mode to ventricular load is limited. When the patient's circulatory state changes, if the assisted blood pressure or flow is not adjusted, some adverse events may occur.

[0005] Previously, most researchers have used a proportional-integral-derivative (PID) controller with a fixed gain to implement these physiological control systems. However, only by adjusting for this specific patient and condition can the best performance be provided. As a result, the control performance cannot be guaranteed under different patients and conditions, which may lead to more dangerous events, such as aspiration and congestion. In recent years, researchers at home and abroad have respectively carried out relevant research on the control system of left ventricular assist devices based on heart rate feedback, flow feedback, and pressure feedback. Their

[0006] The flow control scheme with heart rate feedback has low feedback accuracy, and the heart rate and flow model are non-linear, affecting the control accuracy; although the flow feedback can achieve high feedback accuracy, due to the large volume of the sensor, it is difficult to implant it into the patient's body. The pressure feedback can take into account the volume of the sensor and the feedback accuracy, but it has high requirements for the real-time computing ability of the model. And for most adaptive control strategies, such as fuzzy control, rule sets need to be determined, and the generation of these rule sets is time-consuming and inefficient due to non-linear behavior. Although the active disturbance rejection can respond to physiological demands within a certain range, its adaptive ability is limited and it is difficult to handle the differences between individual patients and dynamic pathological states. Although model predictive control can better track the changes of the reference value, due to its high computational cost, it is difficult to achieve real-time response control in the actual environment. Summary of the Invention

[0007] Aiming at the above technical problems, the purpose of the present invention is to provide a method and system for regulating the rotational speed of a ventricular assist device based on deep learning.

[0008] A method for regulating the rotational speed of a ventricular assist device based on deep learning, comprising:

[0009] Collect the current pressure information and current pump speed information of the left ventricle, send the current pressure information and the current pump speed information into a trained long short-term memory network model, obtain the pump speed information to be regulated through the long short-term memory network model, and send the pump speed information to be regulated to the ventricular assist device.

[0010] Optionally, the training process of the short-term memory network model is:

[0011] S1, obtain a plurality of actual pressure information and actual pump speed information of the left ventricle, send the actual pressure information, the actual pump speed information, and the preset reference pressure information into a preset non-linear model predictive controller to predict the future pressure information, and calculate and obtain the pump speed information to be regulated;

[0012] S2. Feed the actual pressure information, the reference pressure information, the actual pump speed information, the future pressure information, and the pump speed information to be regulated into a long short-term memory network model for training to obtain a trained long short-term memory network model.

[0013] Optionally, before S1, it further includes:

[0014] S0. Establish a coupling model between the ventricular assist device and the human cardiovascular system, and determine that the input of the coupling model is the pump speed signal of the ventricular assist device and the output is the pressure value of the left ventricle.

[0015] The prediction model of the non-linear model predictive controller is a continuous-time non-linear model of the coupling between the ventricular assist device and the human cardiovascular system.

[0016] Optionally, the long short-term memory network model includes an input layer, an LSTM layer, a fully connected layer, and a regression layer. Each time step of the input layer contains 40 features, and the LSTM layer has 128 hidden units.

[0017] Optionally, the long short-term memory network model uses the root mean square error to evaluate the deviation between the predicted value and the reference value.

[0018] Optionally, when training the short-term memory network model, collect the actual pressure information and actual pump speed information under several different physiological environments for training to obtain long short-term memory network models under different physiological environments.

[0019] Before feeding the current pressure information and the current pump speed information into the trained long short-term memory network model, switch the long short-term memory network model to the corresponding current physiological environment.

[0020] Optionally, several different physiological environments include the increase and decrease of systemic vascular resistance (SVR), rest to exercise, and passive body position change.

[0021] Optionally, the predicted pressure information is also obtained through the long short-term memory network model, and a pressure change curve graph is generated from the predicted pressure information and the current pressure information for real-time display.

[0022] A ventricular assist device speed regulation system based on deep learning includes:

[0023] A ventricular assist device, which has a pump head and a pressure sensor, and the pressure sensor is used to collect the current pressure information of the left ventricle.

[0024] A controller, the controller having a rotational speed regulation module, the rotational speed regulation module being connected to the ventricular assist device, the rotational speed regulation module being configured to obtain current pressure information and current pump speed information, send the current pressure information and the current pump speed information into a trained long short-term memory network model, obtain the pump speed information to be regulated through the long short-term memory network model, and send the pump speed information to be regulated to the ventricular assist device.

[0025] Advantages: The present invention has at least one or more of the following advantages:

[0026] 1. The present invention integrates an innovative control method of non-linear model predictive control (NMPC) and long short-term memory network (LSTM) to improve the adaptability and control accuracy of LVAD in actual clinical applications. Specifically, the present invention uses a pressure sensor to collect the left ventricular pressure as the control target, uses NMPC to achieve pressure tracking, integrates the control performance of NMPC through LSTM, solves the problem that NMPC occupies a large amount of computing resources and cannot achieve real-time control. Through the present invention, the monitoring module of the human LVAD can achieve the control target of automatically adjusting the pump speed according to hemodynamic changes, and the LVAD can achieve adaptive control for individual patients.

[0027] 2. In order to make the control scheme adapt to the changes within individual patients and improve the adaptability of the model, the present invention realizes real-time regulation of various dynamic pathological states by considering individual patients in different physiological environments.

[0028] 3. The present invention takes into account three aspects of dynamic adaptability, real-time response ability and individualized regulation strategy. Based on the dynamic pressure tracking of NMPC, the non-linear modeling ability of the LSTM network and the overall optimization design of the fusion control strategy, it not only solves many problems existing in the traditional LVAD constant speed control, meets the individualized needs of patients, improves the stability, safety and clinical practicability of the system, has important theoretical value and engineering significance, but also provides a new research direction for the intelligent and adaptive development of future LVAD control systems. Description of the Drawings

[0029] Figure 1 It is a schematic diagram of a coupling model of the present invention;

[0030] Figure 2 It is a block diagram of a rotational speed regulation strategy of the present invention;

[0031] Figure 3 It is a system structure diagram of the present invention. Detailed Embodiments

[0032] The following will describe in detail the preferred embodiments of the present invention with reference to the accompanying drawings, so as to more clearly understand the purpose, features and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not limitations on the scope of the present invention, but only to illustrate the essential spirit of the technical solution of the present invention.

[0033] In the following description, for the purpose of illustrating various disclosed embodiments, certain specific details are set forth to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the relevant art will recognize that the embodiments can be practiced without one or more of these specific details. In other instances, well-known devices, structures, and techniques associated with the present application may not be shown or described in detail so as not to unnecessarily obscure the description of the embodiments.

[0034] References throughout this specification to "one embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of "in one embodiment" or "in an embodiment" throughout the specification do not necessarily all refer to the same embodiment. Additionally, the particular features, structures, or characteristics may be combined in any manner in one or more embodiments.

[0035] In the following description, in order to clearly show the structure and working mode of the present invention, many directional terms will be used for description. However, words such as "front", "rear", "left", "right", "outer", "inner", "outward", "inward", "up", "down", etc. should be understood as convenient terms and should not be understood as limiting terms.

[0036] Refer to Figure 1 and Figure 2 , an embodiment of the present invention provides a method and system for regulating the rotational speed of a ventricular assist device based on deep learning. The method is used in the rotational speed regulation strategy of the pump head in the ventricular assist device. The method specifically includes:

[0037] Collect the current pressure information and current pump speed information of the left ventricle, send the current pressure information and current pump speed information into the trained long short-term memory network model, obtain the pump speed information to be regulated through the long short-term memory network model, and send the pump speed information to be regulated to the ventricular assist device.

[0038] In the present invention, when predicting according to the current pressure information and current pump speed information, the trained long short-term memory network model (LSTM) is used. That is to say, before actual use, the preset LSTM is trained to obtain a rotational speed regulation strategy applicable to the present invention.

[0039] The collected current pressure information, current pump speed information and other collected data can be preprocessed before being sent into the model, for example, by removing at least one of noise and outliers, to ensure the robustness and accuracy of the subsequent model.

[0040] In one embodiment, the training process of the LSTM of the present invention is as follows:

[0041] S1. Obtain the actual pressure information and actual pump speed information of several left ventricles, send the actual pressure information, actual pump speed information and preset reference pressure information into a preset non-linear model predictive controller to predict the future pressure information, and calculate the pump speed information to be regulated.

[0042] S2. Send the actual pressure information, reference pressure information, actual pump speed information, future pressure information and pump speed information to be regulated into a long short-term memory network model for training to obtain a trained long short-term memory network model.

[0043] In this embodiment, the reference pressure information is the reference left ventricular pressure LVP. Specifically, it is the left ventricular pressure LVP based on a healthy human cardiovascular system model.

[0044] After obtaining the actual pressure information and actual pump speed information for training in this embodiment, it is also preferably preprocessed, for example, by removing at least one of noise and outliers.

[0045] When training the LSTM in this embodiment, the non-linear model predictive controller (NMPC) and LSTM are integrated. Based on NMPC, dynamic pressure tracking is realized, and based on LSTM, the real-time performance of the controller is improved. Finally, the trained LSTM is used to replace NMPC for real-time speed regulation to improve the real-time response ability of the controller.

[0046] In traditional LVAD control methods, fixed-gain PID control or fuzzy control strategies are mostly used, which have problems such as limited response speed and insufficient adjustment ability. The present invention introduces NMPC, as an NMPC pressure tracker, with the left ventricular pressure (LVP) as the target control variable, predicting the future change of left ventricular pressure based on real-time hemodynamic parameters. NMPC includes three basic features: a prediction model, finite horizon rolling optimization, and feedback correction, and dynamically adjusts the pump speed through rolling optimization. By predicting the pressure values in the next N steps, NMPC can accurately track the left ventricular pressure in an optimized manner.

[0047] Although NMPC has good dynamic control performance, its computational cost is high in a real-time environment and it is difficult to meet the fast response requirements of LVAD. Therefore, in the present invention, an LSTM network is combined with NMPC to design an efficient controller to replace the traditional NMPC. The specific advantages include: high efficiency. The LSTM network learns the nonlinear dynamic relationship of LVAD through historical time series data, and completes the prediction of pressure change trend and pump speed regulation at a low computational cost. Nonlinear modeling ability. The LSTM network can effectively capture the nonlinear dynamic characteristics in time series data, making up for the deficiency of the traditional control method in nonlinear modeling of the system. Adaptability. By learning the training data under different patients and physiological states, the LSTM controller has stronger individual adaptation ability and can achieve precise control for the special pathological state of the patient.

[0048] In traditional LVAD control methods, the control objectives are mostly single indicators, and it is difficult to take into account individual physiological differences and dynamic pathological changes. The present invention proposes a fusion control strategy, which realizes the individual adaptive regulation of the left ventricular pressure of the patient by acquiring pressure data and combining the control advantages of NMPC and LSTM. The characteristics of this strategy are as follows: Individualization. Combining the actual physiological state and dynamic needs of the patient, the LVAD pump speed is adjusted to match the hemodynamic characteristics of individual patients. Scalability. Other physiological indicators such as aortic pressure and flow can be further added to the model as auxiliary objectives to enhance the response ability of LVAD to complex pathological states. Clinical practicability. The fusion control strategy effectively reduces the computational complexity of NMPC through the LSTM network, meets the real-time requirements, and at the same time avoids the dependence on implanted flow sensors, improving the feasibility of actual clinical applications.

[0049] In one embodiment, before S1, it further includes:

[0050] S0. Establish a coupling model of the ventricular assist device and the human cardiovascular system, and determine that the input quantity of the coupling model is the pump speed signal of the ventricular assist device, and the output quantity is the pressure value of the left ventricle.

[0051] Therefore, the prediction model of the nonlinear model predictive controller is a continuous-time nonlinear model of the coupling of the ventricular assist device and the human cardiovascular system.

[0052] Refer to Figure 1 , which is a simulated coupling model of the ventricular assist device and the human cardiovascular system. When constructing the coupling model, it is constructed according to the lumped parameter model. The simulation of pressure detection and the implementation of LSTM in Simulink are carried out. This coupling model is based on the first principle, including the left ventricle, systemic circulation, LVAD, and the relevant parameters and state variables are shown in Tables 1 and 2.

[0053] Table 1 List of parameters

[0054]

[0055]

[0056] Table The meanings represented by 26 state variables

[0057]

[0058] According to Kirchhoff's current and voltage laws, the state equations of the coupled model of the human cardiovascular system and the LVAD can be written as follows:

[0059]

[0060]

[0061] Among them, x represents the pressure information at time t, u(t) = ω 2 (t), ω(t) is the pump speed information, with the unit of rpm.

[0062] C(t) = 1 / E(t)

[0063] E(t) = (Emax - Emin)En(tn) + Emin

[0064]

[0065] Among them, Emax and Emin are constants, representing the maximum and minimum values of the left ventricular active elasticity respectively. tn = t / Tmax, Tmax = 0.2 + 0.15tc, where tc is the cardiac cycle. E(t) represents the spontaneous elastic change of the left ventricle during this cardiac cycle. The normal cardiac cycle of the human body is 0.8 s. Therefore, C(t) is a time-varying function, introducing a time-varying term to this system.

[0066] The resistances in the same path where the LVAD is located can be linearly superimposed, and the inductance also satisfies this condition. The pressure difference ΔP of the blood pump has a linear relationship with both the blood flow Q and the derivative Q' of the blood flow. Therefore, the coefficients β0 and β1 can also be linearly superimposed with the resistance and inductance in this path respectively.

[0067] Rd* = Ri + Ro + β0

[0068] Ld* = Li + Lo + β1

[0069] It can be seen from the hydrodynamic characteristics of the axial flow pump that its properties can be expressed by the pressure difference H, the pump flow Q, and the rotational speed ω. The characteristics of the axial flow pump under steady state show that at a certain rotational speed, the pressure difference at the inlet and outlet of the pump can be expressed as a second-order or third-order polynomial related to the pump flow. However, further research reveals a linear relationship between the pump flow and the square of the rotational speed, and its model expression is as follows:

[0070] H = ΔP = Po - Pi = b0Q + b1Q′ + b2ω2

[0071] Where Pi and Po represent the inlet and outlet pressures of the pump respectively, and the three coefficients b0, b1, and b2 are related to the structure of the pump and are preset values.

[0072] In one embodiment, the control method of the non-linear model predictive controller is as follows:

[0073] S11, establish a continuous-time non-linear model of the coupling between the ventricular assist device and the human cardiovascular system, expressed as:

[0074]

[0075] Where x(t) is the state variable of the model, u(t) is the control input vector, u(t) = ω 2 (t), ω(t) is the pump speed information, and f is the non-linear model function;

[0076] S12, discretize the model, expressed as:

[0077] x k+1 = F(x k , u k )

[0078] Where F is the discrete model function obtained by discretizing f, and k is the actual time. S13, finite-time domain rolling optimization:

[0079] Define the prediction time domain N P , and let U be the control input sequence for the next N P time instants:

[0080]

[0081] The prediction sequence of the system state is:

[0082]

[0083] U and X are recursively calculated through the formula x k+1 = F(x k , u k );

[0084] S14. Set constraint conditions including state constraints, control constraints, and terminal constraints:

[0085]

[0086] Among them, x min and x max are the preset minimum and maximum values of the state variable, u min and u max are the preset minimum and maximum values of the control input, y k is the model output vector, y k = x k+1 , y min and y max are the preset minimum and maximum values of the output;

[0087] S15. Define the cost function and the optimization problem:

[0088] Express the cost function including the state x cost, control u cost, and terminal cost of the prediction horizon as:

[0089]

[0090] Among them, x ref is the reference state vector that is expected to be reached at the prediction time step k + i or k + N P , Q is the state cost matrix used to measure the state deviation, u ref is the reference control input vector that is expected to be reached at the prediction time step k + i or k + N P , R is the control cost matrix used to measure the magnitude of the control input, and P is the weight matrix at the end of the prediction horizon used to measure the performance of the model at the end of the prediction horizon;

[0091] The optimization problem is expressed as:

[0092]

[0093] Among them, means to find an optimal control input among all possible control inputs U such that the cost function J reaches the minimum value;

[0094] S16. Receding horizon optimization and feedback:

[0095] Optimize in real time through a numerical solver and apply the control input:

[0096]

[0097] Apply the first input in the optimal control sequence to the actual model x k+1= F(x k , u k ) as follows:

[0098]

[0099] The model updates its state at the actual time k:

[0100]

[0101] Re-obtain the actual state x of the model at the next sampling time k + 1 k+1 , and use x k+1 as the new initial state to repeatedly solve the optimization problem:

[0102]

[0103] Obtain the optimal control sequence at time k + 1:

[0104]

[0105] Apply the first input in the optimal control sequence to the actual model x = F(x k+1 , u k , u k ) as follows:

[0106]

[0107] The model updates its state at the actual time k:

[0108]

[0109] Referring to Figure 2 , the non - linear model predictive controller in this example is used as an NMPC pressure tracker. It can preset pressure constraints (i.e., state constraints), rotational speed constraints (i.e., control constraints), prediction interval (i.e., prediction horizon N P ), control interval (i.e., the length of the control input sequence), input weight (i.e., control cost matrix R), and output weight (i.e., state cost matrix Q) according to the actual control effect. According to the actual pressure information (i.e., the actual LVP) and the reference pressure information (i.e., the desired LVP), it can better predict future pressure values.

[0110] The non - linear model predictive controller of this embodiment includes three basic features: a prediction model, finite - horizon receding optimization, and feedback correction, and can achieve precise tracking of left ventricular pressure. And state constraints and control constraints are introduced in the optimization process to avoid safety hazards such as ventricular aspiration or pulmonary congestion caused by too high or too low pump speed.

[0111] In one embodiment, the long short-term memory network model includes an input layer, an LSTM layer, a fully connected layer, and a regression layer. Each time step of the input layer contains 40 features, and the LSTM layer has 128 hidden units.

[0112] Long Short-Term Memory (LSTM) is a special type of Recurrent Neural Network (RNN) that addresses the problems of long-term dependencies and vanishing gradients in traditional RNNs by designing input gates, forget gates, and output gates. It is one of the mainstream deep learning methods for time series data processing. The LSTM network can effectively capture long-term dependencies and non-linear dynamic characteristics in time series and is widely used in sequence prediction, classification, and regression tasks.

[0113] The present invention predicts the LVAD speed and future pressure based on left ventricular pressure (LVP) data using LSTM. The LSTM is designed with a hierarchical structure of a sequence input layer, an LSTM layer, a fully connected layer, and a regression layer based on time series input signals and is optimized for the joint regression task of time series.

[0114] In the input layer of the LSTM, a feature sequence is input in time steps, and each time step contains 40 features to ensure that the input signal fully represents the patient's physiological state. In the LSTM layer, 128 hidden units are designed, and through the gating mechanism, important information in the time series is selectively remembered or forgotten to effectively model long-term dependencies. The output mode of the LSTM layer is set to "sequence output" (OutputMode = "sequence") to retain the hidden state features of all time steps and be used for jointly predicting future pressure values and pump speeds. After the LSTM layer, the network uses a fully connected layer to map the time series features extracted by the LSTM to the target prediction variables, which are the predictive regression of future N-step pressures and the univariate prediction responsible for the LVAD pump speed. Finally, through the regression layer, the errors between the predicted values and the true values of future N-step pressures and pump speeds are calculated respectively, and the model weights are optimized.

[0115] Compared with traditional Recurrent Neural Networks (RNNs), the LSTM network effectively alleviates the problems of vanishing gradients and exploding gradients in time series by introducing the gating mechanism, while improving the non-linear modeling ability for dynamically changing systems. In addition, the LSTM network is optimized in the selection of the number of hidden layer units and output mode, making it more suitable for regression modeling of time series.

[0116] In one embodiment, the long short-term memory network model uses the Root Mean Square Error (RMSE) as the core metric to evaluate the deviation between the predicted value and the reference value. Its calculation formula is:

[0117]

[0118] Where, is the predicted value of the model, y i is the true value, and n is the number of samples. RMSE can directly reflect the accuracy of the model's prediction performance. A smaller RMSE value means the model has higher prediction accuracy.

[0119] In one embodiment, when training the short-term memory network model, actual pressure information and actual pump speed information under several different physiological environments are collected for training to obtain long short-term memory network models under different physiological environments. Before feeding the current pressure information and current pump speed information into the trained long short-term memory network model, the long short-term memory network model is switched to the corresponding current physiological environment.

[0120] In this embodiment, when training the LSTM, several training data under different physiological environments are collected, and each training data is trained separately to obtain an LSTM controller that can respond in real time under different physiological environments. When performing actual rotational speed regulation, the physiological environment where the patient is currently located is switched to the corresponding LSTM. By considering different physiological states (such as systemic vascular resistance, body position changes, etc.), this embodiment realizes real-time regulation of various dynamic pathological states, overcoming the difficulty of meeting diverse physiological needs in real time when facing the dynamic physiological states of individual patients.

[0121] In one embodiment, several different physiological environments include the rise and fall of systemic vascular resistance (SVR), rest to exercise, and passive body position change.

[0122] In one embodiment, predicted pressure information is also obtained through the long short-term memory network model, and a pressure change curve graph is generated from the predicted pressure information and the current pressure information for real-time display.

[0123] Referring to Figure 3 , the embodiment of the present invention also provides a rotational speed regulation system for a ventricular assist device based on deep learning, including:

[0124] A ventricular assist device, which has a pump head 10 and a pressure sensor 20. The pressure sensor 20 is used to collect the current pressure information of the left ventricle.

[0125] Controller 30, the controller has a rotational speed regulation module, the rotational speed regulation module is connected to the ventricular assist device, the rotational speed regulation module is used to obtain the current pressure information and the current pump speed information, send the current pressure information and the current pump speed information into the trained long short-term memory network model, obtain the pump speed information to be regulated through the long short-term memory network model, and send the pump speed information to be regulated to the ventricular assist device.

[0126] The pressure monitoring end 21 of the pressure sensor 20 is located on one side close to the pump head 10, so as to facilitate real-time monitoring of the blood flow pressure data in the patient's body. As Figure 3 shown, the pressure monitoring end 21 is located at the proximal side of the pump head 10.

[0127] The pump head 10 of the ventricular assist device, the pressure sensor 20 and other structures are prior art and will not be described in detail here.

[0128] The preferred embodiments of the present invention have been described in detail above, but it should be understood that after reading the above teachings of the present invention, those skilled in the art can make various changes or modifications to the present invention. These equivalent forms also fall within the scope defined by the appended claims of this application.

Claims

1. A method for controlling the speed of a ventricular assist device based on deep learning, characterized in that: include: The current pressure information and the current pump speed information of the left ventricle are collected, and the reference pressure information is preset, and the current pressure information and the current pump speed information are sent to the trained long short-term memory network model, and the pump speed information to be regulated is obtained through the long short-term memory network model, and the pump speed information to be regulated is sent to the ventricular assist device.

2. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 1, characterized in that: The training process of the short-term memory network model is: S1, obtaining actual pressure information and actual pump speed information of a plurality of left ventricles, sending the actual pressure information, the actual pump speed information and preset reference pressure information to a preset nonlinear model predictive controller to predict future pressure information, and calculating and obtaining pump speed information to be regulated; S2, sending the actual pressure information, the reference pressure information, the actual pump speed information, the future pressure information and the pump speed information to be regulated into a long short-term memory network model for training to obtain a trained long short-term memory network model.

3. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 2, characterized in that: Prior to S1, this also included: S0, establishing a coupling model between a ventricular assist device and a human cardiovascular system, and determining that the input of the coupling model is a pump speed signal of the ventricular assist device, and the output is a pressure value of the left ventricle.

4. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 3, characterized in that: The prediction model of the nonlinear model predictive controller is a continuous-time nonlinear model of the coupling between the ventricular assist device and the human cardiovascular system after discretization processing.

5. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 1, characterized in that: The long short-term memory network model includes an input layer, an LSTM layer, a fully connected layer and a regression layer, each time step of the input layer contains 40 features, and the LSTM layer has 128 hidden units.

6. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 1, characterized in that: The LSTM model uses root mean square error to evaluate the deviation between the predicted value and the reference value.

7. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 2, characterized in that: When training the short-term memory network model, the actual pressure information and actual pump speed information under several different physiological environments are collected for training to obtain long and short-term memory network models under different physiological environments; Before sending the current pressure information and the current pump speed information into the trained long short-term memory network model, the long short-term memory network model is switched to the corresponding current physiological environment.

8. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 7, characterized in that: Several different physiological settings include rise and fall in systemic vascular resistance, rest to exercise, and passive postural changes.

9. The method for controlling the speed of a ventricular assist device based on deep learning according to claim 1, characterized in that: The predicted pressure information is also obtained through the long short-term memory network model, and the predicted pressure information and the current pressure information are combined to generate a pressure change curve graph for real-time display.

10. A ventricular assist device speed control system based on deep learning, characterized in that: include: A ventricular assist device, wherein the ventricular assist device has a pump head and a pressure sensor, wherein the pressure sensor is used to collect current pressure information of the left ventricle; A controller having a speed control module, wherein the speed control module is connected to the ventricular assist device, and the speed control module is used to obtain current pressure information and current pump speed information, send the current pressure information and the current pump speed information to a trained long short-term memory network model, obtain the pump speed information to be regulated through the long short-term memory network model, and send the pump speed information to be regulated to the ventricular assist device.

Images