JP4787526B2 - Blood pump device - Google Patents

Description

  The present invention relates to a blood pump device for transporting medical blood such as blood.

One type of pump used for artificial hearts is a centrifugal blood pump. The centrifugal blood pump has a housing having a blood inflow port and a blood outflow port, and an impeller that rotates in the housing and feeds blood by centrifugal force during rotation. One of the important parameters for grasping that such a blood pump is functioning correctly with respect to a living body is a flow rate. Although direct flow measurement using a dedicated sensor such as an ultrasonic flow meter can be considered for this flow rate measurement, there is a difficulty in terms of stable continuous monitoring over a long period of time. An increase in the number of parts leads to an increase in the size of the apparatus. For this reason, a method of indirectly measuring the flow rate from the number of rotations of the pump motor, its current, the viscosity and specific gravity of the blood, etc. without a flow meter is expected.
As an indirect measurement method, it is possible to calculate by correcting the flow rate as a function of the rotation speed and the current and adding a viscosity factor thereto. That is, the flow rate is calculated from the motor speed and motor current that can be directly measured. In this case, it is conceivable to obtain a correlation between the flow rate, the rotation speed, and the current and use an approximate expression calculated from the correlation. The approximate expression can be expressed by, for example, the following expression (1).

ここで、ｃｋｌｍは最小自乗法にて算出される係数であり、ｉ［Ａ］はモータ電流、ｖ［ｍＰａ・ｓ（ｃＰ）］は粘度、ｎ［ｒｐｍ］は回転数，ｑ［Ｌ／ｍｉｎ］は流量を示す。

Here, cklm is a coefficient calculated by the method of least squares, i [A] is the motor current, v [mPa · s (cP)] is the viscosity, n [rpm] is the rotation speed, q [L / min ] Indicates the flow rate.

However, when the polynomial as described above is used, if the approximation order is lowered, the accuracy in the entire region basically decreases including data points (sample points in terms of interpolation). Further, when the order of approximation is increased, unnecessary vibration, which is a drawback of polynomial interpolation, is generated, and the accuracy other than the data points is also lowered.
In order to prevent this, there is a method that has a large amount of data in the table and uses the flow rate under the condition closest to the given rotation speed and current as an estimated value, but the area to store the data increases, The increase in the processing for searching for it becomes a load on the pump control device, and in the case of a portable control device, display in real time becomes difficult.
Also, when interpolation is performed by spline interpolation, the problem of accuracy is solved, but there is a problem that the amount of calculation increases.
The objective of this invention is providing the blood pump apparatus provided with the flow volume calculation function which can calculate the blood flow volume (discharge flow volume) easily and reliably, without using a flowmeter.

What achieves the above object is as follows.
(1) A housing having a blood inflow port and a blood outflow port, an impeller that rotates in the housing and feeds blood, a motor for rotating the impeller, a blood measurement data input unit, and an impeller rotational speed measurement Alternatively, the blood pump device includes a calculation function, a motor current value measurement function, and a discharge flow rate calculation unit that calculates a blood discharge flow rate, wherein the discharge flow rate calculation unit is blood at a plurality of blood discharge flow rates measured in advance. Viscosity value, motor current measurement value and motor rotation value as input layer data, blood discharge flow rate related to each input layer data as teacher data, discharge layer related relationship between input layer data and discharge flow rate calculated using neural network Discharge flow rate related relational expression storage unit for storing the formula, the discharge flow rate related relational expression, and input to the blood measurement data input unit Blood viscosity value calculated from input blood viscosity value or input hematocrit value, impeller rotation value obtained from impeller rotation speed measurement or calculation function, motor current measurement value obtained by motor current value measurement function or correction thereof A discharge flow rate calculation function for calculating a blood discharge flow rate at the blood viscosity value and the motor current measurement value and the impeller rotation value,
An input calculated using a neural network with blood viscosity values, motor current measurement values and motor rotation values as input layer data, and pump head values related to each of the input layer data as teacher data, in a number of pump head values measured in advance. It is calculated from the pump head related relational expression storage unit for storing the relational expression between the layer data and the pump head, the pump head related relational expression, the blood viscosity value inputted into the blood measurement data input part or the inputted hematocrit value. The blood viscosity value and the motor current using the blood viscosity value, the impeller rotation value obtained from the impeller rotation number measurement or calculation function, the motor current measurement value obtained by the motor current value measurement function or the correction value thereof. Pump that calculates pump head at measured value and impeller rotation value Degree and the calculation function,
A blood pump device having a fluid output calculation function for calculating a fluid output of the pump using a calculated discharge flow rate calculated by the discharge flow rate calculation function and a calculated pump lift calculated by the pump lift calculation function .

(2) The blood according to (1), wherein the blood pump device is a centrifugal blood pump device, and the impeller rotates in the housing and sends blood by centrifugal force during rotation. Pump device.
(3) The blood pump device includes a housing having a blood inflow port and a blood outflow port, a magnetic body therein, a centrifuge having an impeller that rotates in the housing and feeds blood by centrifugal force during rotation. -Type blood pump unit, a rotor including a magnet for attracting the magnetic body of the impeller of the centrifugal blood pump unit, an impeller rotational torque generating unit including a motor for rotating the rotor, and an impeller position control including an electromagnet The blood pump device according to (1) or (2), wherein the impeller rotates in a non-contact state with respect to the housing.
(4) The blood pump device according to any one of (1) to (3), wherein the blood measurement data input unit is a blood viscosity measurement value input unit, and the blood viscosity value is a blood viscosity measurement value.
(5) The blood measurement data input unit is a hematocrit value input unit, and the blood pump device has a blood viscosity calculation function for calculating blood viscosity from the input hematocrit value. The blood pump device according to any one of the above.
(6) The blood measurement data input unit is a hematocrit value input unit, and the blood pump device has a blood viscosity calculation function for calculating blood viscosity from an input hematocrit value and a specific gravity calculation function for calculating specific gravity. The blood pump device according to any one of (1) to (3) above.
(7) The blood pump device includes a specific gravity input unit or a specific gravity calculation function, and a motor corrected value using the specific gravity input from the specific gravity input unit or calculated by the specific gravity calculation function. The blood pump device according to any one of the above (1) to (6), which has a function of correcting a motor current measurement value as a current measurement value.

(8) The blood pump device includes a correction function based on a motor characteristic, and includes a motor current measurement value correction function in which a value obtained by correcting the motor current measurement value by the motor characteristic correction function is used as a motor current measurement value. The blood pump device according to any one of (1) to (7) above.
(9) The blood pump device includes a calculated discharge flow rate calculated by the discharge flow rate calculation function, a calculated pump lift calculated by the pump lift calculation function, and a shaft torque converted from the motor current measurement value, The blood pump device according to any one of (1) to (8), wherein the blood pump device includes a pump efficiency calculation function that calculates the pump efficiency of the pump using the impeller rotational speed.
(10) The blood pump device includes a calculated discharge flow rate calculated by the discharge flow rate calculation function, a calculated pump lift calculated by the pump lift calculation function, and a shaft torque converted from the motor current measurement value, The blood pump device according to any one of (1) to (9), wherein the blood pump device includes a pump total loss calculation function that calculates a total loss in the pump using the impeller rotational speed.

Moreover, what achieves the said objective is as follows.
(1 1 ) The method for adjusting a blood pump device according to any one of claims 1 to 10 , wherein the flow rate of the pump is measured using a flow sensor in a state where the blood pump device is mounted on the subject, The discharge flow measured using the flow sensor is used as teacher data, and the blood viscosity value or hematocrit value, the motor current measurement value and the motor rotation value at the time of the discharge flow measurement are used as input data, and input is stored using a neural network. The relationship between the layer data and the discharge flow rate is corrected , and the pressure on the inflow side and the outflow side of the pump is measured in a state where the blood pump device is mounted on the subject, and the pump is calculated from the measured pressure value The head is used as teacher data, and the blood viscosity value or hematocrit value, motor current measurement value and motor speed at the time of measurement of the discharge flow rate are measured. Numerical as input data, using a neural network, a method of adjusting the blood pump device for performing a relational expression corrective action to correct the relationship between the input layer data and the pump head for storing.

  (12) The adjustment method of the blood pump device according to (11), wherein the neural network is a three-layer feedforward neural network including three layers of an input layer, an intermediate layer, and an output layer.

Moreover, what achieves the said objective is as follows.
(13) The method for manufacturing a blood pump device according to any one of (1) to ( 10 ), wherein the manufacturing method includes blood viscosity values and motor current measurement values at a plurality of blood discharge flow rates measured in advance. A relational expression calculating step for calculating a relational expression between the input layer data and the discharge flow rate using a neural network, using the numerical value of the motor rotation as the input layer data, the blood discharge flow rate for each input layer data as the teacher data, and using a neural network; A storage step of storing the relational expression between the input layer data obtained in the relational expression calculating step and the discharge flow rate in the discharge flow rate calculation unit , and a blood viscosity value, a motor current measurement value, and a plurality of blood discharge flow rates measured in advance; The numerical value of the motor rotation is used as the input layer data, and the pump head related to each input layer data is used as the teacher data. A pump head related relational expression calculating step for calculating a relational expression between the input layer data and the pump head, and a relational expression between the input layer data and the pump head obtained by the pump head related relational expression calculating step. A method of manufacturing a blood pump device that performs a storing step of storing in a head calculation unit .
(14) The method for manufacturing a blood pump device according to (13), wherein the neural network is a three-layer feedforward neural network including three layers of an input layer, an intermediate layer, and an output layer.

  According to this blood pump device, the blood flow rate (discharge flow rate) can be calculated easily and reliably without using a flow meter. Furthermore, since there is a relational expression of rotation speed, current, viscosity, and flow rate calculated using a neural network, an accurate discharge flow rate can be obtained by calculation as compared with the case where a polynomial regression equation is used. In addition, there is a “generalization ability” that calculates an accurate output when inexperienced data is input, and an accurate flow rate can be predicted.

The blood pump device and the method for producing the blood pump device of the present invention will be described using examples.
FIG. 1 is a block diagram of an embodiment of a blood pump device of the present invention. FIG. 2 is a front view of an example of a blood pump device main body used in the blood pump device of the present invention. FIG. 3 is a plan view of the blood pump device main body shown in FIG. FIG. 4 is a longitudinal sectional view of the blood pump device main body of the embodiment shown in FIG. 5 is a cross-sectional view taken along line AA in FIG.

The blood pump device 1 of the present invention includes a housing 20 having a blood inflow port 22 and a blood outflow port 23, an impeller 21 that rotates in the housing 20 and feeds blood, and a motor 34 for rotating the impeller 21. A blood pump device.
The blood pump device 1 of the present invention calculates a blood measurement data input unit (in other words, a blood viscosity related measurement data input unit) 57, an impeller rotational speed measurement or calculation function, a motor current value measurement function, and a blood discharge flow rate. And a discharge flow rate calculation unit 58.
The discharge flow rate calculation unit 58 uses, as input layer data, blood viscosity values, motor current measurement values, and motor rotation values at a large number of pre-measured blood discharge flow rates, and uses a blood discharge flow rate for each input layer data as teacher data. The discharge flow related relational expression storage unit 60 that stores the relational expression between the input layer data and the discharge flow rate calculated using the above, the discharge flow related relational expression stored in the discharge flow related relational expression storage unit 60, and the blood measurement data input unit The blood viscosity value input to 57 or the blood viscosity value calculated from the input hematocrit value, the impeller rotation value obtained from the impeller rotation speed measurement or calculation function, the motor current measurement value obtained by the motor current value measurement function or its Using the correction value, the blood viscosity value, the motor current measurement value, and the impeller And a discharge flow rate calculation function of calculating a blood discharge flow rate in the rotational speed value.

The discharge flow rate calculation unit 58 uses the blood viscosity value, the motor current measurement value, and the motor rotation value at a large number of pre-measured blood discharge flow rates as input data, and uses each blood discharge flow rate as teacher data, using a neural network. A relational expression storage unit 60 that stores a relational expression (using a sigmoid function) between the input layer data and the discharge flow rate calculated in this way, and the blood viscosity value input to the relational expression, blood measurement data input part 57 or The blood viscosity value calculated from the hematocrit value, the impeller rotation number obtained by the impeller rotation number measurement or calculation function, the motor current measurement value obtained by the motor current value measurement function or the correction value thereof, The blood discharge flow rate at the viscosity value, the motor current measurement value, and the impeller rotation value And a discharge flow rate calculation function to calculate.
1 to 5 includes a housing 20 having a blood inflow port 22 and a blood outflow port 23, and a magnetic body 25 therein, and rotates inside the housing 20, A centrifugal blood pump unit 2 having an impeller 21 for feeding blood by centrifugal force, a rotor 31 having a magnet 33 for attracting the magnetic body 25 of the impeller 21 of the centrifugal blood pump unit 2, and rotating the rotor 31 The impeller rotational torque generating unit 3 including the motor 34 and the impeller position control unit 4 including the electromagnet 41 are provided, and the impeller 21 rotates in a non-contact state with respect to the housing 20.
The blood pump device of the present invention is not limited to the type in which the impeller as described above rotates in a non-contact state. For example, it can be used for a type in which an impeller is joined to a shaft of a motor and the impeller rotates by rotation of the motor. The blood pump device of the present invention is not limited to the centrifugal blood pump device as described above, and may be an axial flow blood pump device or a diagonal flow blood pump device.

As shown in FIGS. 2 to 6, the blood pump device main body 5 of this embodiment includes a housing 20 having a blood inflow port 22 and a blood outflow port 23, and an impeller 21 that rotates in the housing 20 and feeds blood. A blood pump part 2 having an impeller rotational torque generating part (non-control type magnetic bearing constituent part) 3 for the impeller 21, and an impeller position control part (control type magnetic bearing constituent part) 4 for the impeller 21. Prepare.
As shown in FIG. 4, the impeller 21 is held at a predetermined position in the housing 20 by the action of the non-control type magnetic bearing component 3 and the control type magnetic bearing component 4 and normally does not contact the inner surface of the housing. Rotate.
The housing 20 includes a blood inflow port 22 and a blood outflow port 23, and is formed of a nonmagnetic material. A blood chamber 24 communicating with the blood inflow port 22 and the blood outflow port 23 is formed in the housing 20. An impeller 21 is accommodated in the housing 20. The blood inflow port 22 is provided so as to protrude substantially vertically from the vicinity of the center of the upper surface of the housing 20. As shown in FIGS. 3 and 5, the blood outflow port 23 is provided so as to protrude in a tangential direction from the side surface of the housing 20 formed in a substantially cylindrical shape.

As shown in FIG. 5, a disc-like impeller 21 having a through-hole at the center is housed in a blood chamber 24 formed in the housing 20. As shown in FIG. 4, the impeller 21 is formed between a donut plate-like member (lower shroud) 27 that forms a lower surface, and a donut plate-like member (upper shroud) 28 that opens at the center that forms an upper surface. A plurality of (for example, seven) vanes 18. A plurality (seven) blood passages 26 partitioned by the adjacent vanes 18 are formed between the lower shroud and the upper shroud. As shown in FIG. 5, the blood passage 26 communicates with the central opening of the impeller 21, starts from the central opening of the impeller 21, and extends so that the width gradually increases to the outer peripheral edge. In other words, the vane 18 is formed between the adjacent blood passages 26. In this embodiment, each blood passage 26 and each vane 18 are provided at equal angular intervals and in substantially the same shape.
As shown in FIG. 4, the impeller 21 has a plurality (for example, 24) of first magnetic bodies 25 (permanent magnets, driven magnets) embedded therein. In this embodiment, the first magnetic body 25 is embedded in the lower shroud 27. The embedded magnetic body 25 (permanent magnet) attracts the impeller 21 to the side opposite to the blood inflow port 22 by a permanent magnet 33 provided on the rotor 31 of the impeller rotational torque generating unit 3 to be described later, and the rotational torque is rotated by the impeller. It is provided for transmission from the torque generator.

Further, by embedding a certain number of magnetic bodies 25 as in this embodiment, sufficient magnetic coupling with the rotor 31 described later can be ensured. The shape of the magnetic body 25 (permanent magnet) is preferably circular. Alternatively, a ring-shaped magnet may be polarized into multiple poles (for example, 24 poles), in other words, a plurality of small magnets may be arranged in a ring shape with alternating magnetic poles.
Further, the impeller 21 includes a second magnetic body 28 provided in the upper shroud itself or in the upper shroud. In this embodiment, the entire upper shroud is formed of the magnetic body 28. The magnetic body 28 is provided for attracting the impeller 21 to the blood inflow port 22 side by an electromagnet 41 of an impeller position control unit described later. As the magnetic body 28, magnetic stainless steel or the like is used.
The impeller position control unit 4 and the impeller rotational torque generating unit 3 constitute a non-contact magnetic bearing, and the impeller 21 is pulled in an opposite direction so that the appropriate position in the housing 20 that does not contact the inner surface of the housing 20 is obtained. The housing 20 rotates in a non-contact state.

As shown in FIG. 4, the impeller rotational torque generating unit 3 includes a rotor 31 housed in the housing 20 and a motor 34 for rotating the rotor 31. The rotor 31 includes a plurality of permanent magnets 33 provided on the surface on the blood pump unit 2 side. The center of the rotor 31 is fixed to the rotating shaft of the motor 34. The permanent magnets 33 are provided in plural and at equal angles so as to correspond to the arrangement form (number and arrangement position) of the permanent magnets 25 of the impeller 21.
The impeller rotational torque generating unit 3 is not limited to the one provided with the above-described rotor and motor, and may be composed of, for example, a plurality of stator coils for attracting and rotating the permanent magnet 25 of the impeller 21.

As shown in FIGS. 3 and 4, the impeller position control unit 4 includes a plurality of fixed electromagnets 41 for attracting the impeller magnetic body 28, and a position sensor for detecting the position of the impeller magnetic body 28. 42 is provided. Specifically, the impeller position control unit 4 includes a plurality of electromagnets 41 housed in the housing 20 and a plurality of position sensors 42. The plural (three) electromagnets 41 and the plural (three) position sensors 42 of the impeller position control unit are provided at equiangular intervals, and the electromagnet 41 and the position sensor 42 are also provided at equiangular intervals. ing. The electromagnet 41 includes an iron core and a coil. In this embodiment, three electromagnets 41 are provided. The number of electromagnets 41 may be three or more, for example, four. By providing three or more and adjusting these suction forces using the detection result of the position sensor 42, the forces in the rotation axis (z axis) direction of the impeller 21 are balanced and orthogonal to the rotation axis (z axis). The moment about the x and y axes can be controlled.
The position sensor 42 detects a gap interval between the electromagnet 41 and the magnetic body 28, and this detection output is a control unit of a control mechanism (in other words, a controller) 6 that controls the current or voltage applied to the coil of the electromagnet 41. 51. Even if a radial force due to gravity or the like acts on the impeller 21, the magnetic flux shearing force between the permanent magnet 25 of the impeller 21 and the permanent magnet 33 of the rotor 31 and the force between the electromagnet 41 and the magnetic body 28. Since the shearing force of the magnetic flux acts, the impeller 21 is held at the center of the housing 20.

As shown in FIG. 1, the control mechanism 6 includes a power amplifier 52 and a motor control circuit 53 for the magnetic coupling motor 34, a power amplifier 54 for the electromagnet 41, a sensor circuit 55 for the sensor 42, and a sensor. Sensor output monitoring unit (not shown) for monitoring output, control unit 51, power supply unit 56, blood measurement data input unit (specifically, blood viscosity value or hematocrit value) 57, discharge flow rate calculation unit 58, A display unit 59 is provided. The control unit 51 includes a motor current monitoring function and an impeller rotation speed calculation function (motor rotation speed calculation function).
The blood measurement data input unit 57 is, for example, a blood viscosity measurement value input unit. Preferably, as shown in FIG. 1, it is preferable to provide a blood viscosity measurement value input unit and a specific gravity input unit. In this case, blood viscosity and specific gravity are measured by using an external device from blood collected from the user.
The blood measurement data input unit 57 may be a hematocrit value input unit as shown in parentheses in FIG. In this case, the blood pump device has a viscosity calculation function for calculating the viscosity from the input hematocrit value. That is, the discharge flow rate calculation unit 58 has a viscosity calculation function. The viscosity calculation method is as follows.
v = a3Hct3 + a2Hct2 + a1Hct + a0
v [mPa · s] is the viscosity, Hct [%] is the hematocrit value, a0 to a3 = coefficient Furthermore, the blood pump device preferably has a specific gravity calculation function for calculating the specific gravity from the input hematocrit value. That is, the discharge flow rate calculation unit 58 has a specific gravity calculation function. The specific gravity calculation method is as follows.
ρ = b3Hct3 + b2Hct2 + b1Hct + b0
ρ is specific gravity, Hct [%] is hematocrit value, b0 to b3 = coefficient

The blood pump device 1 also has an impeller rotation speed measurement function (motor rotation speed measuring device) or an impeller rotation speed calculation function (motor rotation speed calculation function). Specifically, the control device 6 has a motor rotation number monitoring function.
Further, the correction by the specific gravity of the motor current value can be performed by the following equation.
Motor current correction value = motor current actual measurement value / specific gravity It is preferable that the blood pump device has a temperature correction function. Correction by temperature can be performed by correcting the measured motor current value by temperature. The correction of the motor current value by temperature can be performed by the following equation.
Icomp = f (T, Imeas, a1, a2, .., an)
Icomp: Motor current value after temperature compensation
Imeas: Measured motor current value
f: Conversion function for correction
T: Controller temperature
a1, a2, .. an: Coefficient for correction measured in advance for each product controller

The blood pump device preferably has a correction function based on motor characteristics. The correction of the motor current value by the motor characteristics can be performed by the following equation.
Is ＝ gmotor (Ip, N, c1, c2, .., cn)
gmotor: Function used to convert motor current value
Ip: Measured motor current value
Is: Motor current value after correction by motor characteristics
N: Number of rotations
c1, c2,.., cn: Correction coefficients measured in advance for each product controller By substituting the measured motor current value into Ip in this equation, the motor current value can be corrected.

  Further, the pump device may have a system configuration as shown in FIG. The blood pump device 100 of this embodiment is separated into a blood pump device main body comprising a blood pump device main body 5 and a control mechanism (in other words, a controller) 6 electrically connected thereto, and a discharge flow rate calculator 7. is doing. The controller 6 and the discharge flow rate calculator 7 are capable of data communication by the communication function 8 provided in each. Specifically, the communication function 8 includes a control mechanism side communication interface 88 (controller side communication interface) and a discharge flow rate calculator side communication interface 89. At least, the control mechanism side communication interface 88 (controller side communication interface) has a function of transmitting a motor current value or a related signal and a motor rotation value or a related signal. The discharge flow rate calculator side communication interface 89 has a function of receiving the motor current value or the related signal and the motor rotation value or the related signal transmitted from the control mechanism side communication interface 88. The communication between the two may be either analog communication or digital communication. A known communication format can be used. The difference between the blood pump device 100 of this embodiment and the blood pump device 1 described above is only that the discharge flow rate calculator is separated from the control mechanism, and the other configurations are the same as those of the blood pump device 1 described above. is there.

As shown in FIG. 6, the control mechanism 6 in the blood pump device 100 of this embodiment includes a power amplifier 52 and a motor control circuit 53 for the magnetic coupling motor 34, a power amplifier 54 for the electromagnet 41, and a sensor. 42, a sensor output monitoring unit (not shown) for monitoring the sensor output, a control unit 51, a power supply unit 56, and a control mechanism side communication interface 88. The controller 51 has a motor current monitoring function and an impeller rotation speed calculation function (motor rotation speed calculation function).
The discharge flow rate calculator 7 includes a blood measurement data input unit 57, a discharge flow rate calculation unit 58, a display unit 59, a power supply unit 62, and a discharge flow rate calculator side communication interface 89.
The blood measurement data input unit 57 is, for example, a blood viscosity measurement value input unit. Preferably, as shown in FIG. 6, a blood viscosity measurement value input unit and a specific gravity input unit are preferably provided. In this case, blood viscosity and specific gravity are measured by using an external device from blood collected from the user.
The blood measurement data input unit 57 may be a hematocrit value input unit as shown in parentheses in FIG. In this case, the blood pump device has a viscosity calculation function for calculating the viscosity from the input hematocrit value. That is, the discharge flow rate calculation unit 58 has a viscosity calculation function. The calculation of the discharge flow rate in the discharge flow rate calculation unit may be in any of the above-described embodiments.

Next, the blood viscosity value or hematocrit value, the motor current measurement value, and the motor, which are calculated using the neural network, stored in the blood pump device manufacturing method of the present invention and the flow rate calculation unit of the blood pump device of the present invention. A relational expression (using a sigmoid function) between the input layer data composed of rotational values and the discharge flow rate will be described.
A method of manufacturing a blood pump device according to the present invention includes a housing having a blood inflow port and a blood outflow port, an impeller that rotates in the housing and feeds blood, a motor for rotating the impeller, and blood measurement data An input unit, an impeller rotation speed measurement or calculation function, a motor current value measurement function, a blood viscosity value input from the blood measurement data input unit or a blood viscosity value calculated from an input hematocrit value, the impeller rotation A blood pump device comprising a discharge flow rate calculation unit that calculates a blood discharge flow rate using an impeller rotation value obtained from a number measurement or calculation function, a motor current measurement value obtained by the motor current value measurement function, or a correction value thereof It is a manufacturing method.
The manufacturing method uses a neural network using a blood viscosity value or a hematocrit value, a motor current measurement value, and a motor rotation value as input data, and each blood discharge flow rate as teacher data, in a large number of blood discharge flow rates measured in advance. The discharge flow related relational expression calculating step for calculating the relational expression between the input layer data and the discharge flow rate, and the discharge flow related relational expression between the input layer data and the discharge flow rate obtained by the relational expression calculating step is the discharge flow rate. And a storage step of storing in the calculation unit.

シグモイド関数のゲインαを大きくすると、シグモイド関数はステップ関数に近づき、出力yは{０,１}の値しかとらなくなる。αは、学習をする前に設定しておくパラメータである。急峻な変化をする関数を近似する場合はαを大きくする。流量予測の場合はαの値は小さくして構わない。 Calculation of the relational expression between the input layer data and the discharge flow rate can be performed as follows.
The neural network is configured by using a plurality of artificial elements that simulate biological neurons. Here, the artificial element used in the neural network is not an exact model of a biological neuron, but a simplified engineering model that extracts a specific function. A deterministic analog model is used as the artificial element. This is a model in which the output y is determined as an analog value in the range of 0 ≦ y ≦ 1, and this output is deterministically determined from the input. First, another element or an externally applied signal is input to the model. These values are written as {x ₁ , x ₂ , x ₃ ,… x _N }. This is multiplied by weights {w ₁ , w ₂ , w ₃ ,... W _N } and added to obtain an amount s corresponding to the intracellular potential. The weight corresponds to the synaptic connection weight. When obtaining the output y from the intracellular potential s, a continuous output is obtained using a sigmoid function. The sigmoid function is expressed by the following equation.

When the gain α of the sigmoid function is increased, the sigmoid function approaches a step function, and the output y takes only a value of {0, 1}. α is a parameter set before learning. When approximating a function that changes sharply, α is increased. In the case of flow rate prediction, the value of α may be small.

ここでθは、細胞内電圧の閾値である。 FIG. 7 shows how the sigmoid function changes depending on the gain α. The definitive analog model is a binary model in an extreme model with a sufficiently large gain. A definitive analog model is shown in FIG. The output calculation procedure of this model is expressed by the following equation.

Here, θ is a threshold value of the intracellular voltage.

The discharge flow rate calculation unit 58 of the present invention is configured by a feedforward type neural network. This is a neural network in which the above-described elements are connected as shown in FIG. 9 so that signals flow in one direction from the input side to the output side.
The neural network used in the present invention is a feedforward neural network composed of three layers: an input layer, an intermediate layer, and an output layer. The intermediate layer is preferably one layer, but may be two or more layers. The input layer is composed of three elements, and a blood viscosity value or a hematocrit value, a motor current measurement value, and a motor rotation value are input respectively. The intermediate layer is composed of a plurality of elements and is coupled to each element of the input layer and the output layer. In order to improve the calculation accuracy of the discharge flow rate, it is necessary to increase the number of elements in the intermediate layer. However, since the calculation becomes complicated if the number of elements is increased too much, the number of elements is preferably 3 to 10.

を得る。これを重み付け総和と呼ぶ。次に、s_j ^（１）をシグモイド関数に通して、素子の出力

を得る。中間層素子がj＝1,2,…、KのK個あるとき、これらの全てについて上述の手続きにより出力を求める。
こうして中間層素子の出力を求めた後、出力層素子は、中間層素子の出力y_j ^（１）、出力層素子の出力をy_j ^（2）とすると、これはy_j ^（１）(j=0,1,2,…、K)をその入力として受け取り、これに結線の「重み」w_ij ^（２）をかけて総和し、まず、重み付け総和

を得る。具体的には、今の場合は、j=1である。
続いて、s_j ^（２）をシグモイド関数に通して、素子の出力

を得る。ここで、y_j ^（2）は吐出流量（血液流量）である。具体的には、今の場合は、j=1である。 By using a three-layer feedforward neural network for the discharge flow rate calculation, the blood flow rate (discharge flow rate) can be calculated easily and reliably without using a flow meter, and moreover compared to the case of using a higher-order calculation formula. Can be calculated quickly.
An example of a feedforward type neural network used in the present invention is shown in FIG. The j-th intermediate layer element of the intermediate layer applies the connection weight w _ij ⁽¹⁾ to the input x _i and sums up for i = 0, 1, 2,..., N. First, the intracellular potential (weighted sum)

Get. This is called a weighted sum. Next, let s _j ⁽¹⁾ pass through the sigmoid function and output the element

Get. When there are K intermediate layer elements j = 1, 2,..., K, the output is obtained by the above procedure for all of them.
After obtaining the output of the intermediate layer element in this way, if the output layer element is output y _j ⁽¹⁾ of the intermediate layer element and y _j ⁽²⁾ is the output of the output layer element, this is y _j ⁽¹⁾ (j = 0,1,2, ..., K) is received as its input, and this is multiplied by the "weight" w _ij ⁽²⁾ of the connection and summed. First, the weighted sum

Get. Specifically, in this case, j = 1.
Subsequently, s _j ⁽²⁾ is passed through the sigmoid function to output the element

Get. Here, y _j ⁽²⁾ is the discharge flow rate (blood flow rate). Specifically, in this case, j = 1.

Next, a method for determining the value of the connection weight w of the artificial neural network mounted in the discharge flow rate calculation unit will be described. A sequential update learning method is used to determine w. In the present invention, the connection weight is finely adjusted every time training data is given, using an error back-propagation method that is one of successive update learning methods. Training data related to input / output to be realized is given as a plurality of input / output pairs. In the present invention, the training data includes three inputs of a motor rotation value, a motor current measurement value, and a blood viscosity value (or hematocrit value) used as inputs, and a discharge flow rate (blood flow rate) corresponding to these inputs.
This discharge flow rate is given as a learning teacher signal. Now, assume that there are L pieces of training data, and the m-th of them requires an output y ^(m) for an input (x ₁ ^(m) , x ₂ ^(m) , x ₂ ^(m) ). In this method, it is necessary to obtain calibration data in advance with respect to the motor rotation speed, motor current, blood viscosity (or hematocrit value), and discharge flow rate (blood flow rate) before learning.

For example, the number of revolutions at each viscosity of 1, 2, 3, 4, 5 [mPa · s]: 1200, 1400, 1600, 1800, 2000, 2200, 2400, 2600 [rpm] The motor current value at 2, 3, 4, 5, 6, 7, 8, 9 [L / min] is obtained by experiment. At this time, there are 5 ways of viscosity, 8 ways of rotation, and 9 ways of flow rate, so the number of calibration data is 5 × 8 × 9 = 360 points. Here, the calibration data is acquired with a rotation speed of 200 rpm intervals and a flow rate of 1 L / min, but it is desirable to make the measurement interval as fine as possible in consideration of increasing the accuracy of the discharge flow rate. However, if the amount of calibration data (the number of measurement points) increases, the calibration requires a considerable amount of time, so the above conditions are appropriate for the measurement interval. It is desirable that the motor current measurement value be normalized with the specific gravity of blood. Note that it is desirable to standardize (non-dimensionalize) all the calibration data used for learning. For example, the rotation speed is divided by the maximum rotation speed of 2600 rpm. The flow rate used as the teacher signal is also divided by the maximum flow rate of 9 L / min. Thereby, all the calibration data take values in the range of 0 to 1.
FIG. 11 shows a learning procedure (learning routine) of the error reverse carrying method. The iterative loop inside FIG. 11 is a loop for training data. When training data is selected each time (Step 1 below), parameters are modified so as to reduce the error evaluation scale for the training data (see below). Step 2, 3). The outer loop is repeated until the error rating scale of all training data is sufficiently small.

Hereinafter, the error back propagation method will be described in detail.
<Step 1> Selection of training data The m-th training data is selected as the first procedure in the inner loop of FIG. 11, and the following two steps are executed on this data. The m-th input is represented as (x ₁ ^(m) , x ₂ ^(m) , x ₂ ^(m) ), but the right-hand subscript is used to fix the training data during the two steps. Since (m) is not necessary, it is omitted. Instead, the variable's right shoulder index (m) is used to identify the layer. That is, the subscript (1) on the right side of the variable indicates that the variable is for the first layer (intermediate layer), and the subscript (2) indicates that the variable is for the second layer (output layer). The input layer is the 0th layer.

なお、ネットワーク中の結線の重み全てに、初期値として、乱数などで発生した無作為な値を割り当てておく。ただし、初期値は全て0としてはならない。 <Step 2> Output Calculation FIG. 12 shows a feedforward type neural network composed of three layers: an input layer, an intermediate layer, and an output layer. The input layer consists of nodes (signal relay points) indicated by black dots. The intermediate layer and the output layer are composed of elements (neuron engineering models) indicated by white circles. In the back propagation method, a deterministic analog model is used as an element.
Here, the operation of this element is shown in FIG. If the input of the element is x ₀ , x ₁ ,..., X _N , the output is y, the coupling weight is w ₀ , w ₁ ,..., W _N , and the weighted sum (intracellular voltage) is s, the operation of the element is It can be formulated as follows.

Note that random values generated by random numbers or the like are assigned as initial values to all the connection weights in the network. However, the initial value must not be all zero.

を得る。続いて、s_j ^（1）をシグモイド関数に通して、素子の出力

を得る。中間層素子がj=1,2,…,Nの N個あるとき、これらの全てについて上述の手続きにより中間層素子の出力を求める。出力層素子の出力y₁ ⁽²⁾は、中間層素子の出力y_j ^(１)（j=1,2,…,N）をその入力として受け取り、結線の「重み」w_j１ ^（２）をかけて、重み付け総和

を得る。続いて、s_１ ^（２）をシグモイド関数に通して、素子の出力

を得る。これがネットワークの出力b₁ （＝y_１ ^(２)）となる。式（１４）は出力層素子の出力、すなわち血液吐出流量（無次元）である。式（１４）はシグモイド関数であるため、出力値は０〜１である。したがって、式（１４）の値にスケールファクター（換算係数）をかけたものが実際の血液吐出流量となる。スケールファクターをa[L/min]とすると，ポンプ流量Ｑは以下の式で表される。

After applying the input to the input node, the j-th intermediate layer element adds the connection weight w _ij ⁽¹⁾ to the input x _i (= y _i ⁽⁰⁾ ) and sums up for i = 1, 2, 3 First, the weighted sum

Get. Next, s _j ⁽¹⁾ is passed through the sigmoid function and the output of the element

Get. When there are N intermediate layer elements j = 1, 2,..., N, the output of the intermediate layer element is obtained for all of them by the above procedure. The output y ₁ ⁽²⁾ of the output layer element receives the output y _j ⁽¹⁾ (j = 1, 2,..., N) of the intermediate layer element as its input, and receives the “weight” w _j1 ⁽²⁾ of the connection. Multiply weighted sum

Get. Subsequently, s ₁ ⁽²⁾ is passed through the sigmoid function to output the element

Get. This is the output b ₁ (= y ₁ ⁽²⁾ ) of the network. Expression (14) is the output of the output layer element, that is, the blood discharge flow rate (dimensionless). Since Expression (14) is a sigmoid function, the output value is 0 to 1. Therefore, the actual blood discharge flow rate is obtained by multiplying the value of the equation (14) by the scale factor (conversion coefficient). When the scale factor is a [L / min], the pump flow rate Q is expressed by the following equation.

となる。ここで、ｂは定数である。出力層素子に線形素子を選ぶ場合は、教師信号の規格化とスケールファクターの乗算は必要ない。 In this embodiment, since the teacher signal is normalized by dividing it by the maximum flow rate of 9 L / min, the scale factor is a = 9 L / min. In this embodiment, the output layer element is a nonlinear element (sigmoid function), but may be a linear element. At this time, the equation (14) becomes

It becomes. Here, b is a constant. When a linear element is selected as the output layer element, normalization of the teacher signal and multiplication of the scale factor are not necessary.

と表される。勾配法の原理により、この評価関数を小さくするように、ネットワーク内部の重みを修正する。 Here, the output of this network is compared with the target output of learning. When the target output and t _1, the error evaluation measure E is,

It is expressed. Based on the principle of the gradient method, the weight inside the network is corrected so as to reduce this evaluation function.

<Step 3> Correction of Binding Weight In this example, a case where only the binding weight w is corrected will be described for the sake of simplicity. However, the threshold value θ of the intracellular voltage may be corrected in addition to w. A more desirable function approximation result may be obtained by correcting w and θ.
The procedure for correcting the connection weight is as follows. First, FIG. 13 shows a network flow when the connection weight is corrected (error back propagation mode). The number of nodes, the number of elements, the connection structure between elements, and the weight of the connection are the same as those in FIG. 12 used during output calculation (output calculation mode), but the function of the elements and the direction of signal flow are different.

を求めたあと、メモリに保持されているyを用いてα(1-y)yを計算し、これにuを掛け、

を出力する。出力層に関しては、uの値は、

とする。ｚは左側の層の素子への入力となる。ここで，α(1-y)yは，

なる関係で、シグモイド関数の微分が求まることから出てきたものである。以上のように動作する素子を図１３のように接続する。図１３の結線構造と結線の重みは図１２と全く同様である。図１３の右から、Step２で求めたネットワークの出力b_１と目標出力t_１との誤差b_１−t_１を加える。この入力に基づき、各層で図１５に示した素子機能が働き、素子出力z _j ^(m) が次々と定まる。こうして、誤差逆伝搬モードで各層の素子出力z _j ^(m) を求めた後、これと出力計算モード（Step2）で求めた各層の素子出力y_i ^(m) を用いて、次式により結線の重みを修正する。

w_ij ^{（m＋１）}は、第m層の第i素子と第（m＋1）層の第j素子を結ぶ結線の重みである。以上の方法により、結線の重みw_ij ^{（m＋１）}を修正する。学習は、図１１に図示した学習手順の外側のループを全訓練データの誤差評価尺度（式１５）が十分小さくなるまで繰り返す。
このようにして、人工ニューラルネットの各素子間の結線の重みw_ij ^{（m＋１）}が、上記の学習方法により最終的に決定される。これにより、入力層データと吐出流量との関係式を算出する関係式算出工程が終了する。 FIG. 15 shows element functions in the error back propagation mode. There is a memory inside the element, and its output y (= y _j ^(m) , hereinafter, the subscript is omitted and abbreviated as y) is held in Step 2. Input z _1, z ₂ given from the right side of the _device, ..., the z _N, the weight of the right connection of the elements w _1, w _2, ..., multiplied by w _N, weighted sum

After calculating α (1-y) y using y held in the memory, multiply this by u,

Is output. For the output layer, the value of u is

And z is the input to the left layer element. Where α (1-y) y is

This is because the derivative of the sigmoid function is obtained. The elements operating as described above are connected as shown in FIG. The connection structure and connection weight in FIG. 13 are exactly the same as in FIG. From the right side of FIG. 13, an error b ₁ −t ₁ between the network output b ₁ and the target output t ₁ obtained in Step 2 is added. Based on this input, the element function shown in FIG. 15 works in each layer, and the element output z _j ^(m) is determined one after another. Thus, after obtaining the element output z _j ^(m) of each layer in the error back propagation mode, using this and the element output y _i ^(m) of each layer obtained in the output calculation mode (Step 2), Correct the weight.

w _ij ^{(m + 1)} is a weight of a connection line connecting the i-th element in the m-th layer and the j-th element in the (m + 1) -th layer. The connection weight w _ij ^{(m + 1)} is corrected by the above method. In the learning, the loop outside the learning procedure shown in FIG. 11 is repeated until the error evaluation scale (Equation 15) of all training data becomes sufficiently small.
In this way, the connection weight w _ij ^{(m + 1)} between the elements of the artificial neural network is finally determined by the learning method described above. Thereby, the relational expression calculating step for calculating the relational expression between the input layer data and the discharge flow rate is completed.

And the relational expression consisting of the following expression 11 including the weight term of the connection determined as described above and other expressions 12 to 14-a (or 14-b) for finally calculating the discharge flow rate, It is stored in the relational expression storage unit 60. In other words, in the blood pump device 1 of the present invention, the relational expression storage unit 60 stores Expressions (11) to (14-a) or Expressions (11) to (14-b).
Then, the calculation unit 48 of the blood pump device 1 uses the blood viscosity value calculated from the blood measurement data input unit that is actually given or the blood viscosity value calculated from the input hematocrit value, the impeller rotational speed measurement or calculation function. Using the obtained impeller rotation value, motor current measurement value obtained by the motor current value measurement function or its correction value, and the above relational expressions (11) to (14-a) or (14-b), the discharge flow rate is determined. It has a function to calculate.

Specifically, the blood viscosity value input from the blood measurement data input section that is actually given or the blood viscosity value calculated from the input hematocrit value, the impeller rotation value obtained from the impeller rotation number measurement or calculation function, the motor The weighted sum s _j ⁽¹⁾ of the intermediate layer element is calculated from the following equation (11) using the motor current measurement value obtained by the current value measurement function or its correction value.

そして、下記式１３を用いて、出力層素子の重み付け総和s₁ ^（2）を求める。

Subsequently, the output y _j ⁽¹⁾ of the intermediate layer element is obtained by using this s _j ⁽¹⁾ using the sigmoid function of Expression 12 below.

Then, the weighted sum s ₁ ⁽²⁾ of the output layer element is obtained using the following equation (13 ⁾ .

なお、出力層素子が線形素子の場合には、上記式（１４−ａ）ではなく、下記式１４−ｂを用いて、ポンプ流量Qを算出する。

ここで、bは定数である。
よって、この血液ポンプ装置は、血液粘度値もしくはヘマトクリット値と、モータ電流測定値とモータ回転数値を入力とし、３層フィードフォワード型人工ニューラルネットを用いて得た関係式より、吐出流量（血液流量）を算出して出力する機能を備えている。 When the output layer element is nonlinear, the pump flow rate Q is calculated using the following equation 14-a.

When the output layer element is a linear element, the pump flow rate Q is calculated using the following expression 14-b instead of the above expression (14-a).

Here, b is a constant.
Therefore, this blood pump device inputs the blood viscosity value or hematocrit value, the motor current measurement value, and the motor rotation value as input, and from the relational expression obtained using the three-layer feedforward artificial neural network, the discharge flow rate (blood flow rate) ) Is calculated and output.

Next, a method for adjusting the blood pump device of the present invention will be described.
The blood pump device adjustment method of the present invention is the blood pump device adjustment method described above, wherein the flow rate of the pump is measured using a flow sensor in a state where the blood pump device is mounted on the subject, and the flow sensor The discharge flow rate measured by using the teacher data, blood viscosity value or hematocrit value at the time of discharge flow measurement, motor current measurement value and motor rotation value as input data, using neural network, The relational expression correction work for correcting the relational expression with the discharge flow rate is performed.
This customizes the relational expression for correcting the relational expression between the input layer data to be stored and the discharge flow rate in order to increase the discharge flow rate calculation accuracy immediately after the blood pump is applied to the patient. Specifically, the combination weight w (in some cases, the threshold θ is also changed) is changed.

The storage unit 60 of the blood pump device stores the above-described relational expression obtained by an artificial neural network obtained by so-called offline learning. Then, immediately after applying the blood pump to the patient, so-called online learning is performed.
Specifically, a flow sensor is attached near the outlet of the blood pump, the discharge flow rate is measured, and this is used as teacher data. Further, the blood viscosity value or hematocrit value, the motor current measurement value, and the motor rotation value at the time of the discharge flow rate measurement are used as input layer data, and the discharge flow rate calculation value is calculated using the stored relational expression.
Then, in the same manner as described above, an error back propagation method (back propagation) is performed using the teacher data and the calculated value. At this time, the blood pump device is connected to another computer storing a neural network calculation program, and the discharge flow value, the discharge flow calculation value, and the blood viscosity value at the time of measurement of the discharge flow, which are the teacher data, or The hematocrit value, the motor current measurement value, and the motor rotation value are input to this computer. Then, this computer derives the following expression 21 relating to the coupling weight w from the relational expression between the input layer data stored in the connected blood pump device and the discharge flow rate.

Then, using the above combined weight formula, the measured discharge flow rate is used as teacher data, and the blood viscosity value or hematocrit value, motor current measurement value and motor rotation value at the time of this discharge flow measurement are used as input layer data, A propagation method (back propagation) is performed, and the above equation 21 is recalculated. By the back propagation method, the weight w and the threshold value θ are corrected so that the error between the calculated discharge flow rate and the actual flow rate (teacher signal) is reduced (more precisely, the error evaluation scale E in equation (15) is reduced). We will make corrections so that In this way, learning is repeated so that the error between the calculated discharge flow rate and the actual flow rate becomes small, and the learning is stopped when the error falls within the required specification range (for example, Δ <0.5 L / min). Then, the above-described relational expressions (11) to (14-a) or (14-b) are calculated again using the recalculated coupling weight w _ij ^{(m + 1),} and stored in the calculation unit 58 of the blood pump device. Replace with the one stored in the unit 60. Thereby, highly accurate flow volume prediction suitable for each patient is attained.

Specifically, when a blood pump is applied to a patient during surgery and the discharge flow rate calculation value by the calculation function provided in the blood pump device is compared with the actual measurement value of the flow rate sensor, when the error between the two is large (out of specification range; for example, Error Δ> = 0.5 L / min), and an artificial neural network is used to customize the flow rate calculation formula.
During the operation of the patient (during thoracotomy), the blood pump device compares the calculated discharge flow rate value with the actual measurement value of the flow rate sensor. If the error between the discharge volume calculation value and the measured value by the flow sensor is small and within the required specification error range (for example, Δ <0.5L / min), correct the discharge flow rate relational expression stored in the blood pump device. Without applying to the patient. On the other hand, when the error between the estimated value and the actual measurement value is large (for example, error Δ> = 0.5 L / min), the discharge flow rate related relational expression mounted on the discharge flow rate calculation unit 58 is customized online. Specifically, the number of rotations is set on the console, the flow rate is changed by a flow rate adjustment means such as a pinch cock (clamp), and online learning is performed for multiple operating points, so that the error is reduced. The weight w and the threshold θ are corrected. If the error between the estimated value of the artificial neural network and the measured value of the flow sensor falls within the specification error range, online learning is terminated and the flow sensor is removed from the patient. This method enables more accurate indirect measurement (prediction) of the flow rate. In addition, this method can provide a blood pump device that eliminates the influence of all error factors such as individual differences among patients and individual differences among pump systems.

Further, as a method for manufacturing the blood pump device of the present invention, the blood viscosity value input to the blood measurement data input unit or the blood viscosity value calculated from the input hematocrit value, the impeller rotational speed measurement or the calculation function A method of manufacturing a blood pump device comprising a pump head calculation unit that calculates a pump head using an impeller rotation value obtained, a motor current measurement value obtained by the motor current value measurement function or a correction value thereof, and measured in advance The blood viscosity value or hematocrit value, motor current measurement value and motor rotation value at a large number of blood discharge flow rates as input layer data, and the pump head for each input layer data as teacher data, using a neural network, Pump head related function that calculates the relation between data and pump head And wherein calculating step, may perform a storage step of storing a relational expression of the input layer and the data and the pump head obtained by said pump head related equation calculating step to the pump head calculation unit.
The blood pump device thus obtained has a blood viscosity value or hematocrit value, a motor current measurement value, and a motor rotation value at a plurality of blood discharge flow rates measured in advance as input layer data, and a pump related to each input layer data. The pump head related relational expression storage unit for storing the relational expression between the input layer data calculated using the neural network and the pump head and the pump head related relational expression as the teacher data, and the pump head related relational expression, input to the blood measurement data input unit Blood viscosity value calculated from input blood viscosity value or input hematocrit value, impeller rotation value obtained from impeller rotation speed measurement or calculation function, motor current measurement value obtained by motor current value measurement function or correction thereof Value is used to measure the blood viscosity value and the motor current measurement. Values as well as the ones provided with the pump head calculation function for calculating a pump head in the impeller rotational speed value.

The pump head related relational expression calculating step may be performed independently, but is preferably performed simultaneously with the above-described discharge flow rate related relational expression calculating step. This can be done by changing the configuration of the neural network from 3 layers 3 inputs 1 output to 3 layers 3 inputs 2 outputs. The pump head means a pressure difference between the inflow side and the outflow side of the pump.
The training data necessary for learning the neural network related to the pump head related relation formula calculation process includes three inputs of a motor rotation value, a motor current measurement value, and a blood viscosity value (or hematocrit value) used as inputs, and corresponding outputs. This is the pump head. In this case, the pump head becomes a teacher signal. Since the discharge flow rate calculation unit 58 has a pump head calculation function in addition to the discharge flow rate calculation function, the blood pump device of the present invention has a function of calculating and displaying the pump fluid output and pump efficiency. can do. That is, if the calculated discharge flow rate and the calculated pump head output by the neural network are Q and P, respectively, the fluid output W of the pump is
W = P × Q (22)
Can be calculated as

Further, the motor output [W], the flow rate output [W], the total loss [W], and the pump efficiency η [%] can be calculated by the following equations.
Pump efficiency η = (PQ / TN) × 100 [%] (23)
Fluid output = PQ [W] (24)
Motor output = TN [W] (25)
Total loss in pump = TN-PQ [W] (26)
Pump head: P [Pa] {(760 / 1.013) × 10 ⁻⁵ (mmHg)}
Flow rate: Q [m ³ / s] {60 × 10 ³ (L / min)}
Shaft torque: T [Nm] {(1 / 9.81) × 10 ⁵ (gfcm)}
Number of revolutions: N [rad / s] {60 / (2π) (rpm)}
Here, since the shaft torque T is proportional to the motor current, it can be converted from the measured motor current. The proportionality coefficient needs to be obtained by experiments. With the above calculation method, it is possible to provide a blood pump device having a function of calculating pump efficiency, fluid output, motor output, and total loss in the pump unit. By regularly monitoring this information by an engineer, it is possible to know that there is no abnormality in the pump, and if there is an abnormality, the state of the pump can be known.

  The calculation of the discharge flow rate related relational expression may be performed as follows. In this method, learning is performed in combination with a flow sensor during surgery (opening) of a patient. In this case, the training data is a discharge flow rate (blood flow rate) that is an output corresponding to three inputs of a motor rotation value, a motor current measurement value, and a blood viscosity value (or hematocrit value). The discharge flow rate is a value measured by a flow rate sensor attached to the patient during the operation, and this is used as a teacher signal. Learning is performed by the error back propagation method. The learning time is set to a time that does not become a burden on the patient during the thoracotomy, and it is confirmed that the predicted flow rate sufficiently follows the output of the flow rate sensor. During learning, the motor speed and flow rate are changed within the specified range. The flow rate can be adjusted with a pinch cock (clamp). For example, in the case of the blood pump device of the present invention, it is desirable to perform learning by changing the motor rotation speed in the range of 1200 to 2600 rpm (200 rpm interval) and the rotation speed in the range of 1 to 9 L / min (1 L / min interval). . When this method is adopted, the flow sensor can be removed from the patient's body after learning the number of rotations of the blood pump.

FIG. 1 is a block diagram of an embodiment of a blood pump device of the present invention. FIG. 2 is a front view of an example of a blood pump device main body used in the blood pump device of the present invention. FIG. 3 is a plan view of the blood pump device main body shown in FIG. FIG. 4 is a longitudinal sectional view of the blood pump device main body of the embodiment shown in FIG. 5 is a cross-sectional view taken along line AA in FIG. FIG. 6 is a block diagram of another embodiment of the blood pump device of the present invention. FIG. 7 is an explanatory diagram for explaining the sigmoid function and the relationship between the intracellular potential and the impulse frequency. FIG. 8 is an explanatory diagram for explaining a definitive analog model. FIG. 9 is an explanatory diagram for explaining a feedforward neural network. FIG. 10 is an explanatory diagram illustrating an example of a feedforward neural network. FIG. 11 is an explanatory diagram for explaining the learning procedure of the sequential update learning method. FIG. 12 is an explanatory diagram for explaining a feedforward neural network that learns by the error back propagation method. FIG. 13 is an explanatory diagram for explaining the signal flow in the error back-propagation mode. FIG. 14 is an explanatory diagram for explaining the operation of the element in the output calculation mode. FIG. 15 is an explanatory diagram for explaining the element function in the error back propagation mode.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Blood pump apparatus 2 Blood pump part 3 Impeller rotational torque generation part 4 Impeller position control part 5 Blood pump apparatus main-body part 6 Control mechanism 21 Impeller 25 Magnetic body 31 Rotor 34 Motor 41 Electromagnet 20 Housing 57 Blood measurement data input part 58 Discharge flow rate Calculation unit 59 Display unit 60 Discharge flow related relational expression storage unit

Claims (14)

A housing having a blood inflow port and a blood outflow port, an impeller that rotates in the housing and feeds blood, a motor for rotating the impeller, a blood measurement data input unit, and an impeller rotational speed measurement or calculation function A blood pump device comprising a motor current value measurement function and a discharge flow rate calculation unit for calculating a blood discharge flow rate,
The discharge flow rate calculation unit uses, as input layer data, blood viscosity values, motor current measurement values, and motor rotation values at a number of pre-measured blood discharge flow rates, and the blood discharge flow rate related to each input layer data as teacher data. A discharge flow rate related relational expression storage unit that stores a discharge flow rate related relational expression between the input layer data calculated using the network and the discharge flow rate;
The discharge flow rate relational expression, the blood viscosity value input to the blood measurement data input unit or the blood viscosity value calculated from the input hematocrit value, the impeller rotation value obtained from the impeller rotation speed measurement or calculation function, Using the motor current measurement value obtained by the motor current value measurement function or its correction value, a discharge flow rate calculation function for calculating the blood discharge flow rate at the blood viscosity value and the motor current measurement value and the impeller rotation value,
An input calculated using a neural network with blood viscosity values, motor current measurement values and motor rotation values as input layer data, and pump head values related to each of the input layer data as teacher data, in a number of pump head values measured in advance. It is calculated from the pump head related relational expression storage unit for storing the relational expression between the layer data and the pump head, the pump head related relational expression, the blood viscosity value inputted into the blood measurement data input part or the inputted hematocrit value. The blood viscosity value and the motor current using the blood viscosity value, the impeller rotation value obtained from the impeller rotation number measurement or calculation function, the motor current measurement value obtained by the motor current value measurement function or the correction value thereof. Pump that calculates pump head at measured value and impeller rotation value Degree and the calculation function,
It has a fluid output calculation function for calculating the fluid output of the pump using the calculated discharge flow rate calculated by the discharge flow rate calculation function and the calculated pump lift calculated by the pump lift calculation function. Blood pump device. The blood pump device according to claim 1, wherein the blood pump device is a centrifugal blood pump device, and the impeller rotates in the housing and feeds blood by a centrifugal force at the time of rotation. The blood pump device includes a housing having a blood inflow port and a blood outflow port, a magnetic body therein, a centrifugal blood pump having an impeller that rotates within the housing and that feeds blood by centrifugal force during rotation. A rotor including a magnet for attracting the magnetic body of the impeller of the centrifugal blood pump unit, an impeller rotational torque generating unit including a motor for rotating the rotor, and an impeller position control unit including an electromagnet The blood pump device according to claim 1, wherein the impeller rotates in a non-contact state with respect to the housing. The blood pump device according to any one of claims 1 to 3, wherein the blood measurement data input unit is a blood viscosity measurement value input unit, and the blood viscosity value is a blood viscosity measurement value. 4. The blood measurement data input unit is a hematocrit value input unit, and the blood pump device has a blood viscosity calculation function for calculating blood viscosity from the input hematocrit value. Blood pump device. The blood measurement data input unit is a hematocrit value input unit, and the blood pump device includes a blood viscosity calculation function for calculating blood viscosity from an input hematocrit value and a specific gravity calculation function for calculating specific gravity. 4. The blood pump device according to any one of 3 to 3. The blood pump device includes a specific gravity input unit or specific gravity calculation function, and a motor current measurement value obtained by correcting the motor current measurement value using the specific gravity input by the specific gravity input unit or calculated by the specific gravity calculation function. The blood pump device according to claim 1, further comprising a motor current measurement value correction function. The blood pump device includes a correction function based on motor characteristics, and a motor current measurement value correction function that uses a value obtained by correcting the motor current measurement value using the motor characteristic correction function as a motor current measurement value. The blood pump device according to any one of 1 to 7. The blood pump device includes a calculated discharge flow rate calculated by the discharge flow rate calculation function, a calculated pump lift calculated by the pump lift calculation function, a shaft torque converted from the measured motor current value, and the impeller rotation The blood pump device according to any one of claims 1 to 8, further comprising a pump efficiency calculation function for calculating a pump efficiency of the pump using a number. The blood pump device includes a calculated discharge flow rate calculated by the discharge flow rate calculation function, a calculated pump lift calculated by the pump lift calculation function, a shaft torque converted from the measured motor current value, and the impeller rotation The blood pump device according to any one of claims 1 to 9, further comprising a pump total loss calculation function for calculating a total loss in the pump using a number. The method for adjusting a blood pump device according to any one of claims 1 to 10 , wherein the flow rate of the pump is measured using a flow sensor in a state where the blood pump device is attached to a subject, and the flow sensor is used. The discharge flow rate measured in this way is used as teacher data, and the blood viscosity value or hematocrit value, motor current measurement value, and motor rotation value at the time of the discharge flow measurement are used as input data. In a state where the relational expression with the flow rate is corrected , and the blood pump device is mounted on the subject, the pressure on the inflow side and the outflow side of the pump is measured, and the pump head calculated from the measured pressure value is the teacher data Enter the blood viscosity value or hematocrit value, motor current measurement value, and motor rotation value when measuring the discharge flow rate. And data, using a neural network, a method of adjusting the blood pump which is characterized in that a relational expression corrective action to correct the relationship between the input layer data and the pump head for storing. 12. The blood pump device adjustment method according to claim 11, wherein the neural network is a three-layer feedforward neural network including three layers of an input layer, an intermediate layer, and an output layer. A method for manufacturing a blood pump device according to any one of claims 1 to 10 ,
The manufacturing method uses a blood viscosity value, a motor current measurement value, and a motor rotation value at a large number of pre-measured blood discharge flow rates as input layer data, a blood discharge flow rate related to each input layer data as teacher data, and a neural network And a relational expression calculation step for calculating a relational expression between the input layer data and the discharge flow rate, and a relational expression between the input layer data and the discharge flow rate obtained by the relational expression calculation step is stored in the discharge flow rate calculation unit. Using the neural network, the storage step and the blood viscosity value, motor current measurement value and motor rotation value at a number of pre-measured blood discharge flow rates are used as input layer data, and the pump head related to each input layer data is used as teacher data. , A pump head related relational expression calculating step for calculating a relational expression between the input layer data and the pump head; Manufacturing method of a blood pump apparatus characterized by performing a storage step of storing a relational expression of the input layer and the data and the pump head obtained by pump head related equation calculating step to the pump head calculation unit. The method of manufacturing a blood pump device according to claim 13, wherein the neural network is a three-layer feedforward neural network including three layers of an input layer, an intermediate layer, and an output layer.

Images