JP2023049990A - Blood purification system, control method, control program, learning device, and learning method - Google Patents

Abstract

To provide a blood purification system, etc. capable of executing blood purification more appropriately.SOLUTION: A blood purification system includes: a line through which blood or fluid containing a filtrate flows; a plasma separation device for separating a plasma component from blood flowing through the line; a factor separation device for separating a factor component which causes a disease from the plasma component; a detection part for detecting blood information on blood flowing through the line; a liquid control mechanism for controlling the flow of fluid in the line on the basis of a control parameter; a parameter acquisition part for inputting the detected blood information to a learning model in which learning is executed so as to output a predetermined control parameter when predetermined blood information is input, and acquiring the control parameter output from the learning model; and a control part for controlling the liquid control mechanism on the basis of the acquired control parameter. Learning for the learning model is executed using supervised learning, unsupervised learning, semi-supervised learning, transduction, or multi-task learning.SELECTED DRAWING: Figure 2

Description

The present disclosure relates to a blood purification system, control method, control program, learning device, and learning method.

2. Description of the Related Art A plurality of blood purification systems used for dialysis and the like are installed in a dialysis room provided in a medical facility such as a hospital, and are configured to perform blood purification for a large number of patients in the dialysis room. In such a dialysis room, a server (central control means) is installed to store patient management data (patient weight, blood pressure, etc.) related to blood purification, and the management data is transmitted to individual blood purification systems and displayed. Let

For example, when a patient undergoes dialysis, the body weight and blood pressure before dialysis are first measured by a weight scale or a sphygmomanometer, and transmitted to the central management means and stored as patient-specific information. Based on the unique information of the patient, the central management means obtains the conditions of the patient by calculation using a predetermined arithmetic expression, and transmits the obtained conditions to the blood purification system, thereby performing optimum blood purification for each patient. conduct.

For example, Patent Literature 1 discloses a blood purification system in which a plurality of monitoring devices are installed in a dialysis room in a medical field such as a hospital.

US Pat. No. 6,201,200 discloses a system for monitoring patient parameters and blood dehydration system parameters and identifying system parameters that result in improved patient parameters or worsened patient parameters.

JP 2012-249748 A JP 2014-518692 A

Blood purification systems are required to purify blood more appropriately.

The purpose of the blood purification system, control method, control program, learning device, and learning method is to enable more appropriate blood purification.

A blood purification system according to an embodiment includes a line through which a liquid containing blood or filtrate flows, a plasma separation device that separates plasma components from the blood flowing through the line, and a factor separation device that separates disease-causing factors from the plasma components. a device, a detector that detects blood information about blood flowing in a line, a liquid control mechanism that controls the flow of liquid in the line based on control parameters, and a predetermined control parameter when predetermined blood information is input. A parameter acquisition unit that inputs the blood information detected by the detection unit into the learning model that has been learned to output a control parameter that is output from the learning model, and based on the control parameter that the parameter acquisition unit acquires and a control unit for controlling the liquid control mechanism, the blood information being the degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, and filtrate. pressure, blood flow rate, plasma flow rate, filtrate flow rate, plasma separation device fouling, factor separation device fouling, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and patient connected to line including at least one of blood pressure, and the learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to obtain blood information when blood information is input It is learned to output a control parameter capable of stabilizing the blood condition indicated by the information.

In the blood purification system according to the embodiment, the liquid control mechanism includes a magnetic force regulator that applies a magnetic field to the blood in a predetermined direction, a pump that controls the flow of liquid in the line, or a resistor that applies resistance to the line. Preferably, the applying member is included, and the control parameter includes at least one of the strength of the magnetic field applied by the magnetic force adjuster, the driving amount of the pump, and the magnitude of the resistance applied by the resistance applying member.

In the blood purification system according to the embodiment, a memory for storing the learning model in association with the product data of the plasma separation device, data related to blood flowing in the line, clearance data by the blood purification system, amount of causative substance removed, or antithrombotic property It is preferable to further have a part.

In the blood purification system according to the embodiment, a generation unit that generates a learning model based on the blood information detected by the detection unit before and after the liquid control mechanism is controlled based on the specific control parameter and the specific control parameter. It is preferable to further have

In the blood purification system according to the embodiment, the blood purification system is preferably an extracorporeal circulation blood purification system.

In the blood purification system according to the embodiment, the blood purification system preferably performs apheresis by double plasmapheresis.

A control method according to an embodiment includes a line through which a liquid containing blood or filtrate flows, a plasma separation device that separates a plasma component from the blood flowing through the line, and a factor separation device that separates a disease-causing factor component from the plasma component. a detection unit for detecting blood information about the blood flowing in the line; and a liquid control mechanism for controlling the flow of the liquid in the line based on the control parameter. The blood information detected by the detection unit is input to a learning model that has been learned to output a predetermined control parameter when the information is input, the control parameter output from the learning model is obtained, and the obtained control Based on the parameters, the blood information includes controlling the fluid control mechanism, blood turbulence degree, blood vorticity, heart murmur degree, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure. , blood flow, plasma flow, filtrate flow, plasma separation device fouling, factor separation device fouling, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of the patient connected to the line Including at least one of, the learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning, and when blood information is input, the blood information is learned to output a control parameter capable of stabilizing the blood state shown in .

A control program according to an embodiment includes a line through which a liquid containing blood or filtrate flows, a plasma separation device that separates plasma components from the blood flowing through the line, and a factor separation device that separates disease-causing factors from the plasma components. a control program for a computer included in a blood purification system comprising: a detection unit for detecting blood information about blood flowing in a line; and a liquid control mechanism for controlling the flow of liquid in the line based on a control parameter, inputting the blood information detected by the detection unit to a learning model that has been learned to output a predetermined control parameter when predetermined blood information is input, and acquires the control parameter output from the learning model. , to control the liquid control mechanism based on the acquired control parameters, and the blood information includes the degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure drop, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, plasma separation device fouling, factor separation device fouling, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and The learning model includes at least one of blood pressure of a patient connected to the line, and the learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multi-task learning to input the blood information. is learned to output a control parameter capable of stabilizing the blood condition indicated by the blood information.

The learning device according to the embodiment includes a line through which a liquid containing blood or filtrate flows, a plasma separation device that separates plasma components from the blood flowing through the line, and a factor separation device that separates disease-causing factors from the plasma components. and a detection unit for detecting blood information about blood flowing in a line; and a liquid control mechanism for controlling the flow of liquid in the line based on the control parameter. a data acquisition unit that acquires a combination; a generation unit that uses the combination acquired by the data acquisition unit to generate a learning model trained to output a predetermined control parameter when predetermined blood information is input; and an output control unit for outputting information about the learning model, wherein the blood information includes degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, pressure loss of blood pressure, plasma pressure, and filtrate. pressure, blood flow rate, plasma flow rate, filtrate flow rate, plasma separation device fouling, factor separation device fouling, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and patient connected to line including at least one of blood pressure, and the learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to obtain blood information when blood information is input It is learned to output a control parameter capable of stabilizing the blood condition indicated by the information.

The learning device according to the embodiment further has a communication unit for communicating with a plurality of blood purification systems, and the data acquisition unit acquires combinations by receiving them from the plurality of blood purification systems via the communication unit. is preferred.

In the learning device according to the embodiment, the data acquisition unit controls the liquid control mechanism based on the specific control parameter, and acquires blood information detected by the detection unit before and after controlling the liquid control mechanism based on the specific control parameter. It is preferable to obtain the combination by obtaining

A learning method according to an embodiment comprises a line in which a liquid containing blood or filtrate flows, a plasma separation device that separates plasma components from the blood flowing in the line, and a disease-causing factor component that separates disease-causing factors from the plasma components. Blood information and control parameters in a blood purification system having a factor separation device, a detection unit for detecting blood information about blood flowing in a line, and a liquid control mechanism for controlling the flow of liquid in the line based on control parameters Obtaining a plurality of combinations of, using the obtained combinations, generating a learning model trained to output a predetermined control parameter when predetermined blood information is input, and outputting information about the learning model; The blood information includes the degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure drop, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, plasma a learning model comprising at least one of separation device fouling, factor separation device fouling, hematocrit in blood, albumin concentration in filtrate, degree of hemolysis, and blood pressure of a patient connected to the line; uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to stabilize the blood state indicated in the blood information when the blood information is input. It is learned to output possible control parameters.

A control program according to an embodiment is a computer control program comprising: a line through which liquid including blood or filtrate flows; a plasma separation device for separating plasma components from the blood flowing through the line; A blood purification system having a factor separation device that separates factor components, a detection unit that detects blood information about blood flowing in a line, and a liquid control mechanism that controls the flow of liquid in the line based on a control parameter, Obtaining a plurality of combinations of blood information and control parameters, using the obtained combinations to generate a learning model learned to output a predetermined control parameter when predetermined blood information is input, and relating to the learning model The computer is made to output information, and the blood information is the degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure drop, plasma pressure, filtrate pressure, blood flow rate. , plasma flow rate, filtrate flow rate, plasma separation device fouling, factor separation device fouling, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of the patient connected to the line. Including at least one, the learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning, and when blood information is input, the blood information is indicated It is learned to output control parameters that can stabilize the blood condition.

The blood purification system, control method, control program, learning device, and learning method can perform blood purification more appropriately.

The objects and advantages of the invention may be realized and obtained by means of the elements and combinations particularly pointed out in the claims. Both the foregoing general description and the following detailed description are exemplary and explanatory, and are not limiting of the invention as claimed.

It is a figure showing a schematic structure of management system 100 concerning an embodiment. FIG. 2 is a schematic diagram showing a blood purification unit 14; 1 is a diagram showing a schematic configuration of a blood purification system 40; FIG. FIG. 11 is a schematic diagram showing an example of the data structure of a result table 553; FIG. 3 is a diagram showing a schematic configuration of a server 80; FIG. 4 is a flowchart showing an example of the operation of learning processing of the control device 50; 4 is a flow chart showing an example of control processing operations of the control device 50. FIG. 6 is a flow chart showing an example of the operation of learning processing of the server 80. FIG. FIG. 14 is a schematic diagram showing another blood purification unit 14-2.

Hereinafter, a blood purification system, a control method, a control program, a learning device, and a learning method according to one aspect of the embodiments will be described with reference to the drawings. However, it should be noted that the technical scope of the present invention is not limited to those embodiments, but extends to the invention described in the claims and equivalents thereof.

FIG. 1 is a diagram showing a schematic configuration of a management system 100 according to an embodiment.

As shown in FIG. 1, the management system 100 has one or more blood purification systems 40, a server 80, and the like. Each blood purification system 40 and server 80 are connected via network 70 so as to be able to communicate with each other. Each blood purification system 40 is an extracorporeal circulation blood purification system. Each blood purification system 40 performs apheresis by double filtration plasmapheresis (DFPP). Each blood purification system 40 has a blood purification unit 14, a control device 50, and the like. Network 70 is a wired network such as the Internet or an intranet. The network 70 may be a wireless network such as a wireless LAN (Local Area Network).

FIG. 2 is a schematic diagram showing the blood purification unit 14 included in the blood purification system 40. As shown in FIG.

As shown in FIG. 2, the blood purification unit 14 includes a plasma separation device 1, a blood feed line 3, a blood return line 4, a blood pump 5, a factor separation device 7, a plasma line 8, a plasma liquid feed pump 9, and a filtrate line 10. , filtrate pump 11, magnetic force regulator 16, blood flow detector 17, hematocrit value detector 18, albumin concentration detector 19, first pressure gauges 21 to 23, second pressure gauge 25, third pressure gauge 26, first It has flowmeters 28, 29, a second flowmeter 30, a third flowmeter 31, piping systems 33, 33', (three-way) valves 34, 34', piping systems 35, 35', clamps 36, and the like. A blood purification unit 14 is used to utilize the plasma separation device 1 in a clinical environment. The blood purification unit 14 performs double plasmapheresis for the removal of high molecular weight etiological or etiological-related substances.

The blood supply line 3, the blood return line 4, the plasma line 8, the filtrate line 10, and the piping system 33, 33' are examples of lines through which liquid including blood or filtrate flows, and are blood circuits. As the blood supply line 3, the blood return line 4, the plasma line 8, the filtrate line 10, and the piping systems 33, 33', for example, polyvinyl chloride tubes are used. As the blood supply line 3 and the blood return line 4, for example, tubes having a length and a diameter such that the total volume of liquid contained therein is about 150 ml are used. Other components may be utilized as the blood supply line 3, blood return line 4, plasma line 8, filtrate line 10 and tubing system 33, 33'.

The piping system 33 is a blood circuit and has a blood sampling part 33a from a patient (animal or human). The piping system 33' is a blood circuit and has a blood return portion 33b to the patient. The blood feed line 3 feeds the blood removed from the blood collecting portion 33 a to the plasma separation device 1 via the blood pump 5 . The blood return line 4 sends the blood that has flowed out of the plasma separation device 1 to the blood return section 33b. The plasma line 8 is connected to the plasma outlet side port 1b of the plasma separation device 1, and the filtrate (liquid containing plasma components) flowing out from the plasma outlet side port 1b is sent to the factor separation device 7 via the plasma liquid feed pump 9. send. Filtrate line 10 is connected to factor separation device 7 and sends filtrate flowing out of factor separation device 7 to blood return line 4 via filtrate pump 11 .

The plasma separation device 1 separates plasma components and cell components from blood flowing through each line. The plasma separation device 1 has a plasma outlet side port 1a, a plasma outlet side port 1b, a blood inlet side port 1c, a blood outlet side port 1d, a hollow membrane body 1e, and the like. The blood inlet side port 1c is an inlet for blood removed from a patient. The hollow membrane body 1e is configured by bundling a large number of hollow fiber membranes through which the blood flowing from the blood inlet side port 1c passes. The blood outlet side port 1d is an outlet for the blood that has passed through the hollow membrane body 1e to the outside of the device. The plasma outlet ports 1a and 1b are outlets for the filtrate that has been filtered by the hollow membrane body 1e and permeated to the outside of the hollow membrane body 1e. The plasma outlet side port 1a is a spare port. The plasma component filtered and separated by the hollow membrane body 1e is discharged from the plasma outlet side port 1a or 1b, and the blood rich in cell components is discharged from the blood outlet side port 1d. As the plasma separation device 1, a polyethylene hollow fiber membrane plasma separation device (“Plasmaflow OP-05” manufactured by Asahi Kasei Medical Co., Ltd.) with a nominal pore size of 0.3 μm, an inner diameter of 350 μm, a film thickness of 50 μm, and a membrane area of 0.5 m ² is used. It is possible.

The factor separation device 7 separates disease-causing factor components from the separated plasma components. The factor separation device 7 has a filtrate outlet port 7a, a filtrate outlet port 7b, a plasma outlet port 7c, a plasma inlet port 7d, a hollow membrane body 7e, and the like. The plasma inlet side port 7 d is an inlet for the filtrate that flows out from the plasma outlet side port 1 b of the plasma separation device 1 . The hollow membrane body 7e is configured by bundling a large number of hollow fiber membranes through which the filtrate flowing from the plasma inlet side port 7d passes. Adsorption beads may be used as the hollow membrane body 7e. The plasma outlet side port 7c is an outflow port for disease-causing factor components that have passed through the hollow membrane body 7e. The filtrate outlet ports 7a and 7b are outlets for the filtrate filtered by the hollow membrane body 7e to the outside of the apparatus. The filtrate outlet side port 7b is a spare port. In FIG. 2, the filtrate line 10 is connected to the filtrate outlet port 7a, but it can be connected to the filtrate outlet port 7b in order to make effective use of the hollow membrane or adsorbent beads contained in the factor separation device 7. more preferred. As the factor separation device 7, for example, MONET (registered trademark) manufactured by Fresenius Medical Care can be used.

A blood pump 5 is provided in the blood supply line 3 and controls the flow of blood in the blood supply line 3 . A plasma pump 9 is provided in the plasma line 8 and controls the flow of plasma in the plasma line 8 . A filtrate pump 11 is provided in the filtrate line 10 to control the flow of filtrate in the filtrate line 10 . The blood pump 5 , the plasma feed pump 9 , and the filtrate pump 11 are provided so that their driving amounts (output amounts) can be changed according to control by the control device 50 . A roller pump, for example, is used as the blood pump 5, the plasma feed pump 9 and the filtrate pump 11. FIG. Other known pumps may be used as blood pump 5, plasma feed pump 9 and filtrate pump 11.

Magnetic force adjuster 16 is provided so as to surround a predetermined position of plasma separation device 1 and applies a magnetic field to blood in plasma separation device 1 in a predetermined direction. The magnetic force adjuster 16 is provided so as to surround a predetermined position on the patient's body connected to the blood supply line 3, the blood return line 4, or the blood purification unit 14. A magnetic field may be applied in a predetermined direction to the blood in the body. The magnetic force adjuster 16 has an annular magnet and applies a unidirectional magnetic field in a region inside the annular magnet, parallel or opposite to the direction of blood flow. The magnetic force adjuster 16 is provided so as to change the strength of the applied magnetic field under the control of the control device 50 . The strength of the magnetic field is set to a strength sufficient to reduce the viscosity of the blood by a predetermined amount and/or to reduce turbulence in the blood flow by a predetermined amount at the location of application of the magnetic field. As the magnetic force adjuster 16, for example, an electromagnet, a permanent magnet, a superconducting magnet, or the like is used.

The magnetic force regulator 16 can suppress clogging of blood cell components in the plasma separation device 1, suppress pressure rise in the plasma separation device 1, and reduce blood flow viscosity and blood turbulence. Thereby, the blood purification unit 14 can lower the patient's blood pressure, improve hypertension, and reduce the occurrence of heart murmurs. If the strength of the unidirectional magnetic field of the magnetic force regulator 16 is less than 0.01 tesla, the effect of the magnetic force on red blood cells is low. On the other hand, the magnetic force regulator 16 exceeding 100 Tesla tends to be difficult to handle and expensive. Therefore, the intensity of the unidirectional magnetic field of the magnetic force adjuster 16 is preferably 0.01 tesla or more and 100 tesla or less, more preferably 1 tesla or more and 10 tesla or less.

The blood pump 5 , plasma liquid transfer pump 9 , filtrate pump 11 and magnetic force regulator 16 are examples of a liquid control mechanism that controls liquid flow in each line of the blood purification unit 14 .

A blood flow detector 17 is provided at a predetermined position of the plasma separation device 1 to detect the direction, velocity and/or velocity distribution of blood flow within the plasma separation device 1 . The blood flow detector 17 is provided at a predetermined position of the patient's body connected to the blood supply line 3, the blood return line 4, or the blood purification unit 14, and is connected to the blood supply line 3, the blood return line 4, or the patient's body. may detect the direction, velocity and/or velocity distribution of the blood flow. As the blood flow detector 17, for example, an ultrasonic diagnostic imaging apparatus Aixplorer manufactured by SuperSonic Imagine can be used. The blood purification unit 14 can detect abnormal blood flow in a short time with high accuracy by the blood flow detector 17 .

A hematocrit value detector 18 is provided in the blood return line 4 and measures the hematocrit value in the blood in the blood return line 4 . The hematocrit value is an indicator of blood concentration, and is indicated by the volume ratio of red blood cells to whole blood. In order to confirm whether or not there is an abnormality in the amount of plasma permeating through the plasma separation device 1, the hematocrit value detector 18 is preferably provided near the blood outlet side port 1d. On the other hand, in order to check the difference between the hematocrit value in the blood flowing into the blood purification unit 14 and the hematocrit value in the blood flowing out from the blood purification unit 14, the hematocrit value detector 18 is connected to the blood supply line 3. Preferably, they are provided between the blood collecting portion 33a and the first pressure gauge 23 and between the first flowmeter 29 of the blood return line 4 and the blood return portion 33b. As the hematocrit value detector 18, for example, StatStrip (registered trademark) Hb/Hct manufactured by Nova Biomedical (NOVA (registered trademark) BIOMEDICAL) can be used.

An albumin concentration detector 19 is provided in the filtrate line 10 to measure the albumin concentration in the filtrate in the filtrate line 10 . As the albumin concentration detector 19, for example, a biochemical analyzer (DRI-CHEM NX700 manufactured by Fuji Film Co., Ltd.) can be used.

The first pressure gauge 23 is provided between the blood collecting portion 33 a and the blood pump 5 in the blood supply line 3 and measures the blood pressure in the blood supply line 3 . A first pressure gauge 21 is provided between the blood pump 5 and the plasma separation device 1 in the blood supply line 3 and measures the blood pressure in the blood supply line 3 . The first pressure gauge 22 is provided between the plasma separation device 1 and the blood return section 33b in the blood return line 4 and measures the blood pressure in the blood return line 4. The first pressure gauge 21 and the first pressure gauge 22 can always measure the pressure in and out of the plasma separation device 1 , and the blood pressure by the plasma separation device 1 is measured by the first pressure gauge 21 and the first pressure gauge 22 . of pressure loss can be measured. A second pressure gauge 25 is provided in the plasma line 8 and measures the plasma pressure within the plasma line 8 . A third pressure gauge 26 is provided in the filtrate line 10 to measure the filtrate pressure in the filtrate line 10 . As the first pressure gauges 21 to 23, the second pressure gauge 25 and the third pressure gauge 26, known water pressure gauges such as semiconductor piezoresistive diffusion pressure sensors can be used.

A first flowmeter 28 is provided in the blood supply line 3 and measures the blood flow rate in the blood supply line 3 . A first flow meter 29 is provided in the blood return line 4 and measures the blood flow rate in the blood return line 4 . A second flow meter 30 is provided in the plasma line 8 and measures the plasma flow rate within the plasma line 8 . A third flow meter 31 is provided in the filtrate line 10 to measure the filtrate flow rate in the filtrate line 10 . It is determined by the first flowmeters 28, 29, the second flowmeter 30, and the third flowmeter 31 whether or not the flow rate in each line controlled by the pump or the like in each line is within the set range. It is possible. As the first flowmeters 28, 29, the second flowmeter 30, and the third flowmeter 31, known flowmeters such as Coriolis mass flowmeters, electromagnetic mass flowmeters, and ultrasonic mass flowmeters can be used.

Blood flow detector 17, hematocrit value detector 18, albumin concentration detector 19, first pressure gauges 21 to 23, second pressure gauge 25, third pressure gauge 26, first flowmeters 28, 29, second flowmeter The 30 and the third flowmeter 31 are examples of a detection section that detects blood information about blood flowing through each line of the blood purification unit 14 .

A blood purification unit 14 in a clinical setting may include an anticoagulant injector, an air bubble detector, an alarm function, and the like. In order to use the blood purification system 40 safely, the blood purification system 40 is preferably equipped with a power generator and a battery so that it can operate even in the event of a disaster or power outage.

Blood withdrawn from the patient via the blood collection unit 33 a is sent to the plasma separation device 1 by the blood pump 5 through the blood supply line 3 . The blood sent to the plasma separation device 1 flows into the hollow membrane body 1e from the blood inlet side port 1c and flows out to the blood return line 4 from the blood outlet side port 1d. On the other hand, the filtrate filtered by the hollow membrane body 1e flows out from the plasma outlet side port 1b, passes through the plasma line 8, and is sent to the factor separation device 7 by the plasma liquid sending pump 9. The filtrate sent to the factor separation device 7 flows into the hollow membrane body 7e through the plasma inlet side port 7d, where disease-causing factor components are separated and discharged out of the device through the plasma outlet side port 7c. On the other hand, the filtrate filtered by the hollow membrane body 7 e flows out from the filtrate outlet side port 7 a , passes through the filtrate line 10 , and is sent to the blood return line 4 by the plasma liquid sending pump 9 . The blood flowing out from the blood outlet side port 1d to the blood return line 4 and the filtrate sent to the blood return line 4 from the filtrate outlet side port 7a are returned to the patient via the blood return section 33b.

FIG. 3 is a diagram showing a schematic configuration of the blood purification system 40. As shown in FIG.

The blood purification unit 14 of the blood purification system 40 has a driving device 15, an electrocardiogram measuring device 41, a hemolysis measuring device 42, a pulsometer 43, a sphygmomanometer 44, a blood flow meter 45, etc., in addition to the configuration described above.

The driving device 15 includes one or more motors, drives the blood pump 5, the plasma liquid feeding pump 9, and the filtrate pump 11 according to control signals from the control device 50, and removes the liquid in each line of the blood purification unit 14. control the flow.

The electrocardiogram measuring device 41 is attached to a patient connected to the blood purification unit 14, measures an electrocardiogram (heartbeat waveform) of the patient, and outputs the measured electrocardiogram. As the electrocardiogram measuring device 41, for example, a wearable electrocardiogram measuring device such as Apple Watch (Series 4, 5, 6) can be used. The Apple Watch (Series 4, 5, 6) has a crystal and electrodes and works in conjunction with an ECG application to record an EKG similar to the Lead I ECG. The electrocardiograph 41 constantly detects the heart rhythm and may notify you if an irregular heart rhythm indicative of atrial fibrillation (AFib) is detected.

A hemolysis meter 42 is connected to the patient connected to the blood purification unit 14 and measures the degree of hemolysis of the patient. Hemolysis means destruction of blood cells, in particular destruction of red blood cells. As the hemolysis measuring device 42, a blood leakage detector incorporated in all commercially available dialysis monitoring devices (for example, TR-3300M manufactured by Toray Industries, Inc.) can be used.

A pulse meter 43 and a blood pressure meter 44 are attached to a patient connected to the blood purification unit 14, respectively, and measure the patient's pulse and blood pressure. The blood flow meter 45 is connected to the patient connected to the blood purification unit 14 and measures the circulating blood flow of the patient. As the pulse meter 43, blood pressure meter 44 and blood flow meter 45, known measuring instruments can be used.

The control device 50 is an example of a learning device, and is an information processing device such as a personal computer. The control device 50 has an input device 51, a display device 52, a first communication device 53, an interface device 54, a first storage device 55, a first processing device 56, and the like. The input device 51, the display device 52, the first communication device 53, the interface device 54, the first storage device 55 and the first processing device 56 are interconnected via a CPU (Central Processing Unit) bus or the like.

The input device 51 has a touch panel type input device or an input device such as a keyboard and a mouse, and an interface circuit for acquiring signals from the input device, and outputs an operation signal according to a user's input operation.

The display device 52 has a display including liquid crystal, organic EL (Electro-Luminescence), etc. and an interface circuit for outputting image data to the display, and displays the image data on the display.

The first communication device 53 is an example of a communication unit. The first communication device 53 has a wired communication interface circuit conforming to a communication protocol such as TCP/IP (Transmission Control Protocol/Internet Protocol). The first communication device 53 communicates with the network 70 according to a communication standard such as Ethernet (registered trademark). The first communication device 53 sends data received from the server 80 via the network 70 to the first processing device 56 and sends data received from the first processing device 56 to the server 80 via the network 70 . The first communication device 53 has an antenna for transmitting and receiving wireless signals and a wireless communication interface circuit according to a communication protocol such as a wireless LAN, and communicates with the network 70 according to a communication standard such as a wireless LAN. good too.

The interface device 54 has an interface circuit conforming to a serial bus such as USB (Universal Serial Bus). The interface device 54 includes the driving device 15 of the blood purification unit 14, the magnetic force regulator 16, the blood flow detector 17, the hematocrit value detector 18, the albumin concentration detector 19, the first pressure gauges 21 to 23, and the second pressure gauge. 25, third pressure gauge 26, first flowmeters 28, 29, second flowmeter 30, third flowmeter 31, electrocardiogram measuring device 41, hemolysis measuring device 42, pulse meter 43, sphygmomanometer 44 and blood flow meter 45 etc., and is provided so as to be able to communicate with each connected device. The interface device 54 sends data received from each connected device to the first processing device 56, and transmits data received from the first processing device 56 to each connected device. The interface device 54 may have an interface circuit conforming to a short-range wireless communication standard such as Bluetooth (registered trademark), and may be wirelessly connected to each part of the blood purification unit 14 for communication.

The first storage device 55 is an example of a storage unit. The first storage device 55 includes memory devices such as RAM (Random Access Memory) and ROM (Read Only Memory), fixed disk devices such as hard disks, or portable storage devices such as flexible disks and optical disks. The first storage device 55 also stores computer programs, databases, tables, and the like used for various processes of the control device 50 . The computer program may be installed in the first storage device 55 from a computer-readable portable recording medium using a known setup program or the like. Examples of portable recording media include CD-ROMs (compact disc read only memory) and DVD-ROMs (digital versatile disc read only memory). The computer program may be installed from a predetermined server or the like.

A learning model 551, product data 552, a result table 553, and the like are stored as data in the first storage device 55. FIG. Learning model 551 is a model for controlling blood flow in blood purification unit 14 . Learning model 551 is generated by first processing unit 56 or received from server 80 . The product data 552 is data relating to the blood purification unit 14, and indicates the plasma component and cell component separation performance (filtration performance) of the plasma separation device 1, the factor component separation performance (filtration performance) of the factor separation device 7, and the like. show. The product data 552 is preset by the user using the input device 51 . The result table 553 stores the result of blood purification for each blood purification by the blood purification system 40 . Details of the result table 553 will be described later.

The first processing device 56 operates based on a program stored in advance in the first storage device 55 . The first processing device 56 is, for example, a CPU. As the first processing device 56, a DSP (digital signal processor), LSI (large scale integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), or the like may be used. The first processing device 56 is connected to the input device 51, the display device 52, the first communication device 53, the interface device 54, the first storage device 55, etc., and controls each device. First processing device 56 generates learning model 551 and controls blood flow in blood purification unit 14 using generated learning model 551 .

The first processing device 56 reads the computer program stored in the first storage device 55 and operates according to the read computer program. Thereby, the first processing device 56 functions as a first data acquisition section 561 , a first generation section 562 , a first output control section 563 , a parameter acquisition section 564 and a control section 565 . The first data acquisition unit 561, the first generation unit 562, and the first output control unit 563 are examples of the data acquisition unit, generation unit, and output control unit, respectively.

FIG. 4 is a schematic diagram showing an example of the data structure of the result table 553. As shown in FIG.

The result table 553 associates an identification number (measurement ID), learning model, patient data, product data, blood data, clearance data, causative substance removal amount, antithrombogenicity, etc. for each blood purification by the blood purification system 40. remembered. The learning model is the learning model 551 used in blood purification. As the learning model, identification information or a storage address of the learning model 551 used in blood purification may be stored. The patient data is data related to the patient, such as the name, height, and weight of the patient for whom blood purification was performed. The product data is product data 552 preset by the user.

The blood data is data relating to the blood flowing through each line of the blood purification unit 14, and includes the patient's pulse rate, blood pressure, circulating blood flow rate, blood hematocrit value, degree of hemolysis, and the like before blood purification. The clearance data is data relating to blood purified by the blood purification system 40, and includes the patient's pulse rate, blood pressure, circulating blood flow, blood hematocrit value, degree of hemolysis, and the like after blood purification. Clearance data may further include blood pressure measured during blood purification, blood pressure pressure drop, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, albumin concentration in filtrate, and the like. The causative substance removal amount is the amount of the causative substance removed by the blood purification unit 14 . The antithrombotic property is the ability to suppress activation of the blood coagulation system, and is the degree of thrombi formed in the blood purified by the blood purification unit 14 .

FIG. 5 is a diagram showing a schematic configuration of the server 80. As shown in FIG.

Server 80 is a host computer of control device 50 and is an example of a learning device. The server 80 has a second communication device 81, a second storage device 82, a second processing device 83, and the like. The second communication device 81, the second storage device 82 and the second processing device 83 are interconnected via a CPU bus or the like.

The second communication device 81 is an example of a communication section for communicating with multiple blood purification systems 40 . The second communication device 81 has a wired communication interface circuit conforming to a communication protocol such as TCP/IP. The second communication device 81 communicates with the network 70 according to a communication standard such as Ethernet (registered trademark). The second communication device 81 transmits data received from the control device 50 via the network 70 to the second processing device 83, and transmits data received from the second processing device 83 to the control device 50 via the network 70. . The second communication device 81 has an antenna for transmitting and receiving wireless signals and a wireless communication interface circuit conforming to a communication protocol such as a wireless LAN, and communicates with the network 70 according to a communication standard such as a wireless LAN. good too.

The second storage device 82 includes memory devices such as RAM and ROM, fixed disk devices such as hard disks, portable storage devices such as flexible disks and optical disks, and the like. The second storage device 82 stores computer programs, databases, tables, and the like used for various processes of the server 80 . The computer program may be installed in the second storage device 82 from a computer-readable portable recording medium such as CD-ROM, DVD-ROM, etc. using a known setup program or the like. The computer program may be installed from a predetermined server or the like.

A learning model 821 and the like are stored as data in the second storage device 82 . The learning model 821 is generated by the second processor 83 or received from the controller 50 .

The second processing device 83 operates based on a program stored in advance in the second storage device 82 . The second processing device 83 is, for example, a CPU. A DSP, LSI, ASIC, FPGA, or the like may be used as the second processing device 83 . The second processing device 83 is connected to the second communication device 81, the second storage device 82, etc., and controls each device. The second communication device 81 generates a learning model 821 and transmits it to each control device 50 .

The second processing device 83 reads the computer program stored in the second storage device 82 and operates according to the read computer program. Thereby, the second processing device 83 functions as a second data acquisition section 831 , a second generation section 832 and a second output control section 833 .

FIG. 6 is a flow chart showing an example of the learning process operation of the control device 50 .

An example of the operation of the learning process of the control device 50 will be described below with reference to the flowchart shown in FIG. The operation flow described below is executed mainly by the first processing device 56 in cooperation with each element of the control device 50 based on a program stored in the first storage device 55 in advance.

First, the first data acquisition unit 561 acquires blood information about blood flowing through each line of the blood purification unit 14 (step S101). The blood information includes the degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, plasma separation device 1 Including fouling, fouling of the factor separation device 7, blood hematocrit, filtrate albumin concentration, or degree of hemolysis, pulse, blood pressure or circulating blood flow of the patient. The blood information includes one or more of each parameter described above.

The first data acquisition unit 561 acquires the direction, velocity and/or velocity distribution of the blood flow in the blood purification unit 14 from the blood flow detector 17 via the interface device 54, and based on the acquired information, The degree of turbulence and vorticity of the blood are calculated. For example, the first data acquisition unit 561 calculates the Reynolds number for each blood flow detected by the blood flow detector 17, and determines whether each blood flow is turbulent depending on whether the calculated Reynolds number is equal to or greater than a predetermined threshold. or laminar flow. The first data acquisition unit 561 calculates the ratio of the number of turbulent blood flows to the total number of blood flows detected by the blood flow detector 17 as the degree of turbulence. Further, the first data acquisition unit 561 calculates the velocity vector of the blood flow based on the direction and velocity of the blood flow acquired from the blood flow detector 17, and measures the rotation of the vector field formed by the calculated velocity vector as the vorticity. Calculate as Vorticity is a quantity that describes the rolling behavior of a flow.

The first data acquisition unit 561 acquires the electrocardiogram of the patient connected to the blood purification unit 14 from the electrocardiogram measuring device 41 via the interface device 54, and calculates the degree of heart murmur based on the acquired electrocardiogram. A heart murmur is a heart murmur that is out of sync with the heartbeat and out of rhythm. A heart murmur is a characteristic abnormal sound emitted when blood flows through a heart valve or a blood vessel near the heart (when a valve opens and closes). Abnormal sounds are primarily caused by heart valve defects. The first data acquisition unit 561 calculates, as the degree of heart murmur, the rate of occurrence of irregular heart rhythm suggestive of atrial fibrillation in the heart rhythm indicated in the acquired electrocardiogram. For example, the control device 50 stores the waveform pattern of the heartbeat rhythm of atrial fibrillation in the first storage device 55 in advance. The first data acquisition unit 561 uses a known pattern matching technique to detect a waveform similar to the heart rhythm of atrial fibrillation in the heart rhythm shown in the acquired electrocardiogram. Alternatively, the first data acquisition unit 561 detects P waves corresponding to each R wave in the heart rhythm shown in the acquired electrocardiogram, and determines the heart rhythm suggestive of atrial fibrillation based on the number of disappearances of the P waves. Detect occurrences.

The first data acquisition unit 561 acquires blood pressure from the first pressure gauges 21 to 23 via the interface device 54 . Further, the first data acquisition unit 561 calculates a value obtained by subtracting the blood pressure acquired from the first pressure gauge 22 from the blood pressure acquired from the first pressure gauge 21 as the pressure loss of the blood pressure. The first data acquisition unit 561 also acquires the plasma pressure and the filtrate pressure from the second pressure gauge 25 and the third pressure gauge 26 via the interface device 54, respectively. Also, the first data acquisition unit 561 acquires the blood flow rate, plasma flow rate, and filtrate flow rate from the first flow meters 28 and 29, the second flow meter 30, and the third flow meter 31 via the interface device 54, respectively.

In addition, the first data acquisition unit 561 calculates the filtrate volume of the plasma separation device 1 per unit time from the start of operation of the blood purification unit 14 to the present from the plasma flow rate acquired from the second flowmeter 30. . The first data acquisition unit 561 calculates the current pressure difference between the plasma pressure acquired from the second pressure gauge 25 and the blood pressure acquired from the first pressure gauge 21 . The first data acquisition unit 561 calculates the fouling of the plasma separation device 1 by further dividing the value obtained by dividing the calculated pressure difference from the calculated amount of filtrate by the unit time. Similarly, the first data acquisition unit 561 calculates the filtrate volume of the factor separation device 7 per unit time from the start of operation of the blood purification unit 14 to the present from the filtrate flow rate acquired from the third flowmeter 31. do. The first data acquisition unit 561 calculates the current pressure difference between the filtrate pressure acquired from the third pressure gauge 26 and the plasma pressure acquired from the second pressure gauge 25 . The first data acquisition unit 561 calculates the fouling of the factor separation device 7 by further dividing the value obtained by dividing the calculated pressure difference from the calculated amount of filtrate by the unit time.

The first data acquisition unit 561 acquires the hematocrit value in blood from the hematocrit value detector 18 via the interface device 54 . Also, the first data acquisition unit 561 acquires the concentration of albumin in the filtrate from the albumin concentration detector 19 via the interface device 54 . In addition, the first data acquisition unit 561 obtains the degree of hemolysis of the patient connected to the blood purification unit 14 from the hemolysis measuring instrument 42, the pulse meter 43, the sphygmomanometer 44, and the blood flow meter 45 via the interface device 54, Pulse, blood pressure and circulatory blood flow are obtained.

Next, the first data acquisition section 561 acquires control parameters relating to the flow of liquid in the blood purification unit 14 (step S102). The control parameters include the intensity of the magnetic field applied to the blood in a predetermined direction by the magnetic force adjuster 16, or the drive amount of the blood pump 5, the plasma feed pump 9, or the filtrate pump 11. FIG. Control parameters include one or more of each of the parameters described above.

For example, the control device 50 stores various usable values in the first storage device 55 for each type of control parameter, and the first data acquisition unit 561 acquires the values stored in the first storage device 55. are read out sequentially to acquire the control parameters. Note that the first data acquisition unit 561 may acquire control parameters by generating a random number within a usable value range for each type of control parameter. In addition, the control device 50 stores the optimum value of each control parameter in the first storage device 55 for each type of control parameter, and the first data acquisition unit 561 detects a slight change in the optimum value. The control parameters may be obtained by causing

Next, the first data acquisition unit 561 controls the magnetic force regulator 16 based on the acquired control parameters, or controls the blood pump 5, the plasma liquid transfer pump 9, or the filtrate pump 11 via the driving device 15. (step S103).

Next, the first data acquisition unit 561 acquires blood information in the same manner as in step S101 (step S104). In this way, the first data acquisition section 561 controls the liquid control mechanism of the blood purification unit 14 based on the control parameters acquired in step S102, and the detection section before and after controlling the liquid control mechanism based on the control parameters. Acquire blood information detected by Thereby, the first data acquisition unit 561 acquires a combination of blood information and control parameters.

Next, the first generation unit 562 sets a reward in the learning model 551 based on the blood information acquired by the first data acquisition unit 561 (step S105).

Learning model 551 used in blood purification system 40 is learned by, for example, reinforcement learning. The learning model 551 is learned by, for example, Q-learning as reinforcement learning. The learning model 551 uses the blood purification unit 14 as the environment, the control device 50 as the agent, the blood information as the state, the control of the liquid control mechanism based on the control parameters as the action, and the action that determines the value of control based on each control parameter. have a value function. Each blood information described above is a physical quantity that changes by changing each control parameter described above. That is, there is a correlation between each piece of blood information and each control parameter, and the first generator 562 creates a learning model 551 that can determine an appropriate control parameter according to the state of the blood purification unit 14. can be generated.

In Q-learning, the action value function Q(s, a) representing the action value when the action a is selected in the state s is obtained by using the environmental state s and the action a selected in the state s as independent variables. be learned. In Q-learning, learning is started in a state where the correlation between state s and action a is unknown, and by repeating trial and error to select various actions a in arbitrary state s, action-value function Q repeats Updated. In addition, in Q-learning, when an action a in a certain state s is selected, a reward r is obtained, and an action value function Q is learned. An update formula for the action-value function Q is expressed as the following formula.

In the above formula, s _t and a _t are the state and action at time t, respectively, and the state changes from s _t to s _t+1 by action a _t . r _t+1 is the reward obtained by changing the state from s _t to s _t+1 . The term maxQ means the Q when the action a that becomes the maximum value Q at the time t+1 (considered at the time t) is performed. α and γ are a learning coefficient and a discount rate, respectively, and are arbitrarily set within 0<α≦1 (usually 0.9 to 0.99) and 0<γ≦1 (usually about 0.1).

This update formula is obtained from the evaluation value Q(s _t , a _t ) of the action a _t in a certain state s _t , the evaluation value Q(s _{t + 1} , maxa _t+1 ) is larger, Q(s _t , a _t ) is increased, and if smaller, Q(s _t , a _t ) is also decreased. That is, this update formula brings the value of an action in a state closer to the value of the best action in the next state. Therefore, the action value function is set so that the action value of the action (operating condition) that is most suitable for operating the blood purification unit 14 becomes high, that is, the optimum action (operating condition) of the blood purification unit 14 ) is updated to have a higher action value.

The initial value of each action value function Q is set arbitrarily. The first generation unit 562 uses the number of executions of step S106 as the time in the above formula. The first generation unit 562 identifies the pre-control blood information acquired in step S101 as the state _st , identifies the control based on the control parameter acquired in step S102 as the action _at , and identifies the control based on the control parameter acquired in step S102 as the action at. The blood information after the control is specified as state s _t+1 .

The first generator 562 sets the reward r based on the change in blood information (state) before and after control. The first generation unit 562 calculates the reward r such that the smaller the change (difference) in the blood information after control with respect to the blood information before control, the larger the reward r, and the larger the change (difference), the smaller the reward r. set. For example, one or more thresholds are set in advance for each blood information, and the first generator 562 sets the reward r by comparing the amount of change in each blood information with each threshold. If the amount of change is greater than the maximum value of the threshold, the first generator 562 may set the reward r to zero. In this way, when the blood purification unit 14 is controlled based on a specific control parameter, the first generation unit 562 increases the reward r as the blood condition in the blood purification unit 14 becomes more stable. Set reward r. As a result, the first generation unit 562 can generate the learning model 551 that has been learned to select control parameters that make the blood state in the blood purification unit 14 more stable with respect to the current blood state. can.

Next, the first generation unit 562 updates the action value function based on the combination of the blood information and the control parameter acquired by the first data acquisition unit 561 and the set reward (step S106). The first generator 562 updates the action value function Q(s _t , a _t ) according to the above formula based on the specified state s _t , action a _t , state s _t+1 and the set reward r.

Next, the first generation unit 562 determines whether or not the learning end condition is satisfied (step S107). As learning end conditions, for example, the total update count of the action-value function has reached a predetermined number or more, or the maximum or minimum update count of each action-value function has reached a predetermined number or more. set. If the learning end condition is not satisfied yet, the first data acquisition unit 561 and the first generation unit 562 return the processing to step S101 and repeat the processing of steps S101 to S107. As a result, first data acquisition unit 561 acquires a plurality of combinations of blood information and control parameters, and first generation unit 562 uses each combination acquired by first data acquisition unit 561 to generate an action value function. Update. Note that in the second and subsequent processes, the process of step S101 may be omitted, and the first data acquisition unit 561 may use the blood information after control acquired in the previous step S104 as the blood information before control. .

On the other hand, when the end condition of learning is satisfied, the first generation unit 562 generates a learning model 551 having a combination (action value table) of finally updated values (action values) of each Q(s, a). is generated and stored in the first storage device 55 (step S108). Learning model 551 outputs a control parameter corresponding to an action with the highest action value in that state when predetermined blood information is input as a state, that is, a control parameter that can stabilize the blood information most. have been learned to Note that the learning model 551 may be learned so that when predetermined blood information is input as a state, the action value of each control parameter (each value) in that state is output.

As a result, the blood purification system 40 can automatically create optimal operating conditions for each elapsed time after the start of operation of the blood purification unit 14 . As a result, blood purification system 40 can derive optimal operating conditions, for example, in double plasmapheresis apheresis.

Next, the first output control unit 563 outputs the learning model 551 generated by the first generation unit 562 by transmitting it to the server 80 via the first communication device 53 (step S109), and performs a series of steps. finish. A learning model 551 is an example of information about a learning model. On the other hand, when the server 80 receives the learning model 551 from the control device 50 via the second communication device 81, the server 80 stores the received learning model 551 as the learning model 821 in the second storage device 82, and 81 to other controllers 50 . When receiving the learning model 821 from the server 80 via the first communication device 53 , the other control device 50 stores the received learning model 821 as the learning model 551 in the first storage device 55 . As a result, the management system 100 can share the learning model among a plurality of blood purification systems 40, and can improve the efficiency of generating the learning model. It should be noted that the process of step S109 may be omitted, and the control device 50 may use the learning model 551 generated by its own device only in its own device.

FIG. 7 is a flow chart showing an example of the operation of the control process of the control device 50. As shown in FIG.

An example of the control processing operation of the control device 50 will be described below with reference to the flowchart shown in FIG. The operation flow described below is executed mainly by the first processing device 56 in cooperation with each element of the control device 50 based on a program stored in the first storage device 55 in advance. The control process is executed when the user instructs the start of blood purification using the input device 51 .

First, the parameter acquisition unit 564 acquires the blood information detected by the detection unit of the blood purification unit 14 from the blood purification unit 14 in the same manner as in the process of step S101 in FIG. Store (step S201).

Next, the parameter acquisition unit 564 inputs the acquired blood information to the learning model 551 stored in the first storage device 55, and acquires the control parameters output from the learning model 551 (step S202).

Next, the control unit 565 controls the magnetic force regulator 16 based on the control parameters acquired by the parameter acquisition unit 564, or controls the blood pump 5, the plasma feed pump 9, or the filtrate pump 11 via the driving device 15. is controlled (step S203).

Next, the first data acquisition unit 561 acquires blood information and stores it in the first storage device 55 in the same manner as the process of step S104 in FIG. 6 (step S204).

Next, the first generating unit 562 determines a reward for the learning model 551 (step S205) in the same manner as the process of step S105 in FIG. The first generator 562 sets a reward based on the blood information acquired in step S201 and changes in the blood information acquired in step S204.

Next, the first generator 562 updates the action-value function in the same manner as in step S106 of FIG. 6 (step S206). Based on the blood information acquired in step S201, the blood information acquired in step S204, the control parameter acquired in step S202, and the reward set in step S206, the first generation unit 562 generates an action value Update functions. Note that the processing of steps S204 to S206 is omitted, and the first generation unit 562 does not need to update the learning model 551 in the control processing.

Next, the control unit 565 determines whether or not the user has instructed to end the blood purification using the input device 51 (step S207). If the end of blood purification has not yet been instructed, control unit 565 returns the process to step S201 and repeats the processes of steps S201 to S207.

On the other hand, when the end of blood purification is instructed, control unit 565 stores the result of the control process in result table 553 (step S208).

Control unit 565 reads learning model 551 used in the current blood purification from first storage device 55 . In addition, the control unit 565 receives input of patient data by the user using the input device 51 . Also, the control unit 565 reads the product data 552 of the plasma separation device 1 and the factor separation device 7 from the first storage device 55 . Further, the control unit 565 reads out the blood information first stored in step S201 from the first storage device 55, and obtains it as blood data related to the blood flowing through each line of the blood purification unit . In addition, the control unit 565 reads each piece of blood information stored in step S204 from the first storage device 55 and obtains it as clearance data by the blood purification system 40 . Note that the control unit 565 may acquire only the last blood information stored in step S204 as the clearance data by the blood purification system 40. FIG.

In addition, the control unit 565 receives an input of the patient's causative agent removal amount from the user using the input device 51 . A measuring device for measuring the amount of causative substance removed is connected to the patient connected to the blood purification unit 14, and the control unit 565 receives the amount of the causative substance removed from the patient via the interface 50 from the measuring device. may be obtained.

In addition, the control unit 565 receives an input of the patient's antithrombotic properties from the user using the input device 51 . For example, after blood purification is completed, blood and plasma components are discharged from the blood purification unit 14, and a fixative such as glutaraldehyde for fixing thrombus formed in the hollow membrane body 1e is applied to the blood feed line 3. and is circulated in the blood return line 4 for a predetermined time. After that, the plasma separation device 1 is removed from the blood purification unit 14, and the presence or absence of thrombus formation is evaluated by visually observing the hollow membrane body 1e at the inlet portion of the removed plasma separation device 1. Further, the state of the thrombus inside the hollow membrane body 1e is observed for each predetermined site by a scanning electron microscope. For example, the hollow membrane body 1e is vertically divided into three regions, the filtration side, the center, and the outside (the side opposite to the filtration side), and horizontally divided into three regions, the inlet side, the center, and the outlet side. The following evaluation is performed for each site in the x3 area.

For example, about 15 hollow fiber membranes are randomly extracted from one site, and the cross-sectional area of the thrombus-forming part is measured for each hollow fiber membrane. An average value per book is calculated. Then, the ratio of the average value of the cross-sectional area of the thrombus-forming portion to the average cross-sectional area of the hollow fiber membrane is calculated as the thrombus formation rate for each site. Then, using the thrombus formation rate as an index, the antithrombotic property of the plasma separation device 1 is evaluated. For example, the thrombus formation rate is divided into five stages for each site and scored. A thrombus formation rate of less than 1% is scored as 0 points, a thrombus formation rate of 1% or more and less than 25% is scored as 1 point, and a thrombus formation rate of 25% or more and less than 50% is scored as 2 points. A score of 3 is given when the thrombus formation rate is 51% or more and less than 75%, and a score of 4 is given when the thrombus formation rate is 76% or more. Also, the average score may be calculated for each region in the vertical direction or for each region in the horizontal direction. As a result, the tendency of thrombus formation in the hollow membrane body 1e is grasped for each region. Also, the average value of scores at all sites may be used as the evaluation score for the antithrombotic property of the plasma separation device 1 as a whole. The smaller the score, the higher the antithrombotic property.

The control unit 565 associates the learning model 551, patient data, product data 552, blood data, clearance data, causative substance removal amount, and antithrombotic property with each other, assigns a new measurement ID, and stores it in the result table 553. .

Next, the first output control unit 563 outputs the result of blood purification by displaying it on the display device 52 (step S208), and ends the series of steps. Blood purification results are an example of information about the learning model. As a result, the user can confirm the effect of the learning model 551 , specify the learning model 551 exhibiting a high effect, and share the learning model 551 with other control devices 50 .

Note that the control device 50 may acquire control parameters using a learning model stored in the server 80 instead of using the learning model stored in its own device. In that case, the parameter acquisition unit 564 transmits the blood information to the server 80 via the first communication device 53 in step S202. The second processing device 83 of the server 80 receives the blood information from the control device 50 via the second communication device 81, inputs it to the learning model 821 stored in the second storage device 82, and outputs it from the learning model 821. get the control parameters. The second processing device 83 transmits the obtained control parameters to the control device 50 via the second communication device 81, and the parameter obtaining unit 564 receives the control parameters from the server 80 via the first communication device 53. Obtained by

By using learning model 821 stored in server 80 , controller 50 can appropriately control blood purification system 40 using the latest learning model 821 updated by server 80 . On the other hand, control device 50 can appropriately control blood purification system 40 even when the communication connection with server 80 is disconnected by using learning model 551 stored therein.

As described in detail above, the blood purification system 40 generates the learning model 551 based on the blood information about the blood flowing through the blood purification unit 14 and the control parameters of the blood purification unit 14, and uses the generated learning model 551. to control the blood purification unit 14 . As a result, blood purification system 40 can self-learn, clarify, and provide optimal operating conditions. Therefore, the blood purification system 40 can further stabilize the condition of the blood flowing through the blood purification unit 14, and can purify the blood more appropriately.

In particular, the blood information includes the degree of blood turbulence or blood vorticity. The inventors discovered that blood turbulence (fluid in a turbulent state with random, constantly fluctuating motion) and blood vorticity promote platelet production. When the degree of blood turbulence and the change in blood vorticity are large, fouling of the plasma separation device 1 tends to progress. Since the operation of the blood purification system 80 lasts from several hours to several days, and there is a limit to manual visual work and operation, it is difficult to manually obtain the optimum conditions for the degree of blood turbulence and blood vorticity. be. By using machine learning, the blood purification system 40 can obtain optimum conditions for the degree of blood turbulence and blood vorticity, especially optimum conditions that are difficult for humans to imagine. It is possible to further reduce the ring.

The blood information also includes the degree of heart murmur. If the degree of heart murmur is large, the efficiency of blood removal from the patient tends to be poor. As for patients using an extracorporeal blood treatment device such as the plasma separation device 1, there is a survey result that more than half of the patients diagnosed with ischemic heart disease answered that they had no symptoms. Myocardial infarction is particularly likely during the first year after using an extracorporeal blood treatment device. By using machine learning, the blood purification system 40 can obtain an optimum condition for the degree of heart murmur, especially an optimum condition that is difficult for humans to imagine. It is possible to reduce the possibility of causing Moreover, since the blood purification system 80 has the electrocardiogram measuring device 41, it is possible to make the patient aware of ischemic heart disease.

Also, the control parameter includes the strength of the magnetic field applied to the blood in a predetermined direction. It is difficult to properly control the strength of the magnetic field, since even small changes in the strength of the magnetic field have a large effect on blood flow. In particular, when the strength of the magnetic field changes, the blood flow does not change immediately, but changes after a while, so it is difficult to properly control the strength of the magnetic field. Since the strength of the magnetic field requires delicate adjustment, it is difficult to obtain the optimum conditions by manual visual work and operation. By using machine learning, the blood purification system 40 can acquire the optimum conditions for the strength of the magnetic field, especially the optimum conditions that are difficult for humans to imagine.

As a result of intensive studies and repeated experiments, the inventors have found that the optimum operating conditions can be obtained by providing the blood purification system 40 with a self-learning function. In particular, the blood purification system 40 can enhance the antithrombotic properties of the plasma separation device 1 used for apheresis by double plasmapheresis, extend the lifetime, and reduce fouling. It becomes possible to elucidate and provide operating conditions. In addition, the blood purification system 40 can clarify and provide operating conditions that can extend the lifetime of the factor separation device 7 and reduce fouling. In addition, the blood purification system 40 can clarify and provide operating conditions that can reduce disease-causing substances.

In general, it takes about three days to obtain appropriate operating conditions using a closed-system circulation test device. , it is possible to obtain appropriate operating conditions. In addition, the control device 50 can evaluate the performance of the plasma separation device 1 and the factor separation device 7 while performing filtrate in the blood purification unit 14, and efficiently evaluate the performance of the plasma separation device 1 and the factor separation device 7. can be done. In addition, the control device 50 can set the amount of plasma components, the amount of filtration, etc. in the blood purification unit 14, and collectively manage the operations of the plurality of pumps appropriately.

In addition, the control device 50 that controls the blood purification unit 14 generates the learning model 551, so that the blood purification system 40 can generate the learning model 551 with a simple configuration.

FIG. 8 is a flowchart showing an example of the learning process operation of the server 80 according to another embodiment. In this embodiment, the second data acquisition unit 831, the second generation unit 832, and the second output control unit 833 of the server 80 are examples of the data acquisition unit, generation unit, and output control unit, respectively.

An example of the operation of the learning process of the server 80 will be described below with reference to the flowchart shown in FIG. The operation flow described below is executed mainly by the second processing device 83 in cooperation with each element of the server 80 based on a program stored in advance in the second storage device 82 .

First, the second data acquisition unit 831 acquires a plurality of combinations of blood information and control parameters by receiving them from the control device 50, that is, from the blood purification system 40 via the second communication device 81 (step S301 ). The first data acquisition unit 561 of the control device 50 acquires a plurality of combinations of blood information and control parameters by repeatedly executing the processing of steps S101 to S104 in FIG. Send. The second data acquisition unit 831 acquires a plurality of combinations of blood information and control parameters by receiving them from the control device 50 via the second communication device 81 . Note that the second data acquisition unit 831 may acquire combinations of blood information and control parameters from a plurality of control devices 50 , that is, from a plurality of blood purification systems 40 . Thereby, the second data acquisition unit 831 can efficiently acquire the learning data of the learning model 821 .

Next, the second generation unit 832 determines a reward corresponding to each combination of blood information and control parameters in the same manner as in step S105 of FIG. 6 (step S302).

Next, the second generating unit 832, in the same manner as in the process of step S106 in FIG. 6, for each combination of blood information and control parameters, based on the combination of blood information and control parameters and the set reward, acts Update the value function (step S303).

Next, the second generation unit 832 has a combination (action value table) of finally updated values (action values) of each Q(s, a) in the same manner as the process of step S108 in FIG. A learning model 821 is generated and stored in the second storage device 82 (step S304).

Next, the second output control unit 833 outputs the learning model 821 generated by the second generation unit 832 by transmitting it to each control device 50 via the second communication device 81 (step S305). End the step. A learning model 821 is an example of information about a learning model. On the other hand, when each control device 50 receives the learning model 821 from the server 80 via the first communication device 53 , it stores the received learning model 821 as the learning model 551 in the first storage device 55 .

As described in detail above, blood purification system 40 can more appropriately purify blood even when server 80 generates learning model 821 .

In particular, the management system 100 can centrally manage the learning model used by each blood purification system 40 on the server 80, and suppresses variations in accuracy of the learning model used by each blood purification system 40. becomes possible.

Note that the learning process shown in FIG. 8 may be executed by each control device 50 instead of the server 80 . That is, each control device 50 may acquire a plurality of combinations of blood information and control parameters from other control devices 50 and generate the learning model 551 using the acquired combinations.

FIG. 9 is a schematic diagram showing a blood purification unit 14-2 according to still another embodiment.

The blood purification unit 14-2 has each part of the blood purification unit 14 and is used instead of the blood purification unit 14. FIG. However, blood purification unit 14-2 does not have piping systems 33, 33', (three-way) valves 34, 34', piping systems 35, 35', but instead blood bag 2, resistance imparting members 6, 37, It also has open/close ports 24, 24', 38, 39, and the like. The blood purification unit 14-2 utilizes the plasma separation device 1 in a non-clinical environment that is not used on a patient under substantially the same conditions as in actual use with respect to blood flow, blood pressure, plasma volume, or filtrate volume. used for The blood purification unit 14-2 performs comparative evaluation of the ability of the plasma separation device 1 to inhibit complement activation, antithrombotic property (blood coagulation system activation inhibition ability), lifetime of the plasma separation device 1, and the like. A fouling assessment of the separation device 1 can be performed. In addition, the blood purification unit 14-2 suppresses loss of albumin in the factor separation device 7, and comparatively evaluates the performance of separating disease-causing factors (components) from plasma components, the lifetime of the factor separation device 7, etc. And a fouling assessment of the factor separation device 7 can be performed.

The blood purification unit 14-2 has a closed circuit in which the test liquid circulates through the plasma separation device 1 without contact with air. Human blood (whole blood) is used as the test liquid, but the test liquid is not limited to this, and other liquids such as animal blood or artificial blood containing blood components similar to human blood may be used.

A blood supply line 3 and a blood return line 4 are connected via a blood bag 2 .

A resistance imparting member 6 is provided on the blood return line 4 . The resistance applying member 6 may be provided in the blood supply line 3 instead of or in addition to the blood return line 4 . A resistance applying member 37 is provided in the filtrate line 10 . The resistance applying member 6 and the resistance applying member 37 apply resistance to the blood supply line 3, the blood return line 4, or the filtrate line 10 at each arrangement position to adjust the flow rate or pressure of the test liquid (blood) to the actual use environment. used for matching. In particular, the resistance imparting member 6 functions as a vein model that simulates the peripheral resistance of the human body to impart squeezing resistance to the blood return line 4 and simulates the veins of the human body to adjust the flow of the test liquid. As the resistance applying member 6 and the resistance applying member 37, a clamp capable of changing the magnitude of force (resistance) applied to each line by driving a motor, for example, is used. Various other devices such as valves can be used as the resistance applying member 6 and the resistance applying member 37 . The resistance applying member 6 and the resistance applying member 37 are examples of a liquid control mechanism. The drive device 15 further has a motor for driving the resistance applying member 6 and the resistance applying member 37 .

The open/close ports 24, 24' are stopcocks or the like that can switch between an open state that enables the supply or discharge of the test liquid to the blood feeding line 3 and a closed state that disables the supply or discharge of the test liquid. consists of The open/close ports 24, 24' are set to an open state when the test liquid is sampled during the test or when the test liquid is replaced after the test, and are set to a closed state in other cases. When the open/close ports 24, 24' are closed, the entire circuit of the blood purification unit 14-2 is maintained in a non-air contact state. Similarly, the open/close port 38 is configured with a stopcock or the like that can switch between an open state and a closed state with respect to the filtrate line 10 . The open/close port 39 is configured with a stopcock or the like that can switch between an open state and a closed state with respect to the plasma line 8 .

The blood purification unit 14-2 heats the plasma separation device 1, the blood bag 2, the factor separation device 7 and the whole blood purification unit 14-2 in order to keep the temperatures of various liquids in the blood purification unit 14-2 constant. It is preferable to provide a constant temperature means for regulating the temperature. The constant temperature means has, for example, a water tank in which water is stored and a heater for maintaining the water temperature in the water tank at a predetermined temperature. By placing the blood bag 2 and part of the blood supply line 3 in the water tank, the test liquid flowing through the blood supply line 3 and the blood return line 4 is heated to a temperature similar to that of a human body (approximately 36 to 37°C). A constant temperature can be maintained. Similarly, by placing the factor separation device 7 and part of the plasma line 8 and filtrate line 10 in a water tank, the plasma components flowing through the plasma line 8 and the filtrate line 10 are heated to human body temperature (approximately 36 to 37°C). ) can be maintained at a constant temperature. The constant temperature means may have a heater that maintains the sealed space containing the entire blood purification unit 14-2 at a predetermined temperature.

Each part of the blood purification unit 14-2 has the same differential pressure in order to simulate the effect of the differential pressure due to the height difference of each device provided in the liquid circuit in the blood purification unit 14 used in actual hemofiltration. is arranged so as to have a height difference that generates In addition, each part of the blood purification unit 14-2 is arranged in consideration of the effect of gravity due to the height difference. For example, the blood pump 5 is placed at the highest point. The blood pump 5 is arranged at a height of about 600 mm to 700 mm from the installation surface of the blood purification unit 14-2, which is the lowest point. The plasma separation device 1 and the factor separation device 7 are held with the blood inlet port 1c and the plasma inlet port 7d facing upwards, and the blood outlet port 1d and the plasma outlet port 7c facing downwards. The upper ends of the plasma separation device 1 and the factor separation device 7 are arranged at a height of about 400 mm to 500 mm from the installation surface of the device.

The controller 50 controls the blood purification unit 14-2 in the same manner as the blood purification unit 14 is controlled. However, when the blood purification unit 14-2 is used, the control parameters of the blood purification unit 14-2 include the magnitude of resistance (amount of resistance) applied by the resistance applying member 6 and the resistance applying member 37. FIG. Also, when the blood purification unit 14-2 is used, the blood information does not include the degree of heart murmur, patient's blood pressure, patient's degree of hemolysis, pulse, blood pressure, and circulating blood flow.

Hereinafter, the test method for evaluating the antithrombotic property and lifetime evaluation of the plasma separation device 1 and the test method for evaluating the lifetime of the factor separation device 7 using the blood purification unit 14-2 will be described.

After the plasma separation device 1 and the factor separation device 7 to be tested are installed (set) in the blood purification unit 14-2, the blood feeding line 3 and blood return line 4 are filled with blood as a test liquid. be. The blood pump 5 is driven, and the plasma liquid-feeding pump 9 and the filtrate pump 11 are appropriately driven in sequence according to the state of arrival of plasma components and filtrate to the plasma liquid-feeding pump 9 and the filtrate pump 11 . The blood that has been given a pulsating flow by the blood pump 5 under the same conditions as the actual use environment passes from the blood bag 2 through the blood pump 5, flows into the plasma separation device 1 from the blood inlet side port 1c, and flows into the hollow membrane body. Pass 1e. The blood that has passed through the hollow membrane body 1 e flows from the blood outlet side port 1 d to the blood return line 4 , passes through the resistance applying member 6 , and is returned to the blood bag 2 .

The blood pump 5, plasma feed pump 9, and filtrate pump 11 are continuously driven for a period of time corresponding to the actual use environment (for example, about 3 hours to 3 days). During this time, the first pressure gauges 21 to 23, the second pressure gauge 25, the third pressure gauge 26, the first flowmeters 28, 29, the second flowmeter 30 and the third flowmeter 31 are periodically monitored to separate plasma. Data are acquired on changes over time such as the inlet pressure of the device 1, the outlet pressure, or the differential pressure therebetween. Similarly, for the factor separation device 7, data regarding temporal changes in inlet pressure, filtrate pressure, etc. of the factor separation device 7 are obtained. The controller 50 can determine the lifetimes of the plasma separation device 1 and the factor separation device 7 from the acquired data. When the inlet pressures of the plasma separation device 1 and the factor separation device 7 rise from the initial value (eg, 70 mmHg) to a predetermined value (eg, 150 mmHg), the blood pump 5, The plasma feed pump 9 and the filtrate pump 11 are stopped, the test is terminated, and the elapsed time (operating time) from the start of the test is also acquired as data.

During the test, the blood flowing through the blood supply line 3, the plasma line 8, and the filtrate line 10, the plasma component, and the filtrate are collected as samples at predetermined time intervals, and each time, various components are measured, and the various components are measured over time. Data on changes are also obtained. After the test is completed, the blood and plasma components are discharged from the blood purification unit 14-2, and a fixing solution such as glutaraldehyde for fixing thrombus formed in the hollow membrane body 1e is applied to the blood feed line 3 and the return line. It is circulated within the blood line 4 for a predetermined time. At this time, the driving conditions of the blood pump 5, the plasma feeding pump 9, and the filtrate pump 11 are set to the same conditions as in the blood test. Then, the plasma separation device 1 is taken out from the blood purification unit 14-2, and the blood plasma separation device 1 taken out is evaluated for its antithrombogenicity. As described above, in this embodiment, the evaluation test of the plasma separation device 1 is performed without contact with the atmosphere while the flow rate, pressure, and components of the test liquid are maintained at desired conditions. Under the hood, almost identical tests are performed under the actual use environment via patients.

In addition, embodiment is not limited to what was described above. For example, the control device 50 and the blood purification unit 14 may be configured as an integrated device instead of separate devices. In that case, for example, each device of the blood purification unit 14 is directly connected to the CPU bus or the like of the control device 50, and transmits/receives information to/from the first processing device 56 of the control device 50 via the CPU bus or the like.

Also, the learning model may be learned by a method other than reinforcement learning. For example, the learning model is pretrained by supervised learning such as deep learning. In that case, a plurality of pieces of blood information and a control parameter capable of stabilizing the blood state indicated by each piece of blood information are used as teacher data. The learning model is trained to output control parameters capable of stabilizing the state of blood indicated by each piece of blood information when each piece of blood information is input. Note that the learning model may be learned using unsupervised learning, semi-supervised learning, transduction, multi-task learning, and the like.

When the learning model is learned using unsupervised learning, the learning device (control device 50 or server 80) that generates the learning model controls a specific control parameter and the liquid control mechanism based on the specific control parameter. A combination of the blood information detected before and after the detection is acquired. The learning device acquires combinations of control parameters and blood information in the same manner as in the processing of steps S101 to S104 in FIG. 6 or the processing of step S301 in FIG. The learning device uses a clustering technique such as k-means to classify (group) the blood information detected after controlling the liquid control mechanism based on each control parameter. The operator specifies the group with the most stable blood condition among the classified groups of blood information. The learning device performs supervised learning using, as teacher data, combinations of blood information detected before controlling the liquid control mechanism and control parameters corresponding to blood information belonging to the group with the most stable blood condition. Train the learning model by As a result, the learning device can efficiently extract the control parameters that can stabilize the blood state indicated by the blood information, and can reduce the time required to generate the learning model.

Also, if the learning model is trained using unsupervised learning, the learning device may utilize dimensionality reduction techniques such as principal component analysis to reduce the dimensionality of blood information and/or control parameters. The learning device learns a learning model by supervised learning using teacher data whose dimensionality has been reduced. As a result, the learning device can efficiently extract the control parameters that can stabilize the blood state indicated by the blood information, and can reduce the time required to generate the learning model.

If the learning model is trained using semi-supervised learning, the learning device generates the learning model using transduction, for example. In that case, the learning device generates teacher data by a bootstrap method or the like. First, the learning device combines a plurality of pieces of blood information with control parameters capable of stabilizing the state of blood indicated by each piece of blood information, and combines the plurality of pieces of blood information and the state of blood indicated by each piece of blood information. A combination of control parameters that cannot be stabilized is obtained from the operator. The learning device uses a classification technique such as a support vector machine to classify combinations including control parameters that can stabilize the blood state and combinations including control parameters that cannot stabilize the blood state. Generate a classifier that The learning device further acquires from the operator a combination of a plurality of pieces of blood information and control parameters for which it is unclear whether or not the blood state indicated by each piece of blood information can be stabilized. The learning device uses the generated classifier to extract combinations including control parameters capable of stabilizing the blood state from the acquired combinations. The learning device learns a learning model by supervised learning using combinations including control parameters capable of stabilizing the blood state as teacher data. As a result, the learning device can efficiently extract the control parameters that can stabilize the blood state indicated by the blood information, and can reduce the time required to generate the learning model.

Also, if the learning model is trained using semi-supervised learning, the learning device may generate training data using a graph-based algorithm such as the k-nearest neighbor method. In that case, the learning device combines a plurality of pieces of blood information with control parameters capable of stabilizing the state of blood indicated by each piece of blood information, A combination with a control parameter for which it is unknown whether or not it is possible to stabilize is obtained from the operator. The learning device, for example, by the k nearest neighbor method, selects a control parameter capable of stabilizing the blood state from combinations including control parameters for which it is unknown whether or not the blood state can be stabilized. Extract combinations that are close to the included combinations. The learning device learns a learning model by supervised learning using combinations including control parameters capable of stabilizing the blood state and extracted combinations as teacher data. As a result, the learning device can efficiently extract the control parameters that can stabilize the blood state indicated by the blood information, and can reduce the time required to generate the learning model.

When the learning model is learned using multitask learning, additional parameters are used as teacher data in addition to control parameters that can stabilize the blood state indicated by the plurality of blood information and each blood information. The additional parameter is, for example, blood information detected after the liquid control mechanism is controlled based on the control parameter, or blood information change information before and after the liquid control mechanism is controlled based on the control parameter. The learning device learns the learning model so that when each piece of blood information is input, an additional parameter is output in addition to the control parameter capable of stabilizing the blood state indicated by each piece of blood information. This enables the learning device to calculate control parameters that can bring the blood state closer to the appropriate state.

With these, the management system 100 achieves significant stabilization of blood pressure and blood information other than blood pressure (e.g., degree of blood turbulence, blood vorticity, or degree of heart murmur) to reduce the patient's blood pressure. , it is possible to improve hypertension and reduce the occurrence of heart murmurs.

In addition, the management system 100 realizes remarkable stabilization of blood information, thereby improving the therapeutic effect in a short period of time, reducing the financial burden on the patient, and reducing the mental and physical burden on the patient by reducing the number of hospital visits. It is possible to reduce the burden and reduce the labor burden on medical staff.

In addition, the management system 100 can accumulate information and improve the learning speed by connecting the host computer (server 80) with the blood purification unit 14 and/or the learning device. It is possible to deal with it.

1 Plasma separation device 3 Blood supply line 4 Blood return line 5 Blood pump 6 Resistance imparting member 7 Factor separation device 8 Plasma line 9 Plasma supply pump 10 Filtrate line 11 Filtrate pump 14, 14 -2 blood purification unit, 21 to 23 first pressure gauge, 25 second pressure gauge, 26 third pressure gauge, 28, 29 first flow meter, 30 second flow meter, 31 third flow meter, 37 resistance imparting member , 40 blood purification system, 50 control device, 53 first communication device, 55 first storage device, 551 learning model, 561 first data acquisition unit, 562 first generation unit, 563 first output control unit, 564 parameter acquisition unit , 565 control unit, 80 server, 81 second communication device, 821 learning model, 831 second data acquisition unit, 832 second generation unit, 833 second output control unit

Claims (14)

a line through which a liquid containing blood or filtrate flows;
a plasma separation device for separating plasma components from blood flowing through the line;
a factor separation device for separating disease-causing factor components from the plasma components;
a detection unit that detects blood information about the blood flowing through the line;
a liquid control mechanism for controlling liquid flow in the line based on a control parameter;
The blood information detected by the detection unit is input to a learning model trained to output a predetermined control parameter when predetermined blood information is input, and the control parameter output from the learning model is acquired. a parameter acquisition unit for
a control unit that controls the liquid control mechanism based on the control parameters acquired by the parameter acquisition unit;
The blood information includes degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, and the plasma separation device. fouling of the factor separation device, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of a patient connected to the line;
The learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to stabilize the blood state indicated by the blood information when the blood information is input. is learned to output control parameters that allow
A blood purification system characterized by: The liquid control mechanism includes a magnetic force adjuster that applies a magnetic field to blood in a predetermined direction, a pump that controls the flow of liquid in the line, or a resistance applying member that applies resistance to the line,
2. The control parameter according to claim 1, wherein the control parameter includes at least one of the strength of the magnetic field applied by the magnetic force adjuster, the driving amount of the pump, and the magnitude of resistance applied by the resistance applying member. blood purification system. A storage unit that stores the learning model in association with product data of the plasma separation device, data on blood flowing in the line, clearance data by the blood purification system, amount of causative substance removed, or antithrombogenicity, The blood purification system according to claim 1 or 2. 4. The blood purification system according to any one of claims 1 to 3, further comprising a communication unit that receives said learning model from a learning device. further comprising a generator that generates the learning model based on blood information detected by the detector before and after controlling the liquid control mechanism based on a specific control parameter and the specific control parameter; 4. The blood purification system according to any one of items 1 to 3. The blood purification system according to any one of claims 1 to 5, wherein said blood purification system is an extracorporeal circulation blood purification system. 7. The blood purification system according to claim 6, wherein said blood purification system performs apheresis by double plasmapheresis. a line through which a liquid comprising blood or filtrate flows; a plasma separation device for separating plasma components from the blood flowing through said line; a factor separation device for separating disease-causing factors from said plasma components; A control method for a blood purification system having a detection unit that detects blood information about blood and a liquid control mechanism that controls the flow of liquid in the line based on a control parameter, comprising:
The blood information detected by the detection unit is input to a learning model trained to output a predetermined control parameter when predetermined blood information is input, and the control parameter output from the learning model is acquired. death,
controlling the liquid control mechanism based on the obtained control parameter;
The blood information includes degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, and the plasma separation device. fouling of the factor separation device, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of a patient connected to the line;
The learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to stabilize the blood state indicated by the blood information when the blood information is input. is learned to output control parameters that allow
A control method characterized by: a line through which a liquid comprising blood or filtrate flows; a plasma separation device for separating plasma components from the blood flowing through said line; a factor separation device for separating disease-causing factors from said plasma components; A control program for a computer included in a blood purification system having a detection unit that detects blood information about blood and a liquid control mechanism that controls the flow of liquid in the line based on control parameters,
The blood information detected by the detection unit is input to a learning model trained to output a predetermined control parameter when predetermined blood information is input, and the control parameter output from the learning model is acquired. death,
causing the computer to control the liquid control mechanism based on the acquired control parameters;
The blood information includes degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, and the plasma separation device. fouling of the factor separation device, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of a patient connected to the line;
The learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to stabilize the blood state indicated by the blood information when the blood information is input. is learned to output control parameters that allow
A control program characterized by: a line through which a liquid comprising blood or filtrate flows; a plasma separation device for separating plasma components from the blood flowing through said line; a factor separation device for separating disease-causing factors from said plasma components; A blood purification system having a detection unit that detects blood information about blood, and a liquid control mechanism that controls the flow of liquid in the line based on control parameters, wherein a plurality of combinations of the blood information and the control parameters are detected. a data acquisition unit to acquire;
a generating unit that generates a learning model trained to output a predetermined control parameter when predetermined blood information is input, using the combinations obtained by the data obtaining unit;
an output control unit that outputs information about the learning model,
The blood information includes degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, and the plasma separation device. fouling of the factor separation device, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of a patient connected to the line;
The learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to stabilize the blood state indicated by the blood information when the blood information is input. is learned to output control parameters that allow
A learning device characterized by: further comprising a communication unit for communicating with a plurality of said blood purification systems;
11. The learning device according to claim 10, wherein said data acquisition unit acquires said combination by receiving said combination from said plurality of blood purification systems via said communication unit. The data acquisition unit controls the liquid control mechanism based on specific control parameters, and acquires blood information detected by the detection unit before and after controlling the liquid control mechanism based on the specific control parameters. 11. The learning device according to claim 10, wherein the combination is obtained by: the computer
a line through which a liquid comprising blood or filtrate flows; a plasma separation device for separating plasma components from the blood flowing through said line; a factor separation device for separating disease-causing factors from said plasma components; A blood purification system having a detection unit that detects blood information about blood, and a liquid control mechanism that controls the flow of liquid in the line based on control parameters, wherein a plurality of combinations of the blood information and the control parameters are detected. Acquired,
generating a learning model trained to output a predetermined control parameter when predetermined blood information is input using the acquired combination;
outputting information about the learning model;
The blood information includes degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, and the plasma separation device. fouling of the factor separation device, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of a patient connected to the line;
The learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to stabilize the blood state indicated by the blood information when the blood information is input. is learned to output control parameters that allow
A learning method characterized by: A control program for a computer,
a line through which a liquid comprising blood or filtrate flows; a plasma separation device for separating plasma components from the blood flowing through said line; a factor separation device for separating disease-causing factors from said plasma components; A blood purification system having a detection unit that detects blood information about blood, and a liquid control mechanism that controls the flow of liquid in the line based on control parameters, wherein a plurality of combinations of the blood information and the control parameters are detected. Acquired,
generating a learning model trained to output a predetermined control parameter when predetermined blood information is input using the acquired combination;
causing the computer to output information about the learning model;
The blood information includes degree of blood turbulence, blood vorticity, degree of heart murmur, blood pressure, blood pressure pressure loss, plasma pressure, filtrate pressure, blood flow rate, plasma flow rate, filtrate flow rate, and the plasma separation device. fouling of the factor separation device, blood hematocrit, filtrate albumin concentration, degree of hemolysis, and blood pressure of a patient connected to the line;
The learning model uses supervised learning, unsupervised learning, semi-supervised learning, transduction, or multitask learning to stabilize the blood state indicated by the blood information when the blood information is input. is learned to output control parameters that allow
A control program characterized by:

Images