Dialysis system with motor drive pressure estimation of fluid within patient line

Abstract

A dialysis system with motor driven pressure estimation of fluid within a patient line is disclosed. A peritoneal dialysis ("PD") system includes a fluid pump and a patient line fluidly coupling the fluid pump to an indwelling catheter leading into a peritoneal cavity of a patient. The PD system also includes a motor driver that controls a motor of the fluid pump and transmits an output signal indicative of a load on the motor. The PD system also includes a machine learning algorithm that correlates data related to the output signal from the motor drive with a known fluid pressure within the patient line. A processor of the PD system transmits an input signal to activate the motor driver, receives an output signal from the motor driver, and estimates fluid pressure within the patient line by applying data from the received output signal to a machine learning algorithm.

Description

Technical Field

[0001] The present disclosure generally relates to medical fluid treatment, and particularly to dialysis fluid treatment that requires fluid heating. Background Art

[0002] For various reasons, a person's renal system may fail. Renal failure can produce a variety of physiological disorders. The ability to balance water and minerals or excrete the daily metabolic load is no longer possible. Toxic end products of metabolism (such as urea, creatinine, uric acid, and other substances) may accumulate in the patient's blood and tissues.

[0003] Dialysis is used to treat decreased kidney function, especially renal failure. Dialysis removes waste products, toxins, and excess water from the body that would normally be removed by a properly functioning kidney. Dialysis treatment for replacing kidney function is crucial for many people because this treatment is life-saving.

[0004] One type of renal failure therapy is hemodialysis ("HD"), which typically uses diffusion to remove waste from the patient's blood. A diffusion gradient occurs across a semi-permeable dialyzer between the blood and an electrolyte solution called dialysate or dialysis fluid to cause diffusion.

[0005] Hemofiltration ("HF") is an alternative renal replacement therapy that relies on the convective transport of toxins from the patient's blood. HF is achieved by adding replacement or substitution fluid to the extracorporeal circuit during treatment. During the HF treatment process, the replacement fluid and the fluid that accumulates between the patient and the treatment are ultrafiltered, thus providing a convective transport mechanism that is particularly beneficial for removing medium and large molecules.

[0006] Hemodiafiltration ("HDF") is a treatment modality that combines convective clearance and diffusive clearance. HDF uses dialysis fluid flowing through a dialyzer (similar to standard hemodialysis) to provide diffusive clearance. In addition, a replacement solution is directly provided to the extracorporeal circuit, thus providing convective clearance.

[0007] Most HD, HF, and HDF treatments are performed in a center. There is a trend towards home hemodialysis ("HHD") nowadays, in part because HHD can be performed daily, which provides therapeutic benefits compared to in-center hemodialysis treatments that are typically performed every two or three weeks. Studies have shown that more frequent treatments remove more toxins and wastes and reduce interdialytic fluid overload compared to patients receiving less frequent but potentially longer treatments. Patients receiving more frequent treatments do not experience as many down cycles (fluctuations in fluid and toxins) as in-center patients who have accumulated toxins over two or three days prior to treatment. In some areas, the nearest dialysis center may be many miles from the patient's home, which results in home treatment times taking up much of the day. Treatments at a center near the patient's home may also take up much of the patient's day. HHD can be performed at night or during the day while the patient is relaxed, working, or otherwise being productive.

[0008] Another type of renal failure therapy is peritoneal dialysis ("PD"), which injects a dialysis solution (also known as dialysis fluid) into the peritoneal cavity of the patient via a catheter. The dialysis fluid comes into contact with the peritoneum in the patient's peritoneal cavity. Wastes, toxins, and excess water pass from the patient's bloodstream through the capillaries in the peritoneum and enter the dialysis fluid due to diffusion and osmosis, i.e., an osmotic gradient appears across the peritoneum. Osmotic agents in the PD dialysis fluid provide the osmotic gradient. The used or spent dialysis fluid is drained from the patient's body, thereby removing wastes, toxins, and excess water from the patient's body. This cycle is repeated, for example, multiple times.

[0009] There are various types of peritoneal dialysis therapies, including continuous ambulatory peritoneal dialysis ("CAPD"), automated peritoneal dialysis ("APD"), tidal flow dialysis, and continuous flow peritoneal dialysis ("CFPD"). CAPD is a manual dialysis treatment. Here, the patient manually connects the implanted catheter to the drain tube to allow the used or spent dialysis fluid to drain from the peritoneal cavity. Then, the patient switches the fluid connection so that the patient catheter is connected to a fresh dialysis fluid bag to inject fresh dialysis fluid through the catheter and into the patient's body. The patient disconnects the catheter from the fresh dialysis fluid bag and allows the dialysis fluid to remain in the peritoneal cavity, where the transfer of wastes, toxins, and excess water occurs. After the dwell period, the patient repeats the manual dialysis procedure, for example, four times a day. Manual peritoneal dialysis requires a significant amount of time and effort from the patient, leaving much room for improvement.

[0010] Automated peritoneal dialysis ("APD") is similar to CAPD in that the dialysis treatment includes drain, fill, and dwell cycles. However, an automated PD machine typically performs these cycles automatically while the patient sleeps. The PD machine enables the patient to not have to perform the treatment cycles manually or deliver supplies during the day. The PD machine is fluidly connected to an implanted catheter, a source or bag of fresh dialysis fluid, and a fluid drain tube. The PD machine pumps fresh dialysis fluid from the dialysis fluid source through the catheter and into the patient's peritoneal cavity. The PD machine also allows the dialysis fluid to dwell in the peritoneal cavity and permits the transfer of waste, toxins, and excess water. The source can include multiple liters of dialysis fluid, the multiple liters of dialysis fluid including a number of solution bags.

[0011] The PD machine pumps used or spent dialysis fluid from the patient's peritoneal cavity through the catheter to the drain tube. As in the manual process, several drain, fill, and dwell cycles occur during dialysis. A "last fill" can occur at the end of an APD treatment. The last fill fluid can be retained in the patient's peritoneal cavity until the next treatment begins or can be manually drained at some point during the day.

[0012] In any of the above modalities using an automated machine, component cost, calibration, and maintenance are key design considerations. If components can be eliminated, not only is their cost eliminated, but also the potential calibration of the components and their maintenance and / or replacement. Component removal reduces the weight of the machine and also frees up space within the machine housing or permits the housing to be made smaller.

[0013] For each of the reasons above, it is desirable to provide an APD machine that reduces component cost, calibration, and / or maintenance. SUMMARY OF THE INVENTION

[0014] The present disclosure describes an automated peritoneal dialysis (“PD”) system that provides one or more PD treatment improvements. The system includes a PD machine or cycler. The PD machine is capable of delivering fresh, heated PD fluid to a patient at, for example, 14 kPa (2.0 psig) or higher. The PD machine is capable of removing used PD fluid or effluent from the patient between, for example, -5 kPa (-0.73 psig) and -15 kPa (-2.2 psig) (e.g., -9 kPa (-1.3 psig) or higher). Fresh PD fluid may be delivered to the patient via a dual lumen patient line and is first heated to body fluid temperature, such as 37 °C. The heated PD fluid is then pumped through the fresh PD fluid lumen of the dual lumen patient line to a disposable filter kit that is connected to a patient transfer kit, which in turn is connected to an indwelling catheter leading into the patient's peritoneal cavity. The disposable filter kit is in fluid communication with the fresh and used PD fluid lumens of the dual lumen patient line. The disposable filter kit is provided in one embodiment as a last chance filter for the PD machine and may be heat sterilized between treatments.

[0015] The system may include one or more PD fluid containers or bags that supply fresh PD fluid to the PD machine or cycler. The PD machine or cycler may include internal lines having two-way or three-way valves and at least one PD fluid pump for pumping fresh PD fluid from the one or more PD fluid containers or bags to the patient and for removing used PD fluid from the patient to a house drain or drain container. One or more flexible PD fluid lines lead from the internal lines of the PD machine or cycler to the one or more PD fluid containers or bags. The flexible dual lumen patient line described above leads from the PD machine or cycler to the patient. A flexible drain line leads from the internal lines of the PD machine or cycler to a house drain or drain container. In one embodiment, the system disinfects all internal lines, PD fluid lines, and dual lumen patient lines after a treatment for reuse in the next treatment. Disinfection may involve heat disinfection using the remaining fresh PD fluid.

[0016] The PD machine or cycler also includes different types of sensors that output to the control unit of the machine. Different types of sensors include, for example, temperature sensors, pressure sensors, leak detection sensors, and possibly flow sensors. The pressure sensor detects the PD fluid pressure and is used to control the PD fluid pressure (negative pressure and positive pressure) caused by the PD fluid pump. When a filter kit and a dual - lumen patient line are provided, the filter membrane of the filter kit causes a pressure drop in the fresh PD fluid pressure. Thus, the pressure output by the pump to the filter membrane is greater than the pressure experienced by the patient downstream of the filter membrane due to this pressure drop. Therefore, the PD fluid pressure downstream of the filter membrane is an important pressure to be monitored for use as feedback to control the PD fluid pump so that the PD fluid pressure experienced by the patient is at or below the patient PD fluid pressure limit.

[0017] One or more pressure sensors are positioned to sense the PD fluid pressure in the used PD fluid lumen of the dual - lumen patient line, which is an important pressure downstream of the filter membrane. Thus, a pressure sensor positioned to sense the PD fluid pressure in the fresh PD fluid lumen is not as critical but is used to sense, for example, kinking or blockage of the fresh PD fluid lumen. Thus, it is contemplated to eliminate the pressure sensor positioned to sense the PD fluid pressure in the fresh PD fluid lumen and use the output provided by the driver of the motor used to drive the PD fluid pump to estimate the PD fluid pressure.

[0018] In one embodiment, the PD fluid pump is a piston pump that includes a housing that holds a cylinder within which a piston is actuated via a motor (the motor is controlled by a motor driver, which is considered part of the overall control unit), where the motor drives a motion coupler that is coupled to the piston. The motion coupler converts the rotational motion of the motor into rotational and translational movement of the piston. The motion coupler moves the piston into and out of the cylinder to produce positive pumping pressure and negative pumping pressure, respectively. The motion coupler also rotates the piston within the cylinder to move the PD fluid from the inlet port to the outlet port.

[0019] In one embodiment, the motor is a stepper motor, and the motor driver for the stepper motor provides an output that indicates the load currently seen or experienced by the motor. To estimate the motor load, the motor driver measures the electrical energy flowing into the motor and the electrical energy flowing out of the motor. The difference between the incoming energy and the outgoing energy provides an indication of the mechanical load seen or experienced by the motor. The motor driver measures the portion of the energy fed to the motor that returns to the power supply that powers the stepper motor. This excess energy is measurable and indicates the mechanical load applied to the motor.

[0020] From the perspective of the motor, the load estimation output from the motor driver represents the load angle of the stepper motor, which depends on the external torque applied to the axis of the motor shaft. The stepper motor includes a stationary stator and a rotor that rotates within the stator. When power is supplied to the motor, a magnetic field is applied, where the magnetic field rotationally pulls and pushes the rotor within the stator, resulting in a phase shift between the magnetic field direction of the rotor and the magnetic field direction of the rotating field of the stator. The phase shift is the load angle, i.e., the angle between the magnetic field direction of the rotor and the magnetic field direction of the rotating magnetic field of the stator. From a mathematical perspective, the load estimation output from the motor driver can be a function of the back electromotive force ("EMF") constant inherent in the stepper motor, the coil inductance of the stepper motor, the coil resistance of the stepper motor, the stepping speed (e.g., full steps per second), the load angle, the phase current applied to the motor, and the voltage supplied to the stepper motor.

[0021] The control unit of the PD system of the present disclosure includes software that converts the load estimation output from the motor driver into an accurate pressure value of the positive PD fluid pressure output by the PD fluid pump. This conversion software is developed using a machine learning algorithm or model (such as a Gaussian process regression model or a decision tree model). The machine learning algorithm or model is configured to correlate data that correlates the known PD fluid pressure within the patient line with the load estimation output from the motor driver. This data can be obtained by sampling the load estimation output at discrete intervals. For example, in some embodiments, thirty to fifty data points can be sampled and processed together to obtain the characteristics of the load estimation output from the motor driver, which are used as inputs to the machine learning model.

[0022] The machine learning algorithm includes discrete inputs of the characteristics of the load estimation output from the motor driver. The characteristics can include the minimum value of the sampled data, the maximum value of the sampled data, the average value of the sampled data, the median value of the sampled data, and / or the difference between the maximum and minimum values of the sampled data (e.g., the range). In addition to the load estimation, the machine learning algorithm can also take into account the flow rate and / or upstream pressure of the PD fluid. As discussed in more detail below, the disclosed machine learning algorithm provides an accurate estimate of the PD fluid pressure within the patient line, thereby enabling the removal of the pressure sensor.

[0023] Aspects of the subject matter described herein can be used alone or in combination with one or more other aspects described herein. Without limiting the foregoing description, in a first aspect of the present disclosure that can be combined with any other aspect or a part thereof, a peritoneal dialysis (“PD”) system includes: a housing; a PD fluid pump received by the housing, the PD fluid pump including an actuator (e.g., a piston) actuated by a motor; a patient line fluidly coupling the PD fluid pump to a transfer kit that is connected to an indwelling catheter leading into a peritoneal cavity of a patient; a motor driver configured to control the motor and transmit an output signal indicative of a load on the motor; and a memory storing a machine learning algorithm that correlates data related to the output signal from the motor driver with a known PD fluid pressure within the patient line. The PD system further includes a processor electrically connected to the motor driver and the memory. The processor is configured to: transmit an input signal to activate the motor driver, receive the output signal from the motor driver, estimate the PD fluid pressure within the patient line by applying data from the received output signal to the machine learning algorithm, and cause the motor driver to stop when the estimated PD fluid pressure is higher than a threshold.

[0024] According to a second aspect of the present disclosure that can be combined with any other aspect or a part thereof, the processor is further configured to: receive a plurality of output signals from the motor driver; sample the plurality of output signals at a prescribed rate using a moving window filter; and apply the machine learning algorithm to data from the sampled plurality of output signals within the moving window filter.

[0025] According to a third aspect of the present disclosure that can be combined with any other aspect or a part thereof, the prescribed rate is between 50 Hz and 1000 Hz for sampling between 1 and 100 (preferably about 5 to 7) output signals within the moving window filter.

[0026] According to a fourth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the processor is further configured to determine at least one of the following input variable values as data from the sampled plurality of output signals: (i) a minimum output signal value, (ii) a maximum output signal value, (iii) an average output signal value, (iv) a median output signal value, or (v) a difference between the maximum output signal value and the minimum output signal value; and use at least one of (i) to (v) to apply the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

[0027] According to a fifth aspect of the present disclosure, which can be combined with any other aspect or a part thereof, the processor is further configured to: determine or receive an indication of the flow rate of the PD fluid; and use the flow rate of the PD fluid in combination with at least one of (i) to (v) to apply the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

[0028] According to a sixth aspect of the present disclosure, which can be combined with any other aspect or a part thereof, the processor is further configured to: normalize the flow rate of the PD fluid by dividing the flow rate of the PD fluid by a maximum flow rate; and use the normalized flow rate of the PD fluid in combination with at least one of (i) to (v) to apply the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

[0029] According to a seventh aspect of the present disclosure, which can be combined with any other aspect or a part thereof, the processor is configured to receive the flow rate of the PD fluid from a flow sensor fluidly coupled to the patient line or determine the flow rate of the PD fluid from a programmed PD fluid flow rate.

[0030] According to an eighth aspect of the present disclosure, which can be combined with any other aspect or a part thereof, the machine learning algorithm includes, as inputs, at least one of the flow rate of the PD fluid or the normalized fluid flow rate of the PD fluid; and at least one of (i) a minimum output signal value, (ii) a maximum output signal value, (iii) an average output signal value, (iv) a median output signal value, or (v) a difference between the maximum output signal value and the minimum output signal value.

[0031] According to a ninth aspect of the present disclosure, which can be combined with any other aspect or a part thereof, the machine learning algorithm further includes an upstream PD fluid pressure as an input variable.

[0032] According to a tenth aspect of the present disclosure, which can be combined with any other aspect or a part thereof, the machine learning algorithm is trained using a data set that correlates at least one of the flow rate of the PD fluid or the normalized fluid flow rate of the PD fluid and at least one of (i) to (v) with the PD fluid pressure within the patient line measured by a pressure sensor.

[0033] According to an eleventh aspect of the present disclosure that can be combined with any other aspect or a part thereof, the processor is further configured to: receive an indication of an upstream PD fluid pressure from a pressure sensor located upstream of the PD fluid pump; and use the upstream PD fluid pressure in combination with at least one of (i) to (v) above to apply the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

[0034] According to a twelfth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the upstream PD fluid pressure corresponds to a head height pressure.

[0035] According to a thirteenth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the PD system further includes a filter kit, the filter kit including a hydrophilic filter membrane fluidly coupled between the patient line and the transfer kit. In this aspect, the patient line is a dual - lumen patient line, the dual - lumen patient line including a fresh PD fluid lumen and a used PD fluid lumen, and the processor is configured to estimate the fluid pressure within the fresh PD fluid lumen.

[0036] According to a fourteenth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the motor is a stepper motor, and the output signal represents the load angle of the stepper motor.

[0037] According to a fifteenth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the processor is further configured to apply a smoothing function to a sequence or stream of the estimated PD fluid pressure.

[0038] According to a sixteenth aspect of the present disclosure that can be combined with any other aspect or a part thereof, a PD method for estimating peritoneal dialysis ("PD") fluid pressure includes: storing in a memory of a PD machine a machine learning algorithm that correlates data related to an output signal from a motor driver with a known PD fluid pressure within a patient line. The method further includes transmitting an input signal from a processor of the PD machine to a motor driver of a PD fluid pump to activate the motor driver, which causes the motor to actuate an actuator (such as a piston) for pumping PD fluid from the PD fluid pump to the patient line at a prescribed rate, the patient line being fluidly coupled to a transfer kit that is connected to an indwelling catheter leading to a peritoneal cavity of a patient. The method further includes receiving, in the processor, an output signal from the motor driver indicating a load estimate output of the motor; estimating, via the processor, the PD fluid pressure within the patient line using the machine learning algorithm based on data related to the received output signal; and causing, via the processor, a user interface of the PD machine to display the estimated PD fluid pressure.

[0039] According to a seventeenth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the method further includes, when the estimated PD fluid pressure is higher than a threshold value outside a specified range, stopping the motor driver via the processor.

[0040] According to an eighteenth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the machine learning algorithm includes at least one of Gaussian process regression, linear regression, logistic regression, decision tree, gradient boosting algorithm, random forest algorithm, k-nearest neighbor algorithm, k-means algorithm, support vector machine, naive Bayes algorithm, or a combination thereof.

[0041] According to a nineteenth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the method further includes sampling the output signal at a specified rate using a moving window filter via the processor; and applying the machine learning algorithm to data from the sampled multiple output signals within the moving window filter via the processor.

[0042] According to a twentieth aspect of the present disclosure that can be combined with any other aspect or a part thereof, the specified rate is between 50 Hz and 1000 Hz for sampling between 1 and 100 output signals (preferably between 5 and 7 output signals) within the moving window filter.

[0043] According to a twenty-first aspect of the present disclosure that can be combined with any other aspect or a part thereof, the method further includes determining, via the processor, at least one of the following input variable values as data from the sampled output signals: (i) minimum output signal value, (ii) maximum output signal value, (iii) average output signal value, (iv) median output signal value, or (v) difference between the maximum output signal value and the minimum output signal value; and applying the machine learning algorithm to the sampled multiple output signals within the moving window filter using at least one of (i) to (v) via the processor.

[0044] According to a twenty-second aspect of the present disclosure that can be combined with any other aspect or a part thereof, the method further includes determining or receiving an indication of the flow rate of the PD fluid via the processor; and using the flow rate of the PD fluid in combination with at least one of (i) to (v) via the processor for applying the machine learning algorithm to the sampled multiple output signals within the moving window filter.

[0045] According to a twenty-third aspect of the present disclosure, which can be combined with any other aspect or a part thereof, the method further includes training the machine learning algorithm using a data set that correlates at least one of the flow rate of the PD fluid or the normalized fluid flow rate of the PD fluid, and at least one of (i) to (v) with the PD fluid pressure in the patient line measured by a pressure sensor.

[0046] In a twenty-fourth aspect of the present disclosure, which can be combined with any other aspect or a part thereof, any feature, function, and alternative described in any one or more of Figures 1 to 10 can be combined with any other feature, function, and alternative described in any other description in Figures 1 to 10 .

[0047] Advantages of the present disclosure in accordance with the above aspects and the present disclosure set forth herein are to provide a PD system and an associated method for estimating PD fluid pressure using motor driver output.

[0048] Another advantage of the present disclosure is to provide a PD system and an associated method that eliminates a pressure sensor for measuring the pressure of the PD fluid in the patient line.

[0049] Additional advantages of the present disclosure are to provide a PD system and an associated method for estimating PD fluid pressure using machine learning with the load on a PD fluid pump.

[0050] Additional features and advantages are described in the following detailed description and the drawings, and will be apparent from the following detailed description and the drawings. The features and advantages described herein are not all-inclusive, and in particular, with reference to the drawings and the description, many additional features and advantages will be apparent to those of ordinary skill in the art. Moreover, any particular embodiment need not have all of the improvements or advantages listed herein, and it is expressly contemplated to claim individually advantageous embodiments separately. In particular, the systems of the present disclosure can have any one or more or all of the anti-drip structure and method, PD fluid container emptying structure and method, and pre-discharge patient connection checking structure and method described herein. Further, it should be noted that the language used in the specification is selected primarily for readability and guidance purposes, rather than to limit the scope of the inventive subject matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a schematic diagram of fluid flow of an embodiment of a PD system in a treatment state according to an exemplary embodiment of the present disclosure.

[0052] Figure 2 is a schematic diagram of fluid flow of an embodiment of a PD system in a disinfection state according to an exemplary embodiment of the present disclosure.

[0053] Figure 3 It is a diagram showing features related to estimating the pressure of PD fluid in a double - lumen patient line of a PD system according to an exemplary embodiment of the present disclosure.

[0054] Figure 4 It is a diagram showing a machine - learning algorithm of a PD system for estimating the pressure in a patient line according to an exemplary embodiment of the present disclosure.

[0055] Figure 5 Shows a flowchart according to an exemplary embodiment of the present disclosure, which shows an example procedure of using Figure 4 's machine - learning algorithm to estimate Figures 1 to 3 the pressure of PD fluid in the patient line of a PD system.

[0056] Figure 6 Shows a graph of the output signal flow and the corresponding flow rate of PD fluid for a PD system according to an exemplary embodiment of the present disclosure for Figures 1 to 3 ...

[0057] Figure 7 Shows a flowchart according to an exemplary embodiment of the present disclosure, which shows an example procedure for training Figure 4 's machine - learning model.

[0058] Figure 8 Are graphs according to an exemplary embodiment of the present disclosure, which show the correlation of the output signal, flow rate, and measured PD pressure from a PD fluid pump, and can be included in the training data set for Figure 4 's machine - learning algorithm.

[0059] Figure 9 and Figure 10 Are graphs according to an exemplary embodiment of the present disclosure, which show the accuracy of the machine - learning model. Detailed Description

[0060] This disclosure relates to a PD system that estimates the PD fluid pressure within a patient line downstream of a PD fluid pump. An example PD system is configured to estimate the PD fluid pressure using the load estimate output from the motor driver of the PD fluid pump. To perform the pressure estimation, the PD system uses a machine learning algorithm, such as a Gaussian process regression model or a decision tree model. The machine learning algorithm or model is configured to correlate data that correlates known PD fluid pressures within the patient line with the load estimate output from the motor driver. This data can be obtained by sampling the load estimate output at discrete intervals. For example, in some embodiments, thirty to fifty data points can be sampled and processed together to obtain characteristics of the load estimate output from the motor driver, which are used as inputs to the machine learning model.

[0061] The machine learning algorithm includes discrete inputs of characteristics of the load estimate output from the motor driver. The characteristics can include the minimum value of the sampled data, the maximum value of the sampled data, the average value of the sampled data, the median value of the sampled data, and / or the difference between the maximum and minimum values of the sampled data (e.g., range). In addition to the load estimate, the machine learning algorithm can also take into account the flow rate and / or upstream pressure of the PD fluid. As discussed in more detail below, the disclosed machine learning algorithm provides an accurate estimate of the PD fluid pressure within the patient line, thereby enabling the removal of the pressure sensor.

[0062] This disclosure refers to a PD system that includes a cycler. It should be understood that the systems and methods disclosed herein can be applied to any pressure estimation downstream of a pump. For example, these methods and systems can be directed to downstream pressures from a blood pump or a dialysis fluid pump for hemodialysis machines, hemofiltration machines, hemodiafiltration machines, and continuous renal replacement machines. These systems and methods can also be used for infusion pumps or injection pumps. Additionally, these systems and methods can be used for parenteral nutrition pumps or patient-controlled analgesia pumps.

[0063] Furthermore, although this disclosure refers to PD fluid pressure estimation for eliminating the need for a downstream pressure sensor, it should be understood that PD fluid pressure estimation can also be used for cycler diagnostics. For example, the estimated pressure can be compared to the measured pressure to determine when the non-dynamic load of the pump begins to deviate. When the deviation is significant, the systems and methods disclosed herein can indicate that the PD fluid pump needs repair or replacement.

[0064] System Overview

[0065] Referring now to the drawings, and particularly to Figure 1, a medical system with motor drive pressure estimation of the present disclosure is shown via a peritoneal dialysis (“PD”) system 10. The system 10 includes a PD machine or cycler 20 and a control unit 100 having one or more processors 102, one or more memories 104, a video controller 106, a motor drive 108, and a user interface 110. Alternatively or additionally, the user interface 110 can be, for example, a remote user interface via a tablet or smartphone. The control unit 100 may also include a transceiver and a wired or wireless connection to a network (not shown) (e.g., the Internet) for sending treatment data to and receiving prescription instructions / changes from a doctor or clinician's server interconnected with the doctor's or clinician's computer. In one embodiment, the control unit 100 controls all the current fluid flow and heating components of the system 10 and receives outputs from all the sensors of the system 10. The system 10 in the illustrated embodiment includes, for example, durable and reusable components that come into contact with fresh and used PD fluid, and these components require the PD machine or cycler 20 to be disinfected between treatments, for example, by heat disinfection.

[0066] Figure 1 The system 10 in includes an inline resistance heater 56, reusable supply lines or tubes 52a1 to 52a4 and 52b, an air trap 60 operating with corresponding upper liquid level sensors 62a and lower liquid level sensors 62b, an air trap valve 54d, an exhaust valve 54e located along the exhaust line 52e, a reusable line or pipe 52c, a PD fluid pump 70, temperature sensors 58a and 58b, pressure sensors (78a, shown in dashed lines as it is removed and replaced with motor drive estimation), pressure sensors 78a, 78b1, 78b2, and 78c, reusable patient lines or tubes 52f and 52g having corresponding valves 54f and 54g, a double - lumen patient line 28, a hose reel 80 for retracting the patient line 28, a reusable drain pipe or line 52i extending to the drain line connector 34 and having a drain line valve 54i, and reusable recirculation disinfection pipes or lines 52r1 and 52r2 operating with corresponding disinfection valves 54r1 and 54r2. A third recirculation or disinfection pipe or line 52r3 extends between the disinfection or PD fluid line connectors 30a and 30b for use during disinfection. A fourth recirculation or disinfection pipe or line 52r4 extends between the disinfection connectors 30c and 30d for use during disinfection.

[0067] System 10 also includes PD fluid containers or bags 38a to 38c (e.g., holding the same or different PD fluid formulations), and PD fluid containers or bags 38a to 38c are respectively connected to the distal ends 24e of reusable PD fluid lines 24a to 24c. System 10 also includes a fourth PD fluid container or bag 38d, and the fourth PD fluid container or bag 38d is connected to the distal end 24e of a reusable PD fluid line 24d. The fourth PD fluid container or bag 38d may hold the same or different type of PD fluid (e.g., icodextrin) as that provided in the PD fluid containers or bags 38a to 38c. In one embodiment, the reusable PD fluid lines 24a to 24d extend out through an orifice (not shown) defined or provided by the housing 22 of the PD machine 20.

[0068] The system 10 in the illustrated embodiment includes four sterile or PD fluid line connectors 30a to 30d for respectively connecting to the distal ends 24e of the reusable PD fluid lines 24a to 24d during sterilization. The system 10 also provides a patient line connector 32, and the patient line connector 32 includes an internal cavity, such as a U-shaped cavity, which guides fresh or used dialysis fluid from one PD fluid cavity of the connected distal end 28e of the double-lumen patient line 28 to another PD fluid cavity for sterilization. The reusable supply conduits or lines 52a1 to 52a4 are respectively in communication with the reusable supply lines 24a to 24d. The reusable supply conduits or lines 52a1 to 52a3 respectively operate with valves 54a to 54c to allow PD fluid from a desired PD fluid container or bag 38a to 38c to be drawn into the PD machine 20. In the illustrated example, the three-way valve 94a allows the control unit 100 to select between (i) 2.27% (or other) glucose dialysis fluid from the container or bag 38b or 38c and (ii) icodextrin from the container or bag 38d. In the illustrated embodiment, the icodextrin from the container or bag 38d is connected to the normally closed port of the three-way valve 94a.

[0069] In one embodiment, the system 10 is configured such that during patient fill, the drain line 52i is fluidly connected downstream of the PD fluid pump 70. In this way, if the drain valve 54i fails or leaks in some way during patient fill of the patient P, the fresh PD fluid is pushed down to the disposable drain line 36 instead of the used PD fluid potentially being drawn into the pump 70. The disposable drain line 36 is removed for sterilization in one embodiment, where the drain line connector 34 is covered via a cap 34c to form a closed sterilization loop. The PD fluid pump 70 can be an inherently accurate pump, such as a piston pump, or a less accurate pump, such as a gear pump that operates in cooperation with a flow meter (not shown) to control the flow rate and volume of fresh and used PD fluid.

[0070] System 10 may also include a leak detection pan 82 located at the bottom of the housing 22 of the PD machine 20 and a corresponding leak detection sensor 84 output to the control unit 100. In the example shown, system 10 is provided with an additional pressure sensor 78c upstream of the PD fluid pump 70, which allows the suction pressure of the pump 70 to be measured to assist the control unit 100 in more accurately determining the pump volume. The additional pressure sensor 78c in the illustrated embodiment is positioned along the exhaust pipeline 52e, which may be filled with air or a mixture of air and PD fluid, but should still be at the same negative pressure as the PD fluid located within the PD fluid pipeline 52c.

[0071] Figure 1 System 10 in the example includes redundant pressure sensors 78b1 and 78b2, and the output of one of the redundant pressure sensors 78b1 and 78b2 is used for pump control as discussed herein, while the output of the other pressure sensor is a safety or watchdog output to ensure that the control pressure sensor reads accurately. The pressure sensors 78b1 and 78b2 are positioned along a pipeline including the third recirculation valve 54r3. System 10 may also employ one or more four-way connections marked with an X in Figure 1 which may (i) reduce the total amount and volume of piping that can be reused internally, (ii) reduce the number of valves required, and (iii) minimize the portion of the fluid circuit that is shared by fresh and used PD fluid.

[0072] In Figure 1 System 10 in the example also includes an acid source, such as a citric acid container or bag 66. The citric acid container or bag 66 is selectively fluidly connected to the second three-way valve 94b via a citric acid valve 54m positioned along the citric acid pipeline 52m. In one embodiment, the citric acid pipeline 52m is connected to the normally closed port of the second three-way valve 94b to provide a redundant valve between the citric acid container or bag 66 and the PD fluid circuit during treatment. The redundant valve ensures that no citric acid (or other) reaches the treatment fluid pipeline during treatment. Instead, citric acid (or other acid) is used during disinfection.

[0073] In one embodiment, the control unit 100 uses feedback from any one or more of the pressure sensors 78b1 or 78b2 such that the PD machine 20 can deliver fresh heated PD fluid to the patient at a pressure of, for example, 14 kPa (2.0 psig) or higher. The pressure feedback is used to enable the PD machine 20 to remove used PD fluid or effluent from the patient at a pressure between, for example, -5 kPa (-0.73 psig) and -15 kPa (-2.2 psig) (e.g., -9 kPa (-1.3 psig) or higher (more negative)). The pressure feedback can be used in a proportional, integral, derivative ("PID") pressure routine for pumping fresh and used PD fluid at a desired positive or negative pressure.

[0074] The in-line resistive heater 56 under the control of the control unit 100 can heat the fresh PD fluid to body temperature, such as 37 °C, for delivery to the patient P at a desired flow rate. In an embodiment, the control unit 100 uses feedback from the temperature sensor 58a in a PID temperature routine for pumping the fresh PD fluid to the patient P at a desired temperature.

[0075] Figure 1 Also shown is that the system 10 includes and uses a disposable filter kit 40 that is in fluid communication with the fresh and used PD fluid cavities of the dual-lumen patient line 28. The disposable filter kit 40 includes a disposable connector 42 that attaches to the distal end 28e of the reusable patient line 28. The disposable filter kit 40 also includes a connector 44 that attaches to the transfer kit of the patient. The disposable filter kit 40 also includes a sterilizing-grade hydrophilic filter membrane 46 that further filters the fresh PD fluid. The disposable filter kit 40 is provided as a last-chance filter for the PD machine 20 in one embodiment and is heat sterilized between treatments. Any pathogens that may remain after sterilization (although unlikely) are filtered out of the PD fluid via the hydrophilic membrane 46 of the disposable filter kit 40.

[0076] Figure 1 Shown is the setup of the system 10 for treatment, where the PD fluid containers or bags 38a to 38d are connected via reusable flexible PD fluid lines 24a to 24d, respectively. The dual-lumen patient line 28 is connected to the patient P via the disposable filter kit 40. The disposable drain line 36 is connected to the drain line connector 34. In Figure 1 which, the PD machine or cycler 20 of the system 10 is configured to perform a plurality of patient drain, patient fill, patient dwell, and perfusion procedures as part of a treatment or in preparation for a treatment.

[0077] Figure 2System 10 in the disinfection mode is shown. The PD fluid containers or bags 38a to 38d are removed, and the flexible PD fluid lines 24a to 24d are inserted in a sealed manner into the disinfection or PD fluid line connectors 30a to 30d respectively. The reusable double-lumen patient line 28 is disconnected from the disposable filter kit 40 (which is discarded), and the distal end 28e of the double-lumen patient line 28 is inserted in a sealed manner into the patient line connector 32. The disposable drain line 36 is removed from the drain line connector 34 and discarded. The drain line connector 34 is covered via the cap 34c to form a closed disinfection circuit 90. Figure 2 The PD machine or cycler 20 of the system 10 in is configured to perform a disinfection sequence, such as a heat disinfection sequence, in which fresh PD fluid is heated to a disinfection temperature, such as 70 °C to 90 °C, via the in-line heater 56. The PD fluid pump 70 circulates the heated PD fluid through the closed disinfection circuit 90 for an amount of time appropriate to disinfect the fluid components and lines of the disinfection circuit.

[0078] Example of motor driver pressure estimation

[0079] The hydrophilic filter membrane 46 of the filter kit 40 causes a pressure drop in the fresh PD fluid pressure. The pressure of the fresh PD fluid output by the PD fluid pump 70 through the fresh PD fluid lumen of the double-lumen patient line 28 to the hydrophilic filter membrane 46 is thus greater than the pressure experienced by the patient P downstream of the hydrophilic filter membrane 46 due to this pressure drop. Therefore, the PD fluid pressure downstream of the hydrophilic filter membrane 46 is an important pressure to be monitored for use as feedback to control the PD fluid pump 70, such that the PD fluid pressure experienced by the patient P is at or below the patient PD fluid pressure limits listed above.

[0080] The pressure sensors 78b1, 78b2 are positioned to sense the PD fluid pressure in the used PD fluid lumen of the double-lumen patient line 28, which is an important pressure downstream of the filter membrane. Therefore, the pressure sensor 78a positioned to sense the PD fluid pressure in the fresh PD fluid lumen is not as critical and is instead used to sense, for example, kinking or blockage of the fresh PD fluid lumen. Therefore, it is contemplated to eliminate the pressure sensor 78a (shown in dashed lines) positioned to sense the PD fluid pressure in the fresh PD fluid lumen, and instead use the output signal provided by the motor driver 108 for the motor (e.g., stepper motor) used to drive the PD fluid pump 70 to estimate the PD fluid pressure.

[0081] In one embodiment, the PD fluid pump 70 is a piston pump that includes a housing that holds a cylinder, and a piston that is actuated within the cylinder by an electric motor (the electric motor being under the control of an electric motor driver 108 that is considered to be part of the overall control unit 100), where the electric motor drives a motion coupler that is coupled to the piston. The motion coupler converts the rotational motion of the electric motor into rotational and translational movement of the piston. The motion coupler moves the piston into and out of the cylinder relative to the cylinder to produce positive pumping pressure and negative pumping pressure, respectively. The motion coupler also rotates the piston within the cylinder to move PD fluid from an inlet port of the PD fluid pump 70 to an outlet port.

[0082] In one embodiment, the electric motor is a stepper motor, and the electric motor driver 108 for the stepper motor provides an output signal that indicates the load that the electric motor currently sees or experiences. To estimate the electric motor load, in one embodiment, the electric motor driver 108 measures the electrical energy flowing into the electric motor and the electrical energy flowing out of the electric motor. The difference between the energy flowing in and the energy flowing out provides an indication of the mechanical load seen or experienced by the electric motor. The electric motor driver 108 measures the portion of the energy fed into the electric motor that is returned to the power supply that powers the stepper motor. This excess energy is measurable and indicates the mechanical load applied to the electric motor.

[0083] From the perspective of the electric motor, the load estimate output from the electric motor driver 108 represents the load angle of the stepper motor, which depends on the external torque applied to the axis of the electric motor shaft. A stepper motor includes a stationary stator and a rotor that rotates within the stator. When power is applied to the electric motor, a magnetic field is applied, where the magnetic field rotationally pulls and pushes the rotor within the stator, resulting in a phase shift between the magnetic field direction of the rotor and the magnetic field direction of the rotating field of the stator. The phase shift is the load angle, i.e., the angle between the magnetic field direction of the rotor and the magnetic field direction of the rotating magnetic field of the stator.

[0084] From a mathematical perspective, the load estimate output ("LEO") from the electric motor driver 108 can be a function of the back electromotive force ("EMF") constant inherent in the stepper motor, the coil inductance of the stepper motor, the coil resistance of the stepper motor, the stepping speed (e.g., full steps per second), the load angle, the phase current applied to the electric motor, and the voltage supplied to the stepper motor.

[0085] The electric motor driver 108 is configured to receive one or more input signals from the processor 102 for controlling the pump stroke of the PD fluid pump 70. The one or more input signals can specify the rate and / or duration at which the electric motor driver 108 actuates the electric motor. The input signals can be analog or digital. For an analog signal, the electric motor driver 108 controls the electric motor based on, for example, the amplitude and / or frequency of the input signal. For a digital signal, the electric motor driver 108 uses a look-up table to convert the digital signal into a rate and / or duration for controlling the electric motor.

[0086] Figure 3 FIG. is a diagram showing features related to estimating the pressure of PD fluid in the double-lumen patient line 28 of the PD system 10 according to an exemplary embodiment of the present disclosure. As discussed above in connection with Figure 1 and Figure 2 the PD system 10 includes a PD machine or cycler 20 fluidly coupled to a patient P via a double-lumen patient line 28. The PD machine 20 includes a control unit 100 that includes at least one processor, a memory 104, and a motor driver 108. Additionally, a user interface 100 is communicatively coupled to the control unit 100 to enable a user to control the PD machine 20.

[0087] Figure 3 FIG. shows that the housing 22 of the PD machine 20 is configured to enclose or otherwise support the PD fluid pump 70. As discussed above in connection with Figure 1 and Figure 2 the PD fluid pump 70 is configured to move PD fluid from the PD machine 20 (e.g., from a PD fluid container 38) to the peritoneal cavity of the patient P via the double-lumen patient line 28, a disposable filter kit 40, and / or a transfer kit 302. As discussed above in connection with Figure 1 and Figure 2 the transfer kit 302 may include or be connected to an indwelling catheter leading into the peritoneal cavity of the patient. In some embodiments, the disposable filter kit 40 may be omitted such that the double-lumen patient line 28 is directly fluidly coupled to the transfer kit. Additionally, although reference is made to the double-lumen patient line 28, it should be understood that the double-lumen patient line 28 may be a single lumen where the transfer kit includes a separate connection for receiving fresh PD fluid from the PD machine 20 and pulling used PD fluid into a disposable drain line 36.

[0088] Figure 3 FIG. shows that the PD fluid pump 70 is housed within the housing 22 and is electrically connected to the motor driver 108. In one embodiment, the PD fluid pump 70 is a piston pump that includes a cylinder 304 and a piston 306 that is actuated by a motor 308. The piston 306 is configured to move back and forth within the cylinder 304 to pump fresh PD fluid into the patient line 28. The motor 308 is configured to drive a motion coupler connected to the piston 306 that causes the piston 306 to move into and out of the cylinder 304 to generate positive and negative pumping pressures. The motion coupler converts the rotational motion of the motor 308 into rotational and translational movement of the piston 308. The motion coupler may also cause the piston 306 to rotate within the cylinder 304 to move fresh PD fluid from the inlet port of the PD fluid pump 70 to the outlet port.

[0089] As Figure 3As shown, the processor 102 is configured to transmit an input signal 310 to the motor driver 108 to activate the PD fluid pump 70. The input signal 310 can be analog and / or digital and specifies the pump rate and / or duration. In some embodiments, the input signal 310 specifies the pump rate such that the processor 102 continues to transmit the input signal 310 as long as the PD fluid pump 70 remains active. In another example, the input signal 310 can specify the number of pump strokes to be pumped. In response to the input signal 310, the motor driver 108 transmits a signal to the motor 308. This signal causes the motor 308 to rotate at a specified speed. The motor driver 108 continues to apply a signal to the motor 308 as long as the motor is to be activated.

[0090] The example motor driver 108 is configured to receive feedback indicating the speed / rotation of the motor 308 in addition to the load angle of the motor 308. The speed / rotation of the motor 308 is used by the motor driver 108 as feedback to accelerate or decelerate the motor 308 based on meeting the specified pumping speed and / or duration. The motor driver 108 is configured to convert the load angle into a load estimate value, which is transmitted to the processor 102 as one or more output signals 312 (e.g., load estimate output ("LEO")). It should be understood that the motor driver 108 provides a nearly continuous stream of the output signal 312 as long as the motor 308 is in an operating state. This stream of the output signal 312 enables the processor 102 to determine how the estimated load on the motor 308 changes over time during pumping of the PD fluid. In one embodiment, the PD fluid pump 70 includes a TMC5130 or TMC5160 stepper motor driver produced by Trinamic MotionControl GmbH & Co. KG, and the output signal 312 is a stallGuard TM signal.

[0091] Figure 3 The example memory device 104 includes or stores one or more machine learning algorithms 320, which are configured to estimate the PD fluid pressure within the patient line 28 based on one or more of the output signals 312 from the motor driver 108. The machine learning algorithms 320 can be specified by one or more instructions executable by the processor 102 to perform the operations disclosed herein. As discussed above, the use of the machine learning algorithms 320 enables the removal or elimination of the pressure sensor on the patient line 28, thus saving space and cost.

[0092] Figure 4 is shown in accordance with an example embodiment of the present disclosure Figure 3A diagram of the machine learning algorithm 320. The machine learning algorithm 320 may include Gaussian process regression, linear regression, logistic regression, decision trees, gradient boosting algorithms, random forest algorithms, k-nearest neighbor algorithms, k-means algorithms, support vector machines, naive Bayes algorithms, or combinations thereof. An example machine learning algorithm 320 is configured to receive samples of a plurality of output signals 312 from the processor 102. For example, the processor 102 may be configured to sample the plurality of output signals 312 at a prescribed rate using a moving window feature. The processor 102 then applies the machine learning algorithm 320 to data corresponding to the sampled plurality of output signals within the moving window filter. The prescribed rate may be between 50 Hz and 1000 Hz for sampling between 1 and 100 output signals 312 within the moving window filter. Preferably, the prescribed rate is between 100 Hz and 250 Hz for sampling between 5 and 7 output signals.

[0093] As Figure 4 shown, for each sample of the output signal 312, the processor 102 determines certain characteristics of the sampled data, including for example the minimum value 402 of the output signal, the maximum value 404 of the output signal, the average value 406 of the output signal, the median value 408 of the output signal, and the difference 410 (e.g., range) between the maximum and minimum values of the output signal. In some embodiments, fewer or additional values 402 to 410 may be used as inputs to the machine learning model 320. For example, the median value 406 may be omitted and / or an average derivative value may be used. In some embodiments, the values 402 to 410 may be normalized.

[0094] In some embodiments, the machine learning model 320 may also use the flow rate 412 and / or the upstream pressure value 414. A flow sensor (such as Figure 3 the flow sensor 322 shown on the patient line 28) may be used to determine the flow rate 412. In these embodiments, the flow sensor 322 transmits data indicating the measured flow rate 412 of the PD fluid within the patient line 28 to the processor 102. Additionally or alternatively, the processor 102 determines the flow rate 412 based on the programmed PD fluid flow rate. In this example, the processor 102 may receive program instructions for the PD machine 20 to perform a dialysis treatment. The programmed instructions include for example the flow rate of the PD fluid during the fill phase. When the flow rate 412 is not specified, the processor 102 determines the flow rate 412 based on the volume to be pumped each time when pumping a volume and / or the specified flow rate of the PD motor driver 108. In some cases, the processor 102 is configured to normalize the flow rate 412 by dividing the flow rate by the maximum flow rate. The normalization may be based on the characteristics of the fluid being pumped (such as water or PD fluid).

[0095] In some embodiments, the machine learning model 320 can determine the area under the curve indicative of a sampled set of the output signal 312. The machine learning model 320 can be configured to receive the area associated with one rotation of the PD fluid pump 70. The processor 102 can compute this integral over one rotation to provide a good estimate of the operation of the PD fluid pump 70.

[0096] The upstream pressure value 414 corresponds to the pressure measurement provided by a sensor upstream of the PD fluid pump 70, such as Figure 3 the pressure sensor 324 shown. The use of the upstream pressure provides an absolute pressure to the machine learning algorithm 320, rather than determining a relative pressure.

[0097] As Figure 4 shown, the machine learning algorithm 320 is configured to receive one or more of the input values 402 to 414 of the sampled set of the output signal 312. The machine learning algorithm 320 uses Gaussian process regression and / or decision tree models, for example, to estimate the pressure 416 of the PD fluid within the patient line 28 based on one or more of the input values 402 to 414. In some embodiments, one or more of the input values 402 to 414 are applied to a single machine learning model 320. In other embodiments, the input values 402 to 410 are applied to a first machine learning model, which is then cascaded with a second machine learning model. In this embodiment, the second machine learning model is configured to process the output from the first machine learning model having the flow rate 412 and / or the upstream pressure 414 to generate the estimated pressure 416. In other embodiments, the second machine learning model uses the result having the flow rate from the first machine learning model to generate a result, and the generated result is then provided to a third machine learning model as an input in addition to the upstream pressure.

[0098] Returning to Figure 3 , the processor 102 can display the estimated pressure 416 of the PD fluid within the patient line 28 on the display screen of the user interface 110. Additionally or alternatively, the processor 102 is configured to compare the estimated pressure 416 with a threshold and / or a pressure range. Exceeding the threshold can indicate an occlusion within the patient line 28, a blockage (partial blockage) of the disposable filter kit 40 (when in use), a blockage of the patient catheter, a patient with a full peritoneal cavity, etc. Further, for a failure to meet the minimum range when the PD fluid pump 70 is active, a fluid leak can be indicated. The processor 102 can be configured to cause the user interface 110 to display an alert or warning when the estimated pressure 416 exceeds the threshold and / or is outside the specified range. Further, when the threshold is exceeded or the estimated pressure 416 is outside the specified range, the processor 102 can cause the PD fluid pump 70 to stop operating.

[0099] Figure 5 A flowchart according to an example embodiment of the present disclosure is shown, which shows the use of Figure 4 the machine learning algorithm 320 to estimate Figures 1 to 3 the pressure of the PD fluid in the patient line 28 of the PD machine 20. The example program 500 can be executed by, for example, the processor 102 described in connection with the appendix Figures 1 to 3 or more generally by the control unit 100. Although the program 500 is described with reference to the Figure 5 flowchart shown, it should be understood that many other methods can be used to perform the functions associated with the program 500. For example, the order of many of the blocks can be changed based on the machine learning algorithm used and the number of inputs, certain blocks can be combined with other blocks, and many of the blocks described are optional. The operations described in connection with the program 500 can be specified by one or more instructions stored, for example, in the memory 104 of the PD machine 20.

[0100] When the processor 102 receives the instruction 501 to activate the PD fluid pump 70, the example program begins (block 502). The instruction 501 can include an electronic PD treatment prescription specifying the parameters for performing the PD therapy. The parameters can include the concentration of the PD fluid, the number of PD cycles, the start time, the volume of fresh PD fluid to be provided per cycle, the flow rate of the fresh PD fluid, the dwell time, and / or the volume of spent PD fluid to be removed from the patient per cycle. The processor 102 is configured to use the provided flow rate (and / or pump duration) or determine the flow rate (and / or pump duration) based on the instruction 501. In some embodiments, the instruction 501 can include only a request to start the PD treatment from the patient or the clinician. When the start of the treatment is specified, the processor 102 transmits an input signal 310 to the motor driver 108 such that the motor 308 is actuated (block 504). As discussed above, the input signal 310 can specify the speed and / or duration for actuating the motor 308.

[0101] The processor 102 then receives a stream of output signals 312 from the motor driver 108 indicating the load on the motor 308 (block 506). Figure 6Graph 600 shows an example of the flow of output signal 312 received by processor 102 from motor driver 108 according to an example embodiment of the present disclosure. Each data point within graph 600 specifies a value indicative of the load experienced by motor 108. Processor 102 combines the data from output signal 312 into a time series graph. Processor 102 also applies a moving window filter 602 to sample the data within graph 600. Window filter 602 may sample output signal 312 at a prescribed rate that may be between 50 Hz and 1000 Hz, for sampling between 1 and 100 data points, preferably sampling approximately 5 to 7 data points. Processor 102 is configured to move window filter 602 when a new output signal 312 is received.

[0102] Returning to Figure 5 , processor 102 next determines inputs 402 to 410 (block 508) based on output signal 312. Inputs 402 to 410 are specifically determined based on the output signal 312 data within window filter 602. As discussed above, the inputs include the minimum value 402 of the output signal within window filter 602, the maximum value 404 of the output signal within window filter 602, the average value 406 of the output signal within window filter 602, the median value 408 of the output signal within window filter 602, and the difference 410 (e.g., range) between the maximum and minimum values of the output signal within window filter 602. In some embodiments, processor 102 is also configured to receive a flow input value 412 from flow sensor 322 (or determine flow from instruction 501) and / or receive an upstream pressure input value 414 from pressure sensor 324 (block 510). In some cases, processor 102 may normalize flow input value 412 based on the known maximum flow rate of PD fluid using PD fluid pump 70. Figure 6 Graph 650 shows an example of the flow of PD fluid relative to output signal 312 from motor driver 108.

[0103] Processor 102 then applies inputs 402 to 410, 412, and / or 414 to Figure 4 machine learning algorithm 320 (block 512). Processor 102 may also apply inputs 412 and / or 414 to machine learning algorithm 320. Machine learning algorithm 320 is configured to output an estimated PD fluid pressure 416 of the PD fluid within patient line 28 (block 512). Processor 102 may store the estimated PD fluid pressure 416 to a log file stored in memory 104. Processor 102 may also cause the estimated PD fluid pressure 416 to be displayed by user interface 110 of PD machine 20.

[0104] In some embodiments, the processor 120 compares the estimated PD fluid pressure 416 to a threshold or range (block 514). The threshold can be, for example, 300 kilopascals (“kPa”) (43.5 psig), 400 kPa (58.0 psig), 500 kPa (72.5 psig), etc., which indicates occlusion, a full peritoneal cavity, and / or a partially (or fully) blocked catheter, patient line 28, or filter 40. For example, the range can be from 40 kPa (5.8 psig) to 300 kPa (43.5 psig). In some embodiments, the processor 102 averages (e.g., applies a smoothing function to) the estimated PD fluid pressure 416 over a short duration such as one second, two seconds, etc., to smooth out positive and negative pressure spikes from the pump strokes of the PD fluid pump 70. The processor 102 can then compare the smoothed estimated pressure 416 to the threshold and / or range.

[0105] When the estimated pressure 416 exceeds the threshold and / or is outside the specified range, the processor 102 is configured to cause the motor driver 108 to stop the motor 308 and / or generate a warning / alert 515 (block 516). In some embodiments, the processor 102 can also display the warning / alert on the user interface 110 of the PD machine 20. To cause the motor driver 108 to stop, the processor 102 can stop sending the input signal 310 and / or transmit an input signal 310 specifying that the actuation of the motor 308 will be aborted. When the estimated pressure 416 does not exceed the threshold and / or is within the range, the processor 102 determines whether a command to stop the PD fluid pump 70 has been received, such as at the end of the PD fill phase or the end of PD treatment (block 518). If the PD fluid pump 70 is to continue pumping, the program 500 returns to block 504, where the processor 102 transmits the input signal 310 to the drive motor 108 to keep the motor 308 operating. When a command to stop the PD fluid pump is received, the processor 102 causes the motor driver 108 to stop by suppressing the transmission of the input signal or sending an input signal to abort the motor 308 (block 520). Then, the example program ends.

[0106] Example of machine learning algorithm training

[0107] The example machine learning algorithm 320 is trained before being used on the PD machine 20. As Figure 3 shown, in some embodiments, the PD system 10 can also include a training server 330, which is configured to train the machine learning algorithm 320. The training server 330 is configured to receive inputs 402 to 410 (e.g., characteristics) related to the output signal 312, flow input 412, and / or upstream pressure input 414 in combination with known PD fluid pressures within the patient line 28. As Figure 4As shown, for training, the known PD fluid pressure 418 is combined with the above inputs 402 to 414 and input into the machine learning model 320. The correlation between the inputs 402 and / or 414 and the known PD fluid pressure 418 is processed using, for example, Gaussian process regression, linear regression, logistic regression, decision trees, gradient boosting algorithms, random forest algorithms, k-nearest neighbor algorithms, k-means algorithms, support vector machines, naive Bayes algorithms, or combinations thereof. The training is designed to determine the (weighted) relationship between the inputs and the known fluid pressure 418.

[0108] The training is performed on a data set that can be obtained from multiple cyclers 20 that are the same (or similar models and / or types) as the PD machine 20 used for PD treatment of a patient. The known PD fluid pressure 418 is obtained from a pressure sensor placed downstream of the PD fluid pump 70 (or between the patient line 28 and the disposable filter kit 40) on the patient line 28. The patient line used for training can have a thickness and length similar to that of the patient line 28 used for PD treatment.

[0109] Figure 7 A flowchart according to an example embodiment of the present disclosure is shown, which shows an example program 700 for training Figure 4 the machine learning model 320. The example program 700 can be executed by, for example, Figure 3 a training server 330. Although the program 700 is described with reference to Figure 7 the flowchart shown, it should be understood that many other methods can be used to perform the functions associated with the program 700. For example, the order of many of the blocks can be changed based on the machine learning algorithm used and the number of inputs, some blocks can be combined with other blocks, and many of the blocks described are optional.

[0110] When the training server 330 receives the training data set 701, the example program 700 begins. The training data set 701 includes at least some characteristics (e.g., inputs 402 to 410) of the output signal 312 indicating the motor load from the motor driver 108 during activation of the PD fluid pump (block 702). These characteristics are related to the PD fluid pressure 418 measured when the output signal 312 is generated by the motor driver 108. The training data set 701 can be collected from one or more PD machines 20 having the same (or similar) PD fluid pump. The training data 701 can be sampled from the output signal at a rate that is the same as or similar to the rate during the use of the filter window 602 discussed above.

[0111] As Figure 7 shown, the training server 330 can also receive a training data set 703 that correlates the flow rate and / or the upstream pressure with the known PD fluid pressure 418 (block 704). Figure 8FIG. 800 is a graph of the curves according to an exemplary embodiment of the present disclosure, which illustrate the correlation with the output signal 312, the flow rate 414, and the measured PD pressure 418 from the PD fluid pump 70, which may be included in the training datasets 701 and 703. The flow rate 414 may be measured from Figure 3 sensor 322, or determined based on the operating parameters for performing PD treatment. Figure 8 It is also shown that a smoothing function may be used to filter or smooth the known PD fluid pressure 418 to remove ripples from the pump strokes. In Figure 8 the example shown, there is a strong correlation between the value of the output signal and the pressure when the flow rate varies from 100 millimeters per minute ("ml") to 200 ml per minute.

[0112] Returning to Figure 7 , the training server 330 may remove the training data corresponding to when the PD fluid pump stops (block 706). This data may be indicated by a flow rate of 0 ml / min. The training server 330 then uses at least one machine learning model to determine the relationship, correlation, and correspondence between the inputs 402 to 414 and the known measured PD fluid pressure 418 (block 708). The training server 330 may determine the weights and / or whether some of the inputs are not relevant to the PD fluid pressure.

[0113] Then, the training server 330 may use another dataset to validate the machine learning algorithm 320. For validation, the training server 330 uses the training data associated with the known PD fluid pressure 418. However, the training server 330 uses the inputs 402 to 414 to estimate the PD fluid pressure 416. Then, the training server 330 compares the estimated PD fluid pressure 416 with the measured PD fluid pressure 418 to determine the accuracy of the machine learning model 320.

[0114] Figure 9 FIG. 900 is a graph of a curve according to an exemplary embodiment of the present disclosure, which shows the accuracy of the machine learning model 320. The graph 900 shows the pressure of the PD fluid at 0.5 bar, 1 bar, and 1.5 bar at three separate flow rates. For example, the pressure at the first dataset 902 is at a flow rate of 100 ml / min (milliliters per minute), the pressure at the second dataset 904 is at a flow rate of 200 ml / min, and the pressure at the third dataset 906 is at a flow rate of 300 ml / min. The first data shaded area shows the estimated PD fluid pressure 416. The second data shaded area shows the actual measured PD fluid pressure 418 in the patient line. The graph 900 shows that the estimated PD fluid pressure 416 is almost the same as the measured pressure 418. Figure 10An enlarged region of the graph 900 is shown, which shows the accuracy of the estimated PD fluid pressure. Although Figure 9 and Figure 10 the data in show a slight phase shift in the data due to the filter window 602, the time increment is negligible for the purpose of estimating the PD fluid in the patient line 28 of the PD machine 20.

[0115] Returning to Figure 7 , after the machine learning model 320 is trained, the training server 330 is configured to provide the machine learning model 320 on one or more cyclers 20 (block 710). This can include transmitting the machine learning model 320 over a network to the PD machine 20 for storage in the memory 104. Alternatively, the machine learning model 320 is installed on the PD machine 20 at the time of manufacture.

[0116] Periodically, the training server 330 determines whether additional training data is available (block 712). If no additional training data is available, the example program 700 ends. However, when additional training data exists, the program 700 returns to block 702 to update the machine learning model 320 based on the new data. Then, the training server 330 can send the updated machine learning model 320 to the cycler 20 via, for example, a network.

[0117] Example of PD fluid pump diagnosis

[0118] In some embodiments, the downstream pressure sensor of the PD machine 20 can be retained. However, the processor 102 can still use the machine learning algorithm 320 as a diagnostic check for the PD fluid pump 70. The processor 102 is configured to compare the estimated PD fluid pressure 416 based on the output signal 312 from the motor driver 108 with the PD fluid pressure 418 measured by the pressure sensor. When the deviation exceeds a threshold, this can indicate that the non-dynamic load on the PD fluid pump 70 has changed. Thus, the processor 102 can cause the user interface 110 to display a message indicating that the PD fluid pump 70 should be replaced or repaired. When the deviation is significant (such as 0.5 bar), the processor 102 can prevent the PD treatment from being performed until the PD fluid pump 70 is replaced / repaired.

[0119] Conclusion

[0120] It should be understood that all of the disclosed methods and procedures described herein can be implemented using one or more computer programs or components. These components can be provided as a series of computer instructions on any conventional computer-readable medium, including RAM, ROM, flash memory, magnetic or optical disks, optical memory, or other storage media. These instructions can be configured to be executed by a processor, which, when executing the series of computer instructions, performs or facilitates the performance of all or part of the disclosed methods and procedures.

[0121] It should be understood that various changes and modifications to the exemplary embodiments described herein will be apparent to those skilled in the art. Such changes and modifications can be made without departing from the spirit and scope of the subject matter and without diminishing its intended advantages. Accordingly, these changes and modifications are intended to be covered by the appended claims.

[0122] For example, the PD system 10 need not use redundant or durable components, but instead can alternatively employ a disposable kit having a disposable pumping portion that contacts the corresponding medical fluid. In another example, although a disposable filter kit 40 will not be required as a last-chance filter for a system that does not have heat sterilization, a disposable filter kit 40 can still be provided as a last-chance filter for on-line PD fluid if fresh PD fluid is manufactured on-line during use. Pumping PD fluid with a disposable kit can alternatively be performed via pneumatic pump actuation of a sheet of a disposable cartridge of the disposable kit, via electromechanical pump actuation of a sheet of a disposable cartridge of the disposable kit, or via peristaltic pump actuation of a pumping tube section provided with the disposable kit. In additional examples, although the pump motor 308 is shown as actuating the piston 306, the system 10 can be applied to the pump motor 308 actuating other types of pump actuators for the PD fluid pump 70, such as peristaltic pump actuators, centrifugal pump actuators, gear pump actuators, etc. Further, although the pump motor 308 is described as being a stepper motor, the system 10 can be applied to other types of pump motors, such as AC or DC brushed or brushless motors, servo motors, etc.

[0123] It should be understood that unless the terms "means" or "step" are recited expressly in the claims, 35 U.S.C. 112(f) or pre-AIA 35 U.S.C 112 paragraph 6 is not intended to be invoked. Accordingly, the claims are not meant to be limited to the corresponding structures, materials, or acts described in the specification or its equivalents.

Claims

1. A peritoneal dialysis ("PD") system, comprising: A housing; A PD fluid pump, which is accommodated by the housing, and the PD fluid pump includes an actuator actuated by a motor; A patient line, which fluidly couples the PD fluid pump to a transfer kit, and the transfer kit is connected to an indwelling catheter leading to a peritoneal cavity of a patient; A motor driver, which is configured to control the motor and transmit an output signal indicating a load on the motor; A memory, which stores a machine learning algorithm that associates data related to the output signal from the motor driver with a known PD fluid pressure within the patient line; A processor, which is electrically connected to the motor driver and the memory, and the processor is configured to: Transmit an input signal to activate the motor driver, Receive the output signal from the motor driver, Estimate the PD fluid pressure within the patient line by applying data from the received output signal to the machine learning algorithm, and When the estimated PD fluid pressure is higher than a threshold, stop the motor driver.

2. The PD system according to claim 1, wherein The processor is further configured to: Receive a plurality of output signals from the motor driver; Sample the plurality of output signals at a specified rate using a moving window filter; And Apply the machine learning algorithm to data from the sampled plurality of output signals within the moving window filter.

3. The PD system according to claim 2, wherein The specified rate is between 50 Hz and 1000 Hz and is used to sample between 1 and 100 output signals within the moving window filter.

4. The PD system according to claim 2, wherein, The processor is further configured to determine at least one of the following input variable values as data from the sampled plurality of output signals: (i) the minimum output signal value, (ii) the maximum output signal value, (iii) the average output signal value, (iv) the median output signal value, or (v) the difference between the maximum output signal value and the minimum output signal value; and Use at least one of (i) to (v) for applying the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

5. The PD system according to claim 4, wherein, The processor is further configured to: Determine or receive an indication of the flow rate of the PD fluid; and Use the flow rate of the PD fluid in combination with at least one of (i) to (v) for applying the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

6. The PD system according to claim 5, wherein, The processor is further configured to: Normalize the flow rate of the PD fluid by dividing the flow rate of the PD fluid by the maximum flow rate; and Use the normalized flow rate of the PD fluid in combination with at least one of (i) to (v) for applying the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

7. The PD system according to claim 5, wherein, The processor is configured to receive the flow rate of the PD fluid from a flow sensor fluidly coupled to the patient line or determine the flow rate of the PD fluid from a programmed PD fluid flow rate.

8. The PD system according to claim 5, wherein, The machine learning algorithm includes the following input variables as inputs: at least one of the flow rate of the PD fluid or the normalized fluid flow rate of the PD fluid; and at least one of (i) a minimum output signal value, (ii) a maximum output signal value, (iii) an average output signal value, (iv) a median output signal value, or (v) the difference between the maximum output signal value and the minimum output signal value.

9. The PD system according to claim 8, wherein The machine learning algorithm further includes the upstream PD fluid pressure as an input variable.

10. The PD system according to claim 8, wherein, The machine learning algorithm is trained using a dataset that correlates at least one of the flow rate of the PD fluid or the normalized fluid flow rate of the PD fluid and the at least one of (i) to (v) with the PD fluid pressure within the patient line measured by a pressure sensor.

11. The PD system according to claim 4, wherein, The processor is further configured to: receive an indication of the upstream PD fluid pressure from a pressure sensor located upstream of the PD fluid pump; and use the upstream PD fluid pressure in combination with the at least one of (i) to (v) for applying the machine learning algorithm to the sampled plurality of output signals within the moving window filter.

12. The PD system according to claim 11, wherein, The upstream PD fluid pressure corresponds to a head height pressure.

13. The PD system according to claim 1, further comprising a filter kit including a hydrophilic filter membrane fluidly coupled between the patient line and the transfer kit, Among them, the patient line being a dual - lumen patient line including a fresh PD fluid lumen and a used PD fluid lumen, and wherein the processor is configured to estimate the fluid pressure within the fresh PD fluid lumen.

14. The PD system according to claim 1, wherein, The motor is a stepper motor and the output signal represents the load angle of the stepper motor.

15. The PD system according to claim 1, wherein, The processor is further configured to apply a smoothing function to a sequence or stream of the estimated PD fluid pressure.

16. The PD system according to claim 1, wherein, The actuator includes a piston actuated by the motor.

17. A PD method for estimating peritoneal dialysis ("PD") fluid pressure, the method comprising: storing a machine learning algorithm in a memory of a PD machine, the machine learning algorithm correlating data related to an output signal from a motor driver with a known PD fluid pressure within a patient line; transmitting an input signal from a processor of the PD machine to a motor driver of a PD fluid pump to activate the motor driver, which causes the motor to actuate an actuator for pumping PD fluid from the PD fluid pump to the patient line at a prescribed rate, the patient line being fluidly coupled to a transfer kit that is connected to an indwelling catheter leading to a peritoneal cavity of a patient; receiving, in the processor, an output signal from the motor driver indicating a load estimate output of the motor; estimating, via the processor, the PD fluid pressure within the patient line using the machine learning algorithm based on data related to the received output signal; and causing, via the processor, a user interface of the PD machine to display the estimated PD fluid pressure.

18. The method according to claim 17 further comprises: When the estimated PD fluid pressure is higher than a threshold outside a specified range, causing, via the processor, the motor driver to stop.

19. The method according to claim 17, wherein, The machine learning algorithm includes at least one of Gaussian process regression, linear regression, logistic regression, decision tree, gradient boosting algorithm, random forest algorithm, k-nearest neighbor algorithm, k-means algorithm, support vector machine, naive Bayes algorithm, or a combination thereof.

20. The method according to claim 17, further comprising: Sampling the output signal at a prescribed rate using a moving window filter via the processor; And Applying the machine learning algorithm to data of the sampled multiple output signals within the moving window filter via the processor.

21. The method according to claim 20, wherein, The prescribed rate is between 50 Hz and 1000 Hz and is used to sample between 1 and 100 output signals within the moving window filter.

22. The method according to claim 20, further comprising: Determining, via the processor, at least one of the following input variable values as data from the sampled output signal: (i) minimum output signal value, (ii) maximum output signal value, (iii) average output signal value, (iv) median output signal value, or (v) difference between the maximum output signal value and the minimum output signal value; And Using, via the processor, at least one of (i) to (v) for applying the machine learning algorithm to the sampled multiple output signals within the moving window filter.

23. The method according to claim 22, further comprising: Determining or receiving an indication of the flow rate of the PD fluid via the processor; And Combining the flow rate of the PD fluid with at least one of (i) to (v) via the processor for applying the machine learning algorithm to the sampled multiple output signals within the moving window filter.

24. The method according to claim 22 further comprises: Training the machine learning algorithm using a dataset that associates at least one of the flow rate of the PD fluid or the normalized fluid flow rate of the PD fluid and at least one of (i) to (v) with the PD fluid pressure within the patient line measured by a pressure sensor.

Images