Prediction of impending catheter infusion failure using machine learning

Abstract

In some embodiments, a computer-implemented method of automatically monitoring a status of a catheter infusion is provided. An alarm controller receives a time series of pressure values from a pressure sensor associated with a pump operatively coupled to a catheter. The alarm controller generates one or more features based on the time series of pressure values. The alarm controller provides the one or more features to a machine learning model to generate a predicted confidence of impending failure of the catheter infusion. In response to determining that the predicted confidence of impending failure of the catheter infusion is greater than a predetermined threshold, the alarm controller causes an alert to be presented.

Description

Docket No. 3399-P46WOPREDICTION OF IMPENDING CATHETER INFUSION FAILURE USING MACHINE LEARNINGCROSS-REFERENCE(S) TO RELATED APPLICATION(S)

[0001] This application claims the benefit of Provisional Application No. 63 / 707091, filed October 14, 2024, the entire disclosure of which is hereby incorporated by reference herein for all purposes.BACKGROUND

[0002] The use of catheters for infusion is widespread in medicine. For example, peripheral venous catheters are used for the delivery of fluids, nutrition, medications, and blood; catheters are used for peritoneal dialysis; catheters are used to infuse regional anesthetic; arterial catheters are used for blood sampling and blood pressure monitoring; and so on.

[0003] For infants, peripheral intravenous (“PIV”) catheters have been associated with extravasation, edema, sequestration of medications and caustic solutions, and skin necrosis. Peripherally inserted central venous catheters (“PICC”) have been associated with leaking, obstruction, infection, cardiac tamponade, neurological problems, and even death. Iatrogenic sequelae from PIV's and PICC's have been identified as one of the leading causes of adverse incidents and injury in hospitals, with rates as high as 78% and 62%, respectively. These same adverse sequelae of failed venous and arterial catheter infusions affect other patient populations as well, especially those who are experiencing weakness, cannot speak, or are anesthetized.

[0004] Methods to increase the survival of intravenous infusion sites in patient populations have included avoidance of hyperosmolar solutions, neutralization of total parenteral nutrition, addition of heparin, implementation of inline filtration, antibiotic prophylaxis, and aseptic techniques for insertion and catheter maintenance. Rescue interventions for occluded catheters have included redressing, flushing with saline, heparin, urokinase, and alteplase. Conscientious monitoring for signs of complications has been recommended as an approach to reduce morbidity and mortality related to PIVs. Presently, monitoring systems are dependent upon in-line measures of intraluminal pressure and resistance thresholds to trigger alarms when outside preset limits. However, in a few limited studies, pressure or resistance thresholds have not been shown to be predictive of IV infiltration but may ring only after the extravasation, when the pressure in the tissue compartment has increased. Pressure threshold-based alarms have been shown to be even less effective in neonatalDocket No. 3399-P46WO populations. In neonates, the overall flow rates are relatively low, so at the time of infusion failure, pressure alarm thresholds are often not met. Some technology exists to detect early PIV failure, such as ivWatch®, which measures early failure through optical sensors. This technology is expensive and requires additional instrumentation, but most importantly, this technology shares the drawbacks of other monitoring systems in that it only alerts clinicians after a PIV infusion failure has already occurred. No technology exists with which to predict impending catheter infusion failure.

[0005] Infants cannot speak for themselves and cannot call for help when experiencing the initial signs of discomfort, pain, edema, or erythema, making this problem particularly acute in the neonatal population. They are the most vulnerable patient population and are representative of all who are at risk for catheter infusion iatrogenic injury. Therefore, it would be desirable for an intravenous pump system to be outfitted with an alarm which can alert a health care provider to impending catheter infusion failure in infants and other patients.

[0006] There have been few published studies of monitoring as a way to detect problems with catheter infusions. Phelps et al. studied intraluminal pressure in 44 infants and resistance in 52 infants under one year of age (median, 1 month) and found that intraluminal pressure changes predicted only 25% and resistance changes predicted 4% of PIV infiltrations. Fluid flow in the human intravenous system has been represented by a model which depicts physical devices as ideal resistors, pressure sources, and flow sources, and the venous system as a combination of ordinary and Starling resistors (Philip, 1989). Normal intravenous flows can be represented by the pressure-flow relationship where pressure=flow x resistance. However, it has been suggested that the current infusion devices may not behave as predicted during adverse situations because of physiological changes, which affect the pressure-flow relationship. During extravasation, tissue pressure is lower than venous pressure until the fluid fills the tissue compartment. This may explain why pressure or resistance threshold-based alarms have not previously been shown to be predictive of intravenous infiltration but may ring only after the extravasation, when the pressure in the tissue compartment has increased.SUMMARY

[0007] This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This summary is not intended to identify key features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.Docket No. 3399-P46WO

[0008] In some embodiments, a computer-implemented method of automatically monitoring a status of a catheter infusion is provided. An alarm controller receives a time series of pressure values from a pressure sensor associated with a pump operatively coupled to a catheter. The alarm controller generates one or more features based on the time series of pressure values. The alarm controller provides the one or more features to a machine learning model to generate a predicted confidence of impending failure of the catheter infusion. In response to determining that the predicted confidence of impending failure of the catheter infusion is greater than a predetermined threshold, the alarm controller causes an alert to be presented. In some embodiments, a non-transitory computer-readable medium is provided that has computer-executable instructions stored thereon that, in response to execution by one or more processors of an alarm controller, cause the alarm controller to perform such a method. In some embodiments, a system is provided that comprises a pump configured to be operably coupled to a catheter, a pressure sensor, and a pump system. The pump system is communicatively coupled to the pump, and includes an alarm controller. The alarm controller has at least one processor and a non-transitory computer-readable medium having computer-executable instructions stored thereon that, in response to execution by the at least one processor, cause the alarm system to perform a such a method.

[0009] In some embodiments, a computer-implemented method of training a machine learning model to generate a predicted confidence of impending failure of a catheter infusion is provided. A computing system receives a set of labeled time series of values of pressure data generated by a pressure sensor. The computing system generates features from each labeled time series of values of the set of labeled time series of values to create a set of training data. The computing system trains the machine learning model using the set of training data, and stores the trained machine learning model in a model data store. In some embodiments, a non-transitory computer-readable medium is provided that has computerexecutable instructions stored thereon that, in response to execution by one or more processors of an alarm controller, cause the alarm controller to perform such a method.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The many of the attendant advantages of this invention will become more readily appreciated as the same become better understood by reference to the following detailed description, when taken in conjunction with the accompanying drawings, wherein:

[0011] FIG. 1 is a schematic diagram that illustrates the use of a non-limiting example embodiment of a self-monitoring infusion catheter and pump system according to various aspects of the present disclosure.Docket No. 3399-P46WO

[0012] FIG. 2 is a block diagram that illustrates component details of a non-limiting example embodiment of a self-monitoring catheter infusion system according to various aspects of the present disclosure.

[0013] FIG. 3A to FIG. 3B are a flowchart that illustrates an example embodiment of a method of automatically monitoring a status of catheter infusion according to various aspects of the present disclosure.

[0014] FIG. 4 is a flowchart that illustrates a non-limiting example embodiment of a method of training a machine learning model to generate a predicted confidence of impending failure of a catheter infusion according to various aspects of the present disclosure.

[0015] FIG. 5 is a block diagram that illustrates a non-limiting example embodiment of a computing device appropriate for use as a computing device with embodiments of the present disclosure.DETAILED DESCRIPTION

[0016] In some embodiments of the present disclosure, a self-monitoring catheter infusion system is provided. A time series of pressure values from an in-line pressure sensor may be processed using a machine learning model to detect impending catheter infusion failure. Such techniques can help avoid injury to patients caused by failed catheters, particularly in neonates and other patients that cannot communicate or understand the signs of catheter failure. Such techniques allow the nurse or other care provider the opportunity to intervene before injury or missed therapies occur. Such techniques can also improve the administration of care, in that the automated detection of impending failure, instead of requiring a physical examination to detect failure, allows the pump system to automatically generate alarms, and allows electronic alerts to automatically be transmitted from the pump system to a remote monitoring location. By conducting the machine learning processing of pressure values, the self-monitoring catheter infusion system is given new functionality that was not currently present in catheter infusion systems, at least in that previous systems could not accurately predict or detect impending catheter infusion failure, particularly in neonates.

[0017] FIG. 1 is a schematic diagram that illustrates the use of an example embodiment of a self-monitoring catheter infusion and pump system according to various aspects of the present disclosure. In some respects, the self-monitoring catheter infusion and pump system 100 is similar to other infusion pump systems. For example, the system 100 includes a pumpDocket No. 3399-P46WO system 102 and a pump 104. The pump 104 performs the physical act of pumping fluid through a catheter, while the pump system 102 provides control signals for controlling operation of the pump 104. The catheter 108 is operably coupled to the pump 104, such as having the catheter 108 physically threaded through a mechanism of the pump 104. A proximal end of the catheter 108 is coupled to a fluid bag 110. A distal end of the catheter 108 is inserted into a blood vessel of a patient 90. In use, the pump 104 applies pressure to fluid that enters the catheter from the fluid bag 110, such as medication or parenteral nutrition solution, to cause the fluid to travel through the catheter 108 and into the patient 90. The pump system 102 may have a variety of user interface devices thereon for controlling operation of the pump system 102 such as buttons or dials. The pump system 102 may also include a display 106 for presenting visual indicators of pump status, pressure values, resistance values, configuration values, alerts, warnings, and / or the like. Further details regarding the system 100 are described below.

[0018] FIG. 2 is a block diagram that illustrates component details of an example embodiment of a self-monitoring catheter infusion system according to various aspects of the present disclosure. As shown, the system 100 includes a pump system 102 and a catheter 108. A pump controller (not illustrated) of the pump system 102 controls operation of a pump 104. The pump 104 includes a mechanism (not illustrated) that physically interacts with the catheter 108 to pump fluid through the catheter 108. The pump 104 includes a pressure sensor 208 configured to determine an amount of back pressure that is generated in the distal portion of the catheter 108. This allows the pump system 102 to monitor and control the amount of fluid that is delivered. In some embodiments, the pressure sensor 208 may be located outside of the pump 104, instead of inside of the pump 104 as illustrated.

[0019] In some embodiments, the catheter 108 may be any type of catheter, including but not limited to a peripheral intravenous catheter, a peripherally inserted central catheter (PICC), a surgically placed central venous catheter (e.g., Hickman, Broviac, Swan-Ganz, etc.), a peripheral arterial catheter, a surgically placed arterial catheter, an Extra Corporeal Membrane Oxygenation (ECMO) catheter, an Extra Corporeal Life Support (ECLS) catheter, a dialysis catheter, or a regional anesthetic infusion catheter (e.g., spinal, epidural, perineural, etc ).

[0020] As shown, the pump system 102 includes an alarm controller 210, a memory’ 212. a network interface 214, and one or more alarm presentation devices 216. In some embodiments, both the network interface 214 and the alarm presentation device 216 may be present. In some embodiments, only one of the network interface 214 or the alarm presentation device 216 may be present.Docket No. 3399-P46WO

[0021] In some embodiments, the alarm controller 210 includes one or more computer processors configured to receive signals from the pressure sensor 208 (and, optionally, other sensors that are not illustrated) and to process the signals to determine a state of the catheter 108 and predict impending failure of the catheter infusion. In some embodiments, the processors of the alarm controller 210 may also be configured to cause visual or audible notifications to be presented based on the determined state of the catheter 108. In some embodiments, the alarm controller 210 may include an ASIC, an FPGA, or other hardware device that has been configured to perform the tasks described herein. In some embodiments, the alarm controller 210 may be a general-purpose processor that is configured to read instructions from a computer-readable medium to cause the alarm controller 210 to perform the tasks described herein. The computer- readable medium may be any ty pe of computer-readable medium including but not limited to a ROM, an EEPROM, a flash memory, or a magnetic drive.

[0022] In some embodiments, the memory 212 is configured to store sensor values and other values determined by the alarm controller 210, and to provide the stored values to the alarm controller 210 when requested. In some embodiments, the memory 212 may include any suitable form of computer-readable medium, including but not limited to a flash memory, a magnetic drive, or RAM.

[0023] In some embodiments, the alarm presentation device 216 is configured to receive commands from the alarm controller 210 to present visual or audible notifications. In some embodiments, one or more alarm presentation devices 216 may be present. Some nonlimiting examples of alarm presentation devices 216 include an LCD display, a touchscreen display, an indicator light, an LED display, or a loudspeaker.

[0024] In some embodiments, the network interface 214 is configured to receive commands from the alarm controller 210 for visual or audible notifications to be presented, and to transmit the commands to a remote computing device 218 via a network 92. The remote computing device 218 may then present the notifications on a display or via a loudspeaker. The remote computing device 218 may also store the notifications for future reference. The network 92 may include any suitable types of wired or wireless network communication technologies, including but not limited to Ethernet, USB, Firewire, Bluetooth, Wi-Fi, WiMAX. 3G, 4G, LTE, and the Internet, and the network interface 214 may be configured to communicate via one or more of these technologies. In some embodiments, the remote computing device 218 may be a desktop computing device, a laptop computing device, a mobile personal device (e.g., a tablet computing device, a smartphone, a smart watch, smart eyeglasses, etc.), an electronic medical record system, aDocket No. 3399-P46WO clinical intelligence platform, a server computing device, or a cloud computing service. In some embodiments, the pump system 102 may be located in a room with a patient, and the remote computing device 218 may be located at a remote monitoring station, such as a nurses' station or a telehealth service center.

[0025] A detailed description of processing performed by an example embodiment of the alarm controller 210 is provided below.

[0026] FIG. 3A to FIG. 3B are a flowchart that illustrates an example embodiment of a method of automatically monitoring a status of catheter infusion according to various aspects of the present disclosure. From a start block, the method 300 proceeds to block 302, where a proximal portion of a catheter 108 is operably coupled to a pump system 102. In some embodiments, the proximal portion of the catheter 108 may be fed through or placed within physical pump 104 as illustrated in FIG. 1. Though not illustrated in the flowchart, it is assumed that a furthest proximal end of the catheter 108 is connected to a fluid source to be administered. In some embodiments, operably coupling the catheter 108 to the pump system 102 includes communicatively coupling one or more additional sensors associated with the catheter 108 to the alarm controller 210. The communicative coupling may use any suitable technique, including but not limited to physically coupling conductors to an interface within the pump system 102, wirelessly pairing sensors with the pump system 102, or locating conductors associated with the additional sensors within the pump system such that the pump system 102 can inductively detect the signals generated in the conductors without a direct electrical connection.

[0027] At block 304, the distal end of the catheter 108 is inserted into a patient. Techniques known to those of ordinary skill in the art for inserting catheters 108 into patients may be used. Further, traditional techniques known to those of ordinary skill in the art may be used to verify that the initial insertion of the catheter 108 was successful.

[0028] The method 400 then proceeds through a continuation terminal ("terminal A”) to block 306, where an alarm controller 210 of the pump system 102 receives a time series of pressure values from a pressure sensor 208 associated with a pump 104 of the pump system 102. In some embodiments, the pressure values may be represented by signals that indicate analog values which are converted to digital values by the alarm controller 210. In some embodiments, analog signals may pass through an A / D converter (not illustrated) before being received by the alarm controller 210. At block 308, the alarm controller 210 stores the time series of pressure values in a memory' 212 of the pump system 102.Docket No. 3399-P46WO

[0029] At block 310, the alarm controller 210 generates one or more features based on the time series of pressure values, and stores the one or more features in the memory 212. Any suitable types of features may be used. Some non-limiting examples of suitable features may include, but are not limited to, at least some of the pressure values of the time series of pressure values, a first derivative or a second derivative of the pressure signal, and / or a frequency-domain representation of the pressure signal obtained using a Discrete Fourier Transform (DFT). The derivatives may be determined using finite differencing, or any other suitable technique. The DFT may be determined using a window interval based on a sampling rate of the time series of pressure values. In some embodiments, one or more features that represent the time series of pressure values overall may be included, including but not limited to a minimum pressure value, a maximum pressure value, or an average pressure value.

[0030] The method 300 then proceeds to optional block 312, where the alarm controller 210 receives contextual data and stores the contextual data in the memory' 212. The contextual data may include information relevant to the catheter infusion, and may include one or more of a catheter infusion age. a patient age, a catheter size, a ty pe of infusate, a setting of the pump 104 (e.g.. an IV rate), and / or a type of vessel (e.g., arterial or venous). In some embodiments, some types of contextual data may be received from a sensor. In some embodiments, some types of contextual data may be input by an operator or retrieved from an electronic health record (EHR) system. The contextual data may be individual scalar values, may be provided as a time series of values, or may be provided as any other format. At optional block 314, the alarm controller 210 generates one or more contextual features based on the contextual data, and stores the contextual features in the memory' 212. Optional block 312 and optional block 314 are illustrated as optional because in some embodiments, the method 300 may not receive or use any contextual data, and instead may monitor the status of the catheter infusion based on the time series of pressure values alone.

[0031] The method 300 then advances to a continuation terminal ("terminal B"). From terminal B (FIG. 3B), the method 300 proceeds to block 316, where the alarm controller 210 provides the one or more features and. optionally, the one or more contextual features, as input to a machine learning model to generate a predicted confidence of impending failure.

[0032] Any suitable machine learning architecture for detecting patterns in time series data may be used for the machine learning model. In some embodiments, the machine learning architecture may include, but is not limited to, one or more of a decision tree or an artificial neural network (e.g., a multi-layer perceptron (MLP), a convolutional neural network (CNN), and / or a recurrent neural network (RNN), etc.). In some embodiments, ensembleDocket No. 3399-P46WO strategies including but not limited to one or more of bagging, boosting, and / or stacking may be used to combine results from multiple models. The machine learning model may be trained using any suitable technique, including but not limited to the method 400 illustrated in FIG. 4 and described in further detail below.

[0033] The output of the machine learning model is based on both a prediction of impending failure and a level of certainty in the prediction. Techniques for quantifying model prediction certainty such as Monte Carlo Dropout or Bayesian learning may be used to generate the level of certainty in the prediction. If the machine learning model is more confident that the input features indicate impending failure (or that failure has already occurred), then the output value of the machine learning model may be a higher value. Likewise, if the machine learning model is less confident that the input features indicate impending failure (or that the catheter infusion has not failed), then the output value of the machine learning model may be a lower value, thereby reducing the number of false alarms generated.

[0034] The method 300 then proceeds to decision block 318, where a determination is made regarding whether the output of the machine learning model indicates impending failure. In some embodiments, the determination is made by comparing the predicted confidence of impending failure output by the machine learning model to a threshold confidence value.

[0035] If it is determined that the output of the machine learning model indicates impending failure, then the result of decision block 318 is YES, and the method 300 proceeds to block 320. At block 320. the alarm controller 210 causes an alert to be presented. The alarm controller 210 may cause the alert to be presented by transmitting a command to one or more alarm presentation devices 216 of the pump system 102, by transmitting a command to a remote computing device 218 via the network interface 214 of the pump system 102, or both. The alert itself could be a visual indicator to be presented by an indicator light or a display, a tone to be played by a loudspeaker, or both. In some embodiments, causing the alert to be presented may include automatically ceasing operation of the pump 104 to automatically prevent an injury. The method 300 then proceeds to decision block 322.

[0036] Returning to decision block 318, if it is determined that the output of the machine learning model does not indicate impending failure, then the result of decision block 318 is NO, and the method 300 proceeds directly to decision block 322. At decision block 322, a determination is made regarding whether the method 300 should continue. In someDocket No. 3399-P46WO embodiments, the method 300 may continue as long as the pump 104 is operating. In some embodiments, the method 300 may continue until an alarm is presented. In some embodiments, the method 300 may continue until an operator indicates that the method 300 should cease.

[0037] If it is determined that the method 300 should continue, then the result of decision block 322 is YES, and the method 300 returns to terminal A to continue monitoring the catheter infusion. Otherwise, if it is determined that the method 300 should cease, then the result of decision block 322 is NO. and the method 300 proceeds to an end block and terminates.

[0038] FIG. 4 is a flowchart that illustrates a non-limiting example embodiment of a method of training a machine learning model to generate a predicted confidence of impending failure of a catheter infusion according to various aspects of the present disclosure. The method 400 is performed by a computing system, which may include one or more computing devices 500 as illustrated in FIG. 5 and described in further detail below.

[0039] From a start block, the method 400 proceeds to block 402, where a computing system receives a set of labeled time series of values of pressure data generated by a pressure sensor. The set of labeled time series of values includes a plurality of separate time series of values, with each time series of values being labeled as being associated with a catheter infusion failure or as not being associated with a catheter infusion failure. In some embodiments, a label may be applied by a clinician upon review of a record of why the catheter infusion was ceased. In some embodiments, a label may be applied by automatically analyzing a medical record associated with the catheter infusion represented by the time series of values. In some embodiments, a label may be applied by automatically analyzing data collected from other sensors, including one or more of arterial pressure data, other vital sign data (e.g., heart rate and / or respiration rate), oxygen level data, end-tidal Ch and / or CCh data, near-infrared spectroscopy (NIRS) data, and / or other sensor data collected in an electronic medical record system or another t pe of data analysis or data collection system. In some such embodiments, the label may be determined by comparing one or more sensor values to appropriate thresholds that indicate a presence or absence of catheter infusion failure. In some embodiments, a label may be applied by reviewing the time series of values and determining whether values greater than a pressure threshold value that indicates a failure are present. In some embodiments, the time series of values may be generated by and received from the same pressure sensor 208 to be used when executing the trained machine learning model. In some embodiments, the time series of values may beDocket No. 3399-P46WO generated by a pressure sensor 208 of the same type as a pressure sensor 208 to be used with the trained machine learning model.

[0040] At block 404, the computing system generates features from each labeled time series of values of the set of labeled time series of values to create a set of training data. In some embodiments, similar techniques may be used to generate the features as used at block 310, including but not limited to using at least some of the pressure values of the time series of values as features, generating a first derivative of the pressure signal, generating a second derivative of the pressure signal, or generating a frequency-domain representation of the pressure signal obtained using a Discrete Fourier Transform (DFT). The derivatives may be determined using finite differencing, or any other suitable technique. The DFT may be determined using a window interval based on a sampling rate of the time series of pressure values. In some embodiments, one or more features that represent the time series of pressure values overall may be included, including but not limited to a minimum pressure value, a maximum pressure value, or an average pressure value.

[0041] At optional block 406, the computing system generates additional contextual features for inclusion in the set of training data. In some embodiments, the additional contextual features may include features related to one or more of a pump setting (e.g., an IV rate), a catheter infusion age, a patient age, a catheter size, a ty pe of infusate, and / or a type of vessel (e.g., arterial or venous). The actions of optional block 406 are described as optional because in some embodiments, the machine learning model may be trained without features based on contextual data.

[0042] At optional block 408, the computing system truncates at least one time series of values labeled as associated with a catheter infusion failure at a point where a threshold pressure has been reached that indicates that an infusion failure has already occurred. In some embodiments, a pressure that is greater than a high threshold pressure may be sensed when, e.g., the catheter 108 is clogged or more than a desirable amount of infusion fluid has been infused. In some embodiments, a pressure that is less than a low threshold pressure may be sensed when, e.g., a distal tip of the catheter 108 passes out of a blood vessel. The time series of values may continue after this detectable failure state occurs, particularly if failure of the catheter infusion during which the time series of values was collected had not been detected until well after the failure. Since a goal of the present disclosure is to train the machine learning model to detect patterns that occur before failure of the catheter infusion, allowing a significant number of values to remain in the training data from after the catheter infusion failure may cause the machine learning model to be less effective in detecting prefailure patterns. Accordingly, truncating values from the time series of values from after theDocket No. 3399-P46WO threshold pressure is reached can avoid having a significant number of post-failure values remaining in the time series. It can also lead to more effective training of the machine learning model, and more accurate detection of patterns that appear in the time series of values prior to catheter infusion failure. The actions of optional block 408 are illustrated as optional because in some embodiments, the time series of values provided to the method 400 may be collected in a way that avoids collecting a significant number of post-failure values, or the machine learning model may be trained in a way that is not significantly affected by post-failure values.

[0043] At block 410, the computing system trains the machine learning model using the set of training data. As discussed above, any suitable machine learning model architecture for detecting patterns in time series data may be used, including but not limited to one or more of a decision tree architecture, an artificial neural network, an MLP architecture, a CNN architecture, and / or an RNN architecture. In some embodiments, the RNN architecture may include a long short-term memory (LSTM) model. In some embodiments, the RNN architecture may include a gated recurrent unit (GRU). As also discussed above, in some embodiments, ensemble techniques including but not limited to bagging, boosting, and / or stacking may be used. The machine learning model may be trained using any suitable technique, including but not limited to gradient descent and / or an Adam optimizer.

[0044] In some embodiments, training the machine learning model may include various techniques for improving the training of the model. For example, in some embodiments, the computing system may perform hyperparameter tuning during the training of the machine learning model, which may include tuning hyperparameters including but not limited to one or more of a number of hidden layers, a layer width, or a ty pe of activation function. The hyperparameters may be tuned iteratively using the prediction accuracy on the validation set as the performance metric.

[0045] At block 412. the computing system stores the trained machine learning model in a model data store, and at block 414. the computing system transmits the trained machine learning model to an alarm controller 210. In some embodiments, the computing system may train multiple machine learning models and store the machine learning models in the model data store, and the computing system may determine an appropriate machine learning model to transmit to the alarm controller 210 (e g., a machine learning model trained for the type of pump 104. catheter 108, or pressure sensor 208 associated with the alarm controller 210; a machine learning model trained using similar contextual information as associated with the alarm controller 210, etc.).Docket No. 3399-P46WO

[0046] As used herein, "data store" refers to any suitable device configured to store data for access by a computing device. One example of a data store is a highly reliable, highspeed relational database management system (DBMS) executing on one or more computing devices and accessible over a high-speed network. Another example of a data store is a keyvalue store. However, any other suitable storage technique and / or device capable of quickly and reliably providing the stored data in response to queries may be used, and the computing device may be accessible locally instead of over a network, or may be provided as a cloud-based service. A data store may also include data stored in an organized manner on a computer-readable storage medium, such as a hard disk drive, a flash memory, RAM. ROM, or any other type of computer-readable storage medium. One of ordinary skill in the art will recognize that separate data stores described herein may be combined into a single data store, and / or a single data store described herein may be separated into multiple data stores, without departing from the scope of the present disclosure.

[0047] The method 400 then proceeds to an end block and terminates.

[0048] FIG. 5 is a block diagram that illustrates aspects of an example computing device 500 appropriate for use in a computing system of the present disclosure. The example computing device 500 describes various elements that are common to many different types of computing devices. While FIG. 5 is described with reference to a computing device that is implemented as a device on a network, the description below is applicable to servers, personal computers, mobile phones, smart phones, tablet computers, embedded computing devices, and other devices that may be used to implement portions of embodiments of the present disclosure. Some embodiments of a computing device may be implemented in or may include an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other customized device. Moreover, those of ordinary skill in the art and others will recognize that the computing device 500 may be any one of any number of currently available or yet to be developed devices.

[0049] In its most basic configuration, the computing device 500 includes at least one processor 502 and a system memory 510 connected by a communication bus 508. Depending on the exact configuration and type of device, the system memory 510 may be volatile or nonvolatile memory, such as read only memory' (“ROM”), random access memory (“RAM”), EEPROM, flash memory, or similar memory technology’. Those of ordinary skill in the art and others will recognize that system memory 510 typically stores data and / or program modules that are immediately accessible to and / or currently being operated on by the processor 502. In this regard, the processor 502 may serve as aDocket No. 3399-P46WO computational center of the computing device 500 by supporting the execution of instructions.

[0050] As further illustrated in FIG. 5, the computing device 500 may include a network interface 506 comprising one or more components for communicating with other devices over a network. Embodiments of the present disclosure may access basic services that utilize the network interface 506 to perform communications using common network protocols. The network interface 506 may also include a wireless network interface configured to communicate via one or more wireless communication protocols, such as WiFi, 2G, 3G, LTE, WiMAX, Bluetooth, Bluetooth low energy, and / or the like. As will be appreciated by one of ordinary skill in the art, the network interface 506 illustrated in FIG. 5 may represent one or more wireless interfaces or physical communication interfaces described and illustrated above with respect to particular components of the computing device 500.

[0051] In the example embodiment depicted in FIG. 5, the computing device 500 also includes a storage medium 504. However, services may be accessed using a computing device that does not include means for persisting data to a local storage medium. Therefore, the storage medium 504 depicted in FIG. 5 is represented with a dashed line to indicate that the storage medium 504 is optional. In any event, the storage medium 504 may be volatile or nonvolatile, removable or nonremovable, implemented using any technology capable of storing information such as, but not limited to, a hard drive, solid state drive. CD ROM, DVD, or other disk storage, magnetic cassettes, magnetic tape, magnetic disk storage, and / or the like.

[0052] Suitable implementations of computing devices that include a processor 502, system memory 510, communication bus 508, storage medium 504, and network interface 506 are known and commercially available. For ease of illustration and because it is not important for an understanding of the claimed subject matter, FIG. 5 does not show some of the typical components of many computing devices. In this regard, the computing device 500 may include input devices, such as a keyboard, keypad, mouse, microphone, touch input device, touch screen, tablet, and / or the like. Such input devices may be coupled to the computing device 500 by wired or wireless connections including RF, infrared, serial, parallel, Bluetooth, Bluetooth low energy, USB, or other suitable connections protocols using wireless or physical connections. Similarly, the computing device 500 may also include output devices such as a display, speakers, printer, etc. Since these devices are well known in the art, they are not illustrated or described further herein.Docket No. 3399-P46WO

[0053] The particulars shown herein are by way of example and for purposes of illustrative discussion of example embodiments of the present disclosure only and are presented in the cause of providing what is believed to be the most useful and readily understood description of the principles and conceptual aspects of various embodiments of the disclosure. In this regard, no attempt is made to show structural details in more detail than is necessary for the fundamental understanding of the disclosure, the description taken with the drawings and / or examples making apparent to those skilled in the art how the several forms of the disclosure may be embodied in practice.

[0054] As used herein and unless otherwise indicated, the terms “a” and “an” are taken to mean “one”, “at least one” or “one or more”. Unless otherwise required by context, singular terms used herein shall include pluralities and plural terms shall include the singular.

[0055] Unless the context clearly requires otherwise, throughout the description and the claims, the words ‘comprise’, ‘comprising’, and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense; that is to say, in the sense of “including, but not limited to”. Words using the singular or plural number also include the plural and singular number, respectively. Additionally, the words “herein,” “above,” and “below” and words of similar import, when used in this application, shall refer to this application as a whole and not to any particular portions of the application.

[0056] The description of embodiments of the disclosure is not intended to be exhaustive or to limit the disclosure to the precise form disclosed. While the specific embodiments of, and examples for, the disclosure are described herein for illustrative purposes, various equivalent modifications are possible within the scope of the disclosure, as those skilled in the relevant art will recognize.

[0057] All of the references cited herein are incorporated by reference. Aspects of the disclosure can be modified, if necessary, to employ the systems, functions, and concepts of the above references and application to provide yet further embodiments of the disclosure. These and other changes can be made to the disclosure in light of the detailed description.

[0058] Specific elements of any foregoing embodiments can be combined or substituted for elements in other embodiments. Moreover, the inclusion of specific elements in at least some of these embodiments may be optional, wherein further embodiments may include one or more embodiments that specifically exclude one or more of these specific elements. Furthermore, while advantages associated with certain embodiments of the disclosure have been described in the context of these embodiments, other embodiments may also exhibitDocket No. 3399-P46WO such advantages, and not all embodiments need necessarily exhibit such advantages to fall within the scope of the disclosure.EXAMPLES

[0059] The following paragraphs provide a numbered set of non-limiting example embodiments of the present disclosure.

[0060] Example 1 : A computer-implemented method of automatically monitoring a status of a catheter infusion, the method comprising: receiving, by an alarm controller, a time series of pressure values from a pressure sensor associated with a pump operatively coupled to a catheter; generating, by the alarm controller, one or more features based on the time series of pressure values; providing, by the alarm controller, the one or more features to a machine learning model to generate a predicted confidence of impending failure of the catheter infusion; and in response to determining that the predicted confidence of impending failure of the catheter infusion is greater than a predetermined threshold, causing, by the alarm controller, an alert to be presented.

[0061] Example 2: The computer-implemented method of example 1, wherein generating one or more features based on the time series of pressure values includes at least one of; generating one or more features that include pressure values of the time series of pressure values; determining a first derivative of the pressure values using finite differencing; determining a second derivative of the pressure values using finite differencing; or determining a frequency -domain Discrete Fourier Transform (DFT) using a window interval determined based on a sampling rate of the pressure values.

[0062] Example 3: The computer-implemented method of any one of examples 1-2, wherein the machine learning model includes an artificial neural network.

[0063] Example 4: The computer-implemented method of example 3, wherein the artificial neural network uses a multi-layer perceptron (MLP) architecture, a convolutional neural network (CNN) architecture, or a recurrent neural network (RNN) architecture.

[0064] Example 5: The computer-implemented method of example 4, wherein the RNN uses a long short-term memory (LSTM) architecture or a gated recurrent unit (GRU) architecture.

[0065] Example 6: The computer-implemented method of any one of examples 3-4, wherein the machine learning model uses an ensemble technique.

[0066] Example 7: The computer-implemented method of any one of examples 1-6. wherein providing the one or more features to the machine learning model to generate theDocket No. 3399-P46WO predicted confidence of impending failure of the catheter infusion includes providing additional contextual features to the machine learning model.

[0067] Example 8: The computer-implemented method of example 7, wherein the additional contextual features include features related to one or more of a pump setting, a catheter infusion age, a patient age, an infusate type, a vessel type, or a catheter size.

[0068] Example 9: The computer-implemented method of any one of examples 1-8. wherein the catheter is a peripheral intravenous catheter, a peripherally inserted central catheter (PICC), a surgically placed central venous catheter, a peripheral arterial catheter, a surgically placed arterial catheter, an Extra Corporeal Membrane Oxygenation (ECMO) catheter, an Extra Corporeal Life Support (ECLS) catheter, a dialysis catheter, or a regional anesthetic infusion catheter.

[0069] Example 10: The computer-implemented method of any one of examples 1-9, wherein causing the alert to be presented includes one or more of transmitting a command to one or more alarm presentation devices of a pump system that includes the alarm controller, transmitting the predicted confidence of impending failure of the catheter infusion to a remote computing device for further or alternative presentation, or automatically ceasing operation of a pump associated with the pump system that includes the alarm controller.

[0070] Example 11 : A computer-implemented method of training a machine learning model to generate a predicted confidence of impending failure of a catheter infusion, the method comprising: receiving, by a computing system, a set of labeled time series of values of pressure data generated by a pressure sensor; generating, by the computing system, features from each labeled time series of values of the set of labeled time series of values to create a set of training data; training, by the computing system, the machine learning model using the set of training data; and storing, by the computing system, the trained machine learning model in a model data store.

[0071] Example 12: The computer-implemented method of example 11, further comprising: transmitting, by the computing system, the trained machine learning model to an alarm controller.

[0072] Example 13: The computer-implemented method of any one of examples 11-12, wherein labels of the set of labeled time series of values indicate whether each time series of values is associated with a catheter infusion failure or is not associated with a catheter infusion failure.

[0073] Example 14: The computer-implemented method of example 13, further comprising: for at least one time series of values labeled as associated with a catheterDocket No. 3399-P46WO infusion failure, truncating the time series of values at a point where a threshold pressure has been reached that indicates the failure has already happened.

[0074] Example 15: The computer-implemented method of any one of examples 11-14, wherein the features include at least one of: pressure values of the time series of values; a first derivative of the pressure values of the time series of values determined using finite differencing; a second derivative of the pressure values of the time series of values determined using finite differencing; or a frequency -domain Discrete Fourier Transform (DFT) using a window interval determined based on a sampling rate of the pressure values of the time series of values.

[0075] Example 16: The computer-implemented method of any one of examples 11-15, wherein the machine learning model includes an artificial neural network.

[0076] Example 17: The computer-implemented method of example 16, wherein the artificial neural network uses a multi-layer perceptron (MLP) architecture, a convolutional neural network (CNN) architecture, or a recurrent neural network (RNN) architecture.

[0077] Example 18: The computer-implemented method of example 17, wherein the RNN uses a long short-term memory (LSTM) architecture or a gated recurrent unit (GRU) architecture.

[0078] Example 19: The computer-implemented method of any one of examples 16-17, wherein the machine learning model uses an ensemble technique.

[0079] Example 20: The computer-implemented method of any one of examples 11-19, wherein generating features from each labeled time series of values of the set of labeled time series of values to create a set of training data includes providing additional contextual features for inclusion in the set of training data.

[0080] Example 21 : The computer-implemented method of example 20, wherein the additional contextual features include features related to one or more of a pump setting, a catheter infusion age, a patient age, an infusate type, a vessel type, or a catheter size.

[0081] Example 22: The computer-implemented method of any one of examples 11-21, wherein training the machine learning model using the set of training data includes performing hyperparameter tuning.

[0082] Example 23: The computer-implemented method of example 22, wherein the hyperparameters include one or more of a number of hidden layers, a layer width, or a type of activation function.

[0083] Example 24: A system, comprising: a pump configured to be operably coupled to a catheter; a pressure sensor; a pump system communicatively coupled to the pump, the pumpDocket No. 3399-P46WO system including an alarm controller having at least one processor and a non-transitory computer-readable medium having computer-executable instructions stored thereon that, in response to execution by the at least one processor, cause the alarm system to perform a method of monitoring a catheter infusion performed using the catheter and the pump as recited in any one of example 1 to example 10.

[0084] Example 25: A non-transitory computer-readable medium having computerexecutable instructions stored thereon that, in response to execution by one or more processors of an alarm controller, cause the alarm controller to perform actions of a method as recited in any one of example 1 to example 10.

[0085] Example 26: A non-transitory computer-readable medium having computerexecutable instructions stored thereon that, in response to execution by one or more processors of a computing system, cause the computing system to perform actions of a method as recited in any one of example 11 to example 23.

Claims

Docket No. 3399-P46WOCLAIMSThe embodiments of the invention in which an exclusive property or privilege is claimed are defined as follows:

1. A computer-implemented method of automatically monitoring a status of a catheter infusion, the method comprising: receiving, by an alarm controller, a time series of pressure values from a pressure sensor associated with a pump operatively coupled to a catheter; generating, by the alarm controller, one or more features based on the time series of pressure values; providing, by the alarm controller, the one or more features to a machine learning model to generate a predicted confidence of impending failure of the catheter infusion; and in response to determining that the predicted confidence of impending failure of the catheter infusion is greater than a predetermined threshold, causing, by the alarm controller, an alert to be presented.

2. The computer-implemented method of claim 1. wherein generating one or more features based on the time series of pressure values includes at least one of: generating one or more features that include pressure values of the time series of pressure values; determining a first derivative of the pressure values using finite differencing; determining a second derivative of the pressure values using finite differencing; or determining a frequency -domain Discrete Fourier Transform (DFT) using a window interval determined based on a sampling rate of the pressure values.

3. The computer-implemented method of claim 1, wherein the machine learning model includes an artificial neural network.

4. The computer-implemented method of claim 3. wherein the artificial neural network uses a multi-layer perceptron (MLP) architecture, a convolutional neural network (CNN) architecture, or a recurrent neural network (RNN) architecture.

5. The computer-implemented method of claim 4. wherein the RNN uses a long short-term memory (LSTM) architecture or a gated recurrent unit (GRU) architecture.

6. The computer-implemented method of claim 3, wherein the machine learning model uses an ensemble technique.Docket No. 3399-P46WO7. The computer-implemented method of claim 1. wherein providing the one or more features to the machine learning model to generate the predicted confidence of impending failure of the catheter infusion includes providing additional contextual features to the machine learning model.

8. The computer-implemented method of claim 7, wherein the additional contextual features include features related to one or more of a pump setting, a catheter infusion age, a patient age, an infusate type, a vessel type, or a catheter size.

9. The computer-implemented method of claim 1, wherein the catheter is a peripheral intravenous catheter, a peripherally inserted central catheter (PICC), a surgically placed central venous catheter, a peripheral arterial catheter, a surgically placed arterial catheter, an Extra Corporeal Membrane Oxygenation (ECMO) catheter, an Extra Corporeal Life Support (ECLS) catheter, a dialysis catheter, or a regional anesthetic infusion catheter.

10. The computer-implemented method of claim 1, wherein causing the alert to be presented includes one or more of transmitting a command to one or more alarm presentation devices of a pump system that includes the alarm controller, transmitting the predicted confidence of impending failure of the catheter infusion to a remote computing device for further or alternative presentation, or automatically ceasing operation of a pump associated with the pump system that includes the alarm controller.

11. A computer-implemented method of training a machine learning model to generate a predicted confidence of impending failure of a catheter infusion, the method comprising: receiving, by a computing system, a set of labeled time series of values of pressure data generated by a pressure sensor; generating, by the computing system, features from each labeled time series of values of the set of labeled time series of values to create a set of training data; training, by the computing system, the machine learning model using the set of training data; and storing, by the computing system, the trained machine learning model in a model data store.

12. The computer-implemented method of claim 11, further comprising: transmitting, by the computing system, the trained machine learning model to an alarm controller.Docket No. 3399-P46WO13. The computer-implemented method of claim 11, wherein labels of the set of labeled time series of values indicate whether each time series of values is associated with a catheter infusion failure or is not associated with a catheter infusion failure.

14. The computer-implemented method of claim 13, further comprising: for at least one time series of values labeled as associated with a catheter infusion failure, truncating the time series of values at a point where a threshold pressure has been reached that indicates the failure has already happened.

15. The computer-implemented method of claim 11, wherein the features include at least one of: pressure values of the time series of values; a first derivative of the pressure values of the time series of values determined using finite differencing; a second derivative of the pressure values of the time series of values determined using finite differencing; or a frequency-domain Discrete Fourier Transform (DFT) using a window interval determined based on a sampling rate of the pressure values of the time series of values.

16. The computer-implemented method of claim 11, wherein the machine learning model includes an artificial neural network.

17. The computer-implemented method of claim 16, wherein the artificial neural network uses a multi-layer perceptron (MLP) architecture, a convolutional neural network (CNN) architecture, or a recurrent neural netw ork (RNN) architecture.

18. The computer-implemented method of claim 17, wherein the RNN uses a long shortterm memory (LSTM) architecture or a gated recurrent unit (GRU) architecture.

19. The computer-implemented method of claim 16, wherein the machine learning model uses an ensemble technique.

20. The computer-implemented method of claim 11, wherein generating features from each labeled time series of values of the set of labeled time series of values to create a set of training data includes providing additional contextual features for inclusion in the set of training data.Docket No. 3399-P46WO21. The computer-implemented method of claim 20, wherein the additional contextual features include features related to one or more of a pump setting, a catheter infusion age, a patient age, an infusate type, a vessel type, or a catheter size.

22. The computer-implemented method of claim 11, wherein training the machine learning model using the set of training data includes performing hyperparameter tuning.

23. The computer-implemented method of claim 22, wherein the hyperparameters include one or more of a number of hidden layers, a layer width, or a type of activation function.

24. A system, comprising: a pump configured to be operably coupled to a catheter; a pressure sensor; a pump system communicatively coupled to the pump, the pump system including an alarm controller having at least one processor and a non-transitory computer-readable medium having computer-executable instructions stored thereon that, in response to execution by the at least one processor, cause the alarm system to perform a method of monitoring a catheter infusion performed using the catheter and the pump as recited in any one of claim 1 to claim 10.

25. A non-transitory computer-readable medium having computer-executable instructions stored thereon that, in response to execution by one or more processors of an alarm controller, cause the alarm controller to perform actions of a method as recited in any one of claim 1 to claim 10.

26. A non-transitory computer-readable medium having computer-executable instructions stored thereon that, in response to execution by one or more processors of a computing system, cause the computing system to perform actions of a method as recited in any one of claim 11 to claim 23.

Images