Acoustic reflectometry for clinical condition determination

Abstract

The technology relates to acoustic reflectometry systems and methods. An example system includes a generator that emits acoustic pulses; an acoustic sensor that detects the emitted acoustic pulses and reflections thereof; a processor; and memory storing instructions that, when executed by the processor, causes the system to perform operations. The operations include receive, from the acoustic sensor, acoustic data detected by the acoustic sensor from a patient; receive clinical data associated with the patient; provide at least a portion of the received acoustic data and the clinical data as an input to a trained machine-learning (ML) model; receive, as output from the trained ML model in response to the input, an indication of a clinical condition; and surface the indication of the clinical condition.

Description

ACOUSTIC REFLECTOMETRY FOR CLINICAL CONDITION DETERMINATIONCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of U.S. Provisional Patent Application Serial No. 63 / 625,150, filed January 25, 2024, the entire contents of which is incorporated herein by reference.BACKGROUND

[0002] Medical ventilators offer critical support for patients who require ongoing breathing assistance. An endotracheal tube (ETT) is often used in conjunction with a ventilator, as part of a breathing circuit. The ETT is placed, via intubation, through patient’s mouth into the trachea, bypassing the upper airway. A number of sensors may be used with the ETT to ensure proper positioning within the trachea and to ensure that the patient is receiving adequate ventilation. Measurement data collected by the sensors may be correlated and may provide further information for monitoring the patient’s respiratory condition. It is with respect to this general technical environment that aspects of the present technology disclosed herein have been contemplated. Furthermore, although a general environment is discussed, it should be understood that the examples described herein should not be limited to the general environment identified herein.SUMMARY

[0003] 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 or essential 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. Additional aspects, features, and / or advantages of examples will be set forth in part in the description which follows and, in part, will be apparent from the description, or may be learned by practice of the disclosure.

[0004] In an aspect, the technology relates to an acoustic reflectometry system. The system includes a generator that emits acoustic pulses; an acoustic sensor that detects the emitted acoustic pulses and reflections thereof; a processor; and memory storing instructions that, when executed by the processor, causes the system to perform operations. The operations includereceive, from the acoustic sensor, acoustic data detected by the acoustic sensor from a patient; receive clinical data associated with the patient; provide at least a portion of the received acoustic data and the clinical data as an input to a trained machine -learning (ML) model; receive, as output from the trained ML model in response to the input, an indication of a clinical condition; and surface the indication of the clinical condition.

[0005] In an example, the acoustic data includes at least one of: (1) a composite acoustic-based signal representing reflections of the emitted acoustic pulses and natural sounds generated by the patient, (2) a fdtered acoustic -based signal representing the reflections of the emitted acoustic pulses, or (3) an acoustically measurable parameter generated from detected reflections of the emitted acoustic pulses. In a further example, the acoustic data includes the acoustically measurable parameter, and the acoustically measurable parameter is one of tracheal tube movement, tracheal tube position, obstruction size, or passageway size. In another example, the acoustic data includes the composite acoustic-based signal. In still another example, the acoustic data includes the filtered acoustic-based signal. In a further example, the operations further comprise selecting a particular time window from the filtered acoustic-based signal for use in the input to the ML model, wherein the particular time window corresponds to anatomical reflections from anatomical structures relevant to the clinical condition. In yet another example, the clinical data includes at least one of capnometry data, pulse oximetry data, electrocardiogram data, or patient movement data. In still yet another example, the clinical data includes ventilation data including at least one of a current ventilation mode, a ventilation setting, a pressure value, a flow value, a respiratory rate, work of breathing, or indication of patient-initiated spontaneous breaths. In yet another example, the clinical condition is one of asthma, pneumonia, pleural effusion, atelectasis, tracheal collapse, or a cardiac condition. In still another example, the indication is an alarm.

[0006] In another aspect, the technology relates to an acoustic reflectometry system. The system includes a generator that emits acoustic pulses; an acoustic sensor that detects the emitted acoustic pulses and reflections thereof; a processor; and memory storing instructions that, when executed by the processor, causes the system to perform operations. The operations include receive, from the acoustic sensor, acoustic data detected by the acoustic sensor, wherein the acoustic data includes: a first portion of the acoustic data from a first time window corresponding to reflections, of the emitted acoustic pulses from, a first anatomical structure; a second portion of the acoustic data from a second time window corresponding to reflections,of the emitted acoustic pulses, from a second anatomical structure. The operations further include provide the first portion of the acoustic data to first machine-learning (ML) model trained to classify a first clinical condition; receive, as output from the first ML model, an indication of the first clinical condition; provide the second portion of the acoustic data to a second ML model trained to classify a second clinical condition; receive, as output from the second ML model, an indication of the second clinical condition; and surface the indication of the first clinical condition and the second clinical condition.

[0007] In an example, the first clinical condition is asthma. In a further example, the first anatomical structure is at least one of a small bronchus or bronchiole. In another example, the second clinical condition is at least one pneumonia, atelectasis, or pleural effusion. In a further example, the second anatomical structure is at least one of a secondary bronchus or a tertiary bronchus.

[0008] In another aspect, the technology relates to a method for detecting clinical conditions with an acoustic system. The method includes receiving acoustic data generated from an acoustic sensor coupled to a tracheal tube of a patient being ventilated, wherein the acoustic data includes at least one of: (1) a composite acoustic -based signal representing reflections of emitted acoustic pulses and natural sounds generated by the patient, (2) a filtered acousticbased signal representing the reflections of the emitted acoustic pulses, or (3) an acoustically measurable parameter generated from detected reflections of the emitted acoustic pulses. The method also includes providing at least a portion of the received acoustic data as an input to a trained machine-learning (ML) model; receiving, as output from the trained ML model in response to the input, an indication of a clinical condition; and surfacing the indication of the clinical condition, wherein the clinical condition comprises one of asthma, pneumonia, pleural effusion, atelectasis, or tracheal collapse.

[0009] In an example, the method further includes based on the output from the trained ML model, generating control instructions to change a ventilation setting of the ventilator. In another example, the method further includes receiving clinical data generated from one or more patient sensors connected to a patient, wherein the clinical data includes at least one of capnometry data, pulse oximetry data, electrocardiogram data, patient movement data, a current ventilation mode, a ventilation setting, a pressure value, a flow value, a respiratory rate, work of breathing, or indication of patient-initiated spontaneous breaths.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The following drawing figures, which form a part of this application, are illustrative of aspects of systems and methods described below and are not meant to limit the scope of the disclosure in any manner, which scope shall be based on the claims.

[0011] FIG. 1 depicts an example airway management system.

[0012] FIG. 2A depicts an example acoustic sensor package.

[0013] FIG. 2B depicts a view of the front panel of a monitor associated with the acoustic sensor package of FIG. 2A.

[0014] FIG. 3 depicts an example of a network-connected system fortraining and implemented AI / ML models for determining a clinical endpoint.

[0015] FIG. 4 depicts example acoustic reflectometry signals for two different patients.

[0016] FIG. 5 depicts an example system for determining a clinical endpoint.

[0017] FIG. 6 depicts an example method for determining a clinical endpoint from acoustic reflectometry data.DETAILED DESCRIPTION

[0018] An endotracheal tube (ETT) is a type of breathing tube that is placed through the nose or mouth of a patient and into the trachea. The proximal end of the tube remains outside the patient and is typically connected to a mechanical ventilator to form a patient breathing circuit. The ETT helps maintain patency of the airway, permits positive pressure ventilation; seals off the digestive tract from the trachea (thereby preventing gastric insufflation due to forced air into the stomach), and provides a means for supplying oxygen, anesthesia, or other breathing gases to a patient requiring respiratory support. An ETT is placed in the patient’s airway by a clinician through a process called intubation. Once placed, the ETT may be affixed to the patient by medical tape, elastic harness, or other method, in order to help the distal end of the ETT remain in position in the trachea.

[0019] One approach for monitoring the position of the distal tip of the ETT and detecting ETT obstructions is to use an acoustic reflectometry device. One such device is described in U.S.Pat. No. 10,668,240, titled “Acoustical Guidance and Monitoring System,” (hereinafter “the ’240 Patent”) which is incorporated herein by reference in its entirety. Briefly, the acoustic reflectometry device is connected to the proximal end of the ETT, between the ETT and tubing associated with a ventilator. The device uses a sound pulse generator to transmit acoustic waves (e.g., pulses) from the proximal tip of the ETT towards the distal tip. For instance, the input waveforms can be pulses or other time-limited waveforms. The pulses may consist of any time-limited waveform. These pulses interact with the distal tip of the tube and the patient’s anatomy, and the acoustic pulses are reflected back to the device, which also houses two or more sound-sensing elements (e.g., microphones, acoustic receivers). The received acoustic reflections are processed and analyzed to determine positional drift of the ETT distal tip relative to a baseline position. The received reflections may also be analyzed to determine whether the distal tip of the ETT has entered a larger or smaller diameter respiratory passageway. Obstructions or blockages within the ETT may also be located and characterized based on the reflections of the acoustic pulses. The position and / or blockage information derived from the acoustic analysis is displayed on a monitor associated with the device.

[0020] The sound waves emitted from the acoustic generator interact not only with the ETT but also with other passageways and anatomical structures of the patient. As the sound waves travel through the patient passageways, the sound waves are at least partially reflected at different points in the anatomy, such as branches in passageways (e.g., at various bronchi). The acoustic sensor senses these reflections in addition to the reflections from within the ETT and the tip of the ETT. The acoustic reflections from the anatomy of the patient (referred to herein as anatomical reflections) contain additional information about the condition of the patient. For instance, the timing, strength, and other characteristics of anatomical reflections indicate details about the structural details of the passageways of the patient and, in some cases, the health of those structures.

[0021] In addition to the reflected acoustic signals, the acoustic sensor also senses various sound waves that are generated naturally from the patient. Such sounds may include breathing sounds, cardiac movements (e.g., heart beats), muscle movements, and / or gastrointestinal movements, among others. In some cases, these sounds are effectively ignored or filtered from the acoustic reflectometry signals to more accurately perform the acoustic reflectometry measurement operations, such as determining the ETT tip position or occlusions with in the ETT.

[0022] The raw, or minimally processed signal, generated from the acoustic sensor, however, represents a rich combination of acoustic data. The combination results from the acoustic reflectometry signals, such as the anatomical reflections, and the natural sounds generated from the patient. Accordingly, analysis of such signals provides for additional insights about the clinical conditions of the patient.

[0023] The technology disclosed herein, among other things, analyzes the acoustic data signals to determine or identify potential clinical conditions of the patient in a non-invasive manner that has previously not been possible. Due to both the complexity and richness of the acoustic data, such acoustic data signals are particularly well-suited for in-depth analysis from machinelearning (ML) models and / or artificial intelligence (Al) models that are able to process and account for the complex patterns within the signals. The ML / AI models may further be provided with, and trained with, additional sensor data that is available for the patient (e.g., electrocardiogram (ECG) data, breath rate, temperature, pulse oximetry (SpCh), motion level, etc.) as well as demographic information for the patient. The output from such ML / AI models indicates the possibility of particular types of clinical conditions or endpoints of the patient. For example, determinations of asthma, pneumonia, pleural effusion, atelectasis, tracheal collapse, and / or cardiac conditions, among other clinical conditions, may be determined from ML / AI models.

[0024] FIG. 1 depicts an example of an airway management system or medical ventilation system 100 that includes a ventilator 118, a monitor 146, an acoustic sensor package or device 130, and atracheal tube 108 (e.g., ETT). The tracheal tube 108 is illustrated as an endotracheal tube, which has an inflatable balloon cuff 110 that may be inflated to form a seal against walls 106 of a trachea 104 of a patient 102. However, the tracheal tube 108 may alternatively be uncuffed. The airway management system 100 may be used in conjunction with any other suitable types of tracheal tubes or medical devices. As examples, the airway management system 100 may be utilized with an endotracheal tube, an endobronchial tube, a tracheostomy tube, an introducer, an endoscope, a bougie, a circuit, an airway accessory, a connector, an adapter, a filter, a humidifier, nasal cannula, or a supraglottic mask / tube.

[0025] The system 100 includes devices that facilitate positive pressure ventilation of the patient 102, such as the ventilator 118, which provides mechanical ventilation to the patient 102. For example, the ventilator 118 may provide a gas mixture 120 (e.g., from a source 122of the gas mixture 120) through the acoustic sensor device 130, through the tracheal tube 108, and to lungs 128 of the patient 102, thereby mechanically actuating rest, inspiration, and expiration phases of breathing cycles of the patient 102. The gas mixture 120 may be referred to herein as breathing gases. In some examples, the ventilator 118 includes a gas mixture controller 124 that provides control instructions to cause the ventilator 118 to continuously or intermittently adjust a pressure and / or a composition of the gas mixture 120 provided from the source 122 and to the patient 102. For example, the gas mixture controller 124 may cause the ventilator 118 to direct air, oxygen, or another suitable gas mixture from the source 122 to the lungs 128 of the patient 102.

[0026] The acoustic sensor device 130 is connected to the ventilator 118 through a patient circuit 142, which is depicted as a single-limb circuit (e.g., an inhalation limb) used for providing the gas mixture 120 to the patient 102 for inhalation. In examples where the cuff 110 is not included or is not inflated, the patient 102 may exhale through the nose or mouth (if applicable). In examples where the cuff 110 is inflated, the patient 102 may exhale through the tracheal tube 108, and the exhaled breathing gases may be vented through an exhaust port (not depicted) provided as part of the patient circuit 142. In other examples, the patient circuit 142 may be a dual-limb circuit, where additional tubing is provided for venting exhaled gases from the patient 102 (e.g., an exhalation limb). The exhalation limb may be connected to the ventilator 118, which may provide elements for controlling the exhaled breathing gases. For example, the ventilator 118 may include one or more valves, fdters, and / or other elements used for managing the flow of exhaled breathing gases. In some examples, to support exhalation, another type of tracheal tube 108 (or other device) may be used in place of the depicted tracheal tube 108 (e.g., one of the tracheal tube alternatives described above).

[0027] The ventilator 118 may include a plurality of ventilator sensors 126 that provide measurement data associated with the gas mixture 120. The ventilator sensors 126 may be internal to the ventilator 118 and / or may be coupled to the patient circuit 142. For example, one or more ventilator sensors 126 may provide measurement data associated with flow, pressure, respiratory rate (RR), work of breathing, indications of patient-initiated spontaneous breaths, and / or other respiratory parameters of breathing gases delivered to the patient 102 during inhalation. Additional data may be derived from the measured sensor data. For example, the volume of the delivered gas mixture 120 may be computed using flow measurement data, such as by performing integration of the measured flow data over time.Accordingly, the ventilator 118 may be capable of determining and reporting inhalation volumes (e.g., tidal volume (VT)) derived from measurement data acquired by the ventilator sensors 126. The ventilator may also generate and / or provide data regarding current ventilation settings, such as ventilation mode, positive end-expiratory pressure (PEEP) settings, fraction of inspired oxygen (FiCh) settings, among other types of settings.

[0028] In examples where the tracheal tube 108, patient circuit 142, and ventilator 118 are configured to receive exhaled breathing gases from the patient 102, the ventilator sensors 126 may include additional pressure, flow, and / or other types of sensors for collecting measurements of the exhaled breathing gases. The ventilator 118 may further be capable of determining exhaled breathing volumes as described above.

[0029] The airway management system 100 may also include one or more patient sensors 144 that may be directly or indirectly connected to the patient 102 for collecting additional measurement data. For example, the patient sensors 144 may include a type of pulse oximeter, which provides a noninvasive measurement of oxygen saturation in the patient’s blood (e.g., SpCh). The pulse oximeter may also provide a measurement of the heart rate (HR) of the patient 102 or, in some examples, the patient sensors 144 may include a separate sensor for measuring HR and other information, such as electrocardiogram (ECG) data.

[0030] In another example, the patient sensors 144 may include a sensor or sensing system for measuring the CO2 concentration of the exhaled breathing gases. For instance, the patient sensors 144 may include a capnometry monitor or sensor that samples a portion of the exhaled breathing gases and determines CO2 concentration. In one example, the capnometry monitor may sample exhaled breathing gases via a port provided on the acoustic sensor device 130 or patient circuit 142. The capnometry monitor may provide a measurement of the end-tidal carbon dioxide (EtCCh).

[0031] Additionally or alternatively, the patient sensors 144 may include other types of sensors for collecting measurements from the patient 102. The patient sensors 144 may provide the measured data to the ventilator 118 via one or more sensor cables 145. Sensor data and other information may be transmitted between the ventilator 118 and patient sensors 144 via the sensor cable(s) 145 in analog or digital form, and the transmission may be performed according to any of a wide variety of communications protocols. In some examples, the sensor data may be provided by one or more of the patient sensors 144 to the ventilator 118 via wirelessprotocols, such as Wi-Fi, Bluetooth, or other established or custom wireless transceiver protocol. The sensor data may also or alternatively be communicated to system components other than the ventilator, such as a multi -parameter monitor (MPM).

[0032] In still other examples, the ventilator 118 may include the elements, functions, and / or features for implementing some or all of the measurements performed by the patient sensors 144. In one example, the ventilator 118 may include functions / elements for performing capnometry within the ventilator 118 itself. In other examples, the ventilator 118 may include functions / elements for performing SpCh, HR, and / or other types of measurements.

[0033] Additionally or alternatively, one or more of the patient sensors 144 may be connected to the monitor 146, rather than the ventilator 118. For example, the patient sensors 144 may include a pulse oximeter that provides SpCh data directly to the monitor 146. In further examples, the monitor 146 may be connected to a capnometry monitor or may include the functions / elements for performing capnometry within the monitor 146.

[0034] As illustrated, the acoustic sensor device 130 of the airway management system 100 is coupled to an external or proximal end 114 of the tracheal tube 108. In the illustrated example, the acoustic sensor device 130 may operate as, or be, an adapter that facilitates coupling of the tracheal tube 108 to a patient circuit 142 or hose / tube coupled to the ventilator 118. Other arrangements are also contemplated, such as an acoustic sensor device 130 that is disposed on the tracheal tube 108 or on other components of the breathing circuit.

[0035] The acoustic sensor device 130 includes at least one acoustic generator 132 and at least one acoustic sensor 134 disposed within an adapter housing 136. The acoustic generator 132 is oriented to direct and / or emit incident sound energy 138 (e.g., acoustic pulses, sound, acoustic energy) into the lumen 112 of the tracheal tube 108, which guides the sound energy 138 out of an internal or distal end 116 of the tracheal tube 108 and toward airways of lungs 128 of the patient 102. Further, the acoustic sensor 134 detects reflected sound energy 140 (e.g., detects reflections of the emitted acoustic pulses) or echoes of the incident sound energy 138, back from different positions along the endotracheal tube / and or the airways of the patient. For instance, the emitted acoustic pulses may reflect from items such as obstructions in the tracheal tube, the tip of the tracheal tube, the airways of the lungs 128, and / or other passageways of the body. Accordingly, the acoustic sensor device 130 facilitates acoustic reflectometry techniques that analyze sound pressure waveforms for airway acoustic echoesindicative of airway size. That is, the acoustic generator 132 and the acoustic sensor 134 cooperate to provide sensor signals indicative of a sound pressure waveform having an airway acoustic echo, which the monitor 146 may analyze to determine, among other things, an airway size of the trachea around, or distally located from, the tip of the tracheal tube 108. The values or parameters that can be measured by the acoustic reflectometry system may be referred to as acoustically measurable parameters, and may include parameters such as ETT position, ETT movement, ETT obstruction position and / or size, airway constriction or expansion (e.g., airway size directly distal to the ETT, passageways such as esophagus, bronchus, trachea, upper airway, mouth), among other parameters measurable by the acoustic reflectometry system.

[0036] The acoustic generator 132 in some examples is a speaker or a miniature speaker. However, the acoustic generator 132 may additionally or alternatively include any suitable loudspeakers, buzzers, horns, sounders, and so forth that rely on moving coil, electrostatic, isodynamic, or piezo-electric techniques. Additionally, the sound sensing element(s) 134 (also referred to herein as an acoustic sensor 134) may be a microphone, microphone array, or other sound pressure sensors, in some examples. When implemented as a microphone array, the acoustic sensor 134 and / or the monitor 146 discussed below may be designed to separate the input sound pulses (e.g., distally travelling sound waves) from the output sound pulses (e.g., proximally traveling sound waves). The input sound waves (X) and the output sound waves (Y) may be used to calculate an impulse response (H) of the ETT and the airways (e.g., H=Y / X). Thus, the use of the microphone array allows for taking not only the incident pulse as the input signal, but all of the extraneous echoes from the ventilator side and leveraging them as input energy to the system as well. Further, when implemented as a microphone array, the acoustic sensor 134 may be designed to determine the speed of travel or propagation (sound speed) of the emitted or reflected acoustic pulses in the region between sound sensing elements within the microphone array, or in the region between the acoustic generator 132 and one or more of the sound sensing elements. Other suitably paired components that respectively generate suitable sound energy and receive echoes or reflection of the sound energy may be used in the acoustic sensor device 130. Additional details regarding acoustic reflectometry are described in U.S. Patent No. 9,707,363, titled “System and Method for Use of Acoustic Reflectometry Information in Ventilation Devices,” which is incorporated herein by reference in its entirety, and the ’240 Patent, referenced above. For instance, example components for the acoustic sensor(s) 134 and the acoustic generator 132 are described in the ’240 Patent.

[0037] In some examples, the acoustic sensor device 130 may not include an acoustic generator 132, or the acoustic generator 132 may be permanently or intermittently left unpowered. In such examples, one or more sound-sensitive elements of acoustic sensor 134 may acquire sounds generated naturally by the body of the patient 102. Analysis of these sounds may provide relevant data or information about the health of the patient 102. For example, analysis of naturally occurring, internal body sounds may indicate the presence or onset of congestion, wheezing, pneumonia, and / or other health condition.

[0038] In other examples, when no acoustic pulses are coupled / directed into the lumen 112 by an acoustic generator 132, the acoustic sensor 134 may detect transmitted and / or reflected acoustic energy generated by other sources. For example, the ventilator 118 may inherently generate acoustic energy that couples into the lumen 112 of the tracheal tube 108, and may be suitable for performing some level of acoustic reflectometry, as described above. In some examples, other acoustic sources may provide suitable acoustic energy for performing acoustic reflectometry, such as sources that may be coupled to the patient circuit 142. Acquisition and analysis of acoustic energy provided by sources other than the acoustic generator 132 may include the use of specific filtering and / or data processing that differs from configurations where the acoustic generator 132 is the source of acoustic energy.

[0039] The acoustic sensor device 130 may be communicatively coupled to the monitor 146 via interface cable 158. The interface cable 158 provides a means for acoustic measurement and other data to be transmitted from the acoustic sensor device 130 to the monitor 146. The transmitted data includes acoustic measurement data acquired by the acoustic sensor 134 (such as the types of data described above) that has been processed by elements of the acoustic sensor device 130. For example, the transmitted acoustic measurement data may include data that has been amplified, filtered, digitized (such as by an analog-to-digital converter (ADC)), and / or that has otherwise been processed. In some examples, the acoustic sensor device 130 may transmit unprocessed, or minimally processed, acoustic measurement data. For instance, the acoustic sensor device 130 may transmit the signals acquired by the acoustic sensor 134 in a relatively unprocessed state, such as with minimal (if any) amplification and / or filtering. The acoustic sensor device 130 may also transmit other types of data to the monitor 146 by way of the interface cable 158. For example, elements within the acoustic sensor device 130 may transmit status, information related to error conditions, device settings, configuration information, calibration information, or other types of information.

[0040] Additionally, the monitor 146 may transmit information to the acoustic sensor device 130 via interface cable 158. For instance, the monitor 146 may transmit sensor operating parameters, settings, data, variables, firmware, executable code, or other information or data associated with operation of acoustic sensor device 130. The interface cable 158 may also provide a means for the monitor 146 to provide electrical power to the acoustic sensor device 130. In examples, the interface cable 158 may be implemented as two or more cables, wires, or other type of suitable electrical conductor.

[0041] The monitor 146 may include a communication connection(s) 148 (e.g., input / output ports, communication circuitry, analog and / or interface circuitry, etc.), processing circuitry, such as the processor 150, memory 152, a display 154, and a user interface 156 that may be communicatively coupled to one another. The communication connection(s) 148 may receive data from, and / or transmit data to, the acoustic sensor device 130 via interface cable 158. Data may be transmitted between the communication connection(s) 148 and the sensor device 130 in analog or digital form, and the transmission may be performed according to any of a wide variety of communications protocols. In other examples, the communication connection(s) 148 may communicate with the acoustic sensor device 130 via wireless protocols, such as WiFi, Bluetooth, or other established or custom wireless transceiver protocol.

[0042] In further examples, the communication connection(s) 148 may receive data from, and / or transmit data to, the ventilator 118 via ventilator cable 160. For instance, the ventilator 118 may provide data from ventilator sensors 126 and / or patient sensors 144 to the monitor 146. The ventilator 118 may also provide patient demographic data to the monitor 146, such as the age, gender, height, weight, known medical conditions, and / or other types of patientspecific data. The sensor data and patient demographic data may be used by the monitor 146 as described below.

[0043] The communication connection(s) 148 may also provide data to the ventilator 118 from the monitor 146. For example, the monitor 146 may provide acoustic sensor data to the ventilator 118 and / or may provide other types of data to the ventilator 118, via the ventilator cable 160. As another example, the monitor 146 may provide signals for controlling / adjusting ventilator settings to the ventilator 118.

[0044] In some examples, the communication connection(s) 148 may communicate with the ventilator 118 via wireless protocols, such as Wi-Fi, Bluetooth, or other established or customwireless transceiver protocol. In such examples, the ventilator cable 160 may not be present in the system 100.

[0045] The monitor 146 includes processing circuitry, such as processor 150, which may include any combination of digital and analog circuitry needed for processing signals received from, or transmitted to, the acoustic sensor device 130. Examples of analog processing circuitry may include signal filters, amplifiers, bridge circuits, bias circuits, pulse generators, level translators, ADCs, or other types or combinations of analog circuitry. The processor 150 may include one or more general purpose processors, microprocessors, microcontrollers, digital signal processors (DSPs), graphics processing units (GPUs), or other programmable circuits. In examples, the processor 150 may include any combination of commercially available components, or custom or semi-custom integrated circuits, such as application specific integrated circuits (ASICs). The processor 150 may include elements needed for control or communication with the communication connection(s) 148, memory 152, display 154, user interface 156, and / or other elements of the monitor 146.

[0046] The processor 150 may perform control, interface, communication, or other processing functions by executing instructions that are stored in the memory 152. The memory 152 may include RAM, ROM, electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or other types of storage media. The memory 152 may store instructions, that when executed by the processor 150, cause the monitor 146 or elements thereof to perform the operations described herein. The memory 152 may be used to store data, such as data acquired directly from the acoustic sensor device 130 and / or processed by the processor 150. The memory 152 may be used to store data acquired by other sensors (e.g., patient sensors 144 and / or ventilator sensors 126), patient demographic data, settings, alert thresholds, preferences, or other types of data or information used during operation of the monitor 146.

[0047] The processor 150, and / or other elements of the monitor 146, may receive user input from a user interface 156. In examples, the user interface 156 may include buttons, knobs, soft keys, dials, switches, and / or other forms of user-selectable input that may be made available to the user on the exterior of the monitor 146 (depicted in FIG. 2A). The user interface 156 may allow a clinician to configure the operation of the monitor 146 and / or the acoustic sensordevice 130. For example, the user interface 156 may allow a clinician to enable / disable features and functions of the monitor 146 and / or acoustic sensor device 130, set alert thresholds, respond to alerts, navigate menus, configure settings, store / retrieve measurement data to or from memory 152, and / or provide other types of input to the monitor 146, among other functions. In further examples, the user interface 156 may allow a clinician to enable automatic features of the monitor 146, such as an automatic threshold adjustment feature as described below.

[0048] In some examples, the user interface 156 may provide audible alerts, such as through a speaker coupled to the user interface 156 (depicted in FIG. 2A). For example, the monitor 146 may be configured to provide an audible alert when the monitor detects a clinically relevant airway constriction, obstruction of the tracheal tube 108, and / or for other conditions that require the attention of a clinician. In other examples, the user interface 156 may provide audible alerts for any of a range of possible states and conditions associated with the operation of the monitor 146, acoustic sensor device 130, and / or other portion of the example airway management system 100.

[0049] The monitor 146 also includes a display 154, which may display textual, graphical, and / or other forms of displayable information or data. The information / data may be associated with position of the tracheal tube 108, obstructions within the tracheal tube 108, airway collapse, insufficient ventilation of the patient 102, information related to the respiratory status of the patient 102, data collected or derived from the sensors (e.g., ventilator sensors 126, patient sensors 144, etc.), other configurable settings, alerts, and / or other information associated with the functioning and operation of the monitor 146. The display 154 may be any of a variety of display technologies, such as liquid crystal display (LCD), light emitting diode (LED), organic light emitting diode (OLED), or other display technology. In examples, the display 154 may be a touch-sensitive display (e.g., a capacitive touch-sensitive display) that allows the clinician to provide input through the display 154, such as through a graphical user interface (GUI) provided on the display 154 by the user interface 156. An example GUI is depicted in FIG. 2A.

[0050] In some examples, the monitor 146 may be connected to a network-based host system (depicted in FIG. 3). For example, the communication connection(s) 148 or other elements of the monitor 146 may be capable of establishing connection to a computer network through anyof a variety of wired or wireless communications protocols. In one example, the communication connection(s) 148 may be capable of connecting to a computer network via Wi-Fi protocol (or other wireless protocol). The computer network may include a local area network (LAN), wide area network (WAN), or other type of network. In some examples, the monitor 146 may be connected to the Internet.

[0051] As described herein, the monitor 146 may provide a wide variety of data associated with the example airway management system 100 to the network-based host system. For instance, the monitor 146 may provide data received from the acoustic sensor device 130, clinical data received from the ventilator 118 (e.g., data acquired by patient sensors 144 and / or ventilator sensors 126), patient data received from the ventilator 118, patient data provided by a clinician to the monitor 146, and / or other types of data to the host system. In further examples, the monitor 146 may provide data associated with the configuration of the example airway management system 100. For example, the configuration data may include the types of patient sensors 144 in use in the system 100 (e.g., pulse oximetry, capnometry, etc.), the configuration of the patient circuit 142 (e.g., single-limb or dual-limb), the types of sensor data provided by the ventilator sensors 126 (e.g., flow, pressure, etc.), and / or other configuration data. The network-based host system may use the received data to train one or more ML / AI models, which are then used to determine a clinical endpoint of the patient. In some examples, the host system may provide the trained ML / AO model(s) to the monitor 146 or other system components, which may use the model during operation as described in further detail below.

[0052] In some examples, the ventilator 118 may be similarly capable of connecting and communicating with a network-based host system. For instance, the ventilator 118 may include the same or similar elements as the monitor 146, such as a processor, memory, communication connections, etc., that enable the ventilatorto connect to the computer network and provide data to the host system. Similar to the above description, the ventilator 118 may provide sensor data, patient data, and / or other data to the host system, and may receive recommendations for clinically relevant alert thresholds and / or other types of data / information from the host system.

[0053] In an example scenario, during operation of the airway management system 100, the ventilator 118 provides inhaled breathing gases 120 to the patient 102 during inhalation and receives exhaled breathing gases during exhalation. The ventilator sensors 126 may acquiremeasurement data for flow, pressure, RR, and volume of the inhaled gas mixture 120 and the exhaled breathing gases. The patient sensors 144 may acquire SpCh, EtCCh, ECG, HR, and / or other data from the patient 102. Data acquired from the ventilator sensors 126 and patient sensors 144 may be provided by the ventilator 118 to the monitor 146 via the ventilator cable 160.

[0054] The monitor 146 further receives acoustic data from the acoustic sensor device 130. The acoustic data may indicate the position of the tracheal tube 108 (e.g., relative to a baseline position), any obstructions that may be present in the tracheal tube 108, airway constriction in the vicinity of the tracheal tube distal tip, and / or other data associated with the tracheal tube 108.

[0055] The monitor 146 may also have access to patient demographic data (e.g., age, weight, etc.). The patient demographic data may be provided to the monitor 146 and / or ventilator 118 by a clinician and / or retrieved from a medical database, such as an electronic health record (EHR) database.

[0056] The monitor 146 may transmit the patient data, sensor data, acoustic data, and / or configuration data to a network-connected host system, which analyzes the data using one or more ML / AI models. In other examples, the model may be stored locally and executed on the monitor 146 and / or ventilator 118, among other system components. In some examples, the host system uses the received data to train or update one or more AI / ML models.

[0057] As discussed further below, the AI / ML models receive the acoustic data, sensor data, and / or patient data as input. The AI / ML models then process the received input data to generate an output. The output includes a determination of a clinical condition or endpoint for the patient. For instance, the output may include the likelihood that the patient has a clinical condition such as asthma, pneumonia, pleural effusion, atelectasis, and / or cardiac conditions, among other clinical conditions.

[0058] FIG. 2A depicts a view of an example acoustic sensor package or device 230, which may be similar to, or the same as, acoustic sensor device 130. The acoustic sensor device 230 includes a fitting 261 (e.g., a 15 mm fitting) for connecting tubing associated with a ventilator (such as ventilator 118 and patient circuit 142) . The fitting 261 is connected to a sound-sensing section 262, which includes elements associated with the generation, detection, and processingof acoustic pulses. For example, the sound-sensing section 262 includes an acoustic generator (e.g., acoustic generator 132) and an acoustic sensor (e.g., acoustic sensor 134), among other elements. In some examples, the acoustic generator 132 may be housed within the fitting 261. The sound-sensing section 262 is further connected to a nozzle 263 which may couple to an ETT (e.g., tracheal tube 108).

[0059] The acoustic sensor device 230 includes an internal lumen that is continuous between a proximal opening 259 in the fitting 261 and a distal opening 265 in the nozzle 263. When tubing associated with a patient circuit is connected to the fitting 261 and an ETT is connected to the nozzle 263, inhaled breathing gases (e.g., gas mixture 120) may be provided to the patient (e.g., patient 102) from the ventilator. Inhaled breathing gases pass through the acoustic sensor device 230, from the proximal opening 259 to the distal opening 265. When the patient exhales through the ETT, the exhaled breathing gases pass through the acoustic sensor device 230 from distal opening 265 to the proximal opening 259.

[0060] During operation, the acoustic sensor (located in the sound-sensing section 262) may detect both transmitted and reflected acoustic pulses, along with other sounds such as sounds naturally generated from the patient (e.g., breathing sounds, cardiac sounds, gastrointestinal sounds). The acoustic sensor, and / or elements associated with the acoustic sensor, provide acoustic -based signals that are based on the received pulses. The sound sensing elements generate electrical signals (e.g., the acoustic-based signals) that represent the acoustic signals. These acoustic signals consist of the received incident and reflected waves and are provided by the acoustic receiver and / or its associated elements. As depicted in FIG. 1 and described above, the acoustic -based signals may be provided to an associated monitor (e.g., monitor 146 or 246) via an interface cable (e.g., interface cable 158). In examples, the sound-sensing section 262 may include elements associated with a wireless connection to the associated monitor, such as Bluetooth, Wi-Fi, etc.

[0061] FIG. 2B depicts a front view of an example monitor 246 that receives acoustic-based signals from the acoustic sensor device 230. The monitor 246 may be similar to, or the same as, monitor 146 depicted in FIG. 1 and may include the same or similar elements. For example, the monitor 246 includes portions or elements of a user interface (e.g., user interface 156), such as soft keys 256A, navigation buttons 256B, and an alert silence or response button 256C. Themonitor 246 further includes an alert indicator 256D and a speaker 256E, both of which may be associated with the provision of an alert or other type of notification to a clinician.

[0062] The monitor 246 includes a connection panel 248, which may include one or more ports for connecting cables between the monitor 246 and other components, systems, devices, etc. For example, the interface cable between the monitor 246 and the acoustic sensor device 230 may be connected to the monitor 246 at the connection panel 248. The monitor 246 may also be connected to other devices in an airway management system (e.g., system 100) via one or more cables that attach to the monitor 246 at the connection panel 248. For example, the monitor 246 and may be connected to a ventilator (e.g., ventilator 118) via a ventilator cable (e.g., ventilator cable 160), which attaches to the monitor 246 at the connection panel 248. Other devices in the airway management system (e.g., patient sensors 144) may similarly be connected to the monitor 246 via one or more cables that connect at the connection panel 248. In some examples, the monitor 246 may be connected to a computer network using a suitable networking cable (e.g., an ethemet cable) that connects to the monitor 246 at the connection panel 248.

[0063] The monitor 246 includes a display 254, which may be similar to, or the same as, display 154 described above. The display 254 includes one example of display content 266, which provides information to the clinician.

[0064] The display content 266 includes information associated with the position of the ETT in the trachea of a patient, based on acoustic-based signals provided by the acoustic sensor device 230. In the example depicted in FIG. 2B, the display content 266 indicates an alert condition associated with the position of the distal tip of the ETT. For example, the display content 266 includes a baseline position indicator 268 that provides a numerical indication of a shift in the position of the ETT tip, relative to a baseline position established during an intubation procedure (or established at other times during operation of the monitor 246). As depicted in FIG. 2B, the baseline indicator 268 indicates that the tip of the ETT has shifted by 1.3 cm from baseline. Adjacent to the baseline indicator 268 are directional indicators 269-70 that indicate the direction of the shift from baseline position. As a result of a shift in the tip of the ETT, the baseline indicator 268 and one of the directional indicators 269-70 is illuminated to visually indicate the positional shift. In the example depicted, the baseline indicator 268 and low indicator 269 are illuminated in an alert pattern that indicates that the ETT has shifted to aposition in the trachea that is too low (e.g., the position has crossed a movement threshold). The high indicator 270 is left unilluminated or is illuminated without an alert pattern. In conjunction with indicators 268-69 being illuminated with an alert pattern, the display content 266 may also include an alert message 272 indicating that the ETT is too low in the trachea. The monitor 246 may also provide an audible alert through the speaker 256E.

[0065] In some examples, the ETT tip may shift from its baseline position, but the shift may not exceed a threshold needed to trigger an alert. In such examples, the baseline indicator 268 and directional indicators 269-70 may not be illuminated with an alert pattern and / or may be illuminated with a pattern indicating a positional shift that does not exceed an alert threshold.

[0066] In some examples, the monitor 246 may display a readout of the clinical measures 274A-B on the display 254. For instance, the clinical measures 274A-B may be continually displayed, such as in FIG. 2B, or may be displayed at the onset of an alert condition. In still other examples, the clinical measures 274A-B may not be displayed by the monitor 246.

[0067] FIG. 3 depicts a network-connected system 300 that includes a plurality of patient systems 301-303, a network-based storage 316, a host system 308, and local storage 314. The patient systems 301-303, network-based storage 316, and host system 308 communicate with each other over a network 306. The patient systems 301-303 represent a larger population of patient systems that may be connected to the host system 308 through the network 306. In some examples, the system 300 may include a very large population of patient systems that communicate with the host system 308 as described herein. Each of the patient systems 301- 303 may be connected to different patients at different times, such as when a patient system 301-303 is disconnected from one patient, cleaned, and subsequently used with another patient. Further, in some examples the patient systems 301-303 may operate in different clinics, while in other examples two or more of the patient systems may be co-located.

[0068] The patient systems 301-303 are each representative examples of airway management systems (e.g., example airway management system 100). Each patient system 301-303 includes at least a monitor (e.g., acoustic monitor 146, 246), an acoustic sensor package (e.g., acoustic sensor device 130, 230), and, in some examples, a ventilator (e.g., ventilator 118), which is connected to a patient via a patient circuit (e.g., patient circuit 142). However, in some examples, the patient systems 301-303 may each be configured differently from one another and / or may include different features, functions, and / or elements than each other. Asone example, the patient systems 301-303 may each include a collection of various patient sensors (e.g., patient sensors 144), which may be connected to an MPM, a ventilator, and / or an acoustic monitor. The first patient system 301 may include an SpCh sensor, the second patient system 302 may include an SpCh sensor and a capnometry monitor, and the third patient system 303 may include neither an SpCh sensor nor a capnometry monitor.

[0069] Further, the elements of each patient system 301-303 may be configured to operate differently and / or may provide different functionality. For example, the first and second patient systems 301-302 may include a dual-limb patient circuit, where each ventilator manages both inhaled and exhaled breathing gases. The third patient system 303 may be configured with a single-limb patient circuit, where the ventilator provides breathing gases during inhalation, but does not manage exhaled breathing gases during exhalation. Thus, the patient systems 301-303 may be capable of providing sensor data associated with the inhaled breathing gases (e.g., from ventilator sensors 126), such as inhalation flow, pressure, etc. Whereas the first and second patient systems 301-302 may further be capable of providing sensor data associated with exhaled breathing gases. The patient systems 301-303 (and additional patient systems) may each include different types and numbers of patient sensors that collectively provide a range of clinical data from a population of patients.

[0070] As noted above, the patient systems 301-303 are connected to the host system 308 via network 306. The network 306 may include any of a variety of computer equipment suitable for providing connectivity between the patient systems 301-303, network-based storage 316, host system 308, and / or other resources connected to the network 306. For example, the network 306 may include servers, routers, switches, cables, interconnects, and / or other types of computer equipment for providing connectivity between devices attached to the network 306. The network 306 may be configured as a LAN, WAN (e.g., the Internet), and / or other type of network configuration. In some examples, the network 306 may include computer equipment suitable for providing wireless connections, such as through Wi-Fi and / or other wireless protocols. Thus, in some examples, the patient systems 301-303, host system 308, network-based storage 316, and other resources / devices may connect to the network 306 via a wired connection (e.g., Ethernet cable) or via wireless connection (e.g., Wi-Fi).

[0071] The host system 308 may include one or more computing devices, such as a type of computer and / or server. In examples where the host system 308 includes a plurality ofcomputing devices, the computing devices may be co-located or may be distributed to different physical locations across the network 306. For example, the host system 308 may include a plurality of cloud servers or other type of computing devices arranged as part of one or more computing clusters, data centers, and / or other arrangements. Similarly, the network-based storage 316 may include any of a variety of co-located or distributed storage devices, such as rack-mounted hard disk drives and / or other forms of mass storage. In examples, the networkbased storage 316 may be a type of network-attached storage (NAS), storage area network (SAN), database, and / or other type of storage arrangement. The network-based storage 316 may include computing devices (e.g., computers, servers, etc.) that provide access to the stored data.

[0072] The host system 308 also includes local storage 314, which may be located with, and directly accessible to, the host system 308. For example, the local storage 314 may include internal or external memory storage devices, such as one or more hard disk drives and / or other forms of data storage. The local storage 314 may be distributed and accessible to one or more of the computing devices associated with the host system 308.

[0073] During operation, the host system 308 receives and processes data from the patient systems 301-303. As described above, the data provided by each of the patient systems 301- 303 may vary, depending on the configuration of each patient system 301-303, the types and number of sensors provided within each patient system 301-303, the type of ventilator connected to each patient, and other factors that are specific to each patient system 301-303 and the patients connected thereto.

[0074] To briefly summarize, data provided to the host system 308 by the patient systems 301- 303 may include patient demographic data, clinical data, configuration data, and / or acoustic sensor data. The patient demographic data may include the age, weight, height, gender, body mass index (BMI), health condition (e.g., COPD, heart condition, etc.) and / or other types of patient-specific data. The clinical data may include sensor measurement data associated with patient sensors and / or ventilator sensors. For example, the clinical data may include measurements of SpCh (e.g., from pulse oximetry), EtCCh (e.g., from capnometry), HR, ECG, RR, VT, flow, pressures, and / or data from other types of sensors. The configuration data may include information associated with the devices and / or medical equipment available within each patient system 301-303. For example, the configuration data may include the type ofpatient circuit in use (single-limb or dual-limb), the capabilities of the ventilator, the types of patient sensors available, and the like. The acoustic sensor data may include processed acoustic sensor data and / or unprocessed, or minimally processed, acoustic sensor data. Examples of processed acoustic sensor data include the percentage of ETT obstruction, level of airway constriction, displacement of the ETT tube in the trachea, and / or other conditions associated with the ETT and airway of each patient. Examples of unprocessed or minimally processed acoustic sensor data include unfiltered acoustic-based signals acquired by an acoustic sensor, and / or other types of signals.

[0075] During a training period, the host system 308 provides the data received from the patient systems 301-303 to a training subsystem 310, which may be a type of executable program code, such as software, software applications, utilities, programs, algorithms, firmware, and / or other type of computer instructions. The host system 308 may also provide other training data sets to the training subsystem 310. For example, the host system 308 may have access to, or be provided with, additional clinical data, such as data from one or more clinical studies, clinical databases, and / or data from other clinical sources. In some examples, the additional clinical data may be stored in network-based storage 316 and / or local storage 314.

[0076] The training subsystem 310 may aggregate the received data (including any additional clinical data provided to the host system 308), such as over a large population of patients and patient systems 301-303, and / or over a specified period of time. The aggregated data may be stored in network-based storage 316 and / or local storage 314. The training subsystem 310 may be configured to analyze the aggregated data to determine or identify correlations between variables within the aggregated data. In additional examples, the training subsystem may identify multiple correlations within the aggregated data.

[0077] During the training period, the training subsystem 310 establishes (e.g., trains) one or more ML / AI models 312 using the identified correlations and the aggregated data. The training period may continue until a suitable amount of data has been aggregated to achieve a desired statistical significance (e.g., statistical power) and / or until the trained ML / AI models 312 meet desired performance criteria.

[0078] For example, the received aggregated data may be labeled with known clinical conditions, which may be used for training. The clinical condition labels for the training datamay also include additional details about the known clinical condition, such as the severity of the condition or factors related to the condition (e.g., consolidated lobes). For instance, training data for training of the ML / AI models 312 may be formed from labeled aggregated data from the various patient systems 301-303. As an example, a patient connected to a particular patient system may have a known clinical condition or endpoint. The data aggregated (e.g., patient demographic data, clinical data, acoustic data) from that particular patient system is labeled with particular clinical condition of the patient. Over time, many examples of labeled aggregated data are acquired for different various clinical conditions or endpoints, and those examples may be used as training data for supervised training of the ML / AI models.

[0079] Different subsets of the data types of the aggregated data may be used for different types of clinical conditions. For instance, a first subset of aggregated data may be used for detection of asthma, and a second subset of aggregated data may be used for detection of pneumonia. As an example, the training data for asthma may include at least a first portion of the acoustic reflection signal, an obstruction size within the ETT (determined from the acoustic data), and airway size distal to the ETT tip (determined from the acoustic data). As discussed further below with respect to FIG. 4, different portions of the acoustic reflection signal correspond to anatomical reflections from different segments of the patient’s anatomy. The first portion of the acoustic reflection data, may correspond to the trachea and / or primary bronchi. The primary bronchi may include the left main bronchus and right main bronchus. In other examples, a different portion of the acoustic reflection signal, representing reflections from bronchioles, may be used for the classification of asthma. Additional aggregated data for asthma may also include patient demographic data, movement data, and / or raw acoustic signal data (e.g., including natural sounds from the patient detected by the acoustic sensor).

[0080] As another example, the aggregated data for atelectasis, pneumonia, or pleural effusions may include at least an obstruction size within the ETT (determined from the acoustic data) and a second portion of the acoustic reflection signal. The second portion of the acoustic reflection signal is different than the first portion of the acoustic reflection signal used for asthma. For instance, the second portion of the acoustic reflection signal represents a portion of the acoustic waves that are reflected from lower in the lungs that the first portion of the acoustic reflection signal.

[0081] In some examples, a single ML / AI model is trained for each particular clinical condition or endpoint. For instance, a first ML / AI model may be trained for asthma, and a second ML / AI model may be trained for pneumonia. The different ML / AI models may also accept different inputs, such as inputs on which the models were trained (e.g., the subsets aggregated data for the particular condition). In other examples, some conditions or endpoints may be combined into a single ML / AI model. For instance, a single ML / AI model may be trained to classify the input data as one of pneumonia, atelectasis, and / or pleural effusion.

[0082] Once the ML / AI models 312 are trained, the ML / AI models 312 use newly received data from a patient system 301-303 as input, and provide, as output, determinations of clinical conditions or endpoints for the respective patients. The determinations are generated for each patient system 301-303, based on the input data provided by each patient system 301-303 to the ML / AI models 312. The determinations may be provided to the corresponding patient system 301-303 by the host system 308 via the network 306.

[0083] In examples, the ML / AI models 312 may include neural networks and the like. Some example models may include linear regression models, logistic regression models, decision trees, random forests, support vector machines (SVMs), and / or neural networks. Such models may be trained using supervised and / or semi-supervised training techniques based on labeled training data that indicates the clinical condition and / or endpoint as discussed above.

[0084] In some examples, one or more of the ML / AI models are in the form of a generative Al model. In such examples, training of the generative Al model may include fine-tuning of a fundamental generative Al model, such as the GPT-4 model from OpenAI, BARD from Google, and / or LLaMA from Meta, among other types of generative Al models. As one example, the ML / AI model may be in the form of a deep neural network that utilizes a transformer architecture to process the inputs that are received. The neural network may include an input layer, multiple hidden layers, and an output layer. The hidden layers typically include attention mechanisms that allow the ML / AI model 312 to focus on specific parts of an input, and to generate context-aware outputs.

[0085] The ML / AI model 312 may be a substantially large model that is capable of handling various complex inputs, such as the acoustic signals and other signals discussed herein. For instance, the ML / AI model 312 measured by the number of parameters it has. As one example, the GPT-4 model from OpenAI has billions of parameters. These parameters may be weightsin the neural network that define its behavior, and a large number of parameters allows the model to capture complex patterns in the training data. The training process typically involves updating these weights using gradient descent algorithms, and is computationally intensive, requiring large amounts of computational resources and a considerable amount of time.

[0086] The ML / AI model 312 may operate as a transformer-type neural network. Such an architecture may employ an encoder-decoder structure and self-attention mechanisms to process the input. Initial processing of the input data may include tokenizing the input into tokens that may then be mapped to a unique integer or mathematical representation. The integers or mathematical representations combined into vectors that may have a fixed size. These vectors may also be known as embeddings.

[0087] The initial layer of the transformer model receives the token embeddings. Each of the subsequent layers in the model may uses a self-attention mechanism that allows the model to weigh the importance of each token in relation to every other token in the input. In other words, the self-attention mechanism may compute a score for each token pair, which signifies how much attention should be given to other tokens when encoding a particular token. These scores are then used to create a weighted combination of the input embeddings.

[0088] In some examples, each layer of the transformer model comprises two primary sublayers: the self-attention sub-layer and a feed-forward neural network sub-layer. The selfattention mechanism mentioned above is applied first, followed by the feed-forward neural network. The feed-forward neural network may be the same for each position and apply a simple neural network to each of the attention output vectors. The output of one layer becomes the input to the next. This means that each layer incrementally builds upon the understanding and processing of the data made by the previous layers. The output of the final layer may be processed and passed through a linear layer and a softmax activation function. This outputs a probability distribution over all possible tokens available to the model. The token(s) with the highest probability may then be selected as the output token(s) for the corresponding input token(s).

[0089] The ML / AI models 312 may be stored in network-based storage 316 and / or local storage 314, and accessed as needed by the host system 308, patient systems 301-303, and / or other resources / devices connected to the network 306. Following the training period, the host system may continue to provide newly received data from the patient systems 301-303 to thetraining subsystem 310. For example, newly received data be used by the training subsystem 310 to improve the accuracy of the models and / or to improve performance of one or more of the ML / AI models 312. In some examples, the training subsystem 310 may update or revise the ML / AI models 312 based on the newly received data through reinforcement learning.

[0090] The population of patient systems 301-303 that provide input data to the ML / AI models 312 following the training period may be different than the population of patient systems that provide training data to the training subsystem 310 during the training period. Over time, new patient systems may be added or removed from the population of patient systems 301-303 connected to the host system 308. In some examples, the patient systems 301-303 that provide input data to the ML / AI models 312 following training may be the same patient systems 301- 303 that provided training data during the training period. Additionally, patients connected to the patient systems 301-303 following the training period may different than the patients connected to the patient systems 301-303 during the training period.

[0091] FIG. 4 depicts example acoustic reflectometry signals for two different patients. The acoustic reflectometry signals depicted are signals that are detected by the acoustic sensor for acoustic reflections of the sound emitted from the acoustic generator. As discussed above, different portions of an acoustic reflectometry signal may be used as an input to the ML / AI model depending on the type of condition that is being assessed. The different portions may be different temporal portions of the acoustic signal. For instance, generated sound waves that travel further into the patient before being reflected back to the acoustic sensor arrive at the acoustic sensor later than sound waves that reflect from shallower portions of the patient anatomy or ETT.

[0092] As an example, FIG. 4 depicts a first acoustic reflectometry signal 402, for a first patient, and a second acoustic reflectometry signal 404, for a second patient. The first acoustic reflectometry signal 402 was generated for a patient with a 2.5 mm endotracheal tube, and the second acoustic reflectometry signal 404 was generated for a patient with a 3.0 mm endotracheal tube. The first acoustic reflectometry signal 402 and the second acoustic reflectometry signal 404 may be analyzed by different portions in time at which the signals were received by the acoustic sensor.

[0093] A time scale 406 is provided for reference for both the first acoustic reflectometry signal 402 and the second acoustic reflectometry signal 404. On the time scale, four different discretetime points (to, ti, t2, and ts) are labeled as an example. Example time windows are similarly identified on the time scale 406. For instance, a first time window (W i) occurs between to and ti, and second time window (W2) occurs between ti and t2, a third time window (W3) occurs between time t2 and tv and a fourth time window (W4) occurs between time ts and t4.

[0094] The first time window may correspond to acoustic reflections that occur from a position most proximate to the acoustic sensor, such as reflections that are most likely to occur from within the endotracheal tube or from the tip of the endotracheal tube. The second time window corresponds to acoustic reflections that occur from a position that is further from the acoustic sensor (e.g., more distal from the acoustic sensor) than the reflections of the first time window. For instance, reflection data that occurs within the second time window may correspond to reflections occurring in the trachea and / or the primary bronchi. Accordingly, the acoustic reflections in the second time window may be used, among other things, to assess passageway size of the passageway into which the ETT is positioned. The third time window corresponds to reflections that occur from a position that is even further from the acoustic sensor than the reflections of the second time window. For instance, reflection data that occurs within the third time window may correspond to reflections occurring from the small bronchi and bronchioles, such as the further distal bronchi (e.g., secondary bronchi, tertiary bronchi, bronchioles). The reflection data that occurs within the fourth time window may correspond to reflections occurring from anatomical structures further distal than the structures causing the reflection within the third time window, such as the smallest airways, alveoli, and / or parenchyma.

[0095] As can be seen from the first acoustic reflectometry signal 402 and the second acoustic reflectometry signal 404, the positions of the time windows may vary between patient types and based on the size of the ETT. For instance, the first time window may change based on the size the ETT. For a larger / longer ETT, the first time window is longer. In some examples, the durations of the second through fourth time windows may also be different based on the types of patient (e.g., patient size). Accordingly, the boundaries for the particular time windows may be selected based on the ETT size and / or the patient type.

[0096] The different time windows of the acoustic reflectometry may be used as separate inputs to the ML / AI models depending on the type of clinical condition or endpoint that is to be detected by the ML / AI model. For instance, for an ML / AI model that is trained for classifying asthma, the third time window of the acoustic reflectometry signal (corresponding to smallerbronchi or bronchioles) may be used as an input for the ML / AI model. Similarly, for an ML / AI model that is trained for classifying pneumonia, the third time window of the acoustic reflectometry signal (corresponding to the secondary bronchi, tertiary bronchi, or bronchioles) and / or the fourth time window may be used as input for the ML / AI model.

[0097] FIG. 5 depicts an example system 500 for determining a clinical condition or endpoint. The example system 500 includes an acoustic reflectometry system 502 (e.g., an acoustic sensor device 230 and the acoustic monitor 246), an MPM 504, and a ventilator 506. The acoustic reflectometry system 502, the MPM 504, and the ventilator 506 are in communication with a clinical endpoint subsystem 510. The clinical endpoint subsystem 510 includes an input data selector 512 and one or more ML / AI models. In the example depicted, the clinical endpoint subsystem 510 includes a first ML / AI model 514, a second ML / AI model 516, and an Nth ML / AI model 518.

[0098] The acoustic reflectometry system 502, the MPM 504, and / or the ventilator 506 send the various data discussed herein to the clinical endpoint subsystem 510. For instance, the acoustic data, such as the acoustic reflectometry data and / or the detected sounds from the patient, are provided by the acoustic reflectometry system 502. These detected sounds may be processed to provide frequency-based data to the ML / AI model, such as spectral analysis (periodogram) with segmented frequency bands of interest. The MPM 504 may provide additional sensor data for the patient, such as the SpCh, EtCCh, ECG, HR, movement data, and / or other data measured from a patient sensor. The ventilator 506 provides ventilator-related data to the clinical endpoint subsystem 510. In other examples, the various types of data may be communicated to the clinical endpoint subsystem 510 via different devices. For instance, the sensor data may be communicated to the clinical endpoint subsystem 510 via the acoustic reflectometry system 502, the ventilator 506, and / or directly the patient sensors themselves.

[0099] In some examples, the acoustic reflectometry system 502, MPM 504, and / or the ventilator 506 provide a specific subset of data to the clinical endpoint subsystem 510. For example, the clinical endpoint subsystem 510 may provide a request for a subset of data depending on the types of ML / AI models within the clinical endpoint subsystem 510. In other examples, the acoustic reflectometry system 502, MPM 504, and / or the ventilator 506 provide a stream or batches of available data to the clinical endpoint subsystem 510 for processing by the clinical endpoint subsystem 510.

[0100] The clinical endpoint subsystem 510 may be executed by a standalone computing device and / or be implemented in one or more devices of the system 500. For instance, the clinical endpoint subsystem 510 may be stored and executed by the acoustic reflectometry system 502, the MPM 504, and / or the ventilator 506. In other examples, the clinical endpoint subsystem 510 is hosted on a server that is in communication with the acoustic reflectometry system 502, the MPM 504, and / or the ventilator 506. The server may be located within the same hospital or medical facility as the acoustic reflectometry system 502, the MPM 504, and / or the ventilator 506. In such examples, communication may be accomplished through a hospital or medical facility network. In other examples, the server may be a cloud-based server that is accessible via the Internet. In still other examples, the subcomponents of the clinical endpoint subsystem 510 may be distributed across multiple devices. For instance, one or more of the ML / AI models 514-518 may be hosted on a cloud-based server, while other components, such as the input data selector 512, may reside on other devices, such as devices on the premises of the hospital or medical facility.

[0101] The input data selector 512 of the clinical endpoint subsystem 510 receives the data from the acoustic reflectometry system 502, the MPM 504, and / or the ventilator 506. The input data selector 512 then selects particular subsets of data from the received data that is to be passed to one or more of the ML / AI models 514-518. For instance, each of the ML / AI models 514-518 may require a specific subset of data to be used as an input to determine the particular condition for which the ML / AI model was trained. The input data selector 512 selects that particular data from the received data and provides the subset of data as input to the particular ML / AI models 514-518.

[0102] As an example, the first ML / AI model 514 may be trained to detect if the patient has asthma. The first ML / AI model 514 does so by classifying input data. The particular input data for the first ML / AI model 514 may include at least a first portion of the acoustic reflection signal (e.g., the third time window in FIG. 4 corresponding to smaller bronchi or bronchioles), an obstruction size within the ETT (determined from the acoustic data), and airway size distal to the ETT tip (determined from the acoustic data). Additional aggregated data for asthma may also include patient demographic data, movement data, and / or raw acoustic signal data. Additional sensor data for the patient may also be used. Accordingly, the input data selector 512 may select or filter the received data to include the particular inputs for the first ML / AI model 514. Such selection may include filtering the acoustic data such that only the acousticreflectometry signal from the second time window is provided. The first ML / AI model 514 then classifies the input data, which may result in a classification of asthma or no asthma. In other examples, the classification outputs of the first ML / AI model 514 may not be binary. Rather, the output may be a range of likelihood that the patient has asthma and / or a severity factor related to asthma (or other clinical conditions in other examples).

[0103] As another example, the second ML / AI model 516 may be trained to determine a plurality of different clinical conditions or endpoints, such as pneumonia, atelectasis, and pleural effusion. The input data selector 512 selects the appropriate data for the second ML / AI model 516 based on the type of data on which the second ML / AI model 516 was trained. For instance, the input data selector 512 filters the acoustic reflectometry signal to include acoustic data within the third time window and / or fourth time window of FIG. 4. Additional data may also be selected and provided as input.

[0104] As still another example, the Nth ML / AI model 518 may be trained to determine a cardiac condition. In such examples, the input data selector 512 selects the appropriate data for the Nth ML / AI model 518. In examples of determining cardiac conditions, the acoustic data may be most likely to capture cardiac sounds during pulmonary “quiet” periods of the patient’s breathing. The pulmonary quiet periods may occur at the end of the expiration, such as right before an inhalation is initiated. The input data selector 512 may select these portions of the acoustic data by analyzing the ventilation data as well. For instance, by analyzing the ventilation data, the input data selector 512 is able to identify time segments that correspond to the end of exhalations. The input data selector 512 then selects portions of the acoustic data corresponding to those end-of-exhalation time segments. Additional sensor data and patient demographic data may also be selected based on the data on which the Nth ML / AI model 518 was trained.

[0105] The outputs from the clinical endpoint subsystem 510 include the outputs generated from each of the ML / AI models 514-518. For instance, the outputs may include a first clinical endpoint 520, a second clinical endpoint 522, and a Nth clinical endpoint 524. The clinical endpoints may be a binary determination of a particular endpoint being present for the patient or not. In other examples, the output includes a probability or likelihood of the clinical endpoint or condition being present. Additionally or alternatively, the output may further indicate the severity of the clinical endpoint or condition.

[0106] In some examples, the output may include additional details for a particular condition or endpoint that is identified. For instance, in examples where the clinical condition is pneumonia, the output may also include additional details about the pneumonia condition of the patient. Such additional details may include a number of lobes or a particular identification of which lobes are consolidated. In such examples, the respective ML / AI model that generates the additional details is also trained to produce such an output. For instance, the aggregated data received for training not only includes a label of pneumonia (or no pneumonia) but also an indication of lobe obstruction (which may have been positively confirmed via x-ray or other imaging modalities).

[0107] The outputs (e.g., clinical endpoints 520-524) are then surfaced (e.g., displayed). The outputs may be displayed on any of the devices of the system, such as the acoustic reflectometry system 502, the MPM 504, and / or the ventilator 506. Alternatively or additionally, the outputs may be surfaced on another device, such as a patient terminal or other computing device.

[0108] FIG. 6 depicts an example method 600 for determining a clinical endpoint from acoustic reflectometry data. The method 600 may be performed by computing device and / or a processor within a medical device, such as a monitor, ventilator, etc. For instance, the medical device may include a memory that stores instructions, that when executed by the processor, cause the medical device to perform the operations of method 600.

[0109] At operation 602, acoustic data is received. The acoustic data is generated from an acoustic sensor package and / or monitor, such as those discussed herein that are a part of an acoustic reflectometry system. For instance, an acoustic sensor coupled to an ETT senses sound waves. The sensed sound waves may include natural sounds generated from the patient along with generated sound pulses from an acoustic generator of the package and the reflections thereof.

[0110] The acoustic data that is received may be in various forms. For instance, the acoustic data may be in the form of raw, composite acoustic-based signals that represent the composite sound detected by the acoustic signals. The acoustic data may also be filtered data such that the acoustic data primarily corresponds to the reflections of the generated sound pulses. The acoustic data may also include processed acoustic reflectometry data in the form of acoustically measurable parameters, such as ETT position, ETT movement, ETT obstruction position and / or size, airway constriction or expansion (e.g., airway size directly distal to the ETT,passageways such as esophagus, bronchus, trachea, upper airway, mouth), among other parameters measurable by the acoustic reflectometry system.[oni] At operation 604, clinical data is received. The clinical data includes measurement or data for the patient that is measured or generated from one or more patient sensors. Example sensor data includes SpCh, EtCCh, ECG, HR, patient movement, and / or temperature, among other types of sensor data. The clinical data may also include additional information received from other medical devices, such as ventilator, regarding the current treatment or ventilation of the patient. For instance, the ventilator may provide data relating to the current ventilation settings for the patient as well as measurable parameters of the ventilator

[0112] At operation 606, patient demographic data may be received. The patient demographic data may include data such as age, weight, height, gender, body mass index (BMI), health condition (e.g., COPD, heart condition, etc.) and / or other types of patient-specific data.

[0113] The acoustic data received in operation 604 and / or the clinical data received in operation 606 may be received in an ongoing or substantially continuous basis, such as a stream of data. In other examples, the acoustic data and / or the clinical data may be received on an interval basis and include data for an interval of time. For instance, the interval of time may be less than a minute, 1-5 minutes, 5-30 minutes, or more than 30 minutes of data collected by the various sensors and devices. The frequency at which the data packages are received may be dependent on the type of clinical conditions that are being detected. For clinical conditions that are unlikely to change rapidly, the frequency of receiving the data may be lower (e.g., hourly, twice daily, etc.). In still other examples, the data is received based on changes in data. For instance, for some ventilator settings, they may change infrequently, and the data may be received only upon a change to the data occurring. The patient demographic data received in operation 606, however, is received only once is some examples as the patient demographic data is unlikely to change while the patient is being ventilated.

[0114] At operation 608, one or more subsets of the data received in operations 602-606 are selected. The subsets of the data are based on the particular ML / AI models that are available and being implemented. For example, as discussed above, each of the ML / AI models may have a particular set of input types on which the particular ML / AI model was trained. For instance, when a particular ML / AI model is registered with the system, the ML / AI model may provide the particular inputs that the ML / AI model requires and any other specifications forthose inputs. The received data may then be selected and provided to the ML / AI model(s) based on the registered inputs. In addition, the selection of the data may further include the fdtering and or alteration of the data to provide the particular inputs that are required, such as selecting particular time windows of acoustic reflectometry data, as discussed above. In other examples, the subsets of data may not need to be selected, and the data that is received is either all provided to the ML / AI models and / or the data that is received has been pre-selected and / or pre-filtered by the device providing the data. For example, the acoustic reflectometry system may pre-select a particular time window of data to provide.

[0115] Once the subset(s) of data are identified, the subset(s) of data are provided to the ML / AI models according to the input types required for each of the ML / AI models at operation 610. The ML / AI models then process, at operation 612, the received data to generate one or more outputs including indications of clinical conditions and / or endpoints for the patient for which the data was received. The outputs are then received in operation 614.

[0116] The outputs may also include likelihoods of the clinical conditions, severity of the clinical conditions, and / or additional details about the clinical conditions. In some examples, the outputs may also include a recommended change to one or more ventilation settings based on the detected condition or endpoint. As an example, where the condition is a tracheal collapse or pneumonia, an increase in PEEP may be recommended.

[0117] At operation 616, the indications of the clinical conditions or endpoints that are received in the outputs are surfaced. The indications may be surfaced via a medical device, such as acoustic monitor, a multi-parameter monitor, and / or a ventilator, among other types of medical devices. The indications may also be surfaced as an alarm, which may be a visual and / or audible alarm.

[0118] At operation 618, in some examples, control instructions are generated to change the current ventilation strategy or settings of the ventilator providing ventilation to the patient, the control instructions may be generated based on the outputs from the ML / AI models. For instance, the outputs of the ML / AI models may directly include the changes to ventilation settings that are to be made. In such examples, generation of the control instructions is straightforward. In some examples, however, the recommend setting change may be first surfaced or presented on the ventilator before being implemented. The settings change may then be implemented upon receiving a confirmatory input from a clinician.

[0119] In other examples, the outputs from the ML / AI models do not include the specific ventilation changes that should be implemented based on the indication of the condition or endpoint. In such examples, operation 618 may include applying one or more functions or heuristics to determine changes to the ventilation that should be made. For instance, the heuristics analyze the indication of the clinical condition and current ventilator settings to determine a recommended change. As an example, the condition detected may be pneumonia, and the heuristics may compare a medically recognized ventilator setting range (e.g., PEEP range), for patients with pneumonia having the patient demographics of the patient, with the current setting for the patient (e.g., current PEEP setting). Based on a difference between the recognized ventilator setting range and the current setting, a recommend ventilator setting change may be determined. The recommend ventilator setting change may then be automatically implemented by generating a control instruction. In other examples, the recommended ventilator setting change is first presented on the ventilator and implemented only upon a confirmatory input from the clinician.

[0120] The method 600 then repeats by returning to operations 602 and 604 to receive updated acoustic data and clinical data. As discussed above, after the first performance of the method 600, the patient demographics received in operation 606 may not need to be received again as they are unlikely to change. As discussed above, the rate at which method 600 repeats may be based on the types of conditions that are being detected. For instance, for a tracheal collapse condition, which is likely to occur fairly rapidly, the method 600 may repeat at a high rate (e.g., every 10 seconds or less) such that the corresponding ML / AI model is substantially continuously evaluating and classifying updated acoustic data and / or sensor data. As should be appreciated, the method 600 may be performed concurrently with the ventilation of the patient such that conditions can be detected in real time while treatment options may be implemented for the patient.

[0121] Those skilled in the art will recognize that the methods and systems of the present disclosure may be implemented in many manners and as such are not to be limited by the foregoing aspects and examples. In other words, functional elements being performed by a single or multiple components. In this regard, any number of the features of the different aspects described herein may be combined into single or multiple aspects, and alternate aspects having fewer than or more than all of the features herein described are possible. Functionalitymay also be, in whole or in part, distributed among multiple components, in manners now known or to become known.

[0122] Furthermore, those skilled in the art will recognize that boundaries between the functionality of the above-described operations are merely illustrative. The functionality of multiple operations may be combined into a single operation, and / or the functionality of a single operation may be distributed in additional operations. Some operations may be omitted. Moreover, alternative embodiments may include multiple instances of a particular operation, and the order of operations may be altered in various other embodiments.

[0123] Further, as used herein and in the claims, the phrase “at least one of element A, element B, or element C” is intended to convey any of: element A, element B, element C, elements A and B, elements A and C, elements B and C, and elements A, B, and C. In addition, one having skill in the art will understand the degree to which terms such as “about” or “substantially” convey in light of the measurement techniques utilized herein. To the extent such terms may not be clearly defined or understood by one having skill in the art, the term “about” shall mean plus or minus ten percent.

[0124] Numerous other changes may be made which will readily suggest themselves to those skilled in the art and which are encompassed in the spirit of the disclosure and as defined in the appended claims. While various aspects have been described for purposes of this disclosure, various changes and modifications may be made which are well within the scope of the disclosure. Numerous other changes may be made which will readily suggest themselves to those skilled in the art and which are encompassed in the spirit of the disclosure and as defined in the claims.

[0125] The following examples are illustrative of the techniques described herein.

[0126] Example 1. An acoustic reflectometry system comprising: a generator that emits acoustic pulses; an acoustic sensor that detects the emitted acoustic pulses and reflections thereof; a processor; and memory storing instructions that, when executed by the processor, causes the system to perform operations comprising: receive, from the acoustic sensor, acoustic data detected by the acoustic sensor from a patient; receive clinical data associated with the patient; provide at least a portion of the received acoustic data and the clinical data as an input to a trained machine -learning (ML) model; receive, as output from the trained ML model inresponse to the input, an indication of a clinical condition; and surface the indication of the clinical condition.

[0127] Example 2. The acoustic reflectometry system of Example 1, wherein the acoustic data includes at least one of: (1) a composite acoustic-based signal representing reflections of the emitted acoustic pulses and natural sounds generated by the patient, (2) a fdtered acousticbased signal representing the reflections of the emitted acoustic pulses, or (3) an acoustically measurable parameter generated from detected reflections of the emitted acoustic pulses.

[0128] Example 3. The acoustic reflectometry system of Example 2, wherein the acoustic data includes the acoustically measurable parameter, and the acoustically measurable parameter is one of tracheal tube movement, tracheal tube position, obstruction size, or passageway size.

[0129] Example 4. The acoustic reflectometry system of Example 2, wherein the acoustic data includes the composite acoustic-based signal.

[0130] Example 5. The acoustic reflectometry system of Example 2, wherein the acoustic data includes the fdtered acoustic-based signal.

[0131] Example 6. The acoustic reflectometry system of Example 5, wherein the operations further comprise selecting a particular time window from the fdtered acoustic-based signal for use in the input to the ML model, wherein the particular time window corresponds to anatomical reflections from anatomical structures relevant to the clinical condition.

[0132] Example 7. The acoustic reflectometry system of Example 1, wherein the clinical data includes at least one of capnometry data, pulse oximetry data, electrocardiogram data, or patient movement data.

[0133] Example 8. The acoustic reflectometry system of Example 1, wherein the clinical data includes ventilation data including at least one of a current ventilation mode, a ventilation setting, a pressure value, a flow value, a respiratory rate, work of breathing, or indication of patient-initiated spontaneous breaths.

[0134] Example 9. The acoustic reflectometry system of Example 1, wherein the clinical condition is one of asthma, pneumonia, pleural effusion, atelectasis, tracheal collapse, or a cardiac condition.

[0135] Example 10. The acoustic reflectometry system of Example 9, wherein the indication is an alarm.

[0136] Example 11. An acoustic reflectometry system comprising: a generator that emits acoustic pulses; an acoustic sensor that detects the emitted acoustic pulses and reflections thereof; a processor; and memory storing instructions that, when executed by the processor, causes the system to perform operations comprising: receive, from the acoustic sensor, acoustic data detected by the acoustic sensor, wherein the acoustic data comprises: a first portion of the acoustic data from a first time window corresponding to reflections, of the emitted acoustic pulses from, a first anatomical structure; a second portion of the acoustic data from a second time window corresponding to reflections, of the emitted acoustic pulses, from a second anatomical structure; provide the first portion of the acoustic data to first machine-learning (ML) model trained to classify a first clinical condition; receive, as output from the first ML model, an indication of the first clinical condition; provide the second portion of the acoustic data to a second ML model trained to classify a second clinical condition; receive, as output from the second ML model, an indication of the second clinical condition; and surface the indication of the first clinical condition and the second clinical condition.

[0137] Example 12. The acoustic reflectometry system of Example 11, wherein the first clinical condition is asthma.

[0138] Example 13. The acoustic reflectometry system of Example 12, wherein the first anatomical structure is at least one of a small bronchus or bronchiole.

[0139] Example 14. The acoustic reflectometry system of Example 11, wherein the second clinical condition is at least one pneumonia, atelectasis, or pleural effusion.

[0140] Example 15. The acoustic reflectometry system of Example 14, wherein the second anatomical structure is at least one of a secondary bronchus or a tertiary bronchus.

[0141] Example 16. A method for detecting clinical conditions with an acoustic system, the method comprising: receiving acoustic data generated from an acoustic sensor coupled to a tracheal tube of a patient being ventilated, wherein the acoustic data includes at least one of: (1) a composite acoustic -based signal representing reflections of emitted acoustic pulses and natural sounds generated by the patient, (2) a filtered acoustic -based signal representing thereflections of the emitted acoustic pulses, or (3) an acoustically measurable parameter generated from detected reflections of the emitted acoustic pulses; providing at least a portion of the received acoustic data as an input to a trained machine-learning (ML) model; receiving, as output from the trained ML model in response to the input, an indication of a clinical condition; and surfacing the indication of the clinical condition, wherein the clinical condition comprises one of asthma, pneumonia, pleural effusion, atelectasis, or tracheal collapse.

[0142] Example 17. The method of Example 16, further comprising, based on the output from the trained ML model, generating control instructions to change a ventilation setting of the ventilator.

[0143] Example 18. The method of Example 16, further comprising receiving clinical data generated from one or more patient sensors connected to a patient, wherein the clinical data includes at least one of capnometry data, pulse oximetry data, electrocardiogram data, patient movement data, a current ventilation mode, a ventilation setting, a pressure value, a flow value, a respiratory rate, work of breathing, or indication of patient-initiated spontaneous breaths.

Claims

CLAIMSWhat is claimed is:

1. An acoustic reflectometry system (100) comprising: a generator (132) that emits acoustic pulses; an acoustic sensor (134) that detects the emitted acoustic pulses and reflections thereof; a processor (150); and memory (152) storing instructions that, when executed by the processor, causes the system to perform operations comprising: receive, from the acoustic sensor, acoustic data detected by the acoustic sensor from a patient; receive clinical data associated with the patient; provide at least a portion of the received acoustic data and the clinical data as an input to a trained machine-learning (ML) model; receive, as output from the trained ML model in response to the input, an indication of a clinical condition; and surface the indication of the clinical condition.

2. The acoustic reflectometry system of claim 1, wherein the acoustic data includes at least one of: (1) a composite acoustic-based signal representing reflections of the emitted acoustic pulses and natural sounds generated by the patient, (2) a filtered acoustic-based signal representing the reflections of the emitted acoustic pulses, or (3) an acoustically measurable parameter generated from detected reflections of the emitted acoustic pulses.

3. The acoustic reflectometry system of claim 2, wherein the acoustic data includes the acoustically measurable parameter, and the acoustically measurable parameter is one of tracheal tube movement, tracheal tube position, obstruction size, or passageway size.

4. The acoustic reflectometry system of claim 2, wherein the acoustic data includes the composite acoustic-based signal.

5. The acoustic reflectometry system of claim 2, wherein the acoustic data includes the filtered acoustic-based signal.

6. The acoustic reflectometry system of claim 5, wherein the operations further comprise selecting a particular time window from the filtered acoustic -based signal for use in the input to the ML model, wherein the particular time window corresponds to anatomical reflections from anatomical structures relevant to the clinical condition.

7. The acoustic reflectometry system of claim 1, wherein the clinical data includes at least one of capnometry data, pulse oximetry data, electrocardiogram data, or patient movement data.

8. The acoustic reflectometry system of claim 1, wherein the clinical data includes ventilation data including at least one of a current ventilation mode, a ventilation setting, a pressure value, a flow value, a respiratory rate, work of breathing, or indication of patient- initiated spontaneous breaths.

9. The acoustic reflectometry system of claim 1, wherein the clinical condition is one of asthma, pneumonia, pleural effusion, atelectasis, tracheal collapse, or a cardiac condition.

10. The acoustic reflectometry system of claim 9, wherein the indication is an alarm.

11. An acoustic reflectometry system (100) comprising : a generator (132) that emits acoustic pulses; an acoustic sensor (134) that detects the emitted acoustic pulses and reflections thereof; a processor (150); and memory (152) storing instructions that, when executed by the processor, causes the system to perform operations comprising: receive, from the acoustic sensor, acoustic data detected by the acoustic sensor, wherein the acoustic data comprises: a first portion of the acoustic data from a first time window corresponding to reflections, of the emitted acoustic pulses from, a first anatomical structure;a second portion of the acoustic data from a second time window corresponding to reflections, of the emitted acoustic pulses, from a second anatomical structure; provide the first portion of the acoustic data to first machine-learning (ML) model trained to classify a first clinical condition; receive, as output from the first ML model, an indication of the first clinical condition; provide the second portion of the acoustic data to a second ML model trained to classify a second clinical condition; receive, as output from the second ML model, an indication of the second clinical condition; and surface the indication of the first clinical condition and the second clinical condition.

12. The acoustic reflectometry system of claim 11, wherein the first clinical condition is asthma.

13. The acoustic reflectometry system of claim 12, wherein the first anatomical structure is at least one of a small bronchus or bronchiole.

14. The acoustic reflectometry system of claim 11, wherein the second clinical condition is at least one pneumonia, atelectasis, or pleural effusion.

15. The acoustic reflectometry system of claim 14, wherein the second anatomical structure is at least one of a secondary bronchus or a tertiary bronchus.

Images