Method and system for detecting an exacerbation event in a respiratory waveform

Abstract

A method of detecting an exacerbation event in a respiratory waveform is provided. The method includes: generating a test set of trend features from a respiratory waveform of a subject's breathing with respect to a time window; and detecting for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window. In particular, the test set of trend features includes one or more trend features relating to a respiratory rate, one or more trend features relating to a tidal volume and one or more trend features relating to a modified respiratory duty cycle. The modified respiratory duty cycle is determined based on a ratio of a difference between an inhalation time and an exhalation time to a total breathing time. There is also provided a corresponding system for detecting an exacerbation event in a respiratory waveform.

Description

METHOD AND SYSTEM FOR DETECTING AN EXACERBATION EVENT IN ARESPIRATORY WAVEFORMCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of priority of Singapore Patent Application No. 10202401779T filed on 19 June 2024, the content of which being hereby incorporated by reference in its entirety for all purposes.TECHNICAL FIELD

[0002] The present invention generally relates to a method of detecting an exacerbation event in a respiratory waveform, and a system thereof, and more particularly, with respect to exacerbations of chronic respiratory disease.BACKGROUND

[0003] Exacerbations of chronic cardiopulmonary diseases are a major cause of morbidity and mortality worldwide. There are an estimated 23 million patients with heart failure worldwide, and the prevalence of heart failure in the United States is projected to rise over the next four decades with an estimated 772,000 new heart failure cases projected in the year 2040. Exacerbations of chronic respiratory disease (e.g., Chronic Obstructive Pulmonary Disease (COPD) exacerbations) can accelerate lung function decline and reduce survival. They also lead to a significant rise to the cost of healthcare. COPD exacerbations are an important cause of readmissions with a 30-day readmission rate of approximately 20% and an associated expenditure of an estimated US $15 Billion in annual healthcare spending. For example, Cystic fibrosis (CF), a genetic disorder that affects airways clearance and secretions, has a 30-day readmission rate of approximately 11%.

[0004] Due to the high cost of hospital stays and emergency department visits, more cost- effective “out-of-hospital” healthcare management plans or models have become increasingly appealing. Such healthcare management plans not only provide cost benefits to patients and hospitals, but also increase the ability to provide healthcare to people at home.

[0005] Various methods for predicting risk of exacerbations of chronic respiratory disease have been proposed, for example, using past patient history and course of antibiotics and / or oral corticosteroids, admission to hospital, or questionnaires that document patient-reported change in symptoms. These are in general methods based on a single time point measurement.However, such existing methods, limited by evaluating such a single time point measurement or feature to detect for an exacerbation, were found to be not sufficiently effective or accurate, for example, given that the data did not encompass real-world utilizations or they were sampled at time-points that were disparate and significantly large. There also exist methods for detecting for an exacerbation of chronic respiratory disease based on traditional threshold-type driven analytics on respiratory measurements or features of the respirator}' waveform. However, such existing methods, which rely on predefined thresholds for respiratory measurements or features of the respiratory waveform to detect an exacerbation, were also found to be not sufficiently effective or accurate in detecting for an exacerbation in the respirator}' waveform, as well as being highly inefficient. For example, the use of a simple threshold-based respiratory rate requires defining a baseline for patients and which can often change across the course of medication and treatment.

[0006] A need therefore exists to provide a method of detecting an exacerbation event in a respiratory waveform, as well as a system thereof, that seeks to overcome, or at least ameliorate, one or more deficiencies in conventional exacerbation detection methods, and more particularly, with improved exacerbation detection effectiveness or accuracy. It is against this background that the present invention has been developed.SUMMARY

[0007] According to a first aspect of the present invention, there is provided a method of detecting an exacerbation event in a respiratory waveform using at least one processor, the method comprising: generating a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and detecting for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window, wherein the test set of trend features comprises one or more trend features relating to a respiratory rate, one or more trend features relating to a tidal volume and one or more trend features relating to a modified respiratory duty cycle, the modified respiratory duty cycle being determined based on a ratio of a difference between an inhalation time and an exhalation time to a total breathing time.

[0008] According to a second aspect of the present invention, there is provided a system for detecting an exacerbation event in a respiratory waveform, the system comprising: at least one memory; and at least one processor communicatively coupled to the at least one memory and configured to: generate a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and detecting for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window, wherein the test set of trend features comprises one or more trend features relating to a respiratory rate, one or more trend features relating to a tidal volume and one or more trend features relating to a modified respiratory duty cycle, the modified respiratory duty cycle being determined based on a ratio of a difference between an inhalation time and an exhalation time to a total breathing time.

[0009] According to a third aspect of the present invention, there is provided a computer program product, embodied in one or more non-transitory computer-readable storage mediums, comprising instructions executable by at least one processor to perform the method of detecting an exacerbation event in a respiratory waveform according to the above-mentioned first aspect of the present invention.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Embodiments of the present invention will be better understood and readily apparent to one of ordinary skill in the art from the following written description, by way of example only, and in conjunction with the drawings, in which:FIG. 1 depicts a schematic diagram of a method of detecting an exacerbation event in a respiratory waveform, according to various embodiments of the present invention;FIG. 2 depicts a schematic block diagram of a system for detecting an exacerbation event in a respirator}' waveform, according to various embodiments of the present invention;FIG. 3 illustrates example trend features that may be generated / extracted from an example respiratory waveform with respect to an example time window of 5 days, according to various embodiments of the present invention;FIG. 4 depicts a schematic flow diagram illustrating an example process of the boosted tree method for predicting whether exacerbation exists, according to various embodiments of the present invention;FIG. 5 depicts a Table presenting a univariate analysis performed on the trained exacerbation machine learning model configured as a mixed effects logistic regression model, according to various example embodiments of the present invention; andFIG. 6 depicts a Table presenting the performance statistics of both the mixed effect logistic regression method and the boosted tree (or tree boosting) method with mixed effects, according to various example embodiments of the present invention.DETAILED DESCRIPTION

[0011] Various embodiments of the present invention provide a method of detecting an exacerbation event in a respiratory waveform, and a system thereof, and more particularly, with respect to exacerbations of chronic respiratory disease, such as Chronic Obstructive Pulmonary Disease (COPD) exacerbations.

[0012] As discussed in the background, there exist various methods for detecting for an exacerbation of chronic respiratory disease, including those based on a single time point measurement or feature of the respiratory waveform and those based on traditional thresholdtype driven analytics on respiratory measurements or features of the respiratory waveform. However, such existing methods were found to be not sufficiently effective or accurate in detecting for an exacerbation in the respiratory waveform. In this regard, various embodiments of the present invention provide a method of detecting an exacerbation event in a respiratory waveform, as well as a system thereof, that seeks to overcome, or at least ameliorate, one or more deficiencies in conventional exacerbation detection methods, and more particularly, with improved exacerbation detection effectiveness or accuracy.

[0013] FIG. 1 depicts a schematic diagram of a method 100 of detecting an exacerbation event in a respiratory waveform (with respiratory wave pattern) using at least one processor. The method 100 comprises: generating (at 106) a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and detecting (at 108) for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window. In particular, the test set of trend features comprises one or more trend features relating to a respiratory rate, one or more trend features relating to a tidal volume(or depth of breathing) and one or more trend features relating to a modified respiratory duty cycle. The modified respiratory duty cycle (which may simply be referred to as modified duty cycle) is determined based on a ratio of a difference between an inhalation time and an exhalation time to a total breathing time. In this regard, the total breathing time is a sum of the inhalation time and the exhalation time.

[0014] The method 100 of detecting an exacerbation event in a respiratory waveform according to various embodiments of the present invention advantageously has improved exacerbation detection effectiveness or accuracy. Firstly, trend features are generated or extracted from the respiratory waveform with respect to a time window (c.g., a predefined number of days), thereby advantageously enabling characteristics of trend features of the respiratory waveform in the time window to be captured and utilized in detecting for an exacerbation event in the respiratory waveform. This is in stark contrast to existing methods configured to detect for an exacerbation of chronic respiratory disease based on a single time point measurement or feature of the respiratory waveform as discussed in the background. In addition, the method 100 utilizes an exacerbation machine learning model trained based on trend features of respiratory waveform to provide an exacerbation prediction output to detect for an exacerbation event. In this regard, the exacerbation machine learning model is advantageously able to exploit characteristics of trend features of the respiratory waveform and learn to make a prediction on exacerbation based on characteristics of the trend features. This is in stark contrast to existing methods configured to detect for an exacerbation of chronic respiratory disease based on traditional threshold-type driven analytics on respiratory measurements or features of the respiratory waveform as discussed in the background. Furthermore, the method 100 generates, and detects for an exacerbation event based on, a test set of trend features from the respiratory waveform with respect to a time window, which specifically comprises at least three different types, or at least three different groups of types, of trend features derived from at least three different types of respiratory measurements or variables (or parameters) of the respiratory waveform, respectively, namely, trend feature(s) relating to a respiratory rate, trend feature(s) relating to a tidal volume and trend feature(s) relating to a modified respiratory duty cycle (which is a ratio of a difference between an inhalation time and an exhalation time to a total breathing time). Various embodiments of the present invention found that utilizing such a combination of at least these three different types of trend features, especially with the inclusion of trend feature(s) relating to the modified respiratory duty cycle, for training an exacerbation machine learning model for detecting anexacerbation event results in surprisingly superior exacerbation detection performance, as will be discussed and demonstrated later below in experiments conducted according to various example embodiments of the present invention. Therefore, the method 100 of detecting an exacerbation event in a respiratory waveform according to various embodiments of the present invention advantageously has improved exacerbation detection effectiveness or accuracy. These advantages or technical effects, and / or other advantages or technical effects, will become more apparent to a person skilled in the art as the method 100 of detecting an exacerbation event, as well as the corresponding system for detecting an exacerbation event, is described in more detail according to various embodiments and example embodiments of the present invention.

[0015] In various embodiments, the exacerbation prediction output of the exacerbation machine learning model is an exacerbation probability score for the time window. The exacerbation probability score determined for the time window indicates a likelihood or a degree of confidence of the presence of an exacerbation event in the time window, which may be a number from 0 (corresponding to 0%) to 1 (corresponding to 100%).

[0016] In various embodiments, each trend feature of the one or more trend features relating to the respiratory rate is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the respiratory rate, or a combination of two or more thereof. Similarly, each trend feature of the one or more trend features relating to the tidal volume is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the tidal volume, or a combination of two or more thereof. Similarly, each trend feature of the one or more trend features relating to the modified respiratory duty cycle is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the modified respiratory duty cycle, or a combination of two or more thereof.

[0017] In various embodiments, the test set of trend features further comprises one or more trend features relating to the inhalation time and one or more trend features relating to the exhalation time. Similarly, each trend feature of the one or more trend features relating to the inhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the inhalation time, or a combination of two or more thereof. Similarly, each trend feature of the one or more trend features relating to the exhalation time is generated based on standard deviation, mean,median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the exhalation time, or a combination of two or more thereof. roois] In various embodiments, the respiratory waveform is a continuous respiratory waveform. In various embodiments, the method 100 is performed for each timestep of a series of timesteps to detect for an exacerbation event in the respiratory waveform in a time window associated with the timestep. Accordingly, in various embodiments, the method 100 may be continuously performed (e.g., in a loop) over time for continuously detecting (or monitoring) for an exacerbation event.

[0019] In various embodiments, the respiratory waveform of a subject’s breathing may be generated by a respiratory sensor device in real-time for continuous monitoring of the subject, or may be a prior collected sample. For example, in the case of the continuous monitoring of the subject in real-time, the respiratory sensor device may be a wearable respiratory sensor device.

[0020] In various embodiments, the exacerbation machine learning model is trained based on a training set of trend features extracted from a respiratory waveform with respect to a time window. Similar to the test set of trend features, the training set of trend features comprises one or more trend features relating to the respiratory rate, one or more trend features relating to the tidal volume and one or more trend features relating to the modified respiratory duty cycle. Accordingly, in various embodiments, the exacerbation machine learning model is trained based on training sets of trend features corresponding to the test set of trend features based on which the exacerbation machine learning model is built to detect for an exacerbation event in a respiratory waveform.

[0021] In various embodiments, similar or corresponding to the test set of trend features, each trend feature of the one or more trend features of the training set relating to the respiratory rate is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the respiratory rate, or a combination of two or more thereof. Similarly, each trend feature of the one or more trend features of the training set relating to the tidal volume is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the tidal volume, or a combination of two or more thereof. Similarly, each trend feature of the one or more trend features relating to the modified respiratory duty cycle is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum,cumulative mean or cumulative mean shift with respect to the modified respiratory duty cycle, or a combination of two or more thereof.

[0022] In various embodiments, similar or corresponding to the test set of trend features, the training set of trend features further comprises one or more trend features relating to the inhalation time and one or more trend features relating to the exhalation time. Each trend feature of the one or more trend features relating to the inhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the inhalation time, or a combination of two or more thereof. Each trend feature of the one or more trend features relating to the exhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the exhalation time, or a combination of two or more thereof.

[0023] In various embodiments, the exacerbation machine learning model is a logistic regression model. In various embodiments, the exacerbation machine learning model is a boosted tree model. It will be appreciated by a person skilled in the art that the exacerbation machine learning model is not limited to any particular or specific network architecture or configuration and may be any type of machine learning model (or artificial intelligence (Al) model) as long as the machine learning model is capable of being trained based on trend features to detect or predict for an exacerbation event in a respiratory waveform according to various embodiments of the present invention.

[0024] FIG. 2 depicts a schematic block diagram of a system 200 for detecting an exacerbation event in a respiratory waveform, according to various embodiments of the present invention, corresponding to the above-mentioned method 100 of detecting an exacerbation event in a respiratory waveform as described hereinbefore with reference to FIG. 1 according to various embodiments of the present invention. The system 200 comprises: at least one memory 202; and at least one processor 204 communicatively coupled to the at least one memory 202 and configured to perform the method 200 of detecting an exacerbation event in a respiratory waveform according to various embodiments of the present invention. Accordingly, the at least one processor 204 is configured to: generate a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and detecting for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window. In particular, the test set of trend features comprises one or moretrend features relating to a respiratory rate, one or more trend features relating to a tidal volume and one or more trend features relating to a modified respiratory duty cycle. The modified respiratory duty cycle is determined based on a ratio of a difference between an inhalation time and an exhalation time to a total breathing time. In this regard, the total breathing time is a sum of the inhalation time and the exhalation time.

[0025] It will be appreciated by a person skilled in the art that the at least one processor 204 may be configured to perform various functions or operations through set(s) of instructions (e.g., software modules) executable by the at least one processor 204 to perform various functions or operations. Accordingly, as shown in FIG. 2, the system 200 may comprise: a trend feature generating module (or a trend feature generating circuit) 206 configured to generate a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and an exacerbation detection module (or exacerbation detection module) 208 configured to detect for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window.

[0026] It will be appreciated by a person skilled in the art that the above-mentioned modules are not necessarily separate modules, and two or more modules may be realized by or implemented as one functional module (e.g., a circuit or a software program) as desired or as appropriate without deviating from the scope of the present invention. For example, the trend feature generating module 206 and the exacerbation detection module 208 may be realized (e.g., compiled together) as one executable software program (e.g., software application), which for example may be stored in the at least one memory 202 and executable by the at least one processor 204 to perform the corresponding functions or operations as described herein according to various embodiments of the present invention.

[0027] In various embodiments, the system 200 for detecting an exacerbation event in a respiratory waveform corresponds to the method 100 of detecting an exacerbation event as described hereinbefore with reference to FIG. 1 , therefore, various operations, functions or steps configured to be performed by the least one processor 204 may correspond to various operations, functions or steps of the method 100 described hereinbefore according to various embodiments, and thus need not be repeated with respect to the system 200 for detecting an exacerbation event for clarity and conciseness. In other words, various embodiments described herein in context of methods (e.g., the method 100 of detecting an exacerbation event) are analogously valid for the corresponding systems or devices (e.g., the system 200 for detectingan exacerbation event), and vice versa. For example, in various embodiments, the at least one memory 202 may have stored therein the trend feature generating module 206 and / or the exacerbation detection module 208, which respectively correspond to various operations, functions or steps of the method 100 of detecting an exacerbation event as described hereinbefore according to various embodiments, which are executable by the at least one processor 204 to perform the corresponding operations, functions or steps as described herein.

[0028] A computing system, a controller, a microcontroller or any other system providing a processing capability may be provided according to various embodiments in the present invention. Such a system may be taken to include one or more processors and one or more computer-readable storage mediums. For example, the system 200 for detecting an exacerbation event described hereinbefore may include at least one processor 204 and at least one computer- readable storage medium (or memory) 202 which are for example used in various processing carried out therein as described herein. A memory or computer-readable storage medium used in various embodiments may be a volatile memory, for example a DRAM (Dynamic Random Access Memory) or a non-volatile memory, for example a PROM (Programmable Read Only Memory), an EPROM (Erasable PROM), EEPROM (Electrically Erasable PROM), or a flash memory, e.g., a floating gate memory, a charge trapping memory, an MRAM (Magnetoresistive Random Access Memory) or a PCRAM (Phase Change Random Access Memory).

[0029] In various embodiments, a “circuit” may be understood as any kind of a logic implementing entity, which may be special purpose circuitry or a processor executing software stored in a memory, firmware, or any combination thereof. Thus, in an embodiment, a “circuit” may be a hard-wired logic circuit or a programmable logic circuit such as a programmable processor, e.g., a microprocessor (e.g., a Complex Instruction Set Computer (CISC) processor or a Reduced Instruction Set Computer (RISC) processor). A “circuit” may also be a processor executing software, e.g., any kind of computer program, e.g., a computer program using a virtual machine code, e.g., lava. Any other kind of implementation of various functions or operations may also be understood as a “circuit” in accordance with various other embodiments. Similarly, a “module” may be a portion of a system according to various embodiments in the present invention and may encompass a “circuit” as above, or may be understood to be any kind of a logic-implementing entity therefrom.

[0030] Some portions of the present disclosure may be explicitly or implicitly presented in terms of algorithms and functional or symbolic representations of operations on data within a computer memory. These algorithmic descriptions and functional or symbolic representationsare the means used by those skilled in the data processing arts to convey most effectively the substance of their work to others skilled in the art. An algorithm may be, and generally, conceived to be a self-consistent sequence of steps leading to a desired result.

[0031] The present specification also discloses a system (e.g., which may also be embodied as one or more devices or apparatuses), such as the system 200 for detecting an exacerbation event, for performing various operations, functions or steps of various methods described herein. Such a system may be specially constructed for the required purposes or may comprise a general puipose computer system selectively activated or reconfigured by a computer program stored in the computer system. In general, various algorithms that may be presented herein arc not limited to being implemented or executed by any particular computer system. Alternatively, the construction of more specialized computer system to perform various operations, functions or steps of various methods described herein may be provided as desired or as appropriate without going beyond the scope of the present invention. For example, as described hereinbefore according to various embodiments of the present invention, the system 200 for detecting an exacerbation event may be configured as, or implemented in, a wearable respiratory sensor device for continuous monitoring of a subject in real-time for detecting an exacerbation of chronic respiratory disease, which may thus also be referred to as a wearable exacerbation monitoring or detection device.

[0032] In addition, the present specification also at least implicitly discloses computer program(s) or software / functional module(s), in that it would be apparent to a person skilled in the art that various operations, functions or steps of various methods described herein may be put into effect by computer code. The computer program(s) is not intended to be limited to any particular programming language and implementation thereof, and it will be appreciated by a person skilled in the art that a variety of programming languages and coding thereof may be used to implement the computer program(s). Moreover, the computer program(s) is not intended to be limited to any particular control flow as there are a variety of programming languages which can use different control flows. It will be appreciated by a person skilled in the art that a computer program may be stored on any computer-readable storage medium (non- transitory computer-readable storage medium), such as but not limited to, a magnetic disk, an optical disk or a memory chip. For example, a computer program stored on a computer-readable storage medium may be loaded and executed on a computer system to implement various operations, functions or steps of various methods described herein according to various embodiments of the present invention.

[0033] Accordingly, in various embodiments, there is provided a computer program product, embodied in one or more computer-readable storage mediums (non-transitory computer-readable storage medium), comprising instructions (e.g., the trend feature generating module 206 and / or the exacerbation detection module 208) executable by one or more computer processors to perform a method 100 of detecting an exacerbation event as described hereinbefore with reference to FIG. 1 according to various embodiments of the present invention. Accordingly, various computer programs or software modules described herein may be stored in a computer program product receivable by a system therein, such as the system 200 as shown in FIG. 2, for execution by at least one processor 204 of the system 200 to perform various operations, functions or steps of various methods described herein according to various embodiments of the present invention.

[0034] It will be appreciated by a person skilled in the art that various modules described herein (e.g., the trend feature generating module 206 and / or the exacerbation detection module 208) may be software module(s) realized by computer program(s) or set(s) of instructions executable by a computer processor to perform various functions or operations. Various modules described herein (e.g., the trend feature generating module 206 and / or the exacerbation detection module 208), together with the at least one processor 204 and the at least one memory 202, may also be implemented as hardware module(s) being functional hardware unit(s) designed to perform various functions or operations. More particularly, in the hardware sense, a module is a functional hardware unit designed for use with other components or modules. For example, a module may be implemented using discrete electronic components, or it may form a portion of an entire electronic circuit such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA). Numerous other possibilities exist. It will also be appreciated by a person skilled in the art that a combination of hardware and software modules may be implemented. Furthermore, various operations, functions or steps of various methods described herein may be performed in parallel rather than sequentially as desired or as appropriate (e.g., as long as it does not render the method(s) inoperable or unsatisfactory for its intended purpose).

[0035] It will be appreciated by a person skilled in the ait that the terminology used herein is for the purpose of describing various embodiments only and is not intended to be limiting of the present invention. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” and / or “comprising,” when used in this specification,specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0036] Any reference to an element or a feature herein using a designation such as “first”, “second” and so forth does not limit the quantity or order of such elements or features, unless stated or the context requires otherwise. For example, such designations may be used herein as a convenient way of distinguishing between two or more elements or instances of an element. Thus, a reference to first and second elements does not necessarily mean that only two elements can be employed, or that the first element must precede the second element, unless stated or the context requires otherwise. In addition, a phrase referring to “at least one of’ a list of items refers to any single item therein or any combination of two or more items therein.

[0037] In order that the present invention may be readily understood and put into practical effect, various example embodiments of the present invention will be described hereinafter by way of examples only and not limitations. It will be appreciated by a person skilled in the art that the present invention may, however, be embodied in various different forms or configurations and should not be construed as limited to the example embodiments set forth hereinafter. Rather, these example embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present invention to those skilled in the art.

[0038] As discussed in the background, there exist various methods for detecting for an exacerbation of chronic respiratory disease, including those based on a single time point measurement or feature of the respiratory waveform and those based on traditional thresholdtype driven analytics on respiratory measurements or features of the respiratory waveform. However, such existing methods were found to be not sufficiently effective or accurate in detecting for an exacerbation in the respiratory waveform.

[0039] In contrast, various example embodiments of the present invention provide a method of detecting an exacerbation event in a respirator}' waveform, whereby trend features are generated or extracted from the respiratory waveform with respect to a time window (e.g., a predefined number of days), thereby advantageously enabling characteristics of trend features of the respiratory waveform in the time window to be captured and utilized in detecting for an exacerbation event in the respiratory waveform. This is in stark contrast to existing methods configured to detect for an exacerbation of chronic respiratory disease based on a single time point measurement or feature of the respiratory' waveform as discussed in the background.Accordingly, the method of detecting an exacerbation event according to various example embodiments enables monitoring of physiologic data over time (e.g., long-term physiologic data), such as with respect to a time window of a number of days (e.g., 3 or 5 days), which advantageously enhances predictive ability and provides a personalized assessment. In addition, the method utilizes an exacerbation machine learning model trained based on trend features of respiratory w aveform to provide an exacerbation prediction output to detect for an exacerbation event. In this regard, the exacerbation machine learning model is advantageously able to exploit characteristics of trend features of the respiratory waveform and learn to make a prediction on exacerbation based on characteristics of the trend features. Accordingly, the method according to various example embodiments utilizes machine learning and data-driven modeling to analyze and learn characteristics of trend features over time. This is in stark contrast to existing methods configured to detect for an exacerbation of chronic respiratory disease based on traditional threshold-type driven analytics on respiratory measurements or features of the respiratory waveform as discussed in the background. Accordingly, such existing methods are not able to identify or recognize trends. Furthermore, the method generates, and detects for an exacerbation event based on, a test set of trend features from the respiratory waveform with respect to a time window, which specifically comprises at least three different types, or at least three different groups of types, of trend features derived from at least three different types of respiratory variables (or parameters) of the respiratory waveform, respectively, namely, trend feature(s) relating to a respiratory rate (RR) (e.g., in units of 1 / seconds), trend feature(s) relating to a tidal volume (TV) (e.g., in units of arb) and trend feature(s) relating to a modified respiratory duty cycle (MDC) (which is a ratio of a difference between an inhalation time (Tj) (e.g., in units of seconds) and an exhalation time (Te) (e.g., in units of seconds) to a total breathing time (Tt) (e.g., in units of seconds). In this regard, the total breathing time (Tt) is a sum of the inhalation time (Ti) and the exhalation time (Tc) (i.c., Tt=+ Te~). Various example embodiments of the present invention found that utilizing such a combination of at least these three different types of trend features, especially with the inclusion of trend fcaturc(s) relating to the modified respiratory duty cycle (MDC), for training an exacerbation machine learning model for detecting an exacerbation event results in surprisingly superior exacerbation detection performance. In various example embodiments, the test set of trend features may further comprise trend feature(s) relating to the inhalation time (Tj) and one or more trend features relating to the exhalation time (Te). In various example embodiments, the test set of trend features may further comprise trend feature(s) relating to one or more of Inspiratory-to-Expiratory (I:E) ratio, respiratory rate variability, heart rate, duty cycle (ratio of the inhalation time (Ti) to the total breathing time (Tt)), peripheral oxygen saturation (SpO2), SpO2 variability and subject’s movement. Accordingly, various example embodiments of the present invention provide a method of detecting an exacerbation event in a respiratory waveform, as well as a system thereof, with improved exacerbation detection effectiveness or accuracy. For example, the method may be implemented in a respiratory sensor system or device (e.g., wearable respiratory sensor device) for detecting or identifying an exacerbation of chronic respiratory disease in a cardiopulmonary patient by continuous monitoring of a respiratory waveform of the patient’s breathing in real-time or by examining a prior collected sample of a respiratory waveform of the patient’s breathing.

[0040] In various example embodiments, the method of detecting an exacerbation event in a respiratory waveform comprises: generating a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and detecting for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window. In particular, the test set of trend features comprises one or more trend features relating to a respirator}' rate (RR), one or more trend features relating to a tidal volume (TV) (or log(TV)) and one or more trend features relating to a modified respiratory duty cycle (MDC). The modified respiratory duty cycle is determined based on a ratio of a difference between an inhalation time (Ti, e.g., in units of seconds) and an exhalation time (Te, e.g., in units of seconds) to a total breathing time (Tt, e.g., units of seconds).

[0041] In various example embodiments, preferably to further improve exacerbation detection effectiveness or accuracy, the test set of trend features comprises a respiratory rate set (which may also be referred to as a RR group or family) of trend features relating to the respiratory rate (RR), a tidal volume (TV) set (which may also be referred to as a TV group or family) of trend features relating to the tidal volume (TV) and a modified respiratory duty cycle (MDC) set (which may also be referred to as a MDC group or family) of trend features relating to the modified respiratory duty cycle (MDC). In various example embodiments, the test set of trend features further comprises an inhalation time set (which may also be referred to as an inhalation time group or family) of trend features relating to the inhalation time (Ti) and an exhalation time set (which may also be referred to as an exhalation time group or family) of trend features relating to the exhalation time (Tc).

[0042] Accordingly, in various example embodiments, the method utilizes an exacerbation machine learning model (e.g., based on an algorithm such as a logistic regression algorithm or equation) for determining or predicting an exacerbation event or period (e.g., an exacerbation probability score indicating a likelihood or a degree of confidence of the presence of an exacerbation event) in a cardiopulmonary patient by utilizing a combination of different types of trend features derived from different types of respiratory measurements or variables (or parameters) of the respirator}' waveform. In particular, as described above according to various example embodiments, the test set of trend features (and similarly, the training set of trend features) utilized on the exacerbation machine learning model for detecting an exacerbation event comprises a combination of at least three different types of trend features derived from the respiratory rate (RR), the tidal volume (TV) (or depth of breathing) and the modified respiratory duty cycle (MDC), respectively, extracted from the respiratory waveform (with respiratory wave pattern), which may be expressed as follows:(Equation 1)where Tt= 7) + Te(Equation 2)

[0043] Accordingly, MDC may be expressed as a function (e.g., an embedded function) of Ti, Teand Tt, that is, MDC = f(TitTe~).

[0044] The generation or extraction of the test set of trend features (and similarly, the training set of trend features) for utilizing on the exacerbation machine learning model (i.c., for performing inference by the exacerbation machine learning model based on the test set of trend features generated (or similarly, for training the exacerbation machine learning model based on the training set of trend features generated) will now be described in more detail according to various example embodiments of the present invention.

[0045] A trend feature for a type of respiratory measurement or variable (or parameter) with respect to a time window may be generated or determined using any mathematical operator or function on the respiratory variable with respect to the time window (e.g., a portion or segment or an entire of the time window) as desired or as appropriate as long as a trend characteristic or property of the respiratory variable over time is obtained or captured. In various example embodiments, each trend feature relating to or for a respiratory variable may be generated basedon standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the respirator}' variable, or a combination of two or more thereof. In this regard, various example embodiments found that these mathematical operators or functions are able to extract useful trend characteristics or properties of the respiratory variable over time for utilizing on the exacerbation machine learning model for effectively predicting for an exacerbation event in the time window. It will be appreciated by a person skilled in the art that the present invention is not limited to the above-mentioned mathematical operators and / or functions, and additional or other mathematical operator(s) and / or function(s) may be utilized as desired or as appropriate. It will also be appreciated by a person skilled in the art that the present invention is not limited to any specific mathematical operator or function performed on a respiratory variable with respect to a time window as long as a trend characteristic or property of the respiratory variable over time can be obtained or captured, especially if found to be useful for utilizing on the exacerbation machine learning model for effectively predicting for an exacerbation event in the time window.

[0046] In various example embodiments, as described above, a group or family of trend features may be generated for a type of respiratory variable (e.g., for each type of respirator}' variable to be included in the test set or the training set of trend features), each trend feature of the group or family being generated based on a different mathematical operator or function or a different combination thereof. For illustration purpose, by way of examples only and without limitation, Table 1 below presents an example group or family of example trend features which may be generated for each type of respiratory variable, namely, for tidal volume (TV), modified respiratory duty cycle (MDC), respiratory rate (RR), exhalation time (Te) and inhalation time (Ti).Table 1 — Families o f Trend Features for Various Respiratory Variables

[0047] In Table 1, each example trend feature can be understood based on its name and description provided. For illustration purpose, example trend features will now be described with reference to FIG. 3. In particular, FIG. 3 illustrates example trend features generated from an example respiratory waveform with respect to an example time window of 5 days. As illustrative examples relating to a circle labelled 1 in FIG. 3, a trend feature named “RR_tl / 2 / 3 / 4_mean” or “RR_tl / 2 / 3 / 4_std” means that the trend feature is generated based on a mean (i.e., average) or a standard deviation (std) of the respiratory rate (RR) from day 0 (Do) to day 1 (Di) (for tl), from day 0 (Do) to day 2 (D2) (for t2), from day 0 (Do) to day 3 (D3) (for 13) or from day 0 to day 4 (D4) (for 14), respectively, where day 0 (Do) corresponds to day of the exacerbation event and day 1 (Di) to day 4 (D4) respectively correspond to the nthday prior to the exacerbation event. As an illustrative example relating to a circle labelled 2 in FIG. 3, a trend feature named “RR slope t3” means that the trend feature is generated based on a slope(or gradient) of the best fit line of the respiratory rate (RR) over a time period from day 0 (Do) to day 3 (D3) (for t3). As another illustrative example relating to the circle labelled 2 in FIG. 3, a trend feature named “RR_slope_diff_t4_t3” means that the trend feature is generated based on a difference between a slope (or gradient) of the best fit line of the respiratory rate (RR) over a time period from day 0 (Do) to day 4 (D4) (for t4) and a slope (or gradient) of the best fit line of the respiratory rate (RR) over a time period from day 0 (Do) to day 3 (D3) (for t3). As another illustrative example relating to the circle labelled 3 in FIG. 3, a trend feature named “RR baseline mean” means that the trend feature is generated based on a cumulative mean (i.c., average) of the respiratory rate (RR) from a starting / commcnccmcnt day (c.g., from when a user started using the respiratory sensor device for continuous monitoring / detecting for an exacerbation event). As another illustrative example relating to the circle labelled 3 in FIG. 3, a trend feature named “RR baseline std mean” means that the trend feature is generated based on a mean (i.e., average) of the daily standard deviation of the baseline (i.e., the above- mentioned cumulative average) of the respirator}' rate (RR). As another illustrative example relating to the circle labelled 3 in FIG. 3, a trend feature named “RR baseline slope” means that the trend feature is generated based on a slope (i.e., gradient) of the best fit line of the baseline of the respiratory rate (RR). As another illustrative example relating to the circle labelled 3 in FIG. 3, a trend feature named “RR baseline shift” means that the trend feature is generated based on a difference between the baseline of the respiratory rate (RR) of a day and the day before. As another illustrative example relating to the circle labelled 4 in FIG. 3, a trend feature named “RR std w5 norm” means that the trend feature is generated based on a standard deviation of normalized respiratory rate (RR) over a 5-day window (Do to D4). As another illustrative example relating to the circle labelled 4 in FIG. 3, a trend feature named “RR_max_w5_norm” or a trend feature named “RR_min_w5_norm” means that the trend feature is generated based on a maximum or a minimum of normalized respiratory rate (RR) over a 5-day window (Do to D4). As another illustrative example relating to the circle labelled 5 in FIG. 3, a trend feature named “RR_mean_w5_start_end_diff_norm” or “RR_std_w5_start_end_diff_norm” means that the trend feature is generated based on a difference of the mean (i.e., average) or the standard deviation of normalized respiratory rate (RR) between the start (day 0 (Do)) and the end (day 4 (D4)) of a 5-day window (Do to D4). As another illustrative example relating to the circle labelled 5 in FIG. 3, a trend feature named “RR mean_w5_start_end_percentchange_norm” or“RR_std_w5_start_end_percentchange_norm” means that the trend feature is generated basedon a percentage change of the mean (i.e., average) or the standard deviation of normalized respiratory rate (RR) between the start (day 0 (Do)) and the end (day 4 (D4)) of a 5-day window (Do to D4). As another illustrative example relating to the circle labelled 6 in FIG. 3, a trend feature named “RR_mean_tl / 2 / 3_diff_norm” or “RR_std_tl / 2 / 3_diff_norm” mean that the trend feature is generated based on a difference of the mean (i.e., average) or the standard deviation of normalized respiratory rate (RR) between day 1 (Di) and day 0 (Do) (for ti), between day 2 (D2) and day 0 (Do) (for ti) or between day 3 (D3) and day 0 (Do) (for ts), respectively.

[0048] As an example, for each example group or family, all of the example trend features shown in Table 1 above may be generated and included in the test set or the training set of trend features. As another example, for each example group or family, a subset of the example trend features shown in Table 1 may be generated and included in the test set or the training set of trend features.

[0049] It will be appreciated by a person skilled in the art that the exacerbation machine learning model is not limited to any particular or specific network architecture or configuration and may be any type of machine learning model (or artificial intelligence (Al) model) as long as the machine learning model is capable of being trained based on trend features to detect or predict for an exacerbation event in a respiratory waveform according to various example, embodiments of the present invention. For illustration purpose, two example exacerbation machine learning models will now be described in more detail according to various example embodiments of the present invention, namely, a logistic regression model and a boosted tree model.

[0050] In the case of the exacerbation machine learning model being configured as a logistic regression model (which may be referred to as a mixed effects logistic regression model) according to various example embodiments of the present invention, the logistic regression equation of the logistic regression model may be defined as follows:(Equation 3) where: y denotes an exacerbation event, that is, y = 1 when there is an exacerbation and y = 0 otherwise;X denotes a coefficient vector with xx, x2, ...n;[i denotes a feature vector withandZ denotes random effects with a parameterization of u.

[0051] For example, the logistic regression equation may be utilized during training as follows: logit((Equation 4) where, xltx2, ... xndenote the coefficients (or coefficient elements) of the coefficient vector X — Pn denote the corresponding features (or feature elements) of the feature vector . In particular, the feature elements of the feature vector p arc the trend features of a set (c.g., a training set) of trend features generated from a respiratory waveform of a subject’s breathing.

[0052] When training the logistic regression model based on this logistic regression equation, the objective is to train and optimize the xncoefficients respectively associated with the trend features pn. For example, referring to the illustrative example shown in FIG. 3, for each timestep of a series of timesteps, a training set of trend features (e.g., the illustrative set of example trend features shown in Table 1) may be generated or extracted with respect to a time window (e.g., a predefined or fixed ‘X’ number of days, such as a 5-day window as illustrated in FIG. 3) and is utilized to train the logistic regression model to optimize the coefficients x1, x2, ... xn. Accordingly, an X-day window period may be adopted for scanning across a respiratory waveform to generate a training set of trend features for each timestep (c.g., every 1 second) of a series of timesteps for training the logistic regression model.

[0053] Similarly, during interference when utilizing the trained logistic regression model to predict whether an exacerbation event exist, for each timestep of a series of timesteps, a test set of trend features (e.g., the illustrative set of example trend features shown in Table 1) may be generated or extracted with respect to a time window (e.g., a predefined or fixed ‘X’ number of days, such as a 5-day window as illustrated in FIG. 3) and is utilized for input to the trained logistic regression model to predict whether an exacerbation event exist for the timestep. Accordingly, an X-day window period may be adopted for scanning across a respiratory waveform (e.g., a continuous respiratory waveform generated from a subject’s breathing in realtime) to generate a test set of trend features for each timestep (e.g., every 1 second) of a series of timesteps for continuously detecting or monitoring for an exacerbation event of the subject in real-time.

[0054] In the case of the exacerbation machine learning model being configured as a boosted tree model (which may be referred to as a mixed effects boosted tree model) according to various example embodiments of the present invention, the boosted tree method works by partitioning all groups of trend features into two subsets when finding splits in the tree-buildingalgorithm. FIG. 4 depicts a schematic flow diagram showing how the boosted tree method uses trend features to eventually move from the top node to the final bottom node - eventually deciding if the binary' value of y = 0 or 1 (predicting whether no exacerbation or exacerbation exists).

[0055] Accordingly, in various example embodiments, the method of detecting an exacerbation event in a respiratory waveform is configured to generate at least a respective family of trend features for each of RR, TV and MDC, which collectively may be referred to as a test set of trend features. The test set of trend features may then be input to the trained exacerbation machine learning model to produce an exacerbation prediction output (c.g., an exacerbation probability score for indicating a probability of an exacerbation event or a binary value (c.g., “1” or “0”) for indicating whether an exacerbation event is detected).

[0056] FIG. 5 depicts a Table presenting a univariate analysis performed on the trained exacerbation machine learning model configured as a mixed effects logistic regression model, according to various example embodiments of the present invention. It can be observed that during exacerbations, the respiratory rate (RR) increases, the log tidal volume (log(Vt)) reduces and the modified respiratory duty cycle (MDC) reduces.

[0057] FIG. 6 depicts a Table presenting the performance statistics of both the mixed effect logistic regression method and the boosted tree (or tree boosting) method with mixed effects, according to various example embodiments of the present invention. The average performance or the ROC (receiver-operating characteristic)-AUC (area under the ROC curve) curve (compared with ground truth) can be seen at about 80% for the logistic regression approach and about 95% for the boosted tree method. In particular, from the Table shown in FIG. 6, it can be seen that utilizing the combination of the three different types of trend features, namely, the combination of the respiratory rate (RR), the tidal volume (Vt) and the modified respiratory duty cycle (MDC) for both of the logistic regression method and the boosted tree method for detecting an exacerbation event result in surprisingly superior (significantly improved) exacerbation detection performance, especially with the inclusion of trend features relating to the MDC.

[0058] Accordingly, in various example embodiments, at least trend features relating to the respiratory rate (RR) (or a RR set or family), trend features relating to the tidal volume (TV) (or a TV set or family) and trend features relating to the modified respiratory duty cycle (MDC) (or a MDC set or family) are extracted from a respiratory waveform (with respiratory wave pattern) of a subject with respect to a time window (e.g., a predefined number of days).Advantageously, such trend features can be extracted from “passive” respiratory sensors according to various example embodiments of the present invention such as installed or implemented in a wearable device and thus does not require active usage by users (e.g., patients). For example, the passive respiratory sensors can operate in a continuous mode without requiring the users to specifically engage actively / significantly with the passive respiratory sensors, thus enabling the collection of data points continuously, resulting in the continuous extraction of trend features / signals. In various example embodiments, multiparametric respiratory risk index scores (or exacerbation probability scores) or binary values (e.g., “1” or “0”) for predicting an exacerbation event arc determined (e.g., continuously predicted every 1 second to provide better resolution) based on a combination of trend features relating to at least the respiratory rate, the tidal volume and the modified respiratory duty cycle, which was found to produce significantly enhanced results (accuracy) in detecting for exacerbation events, and is thus also more sensitive with improved exacerbation predictive capability.

[0059] While embodiments of the invention have been particularly shown and described with reference to specific embodiments, it should be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the invention as defined by the appended claims. The scope of the invention is thus indicated by the appended claims and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced.

Claims

CLAIMS1. A method of detecting an exacerbation event in a respiratory waveform using at least one processor, the method comprising: generating a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and detecting for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window, wherein the test set of trend features comprises one or more trend features relating to a respiratory rate, one or more trend features relating to a tidal volume and one or more trend features relating to a modified respiratory duty cycle, the modified respiratory duty cycle being determined based on a ratio of a difference between an inhalation time and an exhalation time to a total breathing time.

2. The method according to claim 1, wherein the exacerbation prediction output of the exacerbation machine learning model is an exacerbation probability score for the time window.

3. The method according to claim 1 or 2, wherein each trend feature of the one or more trend features relating to the respiratory rate is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the respiratory rate, or a combination of two or more thereof, each trend feature of the one or more trend features relating to the tidal volume is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the tidal volume, or a combination of two or more thereof, and each trend feature of the one or more trend features relating to the modified respiratory duty cycle is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the modified respiratory duty cycle, or a combination of two or more thereof.

4. The method according to any one of claims 1 to 3, whereinthe test set of trend features further comprises one or more trend features relating to the inhalation time and one or more trend features relating to the exhalation time, and each trend feature of the one or more trend features relating to the inhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the inhalation time, or a combination of two or more thereof, and each trend feature of the one or more trend features relating to the exhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the exhalation time, or a combination of two or more thereof.

5. The method according to any one of claims 1 to 4, wherein the respiratory waveform is a continuous respiratory waveform, and the method is performed for each timestep of a series of timesteps to detect for an exacerbation event in the respirator}' waveform in a time window associated with the timestep.

6. The method according to any one of claims 1 to 5, wherein the exacerbation machine learning model is trained based on a training set of trend features extracted from a respiratory waveform with respect to a time window, and the training set of trend features comprises one or more trend features relating to the respiratory rate, one or more trend features relating to the tidal volume and one or more trend features relating to the modified respiratory duty cycle.

7. The method according to claim 6, wherein each trend feature of the one or more trend features of the training set relating to the respiratory rate is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the respiratory rate, or a combination of two or more thereof, each trend feature of the one or more trend features of the training set relating to the tidal volume is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the tidal volume, or a combination of two or more thereof, andeach trend feature of the one or more trend features relating to the modified respiratory duty cycle is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the modified respiratory duty cycle, or a combination of two or more thereof.

8. The method according to claim 6 or 7, wherein the training set of trend features further comprises one or more trend features relating to the inhalation time and one or more trend features relating to the exhalation time, each trend feature of the one or more trend features relating to the inhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the inhalation time, or a combination of two or more thereof, and each trend feature of the one or more trend features relating to the exhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the exhalation time, or a combination of two or more thereof.

9. The method according to any one of claims 1 to 8, wherein the exacerbation machine learning model is a logistic regression model.

10. The method according to any one of claims 1 to 8, wherein the exacerbation machine learning model is a boosted tree model.

11. A system for detecting an exacerbation event in a respiratory waveform, the system comprising: at least one memory; and at least one processor communicatively coupled to the at least one memory and configured to: generate a test set of trend features from a respiratory waveform of a subject’s breathing with respect to a time window; and detect for an exacerbation event in the time window based on an exacerbation prediction output of an exacerbation machine learning model based on the test set of tread features generated with respect to the time window,wherein the test set of trend features comprises one or more trend features relating to a respiratory rate, one or more trend features relating to a tidal volume and one or more trend features relating to a modified respiratory duty cycle, the modified respiratory duty cycle being determined based on a ratio of a difference between an inhalation time and an exhalation time to a total breathing time.

12. The system according to claim 11, wherein the exacerbation prediction output of the exacerbation machine learning model is an exacerbation probability score for the time window.

13. The system according to claim 11 or 12, wherein each trend feature of the one or more trend features relating to the respiratory rate is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the respiratory rate, or a combination of two or more thereof, each trend feature of the one or more trend features relating to the tidal volume is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the tidal volume, or a combination of two or more thereof, and each trend feature of the one or more trend features relating to the modified respiratory duty cycle is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the modified respiratory duty cycle, or a combination of two or more thereof.

14. The system according to any one of claims 11 to 13, wherein the test set of trend features further comprises one or more trend features relating to the inhalation time and one or more trend features relating to the exhalation time, and each trend feature of the one or more trend features relating to the inhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the inhalation time, or a combination of two or more thereof, and each trend feature of the one or more trend features relating to the exhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum,cumulative mean or cumulative mean shift with respect to the exhalation time, or a combination of two or more thereof.

15. The system according to any one of claims 11 to 14, wherein the respiratory waveform is a continuous respiratory waveform, and the at least one processor is configured to perform the detection of an exacerbation event in the respiratory waveform for each timestep of a series of timesteps to detect for an exacerbation event in the respirator}' waveform in a time window associated with the timestep.

16. The system according to any one of claims 11 to 15, wherein the exacerbation machine learning model is trained based on a training set of trend features extracted from a respiratory waveform with respect to a time window, and the training set of trend features comprises one or more trend features relating to the respiratory rate, one or more trend features relating to the tidal volume and one or more trend features relating to the modified respiratory duty cycle.

17. The system according to claim 16, wherein each trend feature of the one or more trend features of the training set relating to the respiratory rate is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the respiratory rate, or a combination of two or more thereof, each trend feature of the one or more trend features of the training set relating to the tidal volume is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the tidal volume, or a combination of two or more thereof, and each trend feature of the one or more trend features relating to the modified respiratory duty cycle is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the modified respiratory duty cycle, or a combination of two or more thereof.

18. The system according to claim 16 or 17, wherein the training set of trend features further comprises one or more trend features relating to the inhalation time and one or more trend features relating to the exhalation time,each trend feature of the one or more trend features relating to the inhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the inhalation time, or a combination of two or more thereof, and each trend feature of the one or more trend features relating to the exhalation time is generated based on standard deviation, mean, median, slope of best fit, maximum, minimum, cumulative mean or cumulative mean shift with respect to the exhalation time, or a combination of two or more thereof.

19. The system according to any one of claims 11 to 18, wherein the exacerbation machine learning model is a logistic regression model.

20. The system according to any one of claims 11 to 18, wherein the exacerbation machine learning model is a boosted tree model.

21. A computer program product, embodied in one or more non-transitory computer- readable storage mediums, comprising instructions executable by at least one processor to perform the method of detecting an exacerbation event in a respiratory waveform according to any one of claims 1 to 10.

Images