Asthma severity measurement machine operation method and asthma severity measurement machine

Abstract

A method for stratifying the severity of a patient's asthma includes initially receiving acoustic data corresponding to the patient's sounds from an acoustic sensor and identifying at least one cough sound within the acoustic data by a processor. Whether the patient is present or not, the method further involves determining, by operation of the processor, one or more overall cough sound features of the at least one cough sound for each of one or more distinctive features. The overall cough sound features are then applied to a classifier implemented by the processor and pre-trained with a training set of distinctive features from a population of asthmatic and non-asthmatic subjects. The method then involves monitoring output from the pre-trained classifier to consider the patient's cough sound as indicative of one of several degrees of asthma severity.

Description

[Technical Field]

[0001] Related Applications This application claims priority to Australian Provisional Patent Application No. 2019903000, filed August 19, 2019, the disclosure of which is incorporated herein by reference for all purposes.

[0002] The present invention relates to automatically determining the presence and severity of asthma in a subject. [Background technology]

[0003] Any reference to prior art methods, devices, or documents shall not be deemed to constitute any evidence or admission that they formed or form part of the public general knowledge.

[0004] Asthma is a common pediatric respiratory disease characterized by wheezing, coughing, and difficulty breathing. Asthma severity can be determined by subjective manual scoring systems, such as pulmonary scoring. These systems require significant medical training and expertise to evaluate clinical findings, such as wheezing characteristics and work of breathing. The ability to accurately characterize the severity of asthma in patients is beneficial because the correct treatment and management therapies can be implemented depending on the severity of asthma, for example, in response to bronchodilators.

[0005] Acute asthma is characterized by transient, reversible airflow obstruction and presents with wheezing, dyspnea, and cough that responds to bronchodilators. [3] The underlying pathophysiology is restricted airflow through partially narrowed peripheral and large airways, causing wheezing and increased work of breathing, manifested by increased respiratory rate and signs of accessory muscle use. Wheezing characteristics may change with progressive obstruction; moreover, in severe disease, there may be insufficient airflow to generate any wheezing, resulting in a "silent chest."

[0006] The ability to detect wheezing by auscultation depends on (i) the sound being produced in the lungs with sufficient intensity to reach the chest wall and be detected by the stethoscope, (ii) the clinician having adequate hearing, and (iii) the clinical skill and experience to recognize its defining characteristics.

[0007] Various measures of airway obstruction have been developed. Objective tests include spirometry and Peak Expiratory Flow Rate (PEFR) measurements. [6] In emergency situations, formal spirometry may not be available or practical. PEFR meters have the advantage of being inexpensive and portable. However, young children (<6 years old) and those in severe respiratory distress often cannot adapt to PEFR testing, which is technique- and effort-dependent and requires cooperation. PEFR measurements are not recommended for routine use in children under 12 years of age. [3]

[0008] Clinical rating scales use a combination of clinical features to guide asthma management, measure response to therapy, and determine the urgency of care. The Pulmonary Score (PS) is a commonly used asthma severity scale in clinical practice and is compatible with PEFR measurements [9]. The PS is commonly used in Western Australian hospitals and is widely trusted in routine clinical practice

[10] . The PS utilizes respiratory rate (RR), wheezing characteristics, and accessory (sternocleidomastoid) muscle (AM) use to generate a score out of nine (Table 1). Breathing rate can be easily determined even by lay users with little or no training. However, the assessment and grading of wheezing characteristics and AM use requires significant medical training and expertise. Wheezing assessment requires auscultation with a stethoscope. While the PS can provide essential information in the management of pediatric asthma in all settings, it is not a practical assessment tool in clinical practice or outside of the hospital. The primary objective of this study was to develop a community-deployable technique for assessing asthma severity.

[0009] [Table 1]

[0010] Table 1 (above): Pulmonary score. Clinicians estimate subscores (0-3) based on respiratory rate, wheezing, and accessory muscle use. The overall PS, which indicates the degree of asthma severity in a subject, is the sum of all three subscores. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] U.S. Patent No. 10,098,569 Summary of the Invention [Problem to be solved by the invention]

[0012] There is a need for an automated method for stratifying asthma severity that does not rely on clinical assessment of wheezing severity (auscultation) or accessory muscle use. [Means for solving the problem]

[0013] According to a first aspect of the present invention there is provided a method for stratifying asthma severity in a patient, the method comprising: receiving acoustic data corresponding to sounds from the patient from an acoustic sensor; identifying, by a processor, at least one cough sound in the acoustic data; determining, by a processor, one or more overall cough sound features of at least one cough sound for each of the one or more distinctive features; applying, by a processor, the comprehensive cough sound features to a processor-implemented classifier, the classifier having been pre-trained using a training set of distinctive features from a population of asthmatic and non-asthmatic subjects; monitoring the output from the pre-trained classifier to consider the patient's cough sounds as indicative of one of several degrees of asthma severity; Includes:

[0014] In one embodiment, the one or more comprehensive cough sound features for the one or more characteristic features include values ​​of wavelet features of the cough sound.

[0015] In one embodiment, the degree of severity includes "mild" or "moderate to severe."

[0016] In one embodiment, the method includes applying, by a processor, the patient's respiratory rate, or a value derived therefrom, a respiratory rate value, in addition to the comprehensive cough sound features to a pre-trained classifier to consider the patient's cough sound as indicative of one of several degrees of asthma severity, including "mild," "moderate," and "severe."

[0017] In one embodiment, the respiration rate value comprises a respiration rate-based respiration index that takes into account the age of the patient.

[0018] In one embodiment, the method includes dividing each of the at least one cough sound by a processor into a plurality of segments.

[0019] In one embodiment, the method includes determining, by a processor, a segment feature value for each of a plurality of segments for each of a number of distinctive features.

[0020] In one embodiment, the plurality of segments includes three segments.

[0021] In one embodiment, the method includes applying the segment features to a pre-trained classifier in addition to the overall cough features by a processor.

[0022] In one embodiment, the method includes determining, by the processor, segment feature values ​​for each of a number of distinctive features includes determining values ​​for one or more of the segments for one or more of MFCC1, MFCC2, MFCC3, MFCC4, MFCC6, MFCC9, and MFCC12.

[0023] In one embodiment, the method includes determining, by a processor, a kurtosis value for the first segment.

[0024] In one embodiment, the method includes determining, by a processor, values ​​for each segment for one or more of the following characteristic features: Bispectral score (BGS), Non-Gaussian Score (NGS), the first n formant frequencies (FF), Logarithmic Energy (LogE), Zero crossing (ZCR), Kurtosis, The first n Mel-Frequency Cepstral Coefficients (MFCCs).

[0025] In one embodiment, determining by the processor segment features for each of a number of distinctive features includes determining 21 features for each segment as follows:

[0026] [Table 2]

[0027] In one embodiment, determining by the processor a segment feature value for each of a number of distinctive features includes determining a feature value for each of three segments as follows:

[0028] [Table 3]

[0029] According to another aspect of the invention, there is provided a system including one or more processors arranged to process sounds of a human subject in accordance with the method previously described.

[0030] According to a further aspect of the present invention there is provided an asthma severity machine for determining and indicating asthma severity for a patient, the machine comprising: Electronic memory; at least one processor in communication with the electronic memory configured with instructions stored in the electronic memory; an audio recording assembly in communication with the at least one processor; a human-machine interface in communication with the at least one processor; This electronic memory is processing the patient's digital record to identify at least one cough sound; extracting one or more comprehensive cough sound features for the cough sound for each of the one or more characteristic features; Implementing a pre-trained pattern classifier for asthma severity, the classifier being pre-trained using a training set of distinctive features from a population of asthma and non-asthma subjects; applying the synthetic cough sound features to a pre-trained pattern classifier; operating the human-machine interface to present an asthma severity classification based on the output from the pre-trained pattern classifier; and storing instructions for configuring at least one processor to:

[0031] In one embodiment, the electronic memory stores instructions that configure at least one processor to extract one or more comprehensive cough sound features comprising values ​​of wavelet features of the cough sound.

[0032] In one embodiment, the electronic memory stores instructions that configure at least one processor to operate the human-machine interface to present an asthma severity classification based on output from a pre-trained pattern classifier that includes a classification of "mild" or "moderate to severe."

[0033] In one embodiment, the electronic memory stores instructions that configure at least one processor to apply a respiratory rate value, the patient's respiratory rate or a value derived therefrom, to the pre-trained classifier in addition to the comprehensive cough sound features to operate the human-machine interface to present degrees of asthma severity, including "mild," "moderate," and "severe."

[0034] In one embodiment, the electronic memory stores instructions that configure the at least one processor to divide each at least one cough sound into a plurality of segments.

[0035] In one embodiment, the electronic memory stores instructions that configure at least one processor to, for each of a number of distinctive features, partition the feature quantity into each of a plurality of segments.

[0036] In one embodiment, the electronic memory stores instructions that configure at least one processor to apply the segment features to a pre-trained classifier in addition to the overall cough features.

[0037] In one embodiment, the electronic memory comprises: MFCC1, MFCC2, MFCC3, MFCC4, MFCC6, MFCC9, and MFCC12 instructions configuring the at least one processor to determine characteristic features for one or more of the segments for one or more of the segments.

[0038] In one embodiment, the electronic memory stores instructions that configure at least one processor to determine a kurtosis value for the first segment.

[0039] In one embodiment, the electronic memory stores instructions that configure at least one processor to obtain an estimate of the patient's respiration rate.

[0040] In one embodiment, the electronic memory stores instructions that configure at least one processor to analyze signals from an accelerometer placed on or near the patient to determine the patient's respiratory rate.

[0041] In one embodiment, the electronic memory stores instructions configuring at least one processor to process signals from an acoustic sensor to detect respiratory sounds from the patient to thereby calculate a respiratory rate estimate.

[0042] In one embodiment, the electronic memory stores instructions that configure at least one processor to monitor signals from a lens and CCD assembly positioned to capture images of the rise and fall of the patient's chest to thereby generate a respiratory rate from the images.

[0043] According to another aspect of the present invention, there is provided a method for stratifying asthma severity in a patient, the method comprising: receiving acoustic data from an acoustic sensor placed near the patient; identifying, by a processor, at least one cough sound in the acoustic data; segmenting the cough sound into a plurality of segments by a processor; determining, by a processor, a segment feature value for each of a number of distinctive features; applying, by the processor, the segment features to a pre-trained classifier implemented by the processor; monitoring the output from the pre-trained classifier to consider the patient's cough sounds as indicative of one of several degrees of severity; Includes:

[0044] In another aspect, there is provided an asthma severity indicator assembly for indicating a severity of asthma in a patient, the asthma severity indicator assembly comprising: an audio capture device including a microphone and analog-to-digital conversion circuitry configured to store a digital audio recording of the patient in electronic memory; a cough identification assembly in communication with the memory and arranged to process the digital audio recording to thereby identify portions of the digital audio recording that include sounds of the patient coughing; a cough segmentation assembly responsive to the cough identification assembly and arranged to segment each identified cough into a number of segments; a feature extraction processor in communication with the cough segmentation assembly and arranged to process each of the segments to generate segment features for each of a number of characteristic features; a pattern classifier in communication with the feature extraction processor configured to generate a signal indicative of the patient's cough sound as being one of several degrees of severity; Includes:

[0045] In a further aspect, there is provided an asthma severity machine for determining and indicating the severity of asthma in a patient, the machine comprising: Electronic memory; at least one processor in communication with the electronic memory, the processor being configurable by instructions stored in the electronic memory; an audio recording assembly in communication with the at least one processor; a human-machine interface in communication with the at least one processor; This electronic memory allows the processor to: processing the patient's digital record to identify at least one cough sound; extracting features from the cough sound and from segments thereof; Implementing a pre-trained classifier for asthma severity; applying the features to a pre-trained pattern classifier; operating the human-machine interface to present an asthma severity classification based on the output from the pre-trained pattern classifier; The instruction for performing the above is stored.

[0046] In another aspect, a method is provided for stratifying the severity of a patient's asthma, comprising applying the values ​​of features of the patient's cough sound to a pattern classifier pre-trained using a population of asthma patients with varying degrees of severity.

[0047] In a further aspect, there is provided an asthma severity machine arranged to capture a cough sound from a patient, apply feature values ​​from the cough sound to a pattern classifier that has been pre-trained with a population of asthma patients with varying degrees of asthma severity, and provide a classification based on the output of the pattern classifier.

[0048] Preferred features, embodiments, and variations of the present invention can be understood from the following detailed description, which provides sufficient information for those skilled in the art to practice the invention. The detailed description should not be construed as limiting the scope of the foregoing subject matter of the present invention in any manner. The detailed description makes reference to several drawings, including: [Brief explanation of the drawings]

[0049] [Figure 1] 1 is a flowchart of a method for indicating asthma severity of a patient. [Figure 2] FIG. 1 is a block diagram of an asthma severity machine for presenting a patient's asthma severity. [Figure 3] 3A-3C show exemplary digital waveforms corresponding to patient sounds captured by the machine of FIG. 2. [Figure 4A] FIG. 4 is an external view of the machine of FIG. 2 during recording of the waveform of FIG. 3. [Figure 4B] FIG. 3 is an external view of the machine of FIG. 2 showing the screen for inputting the patient's respiratory rate and age. [Figure 5]3 is a graph showing portions of a recording of a patient's sounds identified as cough sounds by the machine of FIG. 2; [Figure 6] 6 is a graph divided into three segments detailing the cough sound from FIG. 5. [Figure 7] FIG. 3 is an external view of the machine of FIG. 2 presenting a screen showing an overall cough severity rating for each cough and an average severity rating. [Figure 8] 1 is four graphs showing confidence intervals for each of four differently trained logistic regression models for diagnosing asthma severity. DETAILED DESCRIPTION OF THE INVENTION

[0050] FIG. 1 is a flow chart of a method according to a preferred embodiment of the present invention for classifying a patient's asthma as one of "mild," "moderate," or "severe."

[0051] The apparatus used to implement the method includes an asthma stratification machine, which may be referred to as an asthma severity indicator assembly. The asthma stratification machine may be a dedicated assembly including specific circuitry for performing the various operations discussed for classifying asthma. Alternatively, the asthma stratification machine may be a portable computing device, such as a desktop computer or smartphone, containing at least one processor in communication with electronic memory storing instructions that specifically configure the processor to operate to perform the method steps described below. It will be appreciated that the method cannot be performed without either a dedicated machine or an assembly or system consisting of one or more processors in communication with one or more electronic memories storing instructions for specifically configuring the processor to implement the method.

[0052] 2 is a block diagram of an asthma stratification machine 51 implemented using one or more processors and memory of a smartphone. The asthma stratification machine 51 includes at least one processor (or it may simply be referred to as a “processor”) 53 that accesses electronic memory 55. The electronic memory 55 includes an operating system 58, such as the Android® operating system or the Apple iOS operating system, for execution by the processor 53. The electronic memory 55 also includes an asthma stratification software product or “App” 56 according to a preferred embodiment of the present invention. The asthma stratification App 56 includes instructions executable by the processor 53 to thereby specifically configure the processor 53 to process sounds from a patient 52 and present a classification of the patient's 52's asthma severity to a clinician 54 via an LCD touchscreen interface 61. App 56 includes instructions for processor 53 to implement a pattern classifier, such as a trained predictor or decision machine, which in the presently described preferred embodiment of the invention includes a specially trained logistic regression model (LRM) 60. It will be appreciated that in other embodiments, other suitable decision machines may be implemented by processor 53 under control of App 56, such as an artificial neural network or a Bayesian decision machine.

[0053] The processor 53 is in data communication with multiple peripheral assemblies 59-73 via a data bus 57, comprised of metallic conductors along which digital signals 200 are transmitted between the processor 53 and the various peripheral assemblies 59-73, as shown in FIG. 2. Consequently, if necessary, the asthma severity machine 51 can establish voice and data communications with a voice and / or data communications network 81 via a WAN / WLAN assembly 73 and a radio frequency antenna 79. The machine also includes other peripheral assemblies, such as a lens and CCD assembly 59, which provides a digital camera so that images of the patient 52 can be captured as needed. An LCD touchscreen interface 61 comprises a human-machine interface, allowing the clinician 54 to read results and input commands and data into the machine 51. An accelerometer 62 detects accelerations associated with changes in position and orientation and provides corresponding data for processing by the processor 53. A USB port 65 is provided for providing a serial data connection to an external storage device, such as a USB stick, or for creating a cable connection to a data network or an external screen and keyboard. Secondary storage card 64 is also provided with additional secondary storage, if needed, in addition to the internal data storage space facilitated by memory 55. Audio interface 71 couples microphone 75 to data bus 57 and includes an anti-aliasing filter circuit and an analog-to-digital sampler for converting analog electrical waveforms from microphone 75 (which correspond to patient sound waves 39) into digital audio signals 300 (shown in FIG. 3 ), which contain acoustic data 66 that can be stored in memory 55 and processed by processor 53. Audio interface 71 is also coupled to speaker 77. Audio interface 71 includes a digital-to-analog converter for converting the digital audio into an analog signal and an audio amplifier connected to speaker 71 so that audio recorded in memory 55 or secondary storage 64 can be played back for listening by clinician 54.

[0054] In a preferred embodiment, asthma stratification machine 51 is configured by App 56 as a machine for classifying asthma severity in a patient (or it may be referred to herein as "stratifying asthma severity") without the need for external sensors in physical contact with patient 52 or in contact with communications network 81. Furthermore, when acoustic data 66 is stored by machine 51, machine 51 can thereby process the acoustic data to stratify asthma severity without any need for the patient to be present.

[0055] As previously described, the asthma stratification machine 51 shown in FIG. 2 is provided in one embodiment in the form of smartphone hardware custom-configured by the App 56; however, the asthma stratification machine 51 may similarly use several other types of processing assemblies, such as a desktop computer, laptop, or tablet computing device specially programmed with the App 56, or even in a cloud computing environment where the hardware includes a virtual machine. Furthermore, a dedicated circuit asthma stratification machine may also be constructed without a general-purpose processor. For example, such a dedicated machine may have an audio capture device including a microphone and analog-to-digital conversion circuitry configured to store a patient's digital audio recording in electronic memory. The machine may further include a cough identification assembly in communication with the memory and arranged to process the digital audio recording to thereby identify portions of the digital audio recording that include the patient's cough sound. A cough segmentation assembly is provided that is arranged to segment each identified cough into several segments in response to the cough identification assembly. A feature extraction processor is provided that is in communication with the cough segmentation assembly and arranged to process each of the segments to generate segment features for each segment for each of several characteristic features. The dedicated machine further includes hardware implemented with a pattern classifier in communication with the feature extraction processor configured to generate a signal indicative of the patient's cough sounds as being indicative of one of several degrees of asthma severity.

[0056] The steps used by the asthma severity machine 51 to stratify asthma severity in a patient 52, including the instructions that make up the App 56, are shown in the flow chart of FIG. 1 and are described in detail below.

[0057] In Box 3, sounds 39 from the patient's 52's mouth are recorded by microphone 75 and digitized by audio interface 71 to generate digitized signal 300 (shown in FIG. 3 ) that is transmitted along bus 57 to processor 53. Typically, microphone 75 is placed near the patient's mouth, e.g., within one meter, although depending on the microphone's directionality and background noise level, the distance may vary while still ensuring capture of the patient's sounds. Signal 300 includes digitized patient sound data recorded from patient 52 at a sampling frequency Fs, typically 44.1 kHz at 16-bit resolution, and stored as acoustic data 66, e.g., as a file, in memory 55 of asthma severity machine 51. While recording the procedure in Box 3, processor 53 operates LCD touchscreen interface 61 to display screen 10 of messages being recorded, as shown in FIG. 4A . Clinician 54 continues the recording procedure until at least one cough, and preferably two or more, has been captured from patient 52. Once the clinician 54 is satisfied with the number of coughs captured, the clinician then presses the "OK" button on screen 10, which stops the recording operation and causes processor 53 to proceed to box 4 of the flowchart of Figure 1. The patient does not need to be present for any of the operations described in boxes 4-27.

[0058] In box 4, processor 53 operates LCD touchscreen interface 61 to display screen 12, as shown in FIG. 4B, for clinician 54 to enter the patient's age in months along with the patient's 52 respiratory rate. The patient's age and respiratory rate values ​​are later used by processor 53 to calculate a respiratory index value in box 21. It is possible to omit the steps of capturing the respiratory rate and patient's age data in box 4 and calculating the respiratory index in box 21. However, such omission would result in the method being unable to reliably classify a patient as having one of the three categories of "mild," "moderate," or "severe" asthma, but only being able to discriminate between two categories, i.e., "mild" or "moderate to severe."

[0059] In box 5, processor 53 processes the digitized sound recording 300 of patient 52 captured in box 3 to identify coughs CG1,...,CG2 in the sound recording as shown in FIG. n Procedures for identifying cough sounds are known in the prior art, for example in US Pat. No. 10,098,569 to the applicant, the disclosure of which is incorporated herein by reference.

[0060] In box 7, processor 53 generates a signal from the identified cough sounds CG1,...,CG nhave been processed by subsequent processing boxes in a manner that will be described shortly. Because the initial cough sound was not processed, control is transferred from decision box 7 to box 9. In box 9, processor 53 calculates one or more comprehensive cough sound features of the current cough sound for each of the one or more distinctive features. In the presently described embodiment, the comprehensive cough sound features are the values ​​of 13 wavelet features W1,...,W13 based on the Morlet wavelet as detailed in Kosasih, K., U.R. Abeyratne, and V. Swarnkar, Wavelet-Enhanced Cough Analysis for Rapid Pediatric Pneumonia Diagnosis, IEEE Trans on Biomed Eng., 2015, 62(4), pp. 1185-1194, the contents of which are incorporated herein by reference in their entirety.

[0061] In box 11, processor 53 generates the current cough CG i into three segments SS(1), SS(2), and SS(3). In the presently described embodiment, there are three segments, which do not overlap with each other. While this is believed to be the best division scheme, the following procedure may be used with a different number of potentially overlapping segments in other embodiments.

[0062] In box 13, the processor checks whether all three segments SS(i) have been processed in the following boxes 15 and 17. Since the cough sound was not processed earlier, control transfers to box 15.

[0063] In box 15, processor 53 applies a known algorithm to determine, for the current segment SS(i), a feature value for at least one of several features. In the presently described embodiment, the calculated values ​​are: bispectral score (BGS), non-Gaussianity score (NGS), first four formant frequencies (FF), logarithmic energy (LogE), zero crossing (ZCR), kurtosis (Kurt), and 12 Mel-frequency cepstral coefficients (MFCC). The algorithm used by processor 53 is described in Abeyratne, UR et al., "Cough Sound Analysis Can Rapidly Diagnose Pediatric Pneumonia," Annals of Biomedical Engineering, 2013, 41(11), pp. 2448-2462, the disclosure of which, including portions of the instructions constituting App. 56, is incorporated herein by reference in its entirety.

[0064] For each cough segment SS(i), the processor generates 21 unique features in box 15 as set out in Table 2 below.

[0065] [Table 4]

[0066] It will then be appreciated that in the presently described embodiment, the total number of features computed for each cough is 3 (segments) x 21 (values ​​per segment) + 13 (wavelet features for each overall cough) = 76 features in total.

[0067] In box 17, processor 53 stores the values ​​for the current segment in a data structure, e.g., a vector array residing in an allocated portion of memory 55 and including a 1x73 array of real values ​​representing the components of the cough feature vector Fc.

[0068] Once all of the current cough segments have been processed, thereby populating the components comprising the cough feature vector Fc, control transfers from decision box 13 to box 21. As previously alluded to, in box 21 processor 53 calculates the respiratory index BI from the respiratory rate (RR) and patient age data previously captured in box 4.

[0069] The BI is defined as follows: BI=RR-20 (age ≥ 60 months) BI=RR-40 (if age <60 months).

[0070] Therefore, for example, if the age is 45 months and RR=60, BI=60-40=20

[0071] If age = 45 months and RR = 38, BI=38-40=-2

[0072] Preferably, all of the captured values ​​for all features are normalized across the population, i.e., a feature, say BI, is made zero mean and homogeneous variance by subjecting BI to a transformation of the following type: Normalized feature = (feature - mean(feature)) / standard deviation(feature)

[0073] Before applying the features to the classifier, any new subject to be classified is preferably run through these normalization equations.

[0074] The aforementioned paper, Abeyratne, UR et al., Cough sound analysis can rapidly diagnose pediatric pneumonia, Annals of Biomedical Engineering, 2013, 41(11), pp. 2448-2462, discusses the development of respiratory indices.

[0075] Instead of clinician 54 directly inputting the respiratory rate, in other embodiments, machine 51 may be programmed to obtain an estimate of the respiratory rate. For example, machine 51 may be programmed so that when the machine is placed on the patient's chest, the machine analyzes signals from accelerometer 62 to determine the respiratory rate. In another embodiment, machine 51 may be programmed so that when microphone 75 is placed under the nose / mouth of patient 52, processor 53 detects breathing sounds from the patient to generate an estimate of the respiratory rate. In another embodiment, lens and CCD assembly 59 is positioned to capture an image of the rise and fall of patient 39's chest, and microprocessor 53 is programmed to generate the respiratory rate from the image.

[0076] In box 23, processor 53 calculates the total memory size (1×(76+1(BI)+13(W))) inherent in the allocated portion of memory 55. i )) = 1 × 90 array data structure. T contains all 76 components of the Fc-cough vector, as well as components corresponding to the respiratory index value BI and the 13 wavelet values ​​calculated in Box 9.

[0077] In box 25, processor 53 applies test vector T to a pre-trained classifier in the form of a logistic regression model (LRM) 60, which the classifier implements in accordance with instructions including App 56. Logistic regression model 60 has been trained to classify test vector T as indicating one of several degrees of asthma severity. In the presently described embodiment, asthma severity machine 51 classifies each cough as one of three degrees of severity: "mild," "moderate," and "severe." Methods for training LRM 60 are described in subsequent sections of this specification.

[0078] As briefly described above, in other embodiments, processor 53 does not perform boxes 4 and 21, such that the respiratory rate and patient age are not recorded and the respiratory index is not calculated. As a result, test vector T includes only features obtained by processing cough sounds and cough sound segments. It has been found that one such embodiment performs well in interpreting a patient's cough sounds to indicate one of "mild" or "moderate-to-severe" degrees of asthma severity in the patient, but not to indicate one of "mild," "moderate," and "severe" degrees of asthma severity.

[0079] Furthermore, in another embodiment, boxes 13-17 are not implemented by the processor. Instead, the test vector T is calculated using W to distinguish between "mild" and "moderate to severe." i W alone or to distinguish between "mild," "moderate," and "severe" i and the value of BI.

[0080] Control then returns to Box 7 and the described process repeats until all segments of all identified coughs have been processed and tested for asthma severity. Once all cough sounds have been processed, control then transfers from decision Box 7 to Box 27.

[0081] In box 27, processor 53 operates LCD touchscreen interface 61 to generate screen 14 on machine 51 as shown in FIG. 7. Screen 14 displays a severity rating for each identified cough, taking into account the respiratory index. For example, cough No. 1 is shown to have a severity of "mild," and cough No. 3 is shown to have a severity of "moderate." An average severity rating is also provided, e.g., the average severity rating of "mild" is the average severity across all coughs. In this particular example, none of the coughs are classified as "severe," although such a classification is possible.

[0082] The manner in which the LRM 60 of Figure 2 is pre-trained will now be described, including a discussion of the definition of clinical cases.

[0083] a) Acute Asthma: We defined acute asthma as a respiratory event in children with wheezing or decreased breath sounds during auscultation in response to a bronchodilator test (BT). The BT consisted of the administration of an MDI of salbutamol via a spacer up to three times over one hour at the following dosage: 6 puffs for children <6 years and 12 puffs for children >6 years.

[0084] b) Asymptomatic: This group was defined as children without acute respiratory illness or clinically recognizable respiratory symptoms at the time of recruitment, as determined by clinical judgment. By definition, all subjects in the asymptomatic group had a PS score of 0. As expected in a population sample, this group i) subjects who did not report a history of asthma; ii) subjects who reported a history of chronic asthma but were currently taking preventative medication; iii) Subjects who reported a history of chronic asthma but were not currently taking preventative medication It consists of:

[0085] A clinical panel confirmed the acute asthma diagnosis after reviewing all clinical records, including BT reactivity and investigations. Parents / caregivers reported asthma history and current preventive medication use.

[0086] Cough Record: The recording device, in this example, was an iPhone® 6 (Apple, California, USA) held near the patient, i.e., 25–50 cm from the mouth, with the microphone tilted 45 degrees toward the subject.

[14] Sound data were recorded at a sampling rate of Fs = 44.1 kHz with 16-bit resolution. Pediatric research nurses undertook the recordings in a realistic hospital environment, where background noise included talking, crying, medical equipment, footsteps, and door slams. Care was taken to avoid recording coughs in the presence of avoidable distractions, such as coughs from other people or television sounds or loud conversations. However, some recordings included distractions that cannot be avoided in a medical environment.

[0087] Five or more involuntary or spontaneous coughs were recorded from each child. Children under 2 years of age were able to produce only involuntary coughs. Other children with respiratory illnesses produced both spontaneous and / or involuntary coughs. Asymptomatic volunteers were able to produce only spontaneous coughs with the apparent absence of involuntary coughs.

[0088] Calculation of PS score A PS score was calculated for every individual according to the criteria shown in Table 1 (set out in the background section of this specification). Assessment of the degree of accessory muscle (AM) use and wheezing characteristics was performed by a pediatrician and a pediatric nurse experienced in the use of the PS system.

[0089] Clinical Datasets: The dataset was divided into the following groups (Table 3): i. Acute asthma

[0090]

number

[0091] :

[0092]

number

[0093] teeth

[0094]

number

[0095] Let us denote the set of continuous PS values ​​given by:

[0096]

number

[0097] ,

[0098]

number

[0099] and

[0100]

number

[0101] Three different subgroups were considered, represented by: The first subgroup (PS5-9) included subjects with PS = {5, 6, 7, 8, 9}.

[0102]

number

[0103] represents severe airway limitation, the second subgroup (PS2-4)

[0104]

number

[0105] represents a moderate restriction with PS={2, 3, 4}, and the third subgroup

[0106]

number

[0107] (PS 0-1) represented mild (or non-significant) limitations with PS = {0, 1}. These three subgroups were demarcated to represent distinct groupings requiring different clinical management. ii. Asymptomatic

[0108]

number

[0109] To determine whether different medical histories or medication use influence the classification, the inventors divided subjects without active disease (PS=0) into three subgroups:

[0110]

number

[0111] {subjects with no history of viral wheezing / asthma},

[0112]

number

[0113] {subjects with a history of viral wheezing / asthma at the time of recruitment but who were not taking any preventive medication}, and

[0114]

number

[0115] Subjects were divided into those with a history of viral wheezing / asthma and taking preventive medication at the time of recruitment. At the time of recruitment, these subjects had no respiratory disease or clinically recognizable respiratory symptoms. These three groups are considered identical (i.e., asymptomatic, PS=0) when assessing the presence of acute asthma in clinical settings.

[0116] Training set (G1): Cough-based PS classifier design and cross-validation Due to its reliance on subjective and difficult measurements, PS as estimated by the clinical process can be considered merely a gross measure. Taking this into consideration, we first implemented a logistic regression (LR) classifier:

[0117]

number

[0118] Subjects in the combined groups

[0119]

number

[0120] The subjects were trained with the goal of separating them from each other.

[0121]

number

[0122] The groups defined by are the model training and validation sets comparing PS 0-1 with PS ≥ 5. The inventors used a leave-one-out cross validation method to develop and validate the model, which involves using data from all patients in G1 except one to train the model and cough events from the remaining patients to validate the model. This process was systematically repeated so that each patient in G1 was used once as a validation subject. LR is a linear model, which uses several independent features to estimate the probability of a categorical event (the dependent variable). The inventors

[0123]

number

[0124] The dependent variable was set to Y = 1 if there was a subject in G1 that belonged to F, otherwise it was set to Y = 0. The model was based on the independent input feature (i.e., F ca ={f 0c , f1, f2,....f f}), use a regression function to estimate the probability of Y.

[0125]

number

[0126] is derived as:

[0127] The function in (1) gives a continuous output of probability.

[0128]

number

[0129] belonging to or

[0130]

number

[0131] The decision threshold for determining whether a patient belonged to the 2000 group was calculated using a receiver operating curve (ROC).

[0132] In one embodiment, the method for extracting input cough features is summarized as follows. (i) Let x denote the discrete-time sound signal from a random cough event. (ii) Divide x into three equal-sized non-overlapping segments. X i Let denote the i-th segment of x, where i=1, 2, 3. Each segment X i The following features are calculated from

[12] : bispectral score (BGS), non-Gaussianity score (NGS), first four formant frequencies (FF), logarithmic energy (LogE), zero crossing (ZCR), kurtosis (Kurt), and 12 Mel-frequency cepstral coefficients (MFCC). (iii) The total cough event x is used to calculate 13 wavelet features based on the Morlet wavelet as previously described

[13] . A total of 76 features were extracted from each cough event to form the feature vector Fc.

[0133] Regarding point ii) above, i.e., dividing the cough event into three segments, the motivation for dividing the cough in this way will now be explained. The inventors realized that a cough consists of several physiological phases. First, air is inhaled (in-breath), then the glottis closes, creating a high pressure differential across the closure (phases 1-3). The glottis then opens, resulting in a sudden burst of air, creating the sound heard as a cough. The cough (the main bang of the sound) was considered to have three diagnostically significant subsections: the ascending part, the middle part, and the tail part. In the ascending part, sound energy is generated starting from a small level (intensity) and rises rapidly to a near-maximum. In the middle part, sound energy is primarily generated with a slower (less intense) rise to peak energy, followed by another slower (less intense) decline from the peak level. In the third component (the "tail"), several characteristics were observed, such as a slow, sustained decay to lower energy levels, occasionally interrupted by a second, smaller bang (a so-called "cough-grunt"). During their research program, the inventors discovered that best diagnostic results were achieved when coughs were divided into three segments and analyzed, compared to fewer or more than three segments. The decision to use three segments was based on that realization.

[0134] Additionally, the inventors were guided by physics-based considerations. In the first two segments of the main cough, the larger portion of the sound is generated in the upper airway. In the tail section of the cough, while the upper airway is relatively silent, the acoustic response of the lower airway response can be observed in various details. This is not to say that the inventors did not observe lower airway characteristics in the first two segments, but it is likely that the tail of the cough conveys information about the deeper parts of the airway. The inventors were surprised to discover "wheezing" in the cough because, to the best of their knowledge, no other researchers have previously reported such a finding. Wheezing is usually defined by non-coughing respiratory sounds.

[0135] In asthma, airways narrow due to smooth muscle contraction, thus changing their internal size and muscle tone. The narrowed airways resonate as air passes through them at a constant velocity, producing a sound strong enough to reach the outside of the lungs. A lower lung airspace ratio can result in a "longer" cough and, consequently, a longer tail. Wheezing (airway whistling) can also be present in coughs for this reason. Our analysis of segmented coughs helps capture them, allowing us to characterize asthmatic airways. We hypothesize that diseases such as emphysema (which contributes to COPD) result in the collapse of alveoli, the small spherical sacs at the deepest ends of the airways. The lungs lose elasticity, affecting the recoil mechanism that partially "powers" the cough. We hypothesize that the anterior segment of the cough likely conveys this information.

[0136] Another signal processing consideration was that embodiments of the present method involve estimating mathematical features such as MFCCs, formants, and bispectrum. Dividing the cough into segments before calculating these helps to better maintain the stationarity of the signal.

[0137] Nevertheless, the inventors have discovered that a less preferred embodiment of the method can also function without partitioning.

[0138] In connection with using the whole cough event x, it is not necessary to maintain stationarity to calculate the 13 wavelet features based on the Morlet wavelet. Non-segmentation reduces the number of parameters that must be handled and also gives the wavelet a better chance of representing the whole cough signal. For example, when wavelets are extracted on synthetic cough sounds, segmentation does not need to be performed in some embodiments.

[0139] We trained the LR model using several different approaches. In its simplest form, we used only cough features for LR training. To investigate the relative importance of individual clinical subcomponents on the overall clinical PS (Table 1), we also developed an LR model using RR, wheezing severity, and AM as individual inputs. Finally, we augmented cough features with information derived from respiratory rate, on the premise that such information is easily reportable by patients or guardians, even in a home environment. In contrast, AM and wheezing severity estimation require clinical respiratory expertise and are not reportable by laypeople. Because we desired to deploy a technology that could be deployed in the community, we did not see any purpose in using AM or wheezing severity to augment cough features.

[0140] In the LR model, we converted RR into another feature, namely, the respiratory index (BI), defined as

[12] : BI = RR > 20 breaths per minute for ages 60 months or older, and BI = RR > 40 breaths per minute otherwise. This was inspired by the WHO / IMCI algorithm for classifying pediatric pneumonia in resource-poor areas. The standard RR for younger children is known to be higher than that for older children, and the BI attempts to codify this. In our model, AM use, as determined by experienced clinicians, was graded as in the PS scoring system: none, mild increase, increase, and maximum activity (Table 1).

[0141] Independent test set (G2): Once the LR model training was completed following a leave-one-out cross-validation process, one model was selected according to the guidelines in

[12] . We fixed the parameters of this model and ran it on a previously unused test set, i.e., subjects with moderate disease and individuals without clinically recognizable disease.

[0142]

number

[0143] Note that the sets G1 and G2 are mutually exclusive. The definition of G2 makes the independent testing particularly strict.

[0144]

number

[0145] The subjects in include clinically normal subjects who have not been previously tested by the model and who report different disease / medication histories to the model in the training set.

[0146]

number

[0147] The subjects in the second study were also the same. The discriminatory ability of the algorithm to separate cases with similar disease severity (i.e., disease-free vs. mild / moderate), which would be difficult to recognize in clinical practice, is enhanced by the fact that the data used in the second study was previously untested. For these reasons, the results described below are likely an underestimate of the true discriminatory power of the embodiment described with respect to Figure 1.

[0148] result Study population: A total of 224 subjects were included for analysis, of which 103 had no acute respiratory disease and 121 had acute asthma. Table 3 describes the subjects' demographics, cough characteristics, and classification into training and test groups. Subjects with a PS > 7 were not enrolled because they met the exclusion criteria for medical instability. We analyzed a total of 3161 cough episodes.

[0149] The age of subjects in the asymptomatic group was significantly older than that of the asthmatic group overall (p<0.0001). There was no age difference between asthmatic subjects with PS=0,1 and asthmatic subjects with PS=2-4 (p=0.45), nor between the PS=2-4 group and the PS≥5 group (p=0.15).

[0150] [Table 5]

[0151] Leave-one-out validation results for training set G1 (Table 4): The LR model based only on cough features was able to separate the PS ≥ 5 group from the PS ≤ 1 group with a sensitivity of 82.61% and a specificity of 78.38%. The results indicate that cough sounds convey information for the separation of these PS groups.

[0152] As expected for components of the clinical pulmonary score, BI and AM use also separately separated groups PS ≥ 5 from PS ≤ 1. BI alone achieved similar results to cough analysis alone (sensitivity 69%, specificity 81%). Adding BI to cough features improved sensitivity to 91% and specificity to 97%. AM use on its own achieved 100% sensitivity and 94% specificity, which was not enhanced by the addition of a cough model. Because AM use is rarely seen in mild asthma but always present in severe asthma, it is not surprising that this component of PS performed well in this setting. AM use requires significant clinical expertise to interpret and is therefore impractical as a feature for estimating PS in community settings lacking the necessary clinical expertise. Because community deployment is desired, the inventors did not find any additional utility for discriminating disease severity based on AM use.

[0153] [Table 6]

[0154] Results for the independent test set G2: The ability to distinguish disease-free from mild / moderate disease is an essential part of clinical practice and is made more challenging by overlapping disease severity. To investigate the performance of LR models in separating PS 0 from PS 2-4 (moderate disease), we took one LR model and ran it on an independent test set.

[0155]

number

[0156] The purpose was to examine whether the continuous output of the trained LR function in (1) shows any relationship with PS values ​​for moderate disease. Two different groups of asymptomatic subjects (PS=0) were pooled for this analysis (

[0157]

number

[0158] and

[0159]

number

[0160] ), neither of which were used in the G1 test set. The use of the two datasets tests the discriminatory ability of the algorithm to detect asymptomatic respiratory symptoms.

[0161] Table 5 shows the LR power (mean and SD) for all PS values. Figure 1 shows the LR power (mean and 95% CI) from each model tested. The LR models were trained to separate subjects with PS ≥ 5 from those with PS ≤ 1, but the power was significantly different for PS 2-4 and asymptomatic subjects.

[0162]

number

[0163] and

[0164]

number

[0165] and respond appropriately to

[0166] [Table 7]

[0167] The cough-only model separately discriminated between moderate (PS 2–4) and mild (PS 0–1) asthma severity, however its performance decreased above PS = 5 (Table 5).

[0168] Although BI alone did not discriminate between PS0-1 and PS2-4, it distinguished the combined group PS0-4 from PS ≥ 5. Adding BI to the cough-based mathematical features allowed for a clear separation of the three groups PS0-1, PS2-4, and PS ≥ 5 (Figure 1).

[0169] AM use alone was the best single clinical parameter in separating PS<4, but did not distinguish above this level (Table 5, Figure 1).

[0170] Figure 8 shows the mean logistic regression output with 95% confidence intervals (horizontal bars) for subjects with different PS scores. The numbers of data points at PS values ​​0 (G1), 0 (G2), 1, 2, 3, 4, 5, 6, 7, 8, and 9 are 333, 1040, 147, 387, 476, 422, 233, 54, 69, 0, and 0. The fewer data points at PS=6 and 7 contributed to the larger confidence intervals (CI).

[0171] It can be observed from Figure 8 that there was no difference between the LR power of the two asymptomatic groups (PS=0: G1 and G2). These two groups differed significantly in age (mean difference 22 months, SE 7, p<0.0022), but the LR power did not differ for models of cough only, clinical features only, or the combination of cough and clinical features.

[0172] IV. Discussion Results from 224 patients demonstrated that cough conveys useful information about asthma severity. We developed a cough-only model that accurately reflects performance status in children with asymptomatic, mild, and moderate asthma (PS 0–5). Model performance declined above PS 5, requiring a single simplified component (respiratory rate) of the original PS to correct for this. In severe asthma, airway obstruction can progress to the point where little air can pass through the airways, making acoustic analysis relatively unclear in such regions. In such cases, auscultation may similarly fail to detect wheezing. This is captured in the original PS, where the absence of wheezing due to minimal air exchange is given the highest score. Similar to cough analysis, the PS scale has been reported to correlate better with lower degrees of airway obstruction [9].

[0173] When cough features were augmented with routinely acquired clinical features that do not require auscultation or clinical training (evaluation of wheezing severity and AM use), model performance improved significantly at higher PS values. Clinical signs in PS (RR, AM use) reflect respiratory effort, which increases with more severe obstruction until fatigue occurs. Adding RR (as a BI) to the algorithm further separated severe asthma from moderate and mild asthma and provided a key symptom distinction aligned with specific treatment pathways.

[10] Thus, the algorithm is trained to recognize that the combination of reduced breath sounds (indistinct acoustic characteristics) and significantly elevated BI in asthma indicates an increased level of airway obstruction. Both the PS and the algorithm take into account age-dependent variability in BI when determining the degree of disease severity.

[0174] Grading the severity of acute asthma is an essential component of disease management. Clinical practice protocols rely on accurate assessment, and scoring systems such as the Performance Score (PS) are used to help make decisions such as treatment initiation and response, level of clinical monitoring, hospital triage / admission, and discharge planning. However, many scales have not been adequately validated and rely on subjective assessment of symptoms, such as wheezing characteristics and the ratio of expiratory to inspiratory wheezing. A systematic review of 64 studies determined that none of the pediatric asthma scores tested were adequately validated. [7] Similar conclusions were found in preschool-age children. [8]

[0175] Clinically, it is easier to distinguish severe from mild exacerbations, as opposed to distinguishing moderate from mild or no disease. In practical terms, determining increasing RR and AM use accurately identifies severe exacerbations, but these signs are less clear in diagnosing the presence of mild disease. Mild exacerbations typically present with a slight increase in RR and the absence of AM use, leaving wheezing assessment, a technique requiring considerable expertise, as the sole basis for treatment decision-making. Similarly, identifying and grading AM use requires clinical training and is not usable by parents, caregivers, or inexperienced clinicians. In contrast, the current model does not require the input of additional clinical signs for moderate disease (below a PS of 5), requiring only a simple RR count above this level.

[0176] Two groups of asymptomatic children were compared (PS=0). The inventors wanted to determine whether differences in algorithm detection performance were discernible among asymptomatic individuals. The first group consisted of children with chronically controlled asthma on preventive medication, while the second group included a group of children who had previously wheezed or had never wheezed. No differences in LR power were observed between the groups, suggesting a lack of influence from asymptomatic airway disease in the population used by the inventors.

[0177] There were several limitations to this study: There were fewer subjects with severe asthma (PS>5) compared with mild and moderate disease.

[0178] There are inherent problems with using PS as a reference classifier due to subjectivity, low validity, inter-rater reliability, and reproducibility. For this reason, we chose to train the algorithm using the highest and lowest severity data, and the resulting model was tested on lower severity cases (PS<5). Because the number of subjects in these groups is small, it would be beneficial to undertake more extensive studies and, if appropriate, to evaluate the algorithm against objective measures of airway obstruction, including FEV1 and PEFR.

[0179] Although there were age differences between subjects with and without active disease, there were no age differences between subjects with mild acute asthma and those with moderate or moderate-to-severe acute asthma. Because of the difficulty of obtaining spontaneous coughs from asymptomatic children under 2 years of age, control subjects were generally older than subjects with acute asthma. However, children of any age with acute asthma exacerbations tend to cough involuntarily. Interestingly, the two groups with PS = 0 (G1 and G2) differed significantly in age but not in LR power, suggesting that age may not be an important variable in children without acute asthma. Accurate detection of the absence of asthma is important in clinical practice and in determining whether to initiate appropriate treatment.

[0180] Although PS and other similar systems have only been validated in older age groups due to the difficulty of obtaining accurate lung function parameters, they are routinely used in younger children. The embodiments described herein can be used at any age if the child is coughing involuntarily, and from age 2-3 years if voluntary coughing is required. This compares favorably with the lower age limit of 6-7 years for spirometry or peak flow measurements, but difficulties remain in comparing clearly defined age groups.

[0181] Embodiments herein provide an automated system that can be implemented using smartphone hardware specially configured by app instructions to perform methods such as those of the embodiment shown in FIG. 1. This automated system offers advantages including accessibility, portability, and ease of use, making it suitable for in-home monitoring and integration with asthma management plans. Recommended asthma management plans in many schools and childcare centers require caregivers to be able to quickly assess severity based on their observation of wheezing characteristics and respiratory effort assessment (RR and AM use). Measuring respiratory rate is easier for caregivers than assessing the degree of accessory muscle use or the extent of wheezing (Table 1). The embodiments provided herein may improve initiation and adherence to asthma action plans because they only require assessment of cough and RR.

[0182] Additionally, there is potential for use in telemedicine consultations where a clinician is not available to perform the examination during a traditional medical setting and for use in community asthma action plans. Respiratory diseases account for over 30% of telemedicine consultations.

[16] Measurement of RR can be easily added to cough-based models because it is objective and easily calculated by parents, community health workers, or clinicians during a videoconference. Measurement of RR is also valuable in remote medical settings where access to trained health workers is severely limited. In such areas, routines to measure respiratory rate have been well established under the supervision of community health workers using the WHO algorithm to detect childhood pneumonia.

[17]

[0183] It will be noted that in the described embodiment, the classifier is trained using training vectors made from cough sound features from low severity (mild) and highest severity (severe) patients, but the trained classifier is then used to classify cough sound only data (i.e., excluding respiratory index values) between moderate and low severity asthma patients.

[0184] The reason for this is that clinical estimation of pulmonary severity (PS) scores is a difficult task, influenced by subjectivity. Clinicians have difficulty closely agreeing with each other, even on wheezing detection. Therefore, we consider the baseline PS itself a gross index subject to inter-individual variability, making attempts to accurately estimate PS less useful. We believe it is important to be able to separate mild PS from moderate + severe PS, which is both clinically most useful and challenging. Instead of attempting to train a model using a continuum of PS scores, we reasoned that it would be best to use both extremes in training (i.e., the extreme severe end and the extreme mild end) while retaining the moderate component. In this way, we were able to generate a "confidence" region between the two extremes of PS. Mild and severe groups are less likely to be mixed together due to clinical difficulties in PS estimation, and therefore more reliable. In contrast, the mild and moderate groups may be more mixed due to a certain amount of clinical uncertainty, as may the moderate and severe groups. Therefore, the inventors trained on both extremes. A higher logistic output score means it's a severe PS, and a smaller output closer to 0 indicates it's the other end. The inventors then investigated whether they could create a scale that would work equally well in the moderate domain, and found that it did. That was an important discovery.

[0185] It will be appreciated from the preceding discussion that the inventors have considered that cough conveys useful information about asthma severity. In embodiments, automated cough analysis alone separates mild from moderate disease, and in a further embodiment, the addition of a simple RR method allows for further separation of severe disease when compared to the common Asthma Severity Scale (PS) used in children's hospitals in Western Australia. Advantageously, embodiments described herein can support early treatment, hospitalization, and monitoring decisions, as well as the use of asthma action plans in the community. For example, embodiments include providing appropriate treatment and management, such as administering bronchodilators to patients. Methods according to embodiments described herein are advantageous due to their objectivity and accessibility to diverse caregivers with expertise, without requiring clinical training and expertise in assessing wheezing characteristics or AM use. Embodiments, either alone or in combination with RR, enable rapid assessment of asthma severity, which can be translated into clearer management guidance, which may be an improvement over existing asthma scoring systems.

[0186] The disclosure of each of the following references is hereby incorporated by cross-reference in its entirety.

[0187] References 1. Henderson, J. et al., Management of paediatric asthma in general practice, Aust Fam Physician, 2015, 44(6), pp. 349-51. 2. Australian Institute of Health and Welfare, Australian Burden of Disease Study, Impact and Causes of Disease and Mortality in Australia 2011, Australian Burden of Disease Study Series no. 3, BOD 4, 2016, AIHW, Canberra. 3. National Asthma Council of Australia, Australian Asthma Handbook, 1.3 ed, 2017. 4. Levy, ML et al., Wheezing detection, physician and parent documentation and assessment, J Asthma, 2004, pp. 845-53. 5. Elphick, H.E. et al., Validity and reliability of acoustic analysis of infant breathing sounds, Arch Dis Child, 2004, 89(11), pp. 1059–63. 6. Dinakar, C. et al., Clinical tools for assessing asthma control in children, Pediatrics, 2017, 139(1). 7. Bekhof, J., R. Reimink, and P. L. Brand, Systematic review: Inadequate validation of clinical scores for assessing acute respiratory distress in wheezing children, Paediatr Respir Rev, 2014, 15(1), pp. 98–112. 8. Birken, C.S., P.C. Parkin and C. Macarthur, Asthma severity scores for preschool children showed weaknesses in reliability, validity, and responsiveness, J Clin Epidemiol, 2004, 57(11), pp. 1177–81. 9. Smith, S.R., J.D. Baty and D. Hodge, 3rd, Validation of the Lung Score, a Pediatric Asthma Severity Score, Acad Emerg Med, 2002, 9(2), pp. 99-104. 10. Princess Margaret Hospital Emergency Department, Asthma-Paediatric Acute Care Guidelines, 2014, available at http: / / kidshealthwa.com / guidelines / asthma-2 / . 11. Sharan, RV et al., Prediction of spirometry measurements using cough sound features and regression, Physiol Meas, 2018, 39(9), 095001 p. 12. Abeyratne, UR et al., Cough sound analysis can rapidly diagnose pediatric pneumonia, Annals of Biomedical Engineering, 2013, 41(11), pp. 2448-2462. 13. Kosasih, K., U.R. Abeyratne and V. Swarnkar, Wavelet-enhanced cough analysis for rapid pediatric pneumonia diagnosis, IEEE Trans on Biomed Eng., 2015, 62(4), pp. 1185–1194. 14. Sharan, RV et al., Cough sound analysis for diagnosing croup in pediatric patients using biologically inspired features, Conf Proc IEEE Eng Med Biol Soc, 2017, 2017, pp. 4578-4581. 15. Sharan, RV et al., Automatic Croup Diagnosis Using Cough Sound Recognition, IEEE Trans Biomed Eng, 2018. 16. Uscher-Pines L. and A. Mehrotra, Analysis of Teladoc use appears to indicate expanding patient access to care without prior connection to providers, Health Aff (Millwood), 2014, 33(2), pp. 258–64. 17. World Health Organization, Integrated Management of Childhood Illness, Chart Booklet, 2014, Geneva, Switzerland.

[0188] List of abbreviations PS, lung score AM, secondary muscle WA, Western Australia LR, logistic regression CI, confidence interval SD, standard deviation BI, respiratory index RR, respiratory rate PEFR, peak expiratory flow rate FEV1, forced expiratory volume in 1 second

[0189] The preferred feature sets for testing and training are as set out in Table 2, but the inventors have discovered that reduced feature sets work as well, as follows.

[0190] [Table 8]

[0191] In table 6, KRT is an abbreviation for "Kurtosis", MFccn is an abbreviation for "nth MFCC", and WvL is an abbreviation for "Wavelet".

[0192] Terminology As used herein, conditional language such as "can," "could," "might," "may," "e.g.," and the like, among others, is intended to generally convey that some embodiments include certain features, elements, and / or conditions, but not others, unless otherwise specified or understood in the context in which it is used. Thus, such conditional language is not intended to generally imply that features, elements, and / or conditions are required in any way for one or more embodiments, or that one or more embodiments necessarily include logic for determining whether these features, elements, and / or conditions are to be included in or implemented in any particular embodiment, with or without authorial input or direction.

[0193] Depending on the embodiment, some acts, events, or functions of any of the methods described herein may be performed in a different sequence, added, combined, or omitted altogether (e.g., not all described acts or events may be required for performance of the method). Moreover, in some embodiments, acts or events may be performed simultaneously rather than sequentially, for example, via multithreading, interrupt processing, or multiple processors or processor cores.

[0194] The various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented in hardware or software depends upon the particular application and design constraints imposed on the overall system.

[0195] The described functionality may be implemented in various ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

[0196] The various example logic blocks, modules, and circuits described in connection with the embodiments disclosed herein may be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but alternatively, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. Furthermore, a processor may be a discrete integrated circuit including multiple processing cores, which in turn may be several discrete integrated circuits arranged to communicate with each other to cooperatively accomplish processing tasks.

[0197] The blocks of the methods and algorithms described in connection with the embodiments disclosed herein may be embodied directly in hardware, in software modules executed by a processor, or in a combination of both. The software modules may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. Alternatively, the storage medium may be integral to the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. Alternatively, the processor and the storage medium may reside as discrete components in a user terminal.

[0198] Any modules described herein in some embodiments may be implemented as software modules, hardware modules, or a combination thereof. In general, as used herein, the term "module" can refer to logic circuitry embodied in hardware or firmware, or to a collection of software instructions executable on a processor. In addition, modules and components of modules may, in some embodiments, be implemented in analog circuitry.

[0199] While the foregoing detailed description has illustrated, described, and pointed out novel features as applied to various embodiments, it will be understood that various omissions, substitutions, and changes may be made in the form and details of the devices or algorithms shown without departing from the spirit of the disclosure. As will be recognized, some embodiments of the invention described herein may be embodied in a manner that does not provide all of the features and benefits described herein, since some features may be used or practiced separately from other features and benefits. The scope of the various inventions disclosed herein is indicated by the appended claims, rather than by the foregoing description. All changes that come within the meaning and range of equivalency of the claims are intended to be embraced within their scope.

[0200] In accordance with the regulations, the invention has been described in language more or less specific to structural or methodological features. The term "comprises" and variations thereof, such as "comprising" and "comprised of," are used inclusively throughout and do not exclude any additional features.

[0201] Since the means described herein include preferred forms of carrying out the invention, it is to be understood that the invention is not limited to the specific features shown or described. The invention is therefore claimed in any of its forms or modifications within the proper scope of the appended claims as appropriately interpreted by those skilled in the art.

[0202] Throughout this specification and the claims (if any), unless the context requires otherwise, the terms "substantially" or "about" will be understood not to limit the values ​​to the ranges defined by the terms. [Explanation of symbols]

[0203] 10 Message Screen 12 screens 14 screens 39 sounds 51 Asthma Severity Machine 52 patients 53 processors 54 Clinician 55 Electronic Memory 56 App 57 Bus 58 Operating Systems 59 Lens and CCD Assembly 60 Logistic Regression Model (LRM) 61 LCD touchscreen interface 62 Accelerometer 64 Secondary Memory Card 65 USB ports 66 Acoustic Data 67 GPS module 69 Power Adapter Port and Battery Management 71 Audio Interface 73 WAN / WLAN Assembly 75 microphones 77 speakers 79 Radio Frequency Antenna 81 Voice and / or data communications networks 200 digital signals 300 signals

Claims

1. 1. A method of operating an asthma severity machine, the method being performed by an asthma severity machine and comprising a processor, the method comprising: receiving, by the processor, acoustic data corresponding to sounds from the patient from an acoustic sensor; identifying, by the processor, at least one cough sound in the acoustic data; determining, by the processor, one or more overall cough sound features of the at least one cough sound for each of one or more characteristic features; applying, by the processor, the comprehensive cough sound features to a classifier implemented by the processor, the classifier being a logistic regression model or a Bayesian decision machine, pre-trained with a training set of distinctive features from a population of asthmatic and non-asthmatic subjects; monitoring, by the processor, output from the pre-trained classifier to consider the patient's cough sounds as indicative of one of several degrees of asthma severity; dividing each at least one cough sound into three segments by said processor; determining, by the processor, segment feature values ​​for each of the three segments for each of a number of distinctive features; and classifying the patient's cough sound into one of several levels of severity of the asthma based on a probability obtained by applying the segment features to the pre-trained classifier in addition to the overall cough sound features by the processor; The step of determining by the processor a feature amount for each of several characteristic features comprises: Table 1 The feature value for each of the several specific features is a feature value for each segment listed in Table 1 and a comprehensive cough feature value listed in Table 1. Asthma severity machine operation method.

2. 2. The method of claim 1, wherein the one or more comprehensive cough sound features for the one or more characteristic features include values ​​of wavelet features of the cough sound.

3. 3. The method of claim 1 or 2, wherein said degree of severity comprises "mild" or "moderate to severe."

4. applying, by the processor, a respiratory rate value of the patient, or a value derived therefrom, to the pre-trained classifier in addition to the comprehensive cough sound features, in order to consider the patient's cough sound as indicative of one of several degrees of asthma severity; 3. The method of claim 1 or 2, wherein the degrees of asthma severity include "mild", "moderate" and "severe".

5. 5. A method of operating an asthma severity machine according to claim 4, wherein the respiration rate value comprises a respiration index based on the respiration rate that takes into account the age of the patient.

6. determining, by the processor, a segment feature value for each of a number of distinctive features, MFCC1, MFCC2, MFCC3, MFCC4, MFCC6, MFCC9, and MFCC12 2. The method of claim 1, further comprising the step of determining, by said processor, values ​​for one or more of said segments for one or more of:

7. 7. The method of operating an asthma severity machine of claim 6, further comprising the step of determining, by the processor, a kurtosis value for the first segment.

8. determining, by the processor, a segment feature value for each of a number of distinctive features, Bispectral score (BGS), Non-Gaussian Score (NGS), the first n formant frequencies (FF), Logarithmic Energy (LogE), Zero crossing (ZCR), Kurtosis, The first n Mel-Frequency Cepstral Coefficients (MFCCs) 10. The method of claim 1 further comprising determining, by the processor, a value for each segment for one or more of:

9. The step of determining, by the processor, a segment feature value for each of a number of characteristic features, comprises: Table 2 2. The method of claim 1 further comprising the step of determining, by said processor, 21 features for each segment such that:

10. 1. An asthma severity machine for determining and presenting asthma severity for a patient, said asthma severity machine comprising: Electronic memory; at least one processor in communication with said electronic memory configured with instructions stored in said electronic memory; an audio recording assembly in communication with the at least one processor; a human-machine interface in communication with said at least one processor; said electronic memory comprising: processing the patient's digital recording to identify at least one cough sound; extracting one or more comprehensive cough sound features of the cough sound for each of one or more characteristic features; Implementing a pre-trained pattern classifier for asthma severity, the pattern classifier being a logistic regression model or a Bayesian decision machine, pre-trained with a training set of distinctive features from a population of asthma and non-asthma subjects; applying the comprehensive cough sound features to the pre-trained pattern classifier; operating the human-machine interface to present an asthma severity classification based on output from the pre-trained pattern classifier; Segmenting each at least one cough sound into a plurality of segments; MFCC1, MFCC2, MFCC3, MFCC4, MFCC6, MFCC9, and MFCC12 determining characteristic features for one or more of the segments for one or more of the segments; below Table 3 determining a feature value for each of several characteristic features as follows: classifying the patient's cough sound into one of several levels of severity of the asthma based on a probability obtained by applying, by the processor, a feature value for each of the several specific features to the pre-trained classifier in addition to the overall cough sound feature value; and storing instructions that configure the at least one processor to:

11. 11. The asthma severity machine of claim 10, wherein the electronic memory stores instructions for configuring the at least one processor to extract one or more comprehensive cough sound features comprising values ​​of wavelet features of the cough sound.

12. 12. The asthma severity machine of claim 10 or 11, wherein the electronic memory stores instructions that configure the at least one processor to operate the human-machine interface to present an asthma severity classification based on output from the pre-trained pattern classifier, the asthma severity classification comprising a classification of "mild" or "moderate to severe."

13. 13. The asthma severity machine of claim 12, wherein the electronic memory stores instructions for configuring the at least one processor to apply a respiratory rate value of the patient, or a value derived therefrom, to the comprehensive cough sound features as well as to the pre-trained pattern classifier, in order to operate the human-machine interface to present degrees of asthma severity, including "mild," "moderate," and "severe."

14. 11. The asthma severity machine of claim 10, wherein the electronic memory stores instructions that configure the at least one processor to calculate a feature value for each of the plurality of segments for each of a number of characteristic features.

15. 15. The asthma severity machine of claim 14, wherein the electronic memory stores instructions that configure the at least one processor to apply features of the segment in addition to the overall cough sound features to the pre-trained pattern classifier.

16. 16. The asthma severity machine of any one of claims 10, 14 and 15, wherein the electronic memory stores instructions that configure the at least one processor to determine a kurtosis value for a first segment.

Images