Figure - available from: Scientific Reports
This content is subject to copyright. Terms and conditions apply.
Cough frequency monitoring algorithm used in this study. (A) Signal amplitude threshold, structure of variational autoencoder (VAE) and clustering for cough detection. Signal amplitude thresholds were set using data that were determined as coughs by the “labels”, and data units with an amplitude greater than the threshold were categorized into training datasets and test datasets. VAE, a machine learning algorithm with deep learning, consists of a network called an encoder and decoder and can automatically extract and learn multilevel features of coughs in the latent variable space. K-means clustering was used to determine whether the input data units were “cough units” or “non-cough units” from the latent variables. In short, the VAE built a feature extraction network using a training dataset and is clustered by the k-means algorithm. Then, based on the performance of the learned network and clustering results, the test dataset units were automatically labeled as cough or non-cough units. (B) Example of a clustering result in this study. Using the training data, the VAE extracted features of cough in the latent variable space (latent variables Z1 and Z2), and the results were clustered by the k-means algorithm. Based on this algorithm, the area within the red circle was defined as a cough cluster. Our algorithm was created by using Python 3.9 and PyTorch 1.5.1.
Source publication
Objective evaluations of cough frequency are considered important for assessing the clinical state of patients with respiratory diseases. However, cough monitors with audio recordings are rarely used in clinical settings. Issues regarding privacy and background noise with audio recordings are barriers to the wide use of these monitors; to solve the...
Citations
... In this line, automatic cough detection methods using accelerometers are being explored to enable continuous monitoring of cough frequency and severity. Technologies like accelerometer-based bed occupancy detection systems [20,21] or triaxial accelerometers combined with stretchable strain sensors [22] have shown promising results in accurately detecting cough episodes for long-term monitoring [23]. Additionally, advancements in artificial intelligence have demonstrated promising potential for improving the accuracy and efficiency of cough detection systems [24]. ...
Cough is a frequent symptom in many common respiratory diseases and is considered a predictor of early exacerbation or even disease progression. Continuous cough monitoring offers valuable insights into treatment effectiveness, aiding healthcare providers in timely intervention to prevent exacerbations and hospitalizations. Objective cough monitoring methods have emerged as superior alternatives to subjective methods like questionnaires. In recent years, cough has been monitored using wearable devices equipped with microphones. However, the discrimination of cough sounds from background noise has been shown a particular challenge. This study aimed to demonstrate the effectiveness of single-axis acceleration signals combined with state-of-the-art deep learning (DL) algorithms to distinguish intentional coughing from sounds like speech, laugh, or throat noises. Various DL methods (recurrent, convolutional, and deep convolutional neural networks) combined with one- and two-dimensional time and time–frequency representations, such as the signal envelope, kurtogram, wavelet scalogram, mel, Bark, and the equivalent rectangular bandwidth spectrum (ERB) spectrograms, were employed to identify the most effective approach. The optimal strategy, which involved the SqueezeNet model in conjunction with wavelet scalograms, yielded an accuracy and precision of 92.21% and 95.59%, respectively. The proposed method demonstrated its potential for cough monitoring. Future research will focus on validating the system in spontaneous coughing of subjects with respiratory diseases under natural ambulatory conditions.
... Metrics such as specificity (SP) are often reported, but may misrepresent practicality as they are heavily influenced by cough frequency and long periods of silence, and therefore do not contribute useful information in a long-term monitoring scenario [8]. Furthermore, most studies use fixed-length windows, typically on the order of seconds, for features extraction and classification [8], [9]. This time granularity cannot accurately count individual cough events or distinguish cough patterns, as a cough typically lasts 0.3-0.5 s [5]. ...
... However, in samplebased (SB) models, longer windows can hinder the ability of a classifier to detect individual cough events. For example, Otoshi et al. use 5 s windows, and therefore 10 sequential coughs over 10 s would be counted as 2 coughs [9]. ...
... This is because the SB model can only guarantee one cough per window, and a longer window includes more coughs. Therefore, as observed by Otoshi et al. [9], the number of coughs gets further distorted as the window length increases. In contrast, the EB model's segmentation algorithm identifies each individual event within a window and breaks up events too long to be a single cough. ...
Chronic cough disorders are widespread and challenging to assess because they rely on subjective patient questionnaires about cough frequency. Wearable devices running Machine Learning (ML) algorithms are promising for quantifying daily coughs, providing clinicians with objective metrics to track symptoms and evaluate treatments. However, there is a mismatch between state-of-the-art metrics for cough counting algorithms and the information relevant to clinicians. Most works focus on distinguishing cough from non-cough samples, which does not directly provide clinically relevant outcomes such as the number of cough events or their temporal patterns. In addition, typical metrics such as specificity and accuracy can be biased by class imbalance. We propose using event-based evaluation metrics aligned with clinical guidelines on significant cough counting endpoints. We use an ML classifier to illustrate the shortcomings of traditional sample-based accuracy measurements, highlighting their variance due to dataset class imbalance and sample window length. We also present an open-source event-based evaluation framework to test algorithm performance in identifying cough events and rejecting false positives. We provide examples and best practice guidelines in event-based cough counting as a necessary first step to assess algorithm performance with clinical relevance.
... The cough detection algorithm described herein is based on a reusable flexible, accelerometer-based sensor, worn at the suprasternal notch, achieved an accuracy of 98% for cough detection, with 75% sensitivity and greater than 99% specificity in twenty-seven healthy subjects during biomarker validation. This performance is comparable with prior reports but achieves detection via a low-profile chest sensor 28,29 . Automation of objective cough detection by wearable devices has potential benefits for patients for patients with chronic respiratory effects and acute infections. ...
There have been over 769 million cases of COVID-19, and up to 50% of infected individuals are asymptomatic. The purpose of this study aimed to assess the use of a clinical-grade physiological wearable monitoring system, ANNE One, to develop an artificial intelligence algorithm for (1) cough detection and (2) early detection of COVID-19, through the retrospective analysis of prospectively collected physiological data from longitudinal wear of ANNE sensors in a multicenter single arm study of subjects at high risk for COVID-19 due to occupational or home exposures. The study employed a two-fold approach: cough detection algorithm development and COVID-19 detection algorithm development. For cough detection, healthy individuals wore an ANNE One chest sensor during scripted activity. The final performance of the algorithm achieved an F-1 score of 83.3% in twenty-seven healthy subjects during biomarker validation. In the COVID-19 detection algorithm, individuals at high-risk for developing COVID-19 because of recent exposures received ANNE One sensors and completed daily symptom surveys. An algorithm analyzing vital parameters (heart rate, respiratory rate, cough count, etc.) for early COVID-19 detection was developed. The COVID-19 detection algorithm exhibited a sensitivity of 0.47 and specificity of 0.72 for detecting COVID-19 in 325 individuals with recent exposures. Participants demonstrated high adherence (≥ 4 days of wear per week). ANNE One shows promise for detection of COVID-19. Inclusion of respiratory biomarkers (e.g., cough count) enhanced the algorithm's predictive ability. These findings highlight the potential value of wearable devices in early disease detection and monitoring.
... This is an example of how technology may favor ubiquitous breathing monitoring in daily life and help tailor therapies to individual needs and assess their effects. Indeed, some breathing variables are sensitive to clinical deterioration, pain and infections (Nicolò et al., 2020), and the breathing signal may also be used to retrieve other useful information (e.g., number of cough events) (Otoshi et al., 2021). The impact of inhaled corticosteroids on the cough reflex was investigated by Basin et al., who used an animal model to address this issue. ...
... Researchers have been trying to objectively detect common symptoms of various respiratory diseases. A group of researchers has been using the wearable accelerator and Technology stretchable strain sensors placed around the throat and on the chest to detect coughing as a symptom of different respiratory diseases [18]. They have been relying on an auto-encoder to classify cough with sensitivity and specificity .92 and .96 ...
Goal: Millions of people are dying due to respiratory diseases, such as COVID-19 and asthma, which are often characterized by some common symptoms, including coughing. Therefore, objective reporting of cough symptoms utilizing environment-adaptive machine-learning models with microphone sensing can directly contribute to respiratory disease diagnosis and patient care.
Methods:
In this work, we present three generic modeling approaches –
unguided
,
semi-guided
, and
guided
approaches considering three potential scenarios, i.e., when a user has no prior knowledge, some knowledge, and detailed knowledge about the environments, respectively.
Results:
From detailed analysis with three datasets, we find that
guided
models are up to 28% more accurate than the
unguided
models. We find reasonable performance when assessing the applicability of our models using three additional datasets, including two open-sourced cough datasets.
Conclusions:
Though
guided
models outperform other models, they require a better understanding of the environment.
... Cough generally occurs due to deep breathing along with closure of the glottis and colliding with the respiratory muscles [9]. Due to the resistance of the glottis when it opens for a moment it will cause air to enter with a choking sensation resulting in a coughing sound. ...
... Twelve studies (26.1%) were conducted in European countries, such as France [42,43], Germany [44,45], Italy [46,47], Luxembourg [43], Spain [48,49], Switzerland [50], the Netherlands [48,49], and the United Kingdom [47][48][49][51][52][53]. Six studies (13%) originated from countries in the Western Pacific region, including Australia [54,55], Japan [56,57], and South Korea [58,59]. Only 1 study (2.2%) came from the Southeast Asian (China) [60] and Eastern Mediterranean (Qatar) [61] regions. ...
... Three studies (6.5%) included patients undergoing hemodialysis [25,46,61]. Other clinical areas investigated included circadian rhythms [42], cough [57], sarcopenia [58], physical training [39], and rheumatoid arthritis and lupus erythematosus [41]. ...
... Table 3 summarizes the wearable/activity, mobile phone, clinical/biometric, and active data points collected in the studies. [54,59] Heart rate [17, 19-21, 32, 39, 43, 45, 48, 53, 60] Frequency of app use [37] Mobility pattern [37] Body satisfaction [54] Skin conductance [17] Quantity of app use [36] Ultraviolet radiation exposure [28,34,35] Fitness/health motives for exercise [54] Skin temperature [17] Number of days activity monitor data were uploaded to the web-based app [52] Step count [18-22, 26, 27, 29, 30, 39, 43, 46, 56, 59-61] Engagement in binge eating [54] Blood pressure [19,21,43] Call logs [60] Gait parameters [44,51,58] Engagement in dietary restraint [54] Movements in epigastric region [57] Text message logs [60] Anticipatory postural adjustments [51] Immediate mood [60] Expansion of throat skin [57] App usage logs [60] Sit-to-stand duration [51] Patient Health Questionnaire-9 in an app [60] Weight [38,43] GPS location [40,60] Energy expenditure [39,52] Liebowitz Social Anxiety Scale [40] Blood glucose levels [38] Screen on-and-off status [40,60] Sleep duration [19,26,39,48,49,53,56,60] Generalized Anxiety Disorder 7-Item Scale [40] N/A a Ambient audio [40] Sleep efficiency [19,48,49,53,56] Patient Health Questionnaire 8-item scale [40,48,49] N/A Light sensor data [40] Sleep stage [56] Sheehan Disability Scale [40] N/A Telephone call recipient [42] Distance walked [45,56] Responses to daily assessment [59] N/A Moment in time of telephone call [42] Daytime nap duration [24] Meals logged [59] N/A Telephone call duration [42] Daytime nap frequency [24] Intake of green foods logged [59] N/A Articles read [59] Repositioning events [36] Rosenberg Self-Esteem Scale [48] N/A Comments posted [59] Three-dimensional acceleration [17] Weigh-ins logged [59] N/A Number of posts [59] Number of activity monitor wear days across the intervention [52] Self-reported location [31] N/A Messages sent to coaches [59] Number of interactions with wearable sensor [17] Self-reported social context [31] N/A Number of likes [59] Physical activity [16,33,38,41,45,47,48,50,52] Self-reported cannabis use [31] N/A Screen time metrics [24] Number of postural transitions [61] Mental and physical 5-point scale [39] N/A N/A Exercise time [59] Self-reported sleep, hydration, and nutrition [39] N/A N/A ...
Background
Digital phenotyping is the real-time collection of individual-level active and passive data from users in naturalistic and free-living settings via personal digital devices, such as mobile phones and wearable devices. Given the novelty of research in this field, there is heterogeneity in the clinical use cases, types of data collected, modes of data collection, data analysis methods, and outcomes measured.
Objective
The primary aim of this scoping review was to map the published research on digital phenotyping and to outline study characteristics, data collection and analysis methods, machine learning approaches, and future implications.
Methods
We utilized an a priori approach for the literature search and data extraction and charting process, guided by the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-analyses Extension for Scoping Reviews). We identified relevant studies published in 2020, 2021, and 2022 on PubMed and Google Scholar using search terms related to digital phenotyping. The titles, abstracts, and keywords were screened during the first stage of the screening process, and the second stage involved screening the full texts of the shortlisted articles. We extracted and charted the descriptive characteristics of the final studies, which were countries of origin, study design, clinical areas, active and/or passive data collected, modes of data collection, data analysis approaches, and limitations.
Results
A total of 454 articles on PubMed and Google Scholar were identified through search terms associated with digital phenotyping, and 46 articles were deemed eligible for inclusion in this scoping review. Most studies evaluated wearable data and originated from North America. The most dominant study design was observational, followed by randomized trials, and most studies focused on psychiatric disorders, mental health disorders, and neurological diseases. A total of 7 studies used machine learning approaches for data analysis, with random forest, logistic regression, and support vector machines being the most common.
Conclusions
Our review provides foundational as well as application-oriented approaches toward digital phenotyping in health. Future work should focus on more prospective, longitudinal studies that include larger data sets from diverse populations, address privacy and ethical concerns around data collection from consumer technologies, and build “digital phenotypes” to personalize digital health interventions and treatment plans.
... This satisfies an evaluation that considers the symptoms and X-ray images together. On the other hand, in the studies [56] and [57], a novel automatic cough frequency monitoring system combining a triaxial accelerator and a stretchable strain sensor AND the development of a machine learning-based analysis framework to connect multimodal wearable sensor data are presented for the cough and fatigue respectively. Based on these researches, we can obtain more measured and improved responses for the severity of fatigue and cough especially in the first FIS. ...
COVID-19 disease is an outbreak that seriously affected the whole world, occurred in December 2019, and thus was declared a global epidemic by WHO (World Health Organization). To reduce the impact of the epidemic on humans, it is important to detect the symptoms of the disease in a timely and accurate manner. Recently, several new variants of COVID-19 have been identified in the United Kingdom (UK), South Africa, Brazil and India, and preliminary findings have been suggested that these mutations increase the transmissibility of the virus. Therefore, the aim of this study is to construct a support system based on fuzzy logic for experts to help detect of COVID-19 infection risk in a timely and accurate manner and to get a numerical output on symptoms of the virus from every person. The decision support system consists of three different sub and one main Mamdani type fuzzy inference systems (FIS). Subsystems are Common- Serious symptoms (First), Rare Symptoms (Second) and Personal Information (Third). The first FIS has five inputs, fever-time, cough-time, fatigue-time, shortness of breath and chest pain/dysfunction; the second FIS has four inputs, Loss of Taste/Smell, Body Aches, Conjuctivitis, and Nausea/Vomiting/Diarrhea; and the third FIS has three inputs, Age, Smoke, and Comorbidities. Then, we obtain personal risk index of individual by combining the outputs of these subsystems in a final FIS. The results can be used by health professionals and epidemiologists to make inferences about public health. Numerical output can also be useful for self-control of an individual.
This study introduces an automated, non-intrusive method for detecting cough events in clinical settings, using a flexible chest patch with tri-axial acceleration sensors. It aims to achieve remote and automatic identification of coughing in patients with respiratory system diseases, providing clinicians with a tool for the remote and objective recording and recognition of changes in patients' cough patterns. Twenty-five young and healthy persons (hereinafter referred to as healthy adults) and twenty-five clinical diagnosed pneumoconiosis patients (hereinafter referred to as patients) participated in the experiment by wearing a flexible chest patch with an embedded ACC sensor. The top 56% of the higher scoring features were then combined using several feature selection algorithm to perform the cough classification task. The Multi-Criteria Decision Making (MCDM) method was used to select the classifier with the highest scores. The optimized classifier proposed in this paper achieved Accuracy of 87.1%, Precision of 95%, Recall of 79.1%, F1 score of 86.4%, and AUC of 95.4% for recognizing coughs in healthy adults, and Accuracy of 96.1%, Precision of 95%, Recall of 97.4%, F1 score of 96.2%, and AUC of 98.7% for recognizing coughs in patients; and an overall Accuracy of 92% for distinguishing coughs in the combined group of healthy adults and patients. Our study demonstrates the effectiveness of an automated cough recognition system in both pneumoconiosis patients and healthy adults. This approach facilitates daily remote monitoring of cough occurrences in individuals with pneumoconiosis, potentially enhancing the ability of physicians to evaluate the clinical status.
Cough is a common symptom of various respiratory diseases. However, recognizing a subject's cough using acoustic imaging can be challenging due to the complexity of coughing and the vulnerability of acoustic acquisition to external interference. This research aims to address these issues by utilizing a flexible piezoelectric ionic liquid‐doped poly(L‐lactic acid) (PLLA) sensor to gather the pressure signals from the human surface. 1‐butyl‐3‐methylimidazolium bis (trifluoromethylsulfonyl)imide ([BMIm]TFSI) ionic liquid (IL) can enhance the piezoelectric performance of PLLA thin film, the corresponding output voltage, and impact sensitivity are improved to three times and 105.1% for the PLLA/IL thin film with 1 wt% IL, 140 °C annealing 2 h and four times drawing ratio. The effect of IL on the crystallization behavior and piezoelectric properties of thin films during their preparation is investigated. The flexible PLLA/IL sensor is integrated into a wearable electronic system which collects and transmits piezoelectric signals from cough‐induced vibrations to the filter for denoising, then the acquired cough data are decomposed using the Empirical Mode Decomposition method and trained in a multi‐layer perception model to detect human coughing. It's evident that the flexible piezoelectric PLLA/IL sensors can be applied to human coughing recognition and show great potential applications in intelligent healthcare.
