Validation of a Wearable Medical Device for Automatic Diagnosis of OSA against Standard PSG.

STUDY OBJECTIVE: The objective of this study was to assess the accuracy of automatic diagnosis of obstructive sleep apnea (OSA) with a new, small, acoustic-based, wearable technology (AcuPebble SA100), by comparing it with standard type 1 polysomnography (PSG) diagnosis. MATERIAL AND METHODS: This observational, prospective study was carried out in a Spanish hospital sleep apnea center. Consecutive subjects who had been referred to the hospital following primary care suspicion of OSA were recruited and underwent in-laboratory attended PSG, together with the AcuPebble SA100 device simultaneously overnight from January to December 2022. RESULTS: A total of 80 patients were recruited for the trial. The patients had a median Epworth scoring of 10, a mean of 10.4, and a range of 0-24. The mean AHI obtained with PSG plus sleep clinician marking was 23.2, median 14.3 and range 0-108. The study demonstrated a diagnostic accuracy (based on AHI) of 95.24%, sensitivity of 92.86%, specificity of 97.14%, positive predictive value of 96.30%, negative predictive value of 94.44%, positive likelihood ratio of 32.50 and negative likelihood ratio of 0.07. CONCLUSIONS: The AcuPebble SA100 (EU) device has demonstrated an accurate automated diagnosis of OSA in patients undergoing in-clinic sleep testing when compared against the gold-standard reference of in-clinic PSG.

Automatic Identification of Snoring and Groaning Segments in Acoustic Recordings.

Sleep-related breathing disorders have severe impact on the quality of lives of those suffering from them. These disorders present with a variety of symptoms, out of which snoring and groaning are very common. This paper presents an algorithm to identify and classify segments of acoustic respiratory sound recordings that contain both groaning and snoring events. The recordings were obtained from a database containing 20 subjects from which features based on the Mel-frequency cepstral coefficients (MFCC) were extracted. In the first stage of the algorithm, segments of recordings consisting of either snoring or groaning episodes - without classifying them - were identified. In the second stage, these segments were further differentiated into individual groaning or snoring events. The algorithm in the first stage achieved a sensitivity and specificity of 90.5% ±2.9% and 90.0% ±1.6% respectively, using a RUSBoost model. In the second stage, a random forest classifier was used, and the accuracies for groan and snore events were 78.1% ±4.7% and 78.4% ±4.7% respectively.

Assessing the Feasibility of Acoustic Based Seizure Detection.

OBJECTIVE: Long-term monitoring of epilepsy patients outside of hospital settings is impractical due to the complexity and costs associated with electroencephalogram (EEG) systems. Alternative sensing modalities that can acquire, and automatically interpret signals through easy-to-use wearable devices, are needed to help with at-home management of the disease. In this paper, a novel machine learning algorithm is presented for detecting epileptic seizures using acoustic physiological signals acquired from the neck using a wearable device. METHODS: Acoustic signals from an existing database, were processed, to extract their Mel-frequency Cepstral Coefficients (MFCCs) which were used to train RUSBoost classifiers to identify ictal and non-ictal acoustic segments. A postprocessing stage was then applied to the segment classification results to identify seizures episodes. RESULTS: Tested on 667 hours of acoustic data acquired from 15 patients with at least one seizure, the algorithm achieved a detection sensitivity of 88.1% (95% CI: 79%-97%) from a total of 36 seizures, out of which 24 had no motor manifestations, with a FPR of 0.83/h, and a median detection latency of -42 s. CONCLUSION: The results demonstrated for the first time the ability to identify seizures using acoustic internal body signals acquired on the neck. SIGNIFICANCE: The results of this paper validate the feasibility of using internal physiological sounds for seizure detection, which could potentially be of use for the development of novel, wearable, very simple to use, long term monitoring, or seizure detection systems; circumventing the practical limitations of EEG monitoring outside hospital settings, or systems based on sensing modalities that work on convulsive seizures only.

Towards Automatic Identification of Epileptic Recordings in Long-term EEG Monitoring.

Electroencephalogram (EEG) is a crucial tool in the diagnosis and management of epilepsy. The process of analyzing EEG is time consuming leading to the development of seizure detection algorithms to aid its analysis. This approach is limited since it requires seizures to occur during monitoring periods and can often lead to misdiagnosis in cases where seizure occurrence is rare. For such cases, it has been shown that the interictal periods in EEG signals, which is the predominant state in long-term monitoring, can be useful for the diagnosis of epilepsy. This paper presents an algorithm, using the information in interictal periods, to discriminate between long-term EEG recordings of epilepsy patients and healthy subjects. It extracts several time and frequency-time domain features from the signals and classifies them using an ensemble classifier, achieving 100% sensitivity and 98.7% specificity in classifying 267 recordings from 105 subjects. The results demonstrate the feasibility of this approach to reliably identify EEG recordings of epilepsy subjects automatically which can be highly useful to facilitate screening and diagnosis of epilepsy, especially in those parts of the world where there is a lack of trained personnel for interpreting EEG signals.

Accuracy and usability of AcuPebble SA100 for automated diagnosis of obstructive sleep apnoea in the home environment setting: an evaluation study.

OBJECTIVES: Obstructive sleep apnoea (OSA) is a heavily underdiagnosed condition, which can lead to significant multimorbidity. Underdiagnosis is often secondary to limitations in existing diagnostic methods. We conducted a diagnostic accuracy and usability study, to evaluate the efficacy of a novel, low-cost, small, wearable medical device, AcuPebble_SA100, for automated diagnosis of OSA in the home environment. SETTINGS: Patients were recruited to a standard OSA diagnostic pathway in an UK hospital. They were trained on the use of type-III-cardiorespiratory polygraphy, which they took to use at home. They were also given AcuPebble_SA100; but they were not trained on how to use it. PARTICIPANTS: 182 consecutive patients had been referred for OSA diagnosis in which 150 successfully completed the study. PRIMARY OUTCOME MEASURES: Efficacy of AcuPebble_SA100 for automated diagnosis of moderate-severe-OSA against cardiorespiratory polygraphy (sensitivity/specificity/likelihood ratios/predictive values) and validation of usability by patients themselves in their home environment. RESULTS: After returning the systems, two expert clinicians, blinded to AcuPebble_SA100's output, manually scored the cardiorespiratory polygraphy signals to reach a diagnosis. AcuPebble_SA100 generated automated diagnosis corresponding to four, typically followed, diagnostic criteria: Apnoea Hypopnoea Index (AHI) using 3% as criteria for oxygen desaturation; Oxygen Desaturation Index (ODI) for 3% and 4% desaturation criteria and AHI using 4% as desaturation criteria. In all cases, AcuPebble_SA100 matched the experts' diagnosis with positive and negative likelihood ratios over 10 and below 0.1, respectively. Comparing against the current American Academy of Sleep Medicine's AHI-based criteria demonstrated 95.33% accuracy (95% CI (90·62% to 98·10%)), 96.84% specificity (95% CI (91·05% to 99·34%)), 92.73% sensitivity (95% CI (82·41% to 97·98%)), 94.4% positive-predictive value (95% CI (84·78% to 98·11%)) and 95.83% negative-predictive value (95% CI (89·94% to 98·34%)). All patients used AcuPebble_SA100 correctly. Over 97% reported a strong preference for AcuPebble_SA100 over cardiorespiratory polygraphy. CONCLUSIONS: These results validate the efficacy of AcuPebble_SA100 as an automated diagnosis alternative to cardiorespiratory polygraphy; also demonstrating that AcuPebble_SA100 can be used by patients without requiring human training/assistance. This opens the doors for more efficient patient pathways for OSA diagnosis. TRIAL REGISTRATION NUMBER: NCT03544086; ClinicalTrials.gov.

A Systematic Review of Sensing Technologies for Wearable Sleep Staging.

Designing wearable systems for sleep detection and staging is extremely challenging due to the numerous constraints associated with sensing, usability, accuracy, and regulatory requirements. Several researchers have explored the use of signals from a subset of sensors that are used in polysomnography (PSG), whereas others have demonstrated the feasibility of using alternative sensing modalities. In this paper, a systematic review of the different sensing modalities that have been used for wearable sleep staging is presented. Based on a review of 90 papers, 13 different sensing modalities are identified. Each sensing modality is explored to identify signals that can be obtained from it, the sleep stages that can be reliably identified, the classification accuracy of systems and methods using the sensing modality, as well as the usability constraints of the sensor in a wearable system. It concludes that the two most common sensing modalities in use are those based on electroencephalography (EEG) and photoplethysmography (PPG). EEG-based systems are the most accurate, with EEG being the only sensing modality capable of identifying all the stages of sleep. PPG-based systems are much simpler to use and better suited for wearable monitoring but are unable to identify all the sleep stages.

Acoustic Sensing as a Novel Wearable Approach for Cardiac Monitoring at the Wrist.

This paper introduces the concept of using acoustic sensing over the radial artery to extract cardiac parameters for continuous vital sign monitoring. It proposes a novel measurement principle that allows detection of the heart sounds together with the pulse wave, an attribute not possible with existing photoplethysmography (PPG)-based methods for monitoring at the wrist. The validity of the proposed principle is demonstrated using a new miniature, battery-operated wearable device to sense the acoustic signals and a novel algorithm to extract the heart rate from these signals. The algorithm utilizes the power spectral analysis of the acoustic pulse signal to detect the S1 sounds and additionally, the K-means method to remove motion artifacts for an accurate heartbeat detection. It has been validated on a dataset consisting of 12 subjects with a data length of 6 hours. The results demonstrate an accuracy of 98.78%, mean absolute error of 0.28 bpm, limits of agreement between -1.68 and 1.69 bpm, and a correlation coefficient of 0.998 with reference to a state-of-the-art PPG-based commercial device. The results in this proof of concept study demonstrate the potential of this new sensing modality to be used as an alternative, or to complement existing methods, for continuous monitoring of heart rate at the wrist.

Evaluation of features for classification of wheezes and normal respiratory sounds.

Chronic Respiratory Diseases (CRDs), such as Asthma and Chronic Obstructive Pulmonary Disease (COPD), are leading causes of deaths worldwide. Although both Asthma and COPD are not curable, they can be managed by close monitoring of symptoms to prevent worsening of the condition. One key symptom that needs to be monitored is the occurrence of wheezing sounds during breathing since its early identification could prevent serious exacerbations. Since wheezing can happen randomly without warning, a long-term monitoring system with automatic wheeze detection could be extremely helpful to manage these respiratory diseases. This study evaluates the discriminatory ability of different types of feature used in previous related studies, with a total size of 105 individual features, for automatic identification of wheezing sound during breathing. A linear classifier is used to determine the best features for classification by evaluating several performance metrics, including ranksum statistical test, area under the sensitivity--specificity curve (AUC), F1 score, Matthews Correlation Coefficient (MCC), and relative computation time. Tonality index attained the highest effect size, at 87.95%, and was found to be the feature with the lowest p-value when ranksum significance test was performed. Third MFCC coefficient achieved the highest AUC and average optimum F1 score at 0.8919 and 82.67% respectively, while the highest average optimum MCC was obtained by the first coefficient of a 6th order LPC. The best possible combination of two and three features for wheeze detection is also studied. The study concludes with an analysis of the different trade-offs between accuracy, reliability, and computation requirements of the different features since these will be highly useful for researchers when designing algorithms for automatic wheeze identification.

Evaluation of Mel-Frequency Cepstrum for Wheeze Analysis.

Monitoring of wheezes is an integral part of managing Chronic Respiratory Diseases such as asthma and Chronic Obstructive Pulmonary Disease (COPD). Recently, there is a growing interest in automatic detection of wheezes and the use of Mel-Frequency Cepstral Coefficients (MFCC) have been shown to achieve encouraging detection performance. While the successful use of MFCC for identifying wheezes has been demonstrated, it is not clear which MFCC coefficients are actually useful for detecting wheezes. The objective of this paper is to characterize and study the effectiveness of individual coefficients in discriminating between wheezes and normal respiratory sounds. The coefficients have been evaluated in terms of histogram dissimilarity and linear separability. Further, a comparison between the use of single coefficient against other combinations of coefficients is also presented. The results demonstrate MFCC-2 coefficient to be significantly more effective than all the other coefficients in discriminating between wheezes and normal respiratory sounds sampled at 8000 Hz.

An Algorithm for Heart Rate Extraction From Acoustic Recordings at the Neck.

Heart rate is an important physiological parameter to assess the cardiac condition of an individual and is traditionally determined by attaching multiple electrodes on the chest of a subject to record the electrical activity of the heart. The installation and handling complexities of such systems does not prove feasible for a user to undergo a long-term monitoring in the home settings. A small-sized, battery-operated wearable monitoring device is placed on the suprasternal notch at neck to record acoustic signals containing information about breathing and cardiac sounds. The heart sounds obtained are heavily corrupted by the respiratory cycles and other external artifacts. This paper presents a novel algorithm for reliably extracting the heart rate from such acoustic recordings, keeping in mind the constraints posed by the wearable technology. The methodology constructs the Hilbert energy envelope of the signal by calculating its instantaneous characteristics to segment and classify a cardiac cycle into S1 and S2 sounds using their timing characteristics. The algorithm is tested on a dataset consisting of 13 subjects with an approximate data length of 75 h and achieves an accuracy of 94.34%, an RMS error of 3.96 bpm and a correlation coefficient of 0.93 with reference to a commercial device in use.

Characterization Study of Neck Photoplethysmography.

This paper presents a comparison between finger and neck photoplethysmography (PPG) in order to assess the potential and limitations of this, non-conventionally used, body site for application in pulse oximetry. PPG signals were recorded at both sites from healthy subjects to inspect the differences in average waveforms, as well as in oxygen saturation (SpO2) and heart rate (HR) estimation. The results show significant differences in the average PPG pulse waveforms for different contour features such as diastolic or dicrotic notch amplitude, among others. The results show that the HR estimated from signals obtained with the neck sensor are strongly correlated to the output of the reference finger (R=0.862, MAE=1.27 BPM), whereas SpO2 measurements are not that accurately predicted (R=0.129, MAE=11.7%). Spectrograms under different breathing conditions revealed that the respiratory frequency is more predominant in neck PPG than in finger, which has a great potential for respiratory rate (RR) extraction. These are very promising results for the suitability of the neck as an alternative location for monitoring of respiratory diseases, and specifically for sleep apnea.

Pulse oximetry in the neck - a proof of concept.

Oxygen saturation levels are routinely monitored in clinical settings. Pulse oximetry, in transmittance operation mode, is the most common method of estimating oxygen saturation (SpO2). This is inexpensive and non-invasive and thus allows for long-term monitoring. However, it suffers from issues such as signal integrity, reliability and patient comfortability. As a result, there is an interest in exploring other locations on the body where oxygen saturation can be measured reliably. In this paper, a wearable device has been designed to study the feasibility of extracting photoplethysmogram (PPG) signals at the neck in reflectance pulse oximetry mode. It explores the signal integrity and strength compared to other locations as well as the presence of motion artefacts in that location. The results demonstrate that the PPG signals acquired at the neck show a very strong correlation (r=0.82) with the SpO2 values obtained using a commercial device. Further, the SpO2 values are calculated with an accuracy of 98.6%.

Monitoring smoking behaviour using a wearable acoustic sensor.

Smoking is a cause of multiple health problems resulting in diseases which can also be fatal. It is well known that smoking has long-term impact on the health of an individual as well. While a number of studies have looked at the impact of smoking on health and its economic impacts, most of these rely on input from smokers in the form of questionnaires and surveys. Long-term monitoring of smoking habits and behaviour is thus not possible because of the lack of means to do so. This paper proposes the use of a wearable device to monitor breathing signals of subjects. It is shown that the acoustic properties of a smoking breath are different from a non-smoking breath. To encapsulate these differences, several features from a breath segment are extracted and used with a simple classifier to automatically identify smoking breaths. The proposed algorithm detected smoking and non-smoking breaths with average accuracy of 66% and 99% respectively.

ECG artefact identification and removal in mHealth systems for continuous patient monitoring.

Continuous patient monitoring systems acquire enormous amounts of data that is either manually analysed by doctors or automatically processed using intelligent algorithms. Sections of data acquired over long period of time can be corrupted with artefacts due to patient movement, sensor placement and interference from other sources. Owing to the large volume of data these artefacts need to be automatically identified so that the analysis systems and doctors are aware of them while making medical diagnosis. Three important factors are explored that must be considered and quantified for the design and evaluation of automatic artefact identification algorithms: signal quality, interpretation quality and computational complexity. The first two are useful to determine the effectiveness of an algorithm, whereas the third is particularly vital in mHealth systems where computational resources are heavily constrained. A series of artefact identification and filtering algorithms are then presented focusing on the electrocardiography data. These algorithms are quantified using the three metrics to demonstrate how different algorithms can be evaluated and compared to select the best ones for a given wireless sensor network.

A Cough-Based Algorithm for Automatic Diagnosis of Pertussis.

Pertussis is a contagious respiratory disease which mainly affects young children and can be fatal if left untreated. The World Health Organization estimates 16 million pertussis cases annually worldwide resulting in over 200,000 deaths. It is prevalent mainly in developing countries where it is difficult to diagnose due to the lack of healthcare facilities and medical professionals. Hence, a low-cost, quick and easily accessible solution is needed to provide pertussis diagnosis in such areas to contain an outbreak. In this paper we present an algorithm for automated diagnosis of pertussis using audio signals by analyzing cough and whoop sounds. The algorithm consists of three main blocks to perform automatic cough detection, cough classification and whooping sound detection. Each of these extract relevant features from the audio signal and subsequently classify them using a logistic regression model. The output from these blocks is collated to provide a pertussis likelihood diagnosis. The performance of the proposed algorithm is evaluated using audio recordings from 38 patients. The algorithm is able to diagnose all pertussis successfully from all audio recordings without any false diagnosis. It can also automatically detect individual cough sounds with 92% accuracy and PPV of 97%. The low complexity of the proposed algorithm coupled with its high accuracy demonstrates that it can be readily deployed using smartphones and can be extremely useful for quick identification or early screening of pertussis and for infection outbreaks control.

An algorithm for automatic detection of drowsiness for use in wearable EEG systems.

Lack of proper restorative sleep can induce sleepiness at odd hours making a person drowsy. This onset of drowsiness can be detrimental for the individual in a number of ways if it happens at an unwanted time. For example, drowsiness while driving a vehicle or operating heavy machinery poses a threat to the safety and wellbeing of individuals as well as those around them. Timely detection of drowsiness can prevent the occurrence of unfortunate accidents thereby improving road and work environment safety. In this paper, by analyzing the electroencephalographic (EEG) signals of human subjects in the frequency domain, several features across different EEG channels are explored. Of these, three features are identified to have a strong correlation with drowsiness. A weighted sum of these defining features, extracted from a single EEG channel, is then used with a simple classifier to automatically separate the state of wakefulness from drowsiness. The proposed algorithm resulted in drowsiness detection sensitivity of 85% and specificity of 93%.

Algorithm for heart rate extraction in a novel wearable acoustic sensor.

Phonocardiography is a widely used method of listening to the heart sounds and indicating the presence of cardiac abnormalities. Each heart cycle consists of two major sounds - S1 and S2 - that can be used to determine the heart rate. The conventional method of acoustic signal acquisition involves placing the sound sensor at the chest where this sound is most audible. Presented is a novel algorithm for the detection of S1 and S2 heart sounds and the use of them to extract the heart rate from signals acquired by a small sensor placed at the neck. This algorithm achieves an accuracy of 90.73 and 90.69%, with respect to heart rate value provided by two commercial devices, evaluated on more than 38 h of data acquired from ten different subjects during sleep in a pilot clinical study. This is the largest dataset for acoustic heart sound classification and heart rate extraction in the literature to date. The algorithm in this study used signals from a sensor designed to monitor breathing. This shows that the same sensor and signal can be used to monitor both breathing and heart rate, making it highly useful for long-term wearable vital signs monitoring.

An open-source toolbox for standardized use of PhysioNet Sleep EDF Expanded Database.

PhysioNet Sleep EDF database has been the most popular source of data used for developing and testing many automatic sleep staging algorithms. However, the recordings from this database has been used in an inconsistent fashion. For example, arbitrary selection of start and end times from long term recordings, data-hypnogram mismatches, different performance metrics and hypnogram conversion from R&K to AASM. All these differences result in different data sections and performance metrics being used by researchers thereby making any direct comparison between algorithms very difficult. Recently, a superset of this database has been made available on PhysioNet, known as the Sleep EDF Expanded Database which includes 61 recordings. This provides an opportunity to standardize the way in which signals from this database should be used. With this goal in mind, we present in this paper a toolbox for automatically downloading and extracting recordings from the Sleep EDF Expanded database and converting them to a suitable format for use in MATLAB. This toolbox contains functions for selecting appropriate data for sleep analysis (based on our previous recommendations for sleep staging), hypnogram conversion and computation of performance metrics. Its use makes it simpler to start using the new sleep database and also provides a foundation for much-needed standardization in this research field.

Performance-power consumption tradeoff in wearable epilepsy monitoring systems.

Automated seizure detection methods can be used to reduce time and costs associated with analyzing large volumes of ambulatory EEG recordings. These methods however have to rely on very complex, power hungry algorithms, implemented on the system backend, in order to achieve acceptable levels of accuracy. In size, and therefore power-constrained EEG systems, an alternative approach to the problem of data reduction is online data selection, in which simpler algorithms select potential epileptiform activity for discontinuous recording but accurate analysis is still left to a medical practitioner. Such a diagnostic decision support system would still provide doctors with information relevant for diagnosis while reducing the time taken to analyze the EEG. For wearable systems with limited power budgets, data selection algorithm must be of sufficiently low complexity in order to reduce the amount of data transmitted and the overall power consumption. In this paper, we present a low-power hardware implementation of an online epileptic seizure data selection algorithm with encryption and data transmission and demonstrate the tradeoffs between its accuracy and the overall system power consumption. We demonstrate that overall power savings by data selection can be achieved by transmitting less than 40% of the data. We also show a 29% power reduction when selecting and transmitting 94% of all seizure events and only 10% of background EEG.

Automatic sleep staging using state machine-controlled decision trees.

Automatic sleep staging from a reduced number of channels is desirable to save time, reduce costs and make sleep monitoring more accessible by providing home-based polysomnography. This paper introduces a novel algorithm for automatic scoring of sleep stages using a combination of small decision trees driven by a state machine. The algorithm uses two channels of EEG for feature extraction and has a state machine that selects a suitable decision tree for classification based on the prevailing sleep stage. Its performance has been evaluated using the complete dataset of 61 recordings from PhysioNet Sleep EDF Expanded database achieving an overall accuracy of 82% and 79% on training and test sets respectively. The algorithm has been developed with a very small number of decision tree nodes that are active at any given time making it suitable for use in resource-constrained wearable systems.