Showing papers in "Physiological Measurement in 2021"

Classification of 12-lead ECGs: the PhysioNet/Computing in Cardiology Challenge 2020.

Abstract: Objective: Vast 12-lead ECGs repositories provide opportunities to develop new machine learning approaches for creating accurate and automatic diagnostic systems for cardiac abnormalities. \However, most 12-lead ECG classification studies are trained, tested, or developed in single, small, or relatively homogeneous datasets. In addition, most algorithms focus on identifying small numbers of cardiac arrhythmias that do not represent the complexity and difficulty of ECG interpretation. This work addresses these issues by providing a standard, multi-institutional database and a novel scoring metric through a public competition: the PhysioNet/Computing in Cardiology Challenge 2020. Approach: A total of 66361 12-lead ECG recordings were sourced from six hospital systems from four countries across three continents. 43,101 recordings were posted publicly with a focus on 27 diagnoses. For the first time in a public competition, we required teams to publish open-source code for both training and testing their algorithms, ensuring full scientific reproducibility. Main results: A total of 217 teams submitted 1395 algorithms during the Challenge, representing a diversity of approaches for identifying cardiac abnormalities from both academia and industry. As with previous Challenges, high-performing algorithms exhibited significant drops (≈ 10%) in performance on the hidden test data. Significance: Data from diverse institutions allowed us to assess algorithmic generalizability. A novel evaluation metric considered different misclassification errors for different cardiac abnormalities, capturing the outcomes and risks of different diagnoses. Requiring both trained models and code for training models improved the generalizability of submissions, setting a new bar in reproducibility for public data science competitions.

Positive end-expiratory pressure titration with electrical impedance tomography and pressure-volume curve: a randomized trial in moderate to severe ARDS.

Abstract: Objective The aim of the study was to compare titration of positive end-expiratory pressure (PEEP) with electrical impedance tomography (EIT) and with ventilator-embedded pressure-volume (PV) loop in moderate to severe acute respiratory distress syndrome (ARDS). Approach Eighty-seven moderate to severe ARDS patients (arterial oxygen partial pressure to fractional inspired oxygen ratio, PaO2/FiO2 ≤ 200 mmHg) were randomized to either EIT group (n = 42) or PV group (n = 45). All patients received identical medical care using the same general support guidelines and protective mechanical ventilation. In the EIT group, the selected PEEP equaled the airway pressure at the intercept between cumulated collapse and overdistension percentages curves and in the PV group, at the pressure where maximal hysteresis was reached. Main results Baseline characteristics and settings were comparable between the groups. After optimization, PEEP was significantly higher in the PV group (17.4 ± 1.7 versus 16.2 ± 2.6 cmH2O, PV versus EIT groups, p = 0.02). After 48 h, driving pressure was significantly higher in the PV group (12.4 ± 3.6 versus 10.9 ± 2.5 cmH2O, p = 0.04). Lung mechanics and oxygenation were better in the EIT group but did not statistically differ between the groups. The survival rate was lower in the PV group (44.4% versus 69.0%, p = 0.02; hazard ratio 2.1, confidence interval 1·1-3.9). None of the other pre-specified exploratory clinical endpoints were significantly different. Significance In moderate to severe ARDS, PEEP titration guided with EIT, compared with PV curve, might be associated with improved driving pressure and survival rate. Trial registration NCT03112512, 13 April, 2017.

Deep learning based photoplethysmography classification for peripheral arterial disease detection: a proof-of-concept study.

Abstract: Objective.A proof-of-concept study to assess the potential of a deep learning (DL) based photoplethysmography PPG ('DLPPG') classification method to detect peripheral arterial disease (PAD) using toe PPG signals.Approach.PPG spectrogram images derived from our previously published multi-site PPG datasets (214 participants; 31.3% legs with PAD by ankle brachial pressure index (ABPI)) were input into a pretrained 8-layer (five convolutional layers + three fully connected layers) AlexNet as tailored to the 2-class problem with transfer learning to fine tune the convolutional neural network (CNN).k-fold random cross validation (CV) was performed (fork = 5 andk = 10), with each evaluated over k training/validation runs. Overall test sensitivity, specificity, accuracy, and Cohen's Kappa statistic with 95% confidence interval ranges were calculated and compared, as well as sensitivities in detecting mild-moderate (0.5 ≤ ABPI < 0.9) and major (ABPI < 0.5) levels of PAD.Main results.CV with eitherk = 5 or 10 folds gave similar diagnostic performances. The overall test sensitivity was 86.6%, specificity 90.2% and accuracy 88.9% (Kappa: 0.76 [0.70-0.82]) (atk= 5). The sensitivity to mild-moderate disease was 83.0% (75.5%-88.9%) and to major disease was 100.0% (90.5%-100.0%).Significance.Substantial agreements have been demonstrated between the DL-based PPG classification technique and the ABPI PAD diagnostic reference. This novel automatic approach, requiring minimal pre-processing of the pulse waveforms before PPG trace classification, could offer significant benefits for the diagnosis of PAD in a variety of clinical settings where low-cost, portable and easy-to-use diagnostics are desirable.

Agreement of bioelectrical resistance, reactance, and phase angle values from supine and standing bioimpedance analyzers.

Abstract: Objective Bioimpedance devices are commonly used to assess health parameters and track changes in body composition. However, the cross-sectional agreement between different devices has not been conclusively established. Thus, the objective of this investigation was to examine the agreement between raw bioelectrical variables (resistance, reactance, and phase angle at the 50-kHz frequency) obtained from three bioimpedance analyzers. Approach Healthy male (n=76, Mean±SD; 33.8±14.5 years; 83.9±15.1 kg; 179.4±6.9 cm) and female (n=103, Mean±SD; 33.4±15.9 years; 65.6±12.1 kg; 164.9±6.4 cm) participants completed assessments using three bioimpedance devices: supine bioimpedance spectroscopy (BIS), supine single-frequency bioelectrical impedance analysis (SFBIA), and standing multi-frequency bioelectrical impedance analysis (MFBIA). Differences in raw bioelectrical variables between the devices were quantified via one-way analysis of variance for the total sample and for each sex. Equivalence testing was used to determine equivalence between methods. Main results Significant differences in all bioelectrical variables were observed between the three devices when examining the total sample and males only. The devices appeared to exhibit slightly better agreement when analyzing female participants only. Equivalence testing using the total sample as well as males and females separately revealed that resistance and phase angle were equivalent between the supine devices (BIS, SFBIA), but not with the standing analyzer (MFBIA). Significance The present study demonstrated disagreement between different bioimpedance analyzers for quantifying raw bioelectrical variables, with the poorest agreement between devices that employed different body positions during testing. These results suggest that researchers and clinicians should employ device-specific reference values to classify participants based on raw bioelectrical variables, such as phase angle. If reference values are needed but are unavailable for a particular bioimpedance analyzer, the set of reference values produced using the most similar analyzer and reference population should be selected.

Identification of 27 abnormalities from multi-lead ECG signals: an ensembled SE_ResNet framework with Sign Loss function

Abstract: Objective. Cardiovascular disease is a major threat to health and one of the primary causes of death globally. The 12-lead ECG is a cheap and commonly accessible tool to identify cardiac abnormalities. Early and accurate diagnosis will allow early treatment and intervention to prevent severe complications of cardiovascular disease. Our objective is to develop an algorithm that automatically identifies 27 ECG abnormalities from 12-lead ECG databases.Approach. Firstly, a series of pre-processing methods were proposed and applied on various data sources in order to mitigate the problem of data divergence. Secondly, we ensembled two SE_ResNet models and one rule-based model to enhance the performance of various ECG abnormalities' classification. Thirdly, we introduce a Sign Loss to tackle the problem of class imbalance, and thus improve the model's generalizability.Main results. In the PhysioNet/Computing in Cardiology Challenge (2020), our proposed approach achieved a challenge validation score of 0.682, and a full test score of 0.514, placed us 3rd out of 40 in the official ranking.Significance. We proposed an accurate and robust predictive framework that combines deep neural networks and clinical knowledge to automatically classify multiple ECG abnormalities. Our framework is able to identify 27 ECG abnormalities from multi-lead ECG signals regardless of discrepancies in data sources and the imbalance of data labeling. We trained our framework on five datasets and validated it on six datasets from various countries. The outstanding performance demonstrate the effectiveness of our proposed framework.

Bioimpedance analysis as a tool for hemodynamic monitoring: overview, methods and challenges.

Abstract: Recent advances in hemodynamic monitoring have seen the advent of non-invasive methods which offer ease of application and improve patient comfort. Bioimpedance Analysis or BIA is one of the currently employed non-invasive techniques for hemodynamic monitoring. Impedance Cardiography (ICG), one of the implementations of BIA, is widely used as a non-invasive procedure for estimating hemodynamic parameters such as stroke volume (SV) and cardiac output (CO). Even though BIA is not a new diagnostic technique, it has failed to gain consensus as a reliable measure of hemodynamic parameters. Several devices have emerged for estimating CO using ICG which are based on evolving methodologies and techniques to calculate SV. However, the calculations are generally dependent on the electrode configurations (whole body, segmental or localised) as well as the accuracy of different techniques in tracking blood flow changes. Blood volume changes, concentration of red blood cells, pulsatile velocity profile and ambient temperature contribute to the overall conductivity of blood and hence its impedance response during flow. There is a growing interest in investigating limbs for localised BIA to estimate hemodynamic parameters such as pulse wave velocity. As such, this paper summarises the current state of hemodynamic monitoring through BIA in terms of different configurations and devices in the market. The conductivity of blood flow has been emphasized with contributions from both volume and velocity changes during flow. Recommendations for using BIA in hemodynamic monitoring have been mentioned highlighting the suitable range of frequencies (1 kHz-1 MHz) as well as safety considerations for a BIA setup. Finally, current challenges in using BIA such as geometry assumption and inaccuracies have been discussed while mentioning potential advantages of a multi-frequency analysis to cover all the major contributors to blood's impedance response during flow.

Filtering-induced time shifts in photoplethysmography pulse features measured at different body sites: the importance of filter definition and standardization

Abstract: Objective.The waveform of a photoplethysmography (PPG) signal depends on the measurement site and individual physiological conditions. Filtering can distort the morphology of the original PPG signal waveform and change the timing of pulse feature points on PPG signals. We aim to quantitatively investigate the effect of PPG signal morphology (related to measurement site) and type of pulse feature on the filtering-induced time shift (TS).Approach.60 s PPG signals were measured from six body sites (finger, wrist under (volar), wrist upper (dorsal), earlobe, and forehead) of 36 healthy adults. Using infinite impulse response digital filters which are common in PPG signal processing, PPG signals were prefiltered (band-pass, pass and stop bands: >0.5 Hz and 30 Hz for low-pass filter) and then filtered (low-pass, pass and stop bands: 5 Hz). Four pulse feature points were defined and extracted (peak, valley, maximal first derivative, and maximal second derivative). For each subject, overall TS and intra-subject TS variability in feature points were calculated as the mean and standard deviation of TS between prefiltered and filtered PPG signals in 50 cardiac cycles. Statistical testing was performed to investigate the effect of measurement site and type of pulse feature on overall TS and intra-subject TS variability.Main results.Measurement site, type of pulse feature, and their interaction had significant impacts on the overall TS and intra-subject TS variability (p < 0.001 for all). Valley and maximal second derivative showed higher overall TS than peak and maximal first derivative. Finger had higher overall TS and lower intra-subject TS variability than other measurement sites.Significance. Measurement site and type of pulse feature can significantly influence the timing of feature points on filtered PPG signals. Filtering parameters should be quoted to support the reproducibility of PPG-related studies.

Transfer learning from ECG to PPG for improved sleep staging from wrist-worn wearables.

Abstract: Objective To develop a sleep staging method from wrist-worn accelerometry and the photoplethysmogram (PPG) by leveraging transfer learning from a large electrocardiogram (ECG) database. Approach In previous work, we developed a deep convolutional neural network for sleep staging from ECG using the cross-spectrogram of ECG-derived respiration and instantaneous beat intervals, heart rate variability metrics, spectral characteristics, and signal quality measures derived from 5,793 subjects in Sleep Heart Health Study (SHHS). We updated the weights of this model by transfer learning using PPG data derived from the Empatica E4 wristwatch worn by 105 subjects in the `Emory Twin Study Follow-up' (ETSF) database, for whom overnight polysomnographic (PSG) scoring was available. The relative performance of PPG, and actigraphy (Act), plus combinations of these two signals, with and without transfer learning was assessed. Main results The performance of our model with transfer learning showed higher accuracy (1-9 percentage points) and Cohen's Kappa (0.01-0.13) than those without transfer learning for every classification category. Statistically significant, though relatively small, incremental differences in accuracy occurred for every classification category as tested with the McNemar test. The out-of-sample classification performance using features from PPG and actigraphy for four-class classification was Accuracy (Acc)=68.62% and Kappa=0.44. For two-class classification, the performance was Acc=81.49% and Kappa=0.58. Significance We proposed a combined PPG and actigraphy-based sleep stage classification approach using transfer learning from a large ECG sleep database. Results demonstrate that the transfer learning approach improves estimates of sleep state. The use of automated beat detectors and quality metrics means human over-reading is not required, and the approach can be scaled for large cross-sectional or longitudinal studies using wrist-worn devices for sleep-staging.

Automatic cough classification for tuberculosis screening in a real-world environment

Abstract: \textit{Objective:} The automatic discrimination between the coughing sounds produced by patients with tuberculosis (TB) and those produced by patients with other lung ailments. \textit{Approach:} We present experiments based on a dataset of 1358 forced cough recordings obtained in a developing-world clinic from 16 patients with confirmed active pulmonary TB and 35 patients suffering from respiratory conditions suggestive of TB but confirmed to be TB negative. Using nested cross-validation, we have trained and evaluated five machine learning classifiers: logistic regression (LR), support vector machines (SVM), k-nearest neighbour (KNN), multilayer perceptrons (MLP) and convolutional neural networks (CNN). \textit{Main Results:} Although classification is possible in all cases, the best performance is achieved using LR. In combination with feature selection by sequential forward selection (SFS), our best LR system achieves an area under the ROC curve (AUC) of 0.94 using 23 features selected from a set of 78 high-resolution mel-frequency cepstral coefficients (MFCCs). This system achieves a sensitivity of 93\% at a specificity of 95\% and thus exceeds the 90\% sensitivity at 70\% specificity specification considered by the World Health Organisation (WHO) as a minimal requirement for a community-based TB triage test. \textit{Significance:} The automatic classification of cough audio sounds, when applied to symptomatic patients requiring investigation for TB, can meet the WHO triage specifications for the identification of patients who should undergo expensive molecular downstream testing. This makes it a promising and viable means of low cost, easily deployable frontline screening for TB, which can benefit especially developing countries with a heavy TB burden.

Deep learning based classification of unsegmented phonocardiogram spectrograms leveraging transfer learning.

Abstract: Objective.Cardiovascular diseases (CVDs) are a main cause of deaths all over the world. This research focuses on computer-aided analysis of phonocardiogram (PCG) signals based on deep learning that can enable improved and timely detection of heart abnormalities. The two widely used publicly available PCG datasets are from the PhysioNet/CinC (2016) and PASCAL (2011) challenges. The datasets are significantly different in terms of the tools used for data acquisition, clinical protocols, digital storages and signal qualities, making it challenging to process and analyze.Approach.In this work, we have used short-time Fourier transform-based spectrograms to learn the representative patterns of the normal and abnormal PCG signals. Spectrograms generated from both the datasets are utilized to perform four different studies: (i) train, validate and test different variants of convolutional neural network (CNN) models with PhysioNet dataset, (ii) train, validate and test the best performing CNN structure on the PASCAL dataset, as well as (iii) on the combined PhysioNet-PASCAL dataset and (iv) finally, the transfer learning technique is employed to train the best performing pre-trained network from the first study with PASCAL dataset.Main results.The first study achieves an accuracy, sensitivity, specificity, precision and F1 scores of 95.75%, 96.3%, 94.1%, 97.52%, and 96.93%, respectively, while the second study shows accuracy, sensitivity, specificity, precision and F1 scores of 75.25%, 74.2%, 76.4%, 76.73%, and 75.42%, respectively. The third study shows accuracy, sensitivity, specificity, precision and F1 scores of 92.7%, 94.98%, 89.95%, 95.3% and 94.6%, respectively. Finally, the fourth study shows a precision of 96.98% on the noisy PASCAL dataset with transfer learning approach.Significance.The proposed approach employs a less complex and relatively light custom CNN model that outperforms most of the recent competing studies by achieving comparatively high classification accuracy and precision, making it suitable for screening CVDs using PCG signals.

A dilated inception CNN-LSTM network for fetal heart rate estimation.

Abstract: Objective. Fetal heart rate (HR) monitoring is routinely used during pregnancy and labor to assess fetal well-being. The noninvasive fetal electrocardiogram (ECG), obtained by electrodes on the maternal abdomen, is a promising alternative to standard fetal monitoring. Subtraction of the maternal ECG from the abdominal measurements results in fetal ECG signals, in which the fetal HR can be determined typically through R-peak detection. However, the low signal-to-noise ratio and the nonstationary nature of the fetal ECG make R-peak detection a challenging task.Approach. We propose an alternative approach that instead of performing R-peak detection employs deep learning to directly determine the fetal HR from the extracted fetal ECG signals. We introduce a combination of dilated inception convolutional neural networks (CNN) with long short-term memory networks to capture both short-term and long-term temporal dynamics of the fetal HR. The robustness of the method is reinforced by a separate CNN-based classifier that estimates the reliability of the outcome.Main results. Our method achieved a positive percent agreement (within 10% of the actual fetal HR value) of 97.3% on a dataset recorded during labor and 99.6% on set-A of the 2013 Physionet/Computing in Cardiology Challenge exceeding top-performing state-of-the-art algorithms from the literature.Significance. The proposed method can potentially improve the accuracy and robustness of fetal HR extraction in clinical practice.

Onset detection in surface electromyographic signals across isometric explosive and ramped contractions: a comparison of computer-based methods.

Abstract: Objective. Accurate identification of surface electromyography (EMG) muscle onset is vital when examining short temporal parameters such as electromechanical delay. The visual method is considered the 'gold standard' in onset detection. Automatic detection methods are commonly employed to increase objectivity and reduce analysis time, but it is unclear if they are sensitive enough to accurately detect EMG onset when relating them to short-duration motor events.Approach. This study aimed to determine: (1) if automatic detection methods could be used interchangeably with visual methods in detecting EMG onsets (2) if the Teager-Kaiser energy operator (TKEO) as a conditioning step would improve the accuracy of popular EMG onset detection methods. The accuracy of three automatic onset detection methods: approximated generalized likelihood ratio (AGLR), TKEO, and threshold-based method were examined against the visual method. EMG signals from fast, explosive, and slow, ramped isometric plantarflexor contractions were evaluated using each technique.Main results. For fast, explosive contractions, the TKEO was the best-performing automatic detection method, with a low bias level (4.7 ± 5.6 ms) and excellent intraclass correlation coefficient (ICC) of 0.993, however with wide limits of agreement (LoA) (-6.2 to +15.7 ms). For slow, ramped contractions, the AGLR with TKEO conditioning was the best-performing automatic detection method with the smallest bias (11.3 ± 32.9 ms) and excellent ICC (0.983) but produced wide LoA (-53.2 to +75.8 ms). For visual detection, the inclusion of TKEO conditioning improved inter-rater and intra-rater reliability across contraction types compared with visual detection without TKEO conditioning.Significance. In conclusion, the examined automatic detection methods are not sensitive enough to be applied when relating EMG onset to a motor event of short duration. To attain the accuracy needed, visual detection is recommended. The inclusion of TKEO as a conditioning step before visual detection of EMG onsets is recommended to improve visual detection reliability.

Toward practical driving fatigue detection using three frontal EEG channels: a proof-of-concept study.

Abstract: Although various driving fatigue detection strategies have been introduced, the limited practicability is still an obstacle for the real application of these technologies. This study is based on the newly proposed non-hair-bearing (NHB) method to achieve practical driving fatigue detection with fewer channels from NHB areas and more efficient electroencephalogram (EEG) features. EEG data were recorded from 20 healthy subjects (15 males, age = 22.2 ± 3.2 years) in a 90-min simulated driving task using a remote wireless cap. Behaviorally, subjects demonstrated a salient fatigue effect as reflected by a monotonic increase of reaction time. Using a sliding window approach, we determined the vigilant and fatigued states at individual level to reduce the inter-subject differences in behavioral impairment and brain activity. Multiple EEG features, including power-spectrum density (PSD), functional connectivity (FC), and entropy, were estimated in a pair-wise manner, which were set as input for fatigue classification. Intriguingly, this data-driven approach showed that the best classification performance was achieved using three EEG channel pairs located in the NHB area. The mixed features of the frontal NHB area lead to the high within-subject detection rate of driving fatigue (92.7% ± 0.92%) with satisfactory generalizability for fatigue classification across different subjects (77.13% ± 0.85%). Moreover, we found the most prominent contributing features were PSD of different frequency bands within the frontal NHB area and FC within the frontal NHB area and between frontal and parietal areas. In sum, the current work provided objective evidence to support the effectiveness of the NHB method and further improved the performance, thereby moving a step forward towards practical driving fatigue detection in real-world scenarios.

A multiview feature fusion model for heartbeat classification.

Abstract: Objective.An electrocardiogram (ECG) is one of the most common means to diagnose arrhythmia according to different waveforms clinically. Although there are advanced classification methods such as deep learning, the single view feature cannot meet the demand of classification accuracy for new individuals. To this end, a classification model based on multiview fusion was proposed.Approach.First, handcrafted view features were extracted from heartbeats and then deep view features were obtained from the deep learning model. The features of two different perspectives were fused in the fully connected layer, and the random forest classifier was used instead of the Softmax classifier for classification. Notably, Bayesian optimization was utilized in the hyper-parameter tuning of the classifier. The proposed method employed the MIT-BIH database to classify five classes: normal heartbeat (N), left bundle branch block heartbeat (LB), right bundle branch block heartbeat (RB), atrial premature contraction (APC) and premature ventricular contraction (PVC).Main results.The experimental results achieved a higher average accuracy of 98.93%, average precision of 96.92%, average sensitivity of 96.46%, and average specificity of 99.33% in five types of heartbeat classification for inter-patient.Significance.The proposed framework improves the performance of ECG detection for new individuals. And it provides an feasible algorithmic model for single-lead wearable devices with multiview fusion.

Cuffless blood pressure estimation methods: physiological model parameters versus machine-learned features.

Abstract: Objective.For the first time in the literature, this paper investigates some crucial aspects of blood pressure (BP) monitoring using photoplethysmogram (PPG) and electrocardiogram (ECG). In general, the proposed approaches utilize two types of features: parameters extracted from physiological models or machine-learned features. To provide an overview of the different feature extraction methods, we assess the performance of these features and their combinations. We also explore the importance of the ECG waveform. Although ECG contains critical information, most models merely use it as a time reference. To take this one step further, we investigate the effect of its waveform on the performance.Approach.We extracted 27 commonly used physiological parameters in the literature. In addition, convolutional neural networks (CNNs) were deployed to define deep-learned representations. We applied the CNNs to extract two different feature sets from the PPG segments alone and alongside corresponding ECG segments. Then, the extracted feature vectors and their combinations were fed into various regression models to evaluate our hypotheses.Main results.We performed our evaluations using data collected from 200 subjects. The results were analyzed by the mean difference t-test and graphical methods. Our results confirm that the ECG waveform contains important information and helps us to improve accuracy. The comparison of the physiological parameters and machine-learned features also reveals the superiority of machine-learned representations. Moreover, our results highlight that the combination of these feature sets does not provide any additional information.Significance.We conclude that CNN feature extractors provide us with concise and precise representations of ECG and PPG for BP monitoring.

What recording duration is required to provide physiologically valid and reliable dynamic cerebral autoregulation transfer functional analysis estimates

Abstract: Objective. Currently, a recording of 300 s is recommended to obtain accurate dynamic cerebral autoregulation estimates using transfer function analysis (TFA). Therefore, this investigation sought to explore the concurrent validity and the within- and between-day reliability of TFA estimates derived from shorter recording durations from squat-stand maneuvers.Approach. Retrospective analyses were performed on 70 young, recreationally active or endurance-trained participants (17 females; age: 26 ± 5 years, [range: 20-39 years]; body mass index: 24 ± 3 kg m-2). Participants performed 300 s of squat-stands at frequencies of 0.05 and 0.10 Hz, where shorter recordings of 60, 120, 180, and 240 s were extracted. Continuous transcranial Doppler ultrasound recordings were taken within the middle and posterior cerebral arteries. Coherence, phase, gain, and normalized gain metrics were derived. Bland-Altman plots with 95% limits of agreement (LOA), repeated measures ANOVA's, two-tailed paired t-tests, coefficient of variation, Cronbach's alpha, intraclass correlation coefficients, and linear regressions were conducted.Main results. When examining the concurrent validity across different recording durations, group differences were noted within coherence (F(4155) > 11.6,p 0.611), gain (F(4155) 0.440), or normalized gain (F(4155) 0.359) parameters. The Bland-Altman 95% LOA measuring the concurrent validity, trended to narrow as recording duration increased (60 s: < ±0.4, 120 s: < ±0.3, 180 s < ±0.3, 240 s: < ±0.1). The validity of the 180 and 240 s recordings further increased when physiological covariates were included within regression models.Significance. Future studies examining autoregulation should seek to have participants perform 300 s of squat-stand maneuvers. However, valid and reliable TFA estimates can be drawn from 240 s or 180 s recordings if physiological covariates are controlled.

Clinical value of electrical impedance tomography (EIT) in the management of patients with acute respiratory failure: a single centre experience.

Abstract: Objective.We will describe our clinical experience using electrical impedance tomography (EIT) in the management of mechanical ventilation in patients with acute respiratory failure and to determine to which extent EIT-guided positive end-expiratory pressure (PEEP) setting differed from clinically set values.Approach.We conducted a retrospective, observational cohort study performed in a hub centre for the treatment of acute respiratory failure and veno-venous extracorporeal membrane oxygenation (ECMO).Main results.Between January 2017 and December 2019, EIT was performed 54 times in 41 patients, not feasible only in one case because of signal instability. More than 50% was on veno-venous ECMO support. In 16 cases (30%), EIT was used for monitoring mechanical ventilation, i.e. to evaluate recruitability or sigh setting. In 37 cases (70%), EIT was used to set PEEP both with incremental (11 cases in nine patients) and decremental (26 cases, 18 patients) PEEP trial. Clinical PEEP before the decremental PEEP trial (PEEPPRE) was 14.1 ± 3.4 cmH2O and clinical PEEP set by clinicians after the PEEP trial (PEEPPOST) was 13.6 ± 3.1 (p = ns). EIT analyses demonstrated that more hypoxic patients were higher derecruited when compared to less hypoxic patients that were, on the contrary, more overdistended (p < 0.05). No acute effects of PEEP adjustment based on EIT on respiratory mechanics or regional EIT parameters modification were observed.Significance.The variability of EIT findings in our population confirmed the need to provide ventilation settings individually tailored and EIT was confirmed to be an optimal useful clinical bedside noninvasive tool to provide real-time monitoring of the PEEP effect and ventilation distribution.

Agreement between wireless and standard measurements of vital signs in acute exacerbation of chronic obstructive pulmonary disease: a clinical validation study.

Abstract: Objective Wireless sensors for continuous monitoring of vital signs have potential to improve patient care by earlier detection of deterioration in general ward patients. We aimed to assess agreement between wireless and standard (wired) monitoring devices in patients hospitalized with acute exacerbation of chronic obstructive pulmonary disease (AECOPD). Approach Paired measurements of vital signs were recorded with 15 minutes intervals for two hours. The primary outcome was agreement between wireless and standard monitor measurements using the Bland and Altman method to calculate bias with 95% limits of agreement (LoA). We considered LoA of less than ±5 beats/min (bpm) acceptable for heart rate (HR), whereas agreement of peripheral oxygen saturation (SpO2), respiratory rate (RR), and blood pressure (BP) were acceptable if within ±3%-points, ±3 breaths/min (brpm), and ±10 mmHg, respectively. Main results 180 sample-pairs of vital signs from 20 with AECOPD patients were recorded for comparison. The wireless vs standard monitor bias was 0.03 (LoA -3.2 to 3.3) bpm for HR measurements, 1.4% (LoA -0.7 to 3.6%) for SpO2, -7.8 (LoA -22.3 to 6.8) mmHg for systolic BP and -6.2 (LoA -16.8 to 4.5) mmHg for diastolic BP. The wireless vs standard monitor bias for RR measurements was 0.75 (LoA -6.1 to 7.6) brpm. Significance Commercially available wireless monitors could accurately measure heart rate in patients admitted with AECOPD compared to standard wired monitoring. Agreement for SpO2 were borderline acceptable while agreement for RR and BP should be interpreted with caution.

Evaluation of the effectiveness of electrical muscle stimulation on human calf muscles via frequency difference electrical impedance tomography.

Abstract: Objectives. The human skeletal muscle responds immediately under electrical muscle stimulation (EMS), and there is an immediate physiological response in human skeletal muscle. Non-invasive quantitative analysis is at the heart of our understanding of the physiological significance of human muscle changes under EMS. Response muscle areas of human calf muscles under EMS have been detected by frequency difference electrical impedance tomography (fd-EIT).Approach. The experimental protocol consists of four parts: pre-training (pre), training (tra), post-training (post), and relaxation (relax) parts. The relaxation part has three relaxation conditions, which are massage relaxation (MR), cold pack relaxation (CR), and hot pack relaxation (HR).Main results. From the experimental results, conductivity distribution imagesσp(pmeans protocol = pre,tra,post,or relax) are clearly reconstructed byfd-EIT as response muscle areas, which are called theM1response area (composed of gastrocnemius muscle) and theM2response area (composed of the tibialis anterior muscle, extensor digitorum longus muscle, and peroneus longus muscle). A paired samplest-test was conducted to elucidate the statistical significance of spatial-mean conductivities 〈σp〉M1and 〈σp〉M2inM1andM2with reference to the conventional extracellular water ratioβpby bioelectrical impedance analysis. Significance. From thet-test results, 〈σp〉M1and〈σp〉M2have good correlation withβp. In the post-training part, 〈σpost〉 andβpostwere significantly higher than in the pre-training part (n = 24,p < 0.001). The relax-pre difference ratios of spatial-mean conductivity Δ〈σrelax-pre〉 and the relax-pre difference ratios of extracellular water ratio Δβrelax-prein both MR and CR were lower; on the contrary, the Δ〈σrelax-pre〉 and Δβrelax-prein HR were significantly higher than those in post-pre difference ratios of spatial-mean conductivity Δ〈σpost-pre〉 (n = 8,p < 0.05). The reason for the changes in 〈σp〉M1and 〈σp〉M2are caused by the changes in muscle extracellular volumes. In conclusion,fd-EIT satisfactorily evaluates the effectiveness of human calf muscles under EMS.

Imaging slow brain activity during neocortical and hippocampal epileptiform events with electrical impedance tomography.

Abstract: OBJECTIVE Electrical impedance tomography (EIT) is an imaging technique that produces tomographic images of internal impedance changes within an object using surface electrodes. It can be used to image the slow increase in cerebral tissue impedance that occurs over seconds during epileptic seizures, which is attributed to cell swelling due to disturbances in ion homeostasis following hypersynchronous neuronal firing and its associated metabolic demands. In this study, we characterised and imaged this slow impedance response during neocortical and hippocampal epileptiform events in the rat brain and evaluated its relationship to the underlying neural activity. APPROACH Neocortical or hippocampal seizures, comprising repeatable series of high-amplitude ictal spikes, were induced by electrically stimulating the sensorimotor cortex or perforant path of rats anaesthetised with fentanyl-isoflurane. Transfer impedances were measured during ≥30 consecutive seizures, by applying a sinusoidal current through independent electrode pairs on an epicortical array, and combined to generate an EIT image of slow activity. MAIN RESULTS The slow impedance responses were consistently time-matched to the end of seizures and EIT images of this activity were reconstructed reproducibly in all animals (p < 0.03125, N = 5). These displayed foci of activity that were spatially confined to the facial somatosensory cortex and dentate gyrus for neocortical and hippocampal seizures, respectively, and encompassed a larger volume as the seizure progressed. Centre-of-mass analysis of reconstructions revealed that this activity corresponded to the true location of the epileptogenic zone, as determined by EEG recordings and fast neural EIT measurements which were obtained simultaneously. SIGNIFICANCE These findings suggest that the slow impedance response presents a reliable marker of hypersynchronous neuronal activity during epileptic seizures and can thus be utilised for investigating the mechanisms of epileptogenesis in vivo and for aiding localisation of the epileptogenic zone during presurgical evaluation of patients with refractory epilepsies.

Regional lung function measures determined by electrical impedance tomography during repetitive ventilation manoeuvres in patients with COPD.

Abstract: Objective Current standards for conducting spirometry examinations recommend that the ventilation manoeuvres needed in pulmonary function testing are carried out repeatedly during sessions. Chest electrical impedance tomography (EIT) can determine the presence of ventilation heterogeneity during such manoeuvres, which increases the information content derived from such examinations. The aim of this study was to characterise regional lung function in patients with chronic obstructive pulmonary disease (COPD) during repetitive forced full ventilation manoeuvres. Regional lung function measures derived from these manoeuvres were compared with quiet tidal breathing. Approach Sixty hospitalised patients were examined during up to three repeated ventilation manoeuvres. Acceptable spirometry manoeuvres were performed and EIT recordings suitable for analysis obtained in 53 patients (12 women, 41 men; age: 68 ± 12 years (mean ± SD)). Pixel values of tidal volume, forced full inspiratory and expiratory volume in 1 s, and forced inspiratory and expiratory vital capacity were calculated from the EIT data. Spatial ventilation heterogeneity was assessed using the coefficient of variation, global inhomogeneity index, and centres and regional fractions of ventilation. Temporal inhomogeneity was determined by examining the pixel expiration times needed to exhale 50% and 75% of regional forced vital capacity. Main results All EIT-derived measures of regional lung function showed reproducible results during repetitive examinations. Parameters of spatial heterogeneity obtained from quiet tidal breathing were comparable with the measures derived from the forced manoeuvres. Significance Measures of spatial and temporal ventilation heterogeneity obtained in COPD patients by EIT provide comparable findings during repeated examinations within one testing session. Quiet tidal breathing generates similar information on ventilation heterogeneity as forced manoeuvres that demand a high amount of patient effort.

Next-generation ingestible devices: sensing, locomotion and navigation.

Abstract: There is significant interest in exploring the human body's internal activities and measuring important parameters to understand, treat and diagnose the digestive system environment and related diseases. Wireless capsule endoscopy (WCE) is widely used for gastrointestinal (GI) tract exploration due to its effectiveness as it provides no pain and is totally tolerated by the patient. Current ingestible sensing technology provides a valuable diagnostic tool to establish a platform for monitoring the physiological and biological activities inside the human body. It is also used for visualizing the GI tract to observe abnormalities by recording the internal cavity while moving. However, the capsule endoscopy is still passive, and there is no successful locomotion method to control its mobility through the whole GI tract. Drug delivery, localization of abnormalities, cost reduction and time consumption are improvements that can be gained from having active ingestible WCEs. In this article, the current technological developments of ingestible devices including sensing, locomotion and navigation are discussed and compared. The main features required to implement next-generation active WCEs are explored. The methods are evaluated in terms of the most important features such as safety, velocity, complexity of design, control, and power consumption.

Recording and quantifying fetal magnetocardiography signals using a flexible array of optically-pumped magnetometers.

Abstract: Objective Fetal magnetocardiography (fMCG) is a non-invasive biomagnetic technique that provides detailed beat-to-beat fetal heart rate analysis, both in normal rhythm as well as in fetal arrhythmias. New cryogenic-free sensors called optically pumped magnetometers (OPMs) have emerged as a less expensive and more geometrically flexible alternative to traditional Superconducting Quantum Interference Device (SQUID) technology for performing fMCG. The objective of the study was to show the ability of OPMs to record fMCG using flexible geometry while seeking to preserve signal quality, and to quantify fetal heart rate variability (FHRV). Approach Biomagnetic measurements were performed with OPMs in 24 healthy pregnant women with uncomplicated singleton pregnancies between 28 and 38 weeks gestation (GA). A total of 96 recordings were analyzed from OPM data that was collected using sensors placed in two different maternal configurations over the abdomen. The fMCG signals were extracted and the quality of the recordings were quantified by peak amplitudes and signal-to-noise ratio (SNR). R peaks were used to perform both time and frequency domain FHRV analysis. FHRV measures obtained from OPMs were compared descriptively to the same measures obtained from GA-matched existing SQUID data. Main results The fMCG derived from OPMs were observed in 21 of the 24 participants. Higher detection rates (85%) of fMCG signals were observed in the data sets recorded at GA >32 weeks. Peak amplitudes and SNR values were similar between two maternal configurations, but peak amplitudes were significantly higher (p = 0.013) in late GA compared to early GA. FHRV indicators were successfully extracted and their values overlapped substantially with those obtained from SQUID recordings. Significance Taking advantage of the geometric flexibility of the OPMs, we have demonstrated their ability to record and quantify fMCG in different maternal positions as opposed to rigid SQUID configurations.

Fetal ECG extraction using short time Fourier transform and generative adversarial networks.

Abstract: Objective.Fetal ECG (FECG) plays an important role in fetal monitoring. However, the abdominal ECG (AECG) recorded at the maternal abdomen is affected by various noises, making the extraction of FECG a challenging task. The main objective is to present a novel approach to FECG extraction using short time Fourier transform (STFT) and generative adversarial networks (GAN).Methods.Firstly, the AECG signals are transformed from one-dimensional (1D) time domain to two-dimensional (2D) time-frequency domain by using the STFT. Secondly, the 2D-STFT coefficients of FECG are estimated by the GAN model in the time-frequency domain. Finally, after the inverse STFT, the FECG can be reconstructed in the time domain.Main results.Experimental results on two databases demonstrate the effectiveness of the proposed method. Specifically, the SE, PPV andF1of the proposed method on PCDB are 92.37 ± 3.78%, 93.69 ± 3.96% and 93.02 ± 3.81%, respectively. And the SE, PPV andF1on ADFECGDB are 90.32 ± 10.70%, 89.79 ± 9.26% and 90.05 ± 9.81%, respectively.Significance.Unlike the previous studies based on the elimination of maternal ECG in the 1D time domain, the novelty of the proposed method relies on extracting the FECG directly from the AECG in the 2D time-frequency domain. It sheds some light to the topic of FECG extraction.

Obstructive sleep apnea prediction from electrocardiogram scalograms and spectrograms using convolutional neural networks.

Abstract: Objective.In this study, we conducted a comparative analysis of deep convolutional neural network (CNN) models in predicting obstructive sleep apnea (OSA) using electrocardiograms. Unlike other studies in the literature, this study automatically extracts time-frequency features by using CNNs instead of manual feature extraction from ECG recordings.Approach.The proposed model generates scalogram and spectrogram representations by transforming preprocessed 30 s ECG segments from time domain to the frequency domain using continuous wavelet transform and short time Fourier transform, respectively. We examined AlexNet, GoogleNet and ResNet18 models in predicting OSA events. The effect of transfer learning on success is also investigated. Based on the observed results, we proposed a new model that is found more effective in estimation. In total, 152 ECG recordings were included in the study for training and evaluation of the models.Main results.The prediction using scalograms immediately 30 s before potential OSA onsets gave the best performance with 82.30% accuracy, 83.22% sensitivity, 82.27% specificity and 82.95% positive predictive value. The prediction using spectrograms also achieved up to 80.13% accuracy and 81.99% sensitivity on prediction. Per-recording classification suggested considerable results with 91.93% accuracy for prediction of OSA events.Significance.Time-frequency deep features of scalograms and spectrograms of ECG segments prior to OSA events provided reliable information about the possible events in the future. The proposed CNN model can be used as a good indicator to accurately predict OSA events using ECG recordings.

AECG-DecompNet: abdominal ECG signal decomposition through deep-learning model.

Abstract: Objective.The accurate decomposition of a mother's abdominal electrocardiogram (AECG) to extract the fetal ECG (FECG) is a primary step in evaluating the fetus's health. However, the AECG is often affected by different noises and interferences, such as the maternal ECG (MECG), making it hard to evaluate the FECG signal. In this paper, we propose a deep-learning-based framework, namely 'AECG-DecompNet', to efficiently extract both MECG and FECG from a single-channel abdominal electrode recording.Approach.AECG-DecompNet is based on two series networks to decompose AECG, one for MECG estimation and the other to eliminate interference and noise. Both networks are based on an encoder-decoder architecture with internal and external skip connections to reconstruct the signals better.Main results.Experimental results show that the proposed framework performs much better than utilizing one network for direct FECG extraction. In addition, the comparison of the proposed framework with popular single-channel extraction techniques shows superior results in terms of QRS detection while indicating its ability to preserve morphological information. AECG-DecompNet achieves exceptional accuracy in theprecisionmetric (97.4%), higher accuracy inrecallandF1metrics (93.52% and 95.42% respectively), and outperforms other state-of-the-art approaches.Significance.The proposed method shows a notable performance in preserving the morphological information when the FECG within the AECG signal is weak.

Adaptive threshold split Bregman algorithm based on magnetic induction tomography for brain injury monitoring imaging.

Abstract: Objective. Traditional magnetic induction tomography (MIT) algorithms have problems in reconstruction, such as large area error (AE), blurred boundaries of reconstructed targets, and considerable image noise (IN). As the size and boundary of a lesion greatly affect the treatment plan, more accurate algorithms are necessary to meet clinical needs.Approach. In this study, adaptive threshold split Bregman (ATSB) is proposed for brain injury monitoring imaging in MIT. We established a 3D brain MIT simulation model with the actual anatomical structure and a phantom model and obtained the reconstructed images of single targets in different positions and multiple targets, using the Tikhonov, eigenvalue threshold regularisation (ETR), split Bregman (SB), and ATSB algorithms.Main results. Compared with the Tikhonov and ETR algorithms, the ATSB algorithm reduced the AE by 95% and the IN by 17% in a simulation and reduced the AE by 87% and IN by 6% in phantom experiments. Compared with the SB algorithm, the ATSB algorithm can reduce the difficulty of adjusting parameters and is easier to use in clinical practice. The simulation and phantom experiments results showed that the ATSB algorithm could reconstruct the target size more accurately and could distinguish multiple targets more effectively than the other three algorithms.Significance. The ATSB algorithm could improve the image quality of MIT and better meet the needs of clinical applications and is expected to promote brain injury monitoring imaging via MIT.

A novel method for continuous blood pressure estimation based on a single-channel photoplethysmogram signal.

Abstract: OBJECTIVE Currently, continuous blood pressure (BP) measurements are mostly based on multi-sensor combinations and datasets with limited BP ranges. Besides, most BP-related features derive from the photoplethysmogram (PPG) signal. The mechanism of PPG signal formation is not considered. We aimed to design a noninvasive and continuous method for estimation of BP using a single PPG sensor, which takes the mechanism of PPG signal formation into account. APPROACH We prepared a dataset containing PPG signals for 294 patients from three public databases for constructing the BP estimation model. The features used in the model consisted of two types: novel features based on a multi-Gaussian model and existing features. The multi-Gaussian model fitted the different components (i.e. the main wave, the dicrotic wave and the tidal wave) of the PPG signal. Ensemble machine learning algorithms were applied to estimate systolic blood pressure (SBP) and diastolic blood pressure (DBP). When partitioning the dataset, there was an overlap between the training set and the testing set. MAIN RESULTS Datasets with a wide-range of SBP and DBP values (SBP ranging from 74 to 229 mmHg and DBP ranging from 26 to 141 mmHg) were used to evaluate our method. The mean and standard deviation of error for SBP and DBP estimations were -0.21 ± 5.21 mmHg and -0.19 ± 3.37 mmHg, respectively. The model performance fully met the Association for the Advancement of Medical Instrumentation standard and was grade 'A' on the British Hypertension Society standard. SIGNIFICANCE The multi-Gaussian model could be used to estimate BP, and our method was able to track a wide range of BP accurately. In addition our method is based on a single PPG sensor, making it very convenient.

Regions of interest selection and thermal imaging data analysis in sports and exercise science: a narrative review.

Abstract: OBJECTIVE Infrared thermography (IRT) is a non-invasive, contactless and low-cost technology that allows recording of the radiating energy that is released from a body, providing an estimate of its superficial temperature. Thanks to the improvement of infrared thermal detectors, this technique is widely used in the biomedical field to monitor the skin temperature for different purposes (e.g. assessing circulatory diseases, psychophysiological state, affective computing). Particularly, in sports and exercise science, thermography is extensively used to assess sports performance, to investigate superficial vascular changes induced by physical exercise, and to monitor injuries. However, the methods of analysis employed to treat IRT data are not standardized, and hence introduce variability in the results. APPROACH This review focuses on the methods of analysis currently used for thermal imaging in sports and exercise science. MAIN RESULTS Firstly, the procedures employed for the selection of regions of interest (ROIs) from anatomical body districts are reviewed, paying attention also to the potentialities of morphing algorithms to increase the reproducibility of thermal results. Secondly, the statistical approaches utilized to characterize the temperature frequency and spatial distributions within ROIs are investigated, showing their strengths and weaknesses. Moreover, the importance of employing tracking methods to analyze the temporal thermal oscillations within ROIs is discussed. Thirdly, the capability of employing procedures of investigation based on machine learning frameworks on thermal imaging in sports science is examined. SIGNIFICANCE Finally, some proposals to improve the standardization and the reproducibility of IRT data analysis are provided, in order to facilitate the development of a common database of thermal images and to improve the effectiveness of IRT in sports science.

Feature extraction of EEG signals based on functional data analysis and its application to recognition of driver fatigue state

Abstract: Objective Our objective is to study how to obtain features which can reflect the continuity and internal dynamic changes of electroencephalography (EEG) signals and study an effective method for fatigued driving state recognition based on the obtained features. Approach A method of EEG signalfeature extraction based on functional data analysis is proposed. Combined with kernel principal component analysis method, the obtained features are applied to the recognition of driver fatigue state, and a corresponding recognition model of fatigued driving state is constructed. Main results The recognition model is tested on the real collected driver fatigue EEG signals by selecting a suitable classifier. The test results show that the proposed driver fatigue state recognition method has good recognition effect, especially on the classifier based on decision tree, with an average accuracy of 99.50%. Significance The extracted features well reflect the continuityand internal dynamic changes of the EEG signals, and it is of great significance and application value to study an effective method of fatigued driver state recognition based on the features.