Figure - available from: Frontiers in Cardiovascular Medicine
This content is subject to copyright.
Data collection setup: (i) a chest-mounted single channel wireless ECG monitor collecting ECG and inertial (movement) measurements, and (ii) the mCardia mobile application for collection of patient-reported data (28).
Source publication
ECG is a non-invasive tool for arrhythmia detection. In recent years, wearable ECG-based ambulatory arrhythmia monitoring has gained increasing attention. However, arrhythmia detection algorithms trained on existing public arrhythmia databases show higher FPR when applied to such ambulatory ECG recordings. It is primarily because the existing publi...
Citations
... Multimodal ECG datasets are often publicly available, which allows researchers and developers to benchmark their algorithms against existing methods and collaborate with other experts in the field. In this research, we utilized datasets such as the China Physiological Signal Challenge 2018 (CPSC2018), the St. Petersburg INCART 12-lead Arrhythmia Database, the Georgia 12-lead ECG Challenge Database (CinC2020), the PhysioNet Computing in Cardiology Challenge 2017 (CinC2017), and the Contextual Arrhythmia Database (CACHET-CADB) [30] as shown in Fig. 2. The above databases have two data formats: header files (.hea) and mat files (.mat). The mat file consists of patient ECG signals. ...
In healthcare, the persistent challenge of arrhythmias, a leading cause of global mortality, has sparked extensive research into the automation of detection using machine learning (ML) algorithms. However, traditional ML and AutoML approaches have revealed their limitations, notably regarding feature generalization and automation efficiency. This glaring research gap has motivated the development of AutoRhythmAI, an innovative solution that integrates both machine and deep learning to revolutionize the diagnosis of arrhythmias. Our approach encompasses two distinct pipelines tailored for binary-class and multi-class arrhythmia detection, effectively bridging the gap between data preprocessing and model selection. To validate our system, we have rigorously tested AutoRhythmAI using a multimodal dataset, surpassing the accuracy achieved using a single dataset and underscoring the robustness of our methodology. In the first pipeline, we employ signal filtering and ML algorithms for preprocessing, followed by data balancing and split for training. The second pipeline is dedicated to feature extraction and classification, utilizing deep learning models. Notably, we introduce the ‘RRI-convoluted transformer model’ as a novel addition for binary-class arrhythmias. An ensemble-based approach then amalgamates all models, considering their respective weights, resulting in an optimal model pipeline. In our study, the VGGRes Model achieved impressive results in multi-class arrhythmia detection, with an accuracy of 97.39% and firm performance in precision (82.13%), recall (31.91%), and F1-score (82.61%). In the binary-class task, the proposed model achieved an outstanding accuracy of 96.60%. These results highlight the effectiveness of our approach in improving arrhythmia detection, with notably high accuracy and well-balanced performance metrics.
... Within the discussion, further focus will set on the results of the MAD metrics, because, to the best of our knowledge, there is no validation and calibration study using the MAI metric so far. Nevertheless, this metric is also important as it uses a bandpass filter that ensures that accelerations that do not come from physical movements tend to be filtered out and is more and more used in studies [74,75], so it was decided to determine the cut-off points for both MAD and MAI metrics. In comparison to Aittasalo et al. [70] who used the MAD metric for a hip worn accelerometer in children aged 13-15 years, our cut-off points differ especially in Cut 1 (SB-LPA: 52.9 vs. 26.9 ...
Background
To assess physical activity (PA) there is a need of objective, valid and reliable measurement methods like accelerometers. Before these devices can be used for research, they need to be calibrated and validated for specific age groups as the locomotion differs between children and adults, for instance. Therefore, the aim of the present study was the calibration and validation of the Move4 accelerometer for children aged 8–13 years.
Methods
53 normal weighted children (52% boys, 48%girls) aged 8–13 years (mean age = 10.69 ± 1.46, mean BMI = 17.93 kg/m− 2, 60th percentile), wore the Move4 sensor at four different body positions (thigh, hip, wrist and the Move4ecg including heart rate measurement at the chest). They completed nine activities that considered the four activity levels (sedentary behavior (SB), light PA (LPA), moderate PA (MPA) and vigorous PA (VPA)) within a test-retest design. Intensity values were determined using the mean amplitude deviation (MAD) as well as the movement acceleration intensity (MAI) metrics. Determination of activities and energy expenditure was validated using heart rate. After that, cut-off points were determined in Matlab by using the Classification and Regression Trees (CART) method. The agreement for the cut-off points between T1 and T2 was analyzed.
Results
MAD and MAI accelerometer values were lowest when children were lying on the floor and highest when running or doing jumping jacks. The mean correlation coefficient between acceleration values and heart rate was 0.595 (p = 0.01) for MAD metric and 0.611 (p = 0.01) for MAI metric, indicating strong correlations. Further, the MAD cut-off points for SB-LPA are 52.9 mg (hip), 62.4 mg (thigh), 86.4 mg (wrist) and 45.9 mg (chest), for LPA-MPA they are 173.3 mg (hip), 260.7 mg (thigh), 194.4 mg (wrist) and 155.7 mg (chest) and for MPA-VPA the cut-off points are 543.6 mg (hip), 674.5 mg (thigh), 623.4 mg (wrist) and 545.5 mg (chest). Test-retest comparison indicated good values (mean differences = 9.8%).
Conclusion
This is the first study investigating cut-off points for children for four different sensor positions using raw accelerometer metrics (MAD/MAI). Sensitivity and specificity revealed good values for all positions. Nevertheless, depending on the sensor position, metric values differ according to the different involvement of the body in various activities. Thus, the sensor position should be carefully chosen depending on the research question of the study.
... A further breakdown of the number of each of these rhythm types present is listed in Table II. To evaluate the model's generalisability on a completely unseen dataset, recorded using a different ECG device and demographics, we also use the publicly available dataset CACHET-CADB [46]. It consists of 1602 10-second singlechannel ECG samples and 3-channel ACC data, sampled at 1024Hz and 64Hz, respectively. ...
... It consists of 1602 10-second singlechannel ECG samples and 3-channel ACC data, sampled at 1024Hz and 64Hz, respectively. The dataset comes from 24 subjects and includes 4 different ECG rhythm classes, with their breakdown being 747 AF, 615 NSR, 221 Noise, and 19 episodes of "Other" rhythm types [46]. ...
p>Atrial fibrillation (AF) is a common cardiac arrhythmia causing severe complications if left untreated. Due to its sporadic nature, early detection often requires longitudinal ambulatory electrocardiogram (ECG) screening. Recently, deep learning (DL) has gained prominence in analysing long-term ECG and automating AF detection. However, like any medical classification problem, obtaining diverse labelled ECG data for DL model training is expensive and time-consuming. This paper proposes a semi-supervised learning (SSL) based AF classification model employing a variational auto-encoder (VAE). It leverages varying amounts of labelled and unlabelled ECG data to optimise the AF detection performance on ambulatory ECG. As ambulatory contexts under free-living conditions influence ECG recordings, we incorporate context via accelerometry data and experiment with its influence on model performance. Our proposed SSL model was trained on ECG data from 72,003 unique patients and can classify between sinus rhythms, AF, and other arrhythmias. Experimental results on our unseen test dataset and the publicly available CACHET-CADB dataset clearly demonstrate the model’s generalisability, achieving an accuracy of over 91% with just 20% of the training set being labelled. With extensive experiments, our study exhibits the ability of SSL to improve AF detection from ambulatory ECG using small amounts of labelled data.</p
... A further breakdown of the number of each of these rhythm types present is listed in Table II. To evaluate the model's generalisability on a completely unseen dataset, recorded using a different ECG device and demographics, we also use the publicly available dataset CACHET-CADB [46]. It consists of 1602 10-second singlechannel ECG samples and 3-channel ACC data, sampled at 1024Hz and 64Hz, respectively. ...
... It consists of 1602 10-second singlechannel ECG samples and 3-channel ACC data, sampled at 1024Hz and 64Hz, respectively. The dataset comes from 24 subjects and includes 4 different ECG rhythm classes, with their breakdown being 747 AF, 615 NSR, 221 Noise, and 19 episodes of "Other" rhythm types [46]. ...
p>Atrial fibrillation (AF) is a common cardiac arrhythmia causing severe complications if left untreated. Due to its sporadic nature, early detection often requires longitudinal ambulatory electrocardiogram (ECG) screening. Recently, deep learning (DL) has gained prominence in analysing long-term ECG and automating AF detection. However, like any medical classification problem, obtaining diverse labelled ECG data for DL model training is expensive and time-consuming. This paper proposes a semi-supervised learning (SSL) based AF classification model employing a variational auto-encoder (VAE). It leverages varying amounts of labelled and unlabelled ECG data to optimise the AF detection performance on ambulatory ECG. As ambulatory contexts under free-living conditions influence ECG recordings, we incorporate context via accelerometry data and experiment with its influence on model performance. Our proposed SSL model was trained on ECG data from 72,003 unique patients and can classify between sinus rhythms, AF, and other arrhythmias. Experimental results on our unseen test dataset and the publicly available CACHET-CADB dataset clearly demonstrate the model’s generalisability, achieving an accuracy of over 91% with just 20% of the training set being labelled. With extensive experiments, our study exhibits the ability of SSL to improve AF detection from ambulatory ECG using small amounts of labelled data.</p
... The fit of wearable device contacts is also a common problem. For this reason, many studies have expanded the relevant algorithms and databases so that testing devices can be adapted to self-testing using devices such as wearables to improve signal-to-noise ratios outside the hospital, exclude motion artifacts [87,88], and more accurately determine the occurrence of phenomena such as arrhythmias [89]. different electrical signals and form regular ECG curves [85]. ...
... The fit of wearable device contacts is also a common problem. For this reason, many studies have expanded the relevant algorithms and databases so that testing devices can be adapted to self-testing using devices such as wearables to improve signal-to-noise ratios outside the hospital, exclude motion artifacts [87,88], and more accurately determine the occurrence of phenomena such as arrhythmias [89]. Wearable ECG allows the detection of sleep apnea. ...
Sleep is an essential physiological activity, accounting for about one-third of our lives, which significantly impacts our memory, mood, health, and children’s growth. Especially after the COVID-19 epidemic, sleep health issues have attracted more attention. In recent years, with the development of wearable electronic devices, there have been more and more studies, products, or solutions related to sleep monitoring. Many mature technologies, such as polysomnography, have been applied to clinical practice. However, it is urgent to develop wearable or non-contacting electronic devices suitable for household continuous sleep monitoring. This paper first introduces the basic knowledge of sleep and the significance of sleep monitoring. Then, according to the types of physiological signals monitored, this paper describes the research progress of bioelectrical signals, biomechanical signals, and biochemical signals used for sleep monitoring. However, it is not ideal to monitor the sleep quality for the whole night based on only one signal. Therefore, this paper reviews the research on multi-signal monitoring and introduces systematic sleep monitoring schemes. Finally, a conclusion and discussion of sleep monitoring are presented to propose potential future directions and prospects for sleep monitoring.
