Figure 1 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Experimental procedure according to two physical conditions, three body postures, and six movement speeds. The experiment lasted for a total of 63 min and sufficient rest was given between each task.
Source publication
Cardiac activity has been monitored continuously in daily life by virtue of advanced medical instruments with microelectromechanical system (MEMS) technology. Seismocardiography (SCG) has been considered to be free from the burden of measurement for cardiac activity, but it has been limited in its application in daily life. The most important issue...
Citations
... Methods to estimate HR from BCG or SCG include signal processing [8], [12] and data-driven approaches [8], [13]- [16]. Data-driven approaches are more favorable as they can be adapted and/or specialized to different datasets or sensing modalities. ...
... Some of the more recent data-driven approaches are deep learning (DL)-based. One common variant [13], [15] of these methods first removes high frequency oscillations through convolutional filters intended to retain cardiac pulsation. Next, dense layers are used to regress HR. ...
... However, we notice that these dense regressors are prone to overfitting and do not generalize well to unseen subjects in a preliminary analysis in our dataset; dense layers may contain myriad parameters, increasing the amount of data required to ensure generalizability. Unfortunately, except for [15], some of the aforementioned DL-based methods [13], [14] do not split the training and testing sets by subjects, violating the identically and independently distributed assumption [17] and misinforming the generalizability of the model. Therefore, it remains uncertain whether dense layer-based DL models are generalizable to unseen subjects' data. ...
Estimating heart rate (HR) from the
seismocardiogram (SCG) signal can provide a more unobtrusive
alternative for long-term HR monitoring where the gold standard
electrocardiogram (ECG) signal is less favorable. Deep learning
(DL) approaches have demonstrated promise in accomplishing
this and are attractive due to their flexible and data-driven nature.
However, current dense layer-based DL approaches lack a
carefully designed regressor to estimate the HR in the SCG input
and may overfit in low-data regime. It is also uncertain how well
most of these DL approaches can generalize to unseen subjects, as
evaluation has primarily taken place with no separation of
subjects between training and testing data. In this work, we
address these limitations by designing a DL approach for SCG HR
estimation that leverages our proposed Dominant Frequency
Regressor (DFR) with a Fast Fourier Transform layer and test the
model performance using leave-one-subject-out cross validation.
Specifically, we measure ECG and SCG from 19 subjects using our
custom-built wearable patch. Here, ECG is used for training and
regularization only, and SCG is used for training and inference.
These signals are bandpass filtered, segmented into 60s windows,
and normalized. Next, the proposed DFR-based DL model is
applied and regularized by domain adversarial training. We
report a mean absolute error (MAE) of 1.42±1.66 beats per minute
(bpm) for HR with a range of 63-104 bpm. The proposed method
can augment existing methods or be adapted for other problems
of similar nature owning to its superiority to the dense layersbased
alternatives. This work can lead to accurate, real-time
algorithms for estimating 60s HR from a single-chest worn
accelerometer only, which could be embedded in textiles.
... This model obtained 92.7% test accuracy for ImageNet [34], winning it the ImageNet 2014 competition. To date, VGG16 is still considered an excellent vision model architecture, and has successfully been used in many real-world applications [10,35,36]. ...
With rapid advancements in in-vehicle network (IVN) technology, the demand for multiple advanced functions and networking in electric vehicles (EVs) has recently increased. To enable various intelligent functions, the electrical system of existing vehicles incorporates a controller area network (CAN) bus system that enables communication among electrical control units (ECUs). In practice, traditional network-based intrusion detection systems (NIDSs) cannot easily identify threats to the CAN bus system. Therefore, it is necessary to develop a new type of NIDS—namely, on-the-move Intrusion Detection System (OMIDS)—to categorise these threats. Accordingly, this paper proposes an intrusion detection model for IVNs, based on the VGG16 classifier deep learning model, to learn attack behaviour characteristics and classify threats. The experimental dataset was provided by the Hacking and Countermeasure Research Lab (HCRL) to validate classification performance for denial of service (DoS), fuzzy attacks, spoofing gear, and RPM in vehicle communications. The proposed classifier’s performance was compared with that of the XBoost ensemble learning scheme to identify threats from in-vehicle networks. In particular, the test cases can detect anomalies in terms of accuracy, precision, recall, and F1-score to ensure detection accuracy and identify false alarm threats. The experimental results show that the classification accuracy of the dataset for HCRL Car-Hacking by the VGG16 and XBoost classifiers (n = 50) reached 97.8241% and 99.9995% for the 5-subcategory classification results on the testing data, respectively.
... Although the use of machine learning in seismocardiography has been reported in several studies [23]- [28], the use of semantic segmentation for heartbeat detection in seismocardiograms with deep neural networks was mentioned only in [28] to the best of our knowledge. ...
Heartbeat detection is an essential part of the
analysis of cardiac signals because it is recognized as a
representative measurement of cardiac function. The gold
standard of heartbeat detection is detecting the QRS com-
plexes in electrocardiograms. Due to the development of sen-
sors and information and communication technologies (ICT),
seismocardiography (SCG), becomes a viable alternative to
electrocardiography for heart rate monitoring. In this work,
we propose a system for detecting the heartbeat based on
seismocardiograms using deep learning methods. The study
was conducted on a publicly available data set (CEBS) which
contains simultaneous measurements of ECG, breathing signal,
and seismocardiograms. Our approach to heartbeat detection in
seismocardiograms uses a model based on convolutional neural
network that is based on ResNet and contains a squeeze-and-
excitation unit. The created model scored state-of-the-art results
(Jaccard and F1 score above 97%) on the test dataset, which
proves its high reliability.
... Using the extreme learning machine, the error rate is reduced significantly, from 45.5% to 12.5%, when compared to the dispersion-maximum method by Choe and Cho (2017). Lee and Whang (2018) use an 11-layer CNN for heart rate estimation from body movements recorded with a wearable six-axis accelerometer. In an evaluation on measurements from 30 patients in 12 different scenarios, the CNN dramatically outperformed the group's traditional signal processing approach. ...
... The reviews highlight the great potential of deep learning for biosignal processing. However, in a number of publications (such as Lee and Whang (2018) (2018)). In this work, we are therefore also investigating the extent to which increasing the model size can produce convincing performance gains. ...
... Although FCNs are more prevalent in image segmentation applications (Springenberg et al 2014, Shelhamer et al 2017), the same concept can also be applied to regression or classification models. Many works described above use convolutional and pooling layers followed by a sequence of dense layers (Kiranyaz et al 2016, Acharya et al 2017, Lu et al 2018, Lee and Whang 2018. In contrast, FCNs use convolutional layers with larger strides and/or without padding to reduce the input dimensions. ...
Objective:
Ballistocardiography (BCG) is an unobtrusive approach for cost-effective and patient-friendly health monitoring. In this work, deep learning methods are used for heart rate estimation from BCG signals and are compared against five digital signal processing methods found in literature.
Approach:
The models are evaluated on a dataset featuring BCG recordings from 42 patients, acquired with a pneumatic system. Several different deep learning architectures, including convolutional, recurrent and a combination of both are investigated. Besides model performance, we are also concerned about model size and specifically investigate less complex and smaller networks.
Main results:
Deep learning models outperform traditional methods by a large margin. Across 14 patients in a held-out testing set, an architecture with stacked convolutional and recurrent layers achieves an average mean absolute error (MAE) of 2.07 beat/min, whereas the best-performing traditional method reaches 4.24 beat/min. Besides smaller errors, deep learning models show more consistent performance across different patients, indicating the ability to better deal with inter-patient variability, a prevalent issue in BCG analysis. In addition, we develop a smaller version of the best-performing architecture, that only features 8283 parameters, yet still achieves an average MAE of 2.32 beat/min on the testing set.
Significance:
This is the first study that applies and compares different deep learning architectures to heart rate estimation from bed-based BCG signals. Compared to signal processing algorithms, deep learning models show dramatically smaller errors and more consistent results across different individuals. The results show that using smaller models instead of excessively large ones can lead to sufficient performance for specific biosignal processing applications. Additionally, we investigate the use of fully convolutional networks for 1D signal processing, which is rarely applied in literature.
... Thereafter, the sensor fusion basketball shooting posture dataset was finally formed. 3 Journal of Sensors VGG16 deep learning model showed excellent performance in image classification, and the VGG16 model with onedimensional convolution kernels had been used to classify the one-dimensional data obtained by using the accelerometer and gyroscope [26], in this study, we used the onedimensional convolution kernels VGG16 deep learning model to classify sensor fusion basketball shooting postures. ...
In recent years, with the development of wearable sensor devices, research on sports monitoring using inertial measurement units has received increasing attention; however, a specific system for identifying various basketball shooting postures does not exist thus far. In this study, we designed a sensor fusion basketball shooting posture recognition system based on convolutional neural networks. The system, using the sensor fusion framework, collected the basketball shooting posture data of the players' main force hand and main force foot for sensor fusion and used a deep learning model based on convolutional neural networks for recognition. We collected 12,177 sensor fusion basketball shooting posture data entries of 13 Chinese adult male subjects aged 18-40 years and with at least 2 years of basketball experience without professional training. We then trained and tested the shooting posture data using the classic visual geometry group network 16 deep learning model. The intratest achieved a 98.6% average recall rate, 98.6% average precision rate, and 98.6% accuracy rate. The intertest achieved an average recall rate of 89.8%, an average precision rate of 91.1%, and an accuracy rate of 89.9%.
... The basic principle of the convolutional neural network [12] [13]: Input the recognized target image, convolution operation to obtain the feature image; ...
... In 2018 Vega-Martínez et al. published a review article on the heart rate measurements mentions the use of gyrocardiography as a method for monitoring the heart rate which may be used to perform HRV analysis [62]. Standalone heart beat detection methods published in 2018 include the methods proposed by Lee et al. [37,38], Hernandez and Cretu [57] and Kaisti et al. [58] with complementary data set "Mechanocardiograms with ECG Reference" published on IEEE DataPort [63]. ...
Gyrocardiography (GCG) is a non-invasive technique of analyzing cardiac vibrations by a MEMS (microelectromechanical system) gyroscope placed on a chest wall. Although its history is short in comparison with seismocardiography (SCG) and electrocardiography (ECG), GCG becomes a technique which may provide additional insight into the mechanical aspects of the cardiac cycle. In this review, we describe the summary of the history, definition, measurements, waveform description and applications of gyrocardiography. The review was conducted on about 55 works analyzed between November 2016 and September 2020. The aim of this literature review was to summarize the current state of knowledge in gyrocardiography, especially the definition, waveform description, the physiological and physical sources of the signal and its applications. Based on the analyzed works, we present the definition of GCG as a technique for registration and analysis of rotational component of local cardiac vibrations, waveform annotation, several applications of the gyrocardiography, including, heart rate estimation, heart rate variability analysis, hemodynamics analysis, and classification of various cardiac diseases.
... As SCG and GCG are mutually orthogonal, they are inherently susceptible to different noise characteristics, which enables a deeper analysis of cardiac activity [14]. GCG offers the potential for novel insights into cardiac fiducial points [7,31], higher fidelity for certain types of motion artifact [31][32][33][34], and can assist in SCG beat detection using kinetic energy envelopes [34,35]. ...
... As SCG and GCG are mutually orthogonal, they are inherently susceptible to different noise characteristics, which enables a deeper analysis of cardiac activity [14]. GCG offers the potential for novel insights into cardiac fiducial points [7,31], higher fidelity for certain types of motion artifact [31][32][33][34], and can assist in SCG beat detection using kinetic energy envelopes [34,35]. ...
Gyrocardiography (GCG) is a non-invasive technique of analyzing cardiac vibrations by a MEMS (micro electromechanical) gyroscope placed on a chest wall. Although its history is short in comparison with seismocardiography (SCG) and electrocardiography (ECG), GCG becomes a technique which may provide additional insight into the mechanical aspects of the cardiac cycle. In this review, we describe the summary of the history, definition, measurements, waveform description and applications of gyrocardiography. The review was conducted on about 55 works analyzed between November 2016 and September 2020. The analyzed works demonstrate the definition of GCG, waveform annotation, several applications of the gyrocardiography, including, heart rate estimation, heart rate variability analysis, hemodynamics analysis, classification of various cardiac diseases.
... W sejsmokardiogramach wpływ wysiłku fizycznego jest widoczny jako wzrost amplitudy i mocy średniokwadratowej sygnału, co jest skorelowane ze wzrostem objętości wyrzutowej [90]. Jednym z ograniczeń w stosowaniu sejsmokardiografii jest wrażliwość na artefakty ruchowe oraz ich wpływ na morfologię sygnału [75,119]. Yang i wsp. ...
... gdzie S 1 oraz S 2 są analizowanymi wielkościami. W kontekście sejsmokardiogramów i żyrokardiogramów wykresy Blanda-Altmana były wykorzystywane do opisu jakości detekcji uderzeń serca [115,119,197], analizy okresu przedwyrzutowego [225] oraz analizy HRV [115,117,197]. Rysunek 4.7: Wykresy korelacji indeksów analizy czasowej HRV obliczonych na oknach o stałej szerokości dla elektrokardiogramów i sejsmokardiogramów. ...
Choroby układu krążenia są główną przyczyną zgonów w Polsce, Europie i na świecie. Wczesna diagnoza oraz rozpoczęcie leczenia pozwala zmniejszyć liczbę zgonów oraz wydłużyć życie. Często stosowanym badaniem stanu układu krążenia jest monitorowanie tętna. Złotym standardem monitorowania tętna jest ciągła rejestracja elektrokardiogramu (EKG), jednakże prowadzone są badania nad alternatywnymi metodami, które nie wymagają użycia elektrod, takich jak sejsmo-kardiografia (SKG) i żyrokardiografia (ŻKG). Sejsmokardiografia jest analiząwibracji związanych z pracą serca za pomocą akcelerometru, natomiast żyro-kardiografia polega na rejestracji zmian prędkości kątowej żyroskopem.Głównym celem pracy jest opracowanie algorytmów detekcji uderzeń serca w sygnałach EKG, SKG i ŻKG i przeprowadzenie badań nad użytecznością sejsmokardiografii i żyrokardiografii w analizie zmienności rytmu serca (HRV), a główna teza badawcza pracy brzmi: wyniki analizy zmienności rytmu serca na podstawie sejsmokardiogramów i żyrokardiogramów nie różnią się istotnie od wyników uzyskanych na podstawie elektrokardiogramów i można ustalić czas otwarcia zastawki aortalnej. W pracy wykorzystano 35 zarejestrowanych elektrokardiogramów, sejsmokardiogramów i żyrokardiogramów, wśród których jest 29 sygnałów z bazy „Mechanocardiograms with ECG reference” dostępnej w IEEE DataPort oraz6 sygnałów zarejestrowanych opisanym w pracy rejestratorem. Rejestrator składa się z inercyjnego układu pomiarowego LSM9DS1, moduł do rejestracji elektro-kardiogramu AD8232 oraz Arduino Micro odpowiedzialnego za pobieranie próbek sygnałów i przekazywanie do komputera.Na każdym sygnale wykonano niezależnie detekcję tętna na odprowadzeniu II elektrokardiogramu, osi Z sejsmokardiogramu oraz osi Y żyrokardiogramuoraz obliczono odstępy między kolejnymi uderzeniami serca dla elektrokardiogramów, sejsmokardiogramów i żyrokardiogramów i przeprowadzono analizy HRVw dwóch wariantach: na sygnale HRV podzielonym na nakładające się okna o stałej szerokości oraz na całych sygnałach HRV. Długość odstępów między kolejnymi uderzeniami serca w trzech analizowanych sygnałach (EKG, SKG, ŻKG) nie różni się znacząco, jednakże wpływa na wyniki analizy HRV. Pomimo to obliczone wartości średnie i odchylenia standardowego, współczynniki korelacji liniowej, błędy bezwzględne oraz wy-kresy Blanda-Altmana wskazują na wysoką zgodność wskaźników HRV obliczonych na elektrokardiogramach i sejsmokardiogramach oraz elektrokardiogramach i żyrokardiogramach. Dla pierwszego wariantu analizy HRV zauważono nieco silniejszą korelację i większą zgodność współczynników HRV obliczonychdla EKG, SKG oraz ŻKG niż dla drugiego wariantu.Efektem pracy jest opracowanie rejestratora EKG, SKG i ŻKG opartego na Arduino oraz potwierdzenie faktu, że można wykonać analizę HRV na sejsmokardiogramach i żyrokardiogramach, a jej wyniki nie będą istotnie różne od wyników uzyskanych na elektrokardiogramach.
... As the two measurement methods are mutually orthogonal, they are inherently susceptible to different noise characteristics, which enables a deeper analysis when combining the information from both. GCG offers the potential for novel insights into cardiac fiducial points [22,28], higher fidelity for certain types of motion artifact [29][30][31], and can assist in SCG beat detection using kinetic energy envelopes [30,32]. In this paper, we have modified our previously developed autocorrelated differential algorithm (ADA) [33] to process GCG in parallel with SCG and thereby exploit the content-rich information in VCG signals. ...
Cardiography is an indispensable element of health care. However, the accessibility of at-home cardiac monitoring is limited by device complexity, accuracy, and cost. We have developed a real-time algorithm for heart rate monitoring and beat detection implemented in a custom-built, affordable system. These measurements were processed from seismocardiography (SCG) and gyrocardiography (GCG) signals recorded at the sternum, with concurrent electrocardiography (ECG) used as a reference. Our system demonstrated the feasibility of non-invasive electro-mechanical cardiac monitoring on supine, stationary subjects at a cost of $100, and with the SCG–GCG and ECG algorithms decoupled as standalone measurements. Testing was performed on 25 subjects in the supine position when relaxed, and when recovering from physical exercise, to record 23,984 cardiac cycles at heart rates in the range of 36–140 bpm. The correlation between the two measurements had r2 coefficients of 0.9783 and 0.9982 for normal (averaged) and instantaneous (beat identification) heart rates, respectively. At a sampling frequency of 250 Hz, the average computational time required was 0.088 s per measurement cycle, indicating the maximum refresh rate. A combined SCG and GCG measurement was found to improve accuracy due to fundamentally different noise rejection criteria in the mutually orthogonal signals. The speed, accuracy, and simplicity of our system validated its potential as a real-time, non-invasive, and affordable solution for outpatient cardiac monitoring in situations with negligible motion artifact.
