Performance Comparison of Gyrocardiogram and Seismocardiogram

Signals in Valvular Heart Disease Assessment

Ecem Erin

and Beren Semiz

2 a

Department of Physics, Bogazici University, Istanbul, Turkey

Department of Electrical and Electronics Engineering, Koc University, Istanbul, Turkey

Keywords:

Valvular Heart Disease, Seismocardiogram, Gyrocardiogram, Cardiovascular Health Monitoring, Biomedical

Signal Processing.

Abstract:

Cardiovascular diseases have been identiﬁed as one of the leading causes of mortality worldwide. Among

these diseases, valvular heart diseases (VHDs) have a greater impact on the population. The existing methods

for VHD assessment are expensive and only applicable within clinical environments. Hence, there is a need for

accessible and cost-efﬁcient systems to provide continuous VHD monitoring. As stenosis and regurgitation

are characterized by the change in the blood ﬂow patterns, it was hypothesized that angular acceleration

(gyrocardiogram, GCG) could capture the differences in blood ﬂow and changes in cardiovascular parameters

better than linear acceleration (seismocardiogram, SCG). In this work, a publicly available dataset including

36 patients with stenosis and 44 patients with regurgitation was used. The SCG and GCG signals were ﬁrst

divided into 10- second long segments. From each segment, ﬁve features were extracted from all axes and

used to train the SCG- and GCG-based XGBoost models. Overall, the GCG-based model resulted in better

performance in distinguishing between the stenosis and regurgitation cases: the precision, recall and accuracy

values were 94.7, 94.5, and 94.5 for the SCG, and 96.0, 95.9 and 95.9 for the GCG, respectively. Predictive

performances of SCG and GCG models on the cardiovascular parameters were also investigated and resulted

in (SCG, GCG) mean absolute percent errors of (19.4, 20.6), (15.5, 14.5), (12.0, 13.1) for ejection fraction,

left ventricular end diastolic dimension and left ventricle posterior wall thickness, respectively. These results

showed that in addition to SCG, GCG could also be used for VHD evaluation and potentially be employed in

continuous monitoring systems.

1 INTRODUCTION

According to the 2020 report by the World Health Or-

ganization (WHO), cardiovascular diseases have been

identiﬁed as one of the leading causes of mortality

worldwide (WHO, 2020). Among cardiovascular dis-

eases, valvular heart diseases (VHDs) have a greater

impact on the population and result in higher mortal-

ity rates (Go et al., 2013). In the heart, four main

heart valves are present: tricuspid valve, aortic valve,

mitral valve, and pulmonary valve (Klabunde, 2011;

Svensson, 2008). VHDs primarily emerge due to the

impairments in these valves. These impairments can

be group under two main categories: stenosis and re-

gurgitation, which can affect any of the aforemen-

tioned valves. In the case of stenosis, the valve open-

ing narrows, leading to inadequate blood outﬂow. On

the other hand, regurgitation refers to the valve’s fail-

https://orcid.org/0000-0002-7544-5974

ure to prevent the backward ﬂow of blood (Svensson,

2008).

While echocardiography, magnetic resonance

imaging, cardiac catheterization, and computed to-

mography can be used to monitor VHDs, these meth-

ods are costly and only available in clinical settings

(Svensson, 2008). Hence, there is a need for accessi-

ble and cost-efﬁcient systems to provide non-invasive

and continuous VHD monitoring. Recent advance-

ments in wearable sensor research have paved the way

to collect physiological signals from the body non-

invasively. As these signals directly originate from

the underlying anatomy, they can provide valuable in-

formation regarding the current physiological status

of the subject. Among these physiological signals,

seismocardiogram (SCG) has been widely leveraged

in wearable system design. SCG originates from the

contraction of the heart and corresponds to the result-

ing micro chest vibrations (Inan et al., 2014). Mostly,

786

Erin, E. and Semiz, B.

Performance Comparison of Gyrocardiogram and Seismocardiogram Signals in Valvular Heart Disease Assessment.

DOI: 10.5220/0012441700003657

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 1, pages 786-792

ISBN: 978-989-758-688-0; ISSN: 2184-4305

tri-axial accelerometers are used to collect the SCG

signal in three directions: lateral, head-to-foot and

dorso-ventral. On the other hand, recent studies have

shown that utilizing gyroscope-based vibrations could

offer enhanced capabilities in detecting heart activ-

ity and blood movement compared to accelerometers

(Yang et al., 2017; Yang and Tavassolian, 2017; Ja-

fari Tadi et al., 2017; Shandhi et al., 2019). However,

the performance of gyrocardiogram (GCG) in VHD

assessment is still an open research question.

From physics perspective, accelerometers mea-

sure linear acceleration; whereas gyroscope can cap-

ture dynamic angular velocity in three axes (yaw,

pitch, and roll) (Faisal et al., 2019). Considering that

the stenosis and regurgitation impairments exhibit dif-

ferent blood ﬂow characteristics, it was hypothesized

that GCG could capture the differences in blood ﬂow

and changes in cardiovascular parameters better than

SCG. The contributions of this study can thus be listed

as follows: for the ﬁrst time to the best of our knowl-

edge, (i) the performance of GCG in VHD assess-

ment was studied and compared with the SCG perfor-

mance, (ii) contributions of the spectral features and

individual SCG and GCG axes (X, Y, and Z) in VHD

assessment were investigated, (iii) predictive perfor-

mances of SCG and GCG models on the cardiovascu-

lar parameters (ejection fraction, left ventricular end

diastolic dimension and left ventricle posterior wall

thickness) under stenosis and regurgitation conditions

were investigated.

2 METHODS

2.1 Dataset Description

In this study, An Open-Access Database for the Eval-

uation of Cardio-Mechanical Signals From Patients

With Valvular Heart Disease, which includes seismo-

cardiogram (SCG), gyrocardiogram (GCG) and elec-

trocardiogram (ECG) signals, was used (Yang et al.,

2021). The participants were identiﬁed as having dif-

ferent forms of valvular heart diseases (VHD), includ-

ing mitral valve stenosis (MS), aortic stenosis (AS),

aortic valve regurgitation (AR), mitral valve regurgi-

tation (MR), and tricuspid valve regurgitation (TR).

Each speciﬁc VHD was indicated by a label whether

if the disease is present.

The SCG, ECG and GCG signals were col-

lected using commercially available Shimmer system

(Shimmer 3 ECG module, Shimmer Sensing, United

Kingdom) while the subject was in supine position.

For the analysis, only the tri-axial SCG and GCG sig-

nals were used. The X, Y, and Z directions of the SCG

and GCG signals were representing the vibrations in

the lateral, head-to-foot, and dorso-ventral directions,

respectively.

Figure 1: Study Overview.

2.2 Pre-Processing and Data

Preparation

As the goal was to investigate the performance of

SCG and GCG in stenosis and regurgitation assess-

ment, two different sub-datasets were generated: one

having only SCG signals, the other one including only

the GCG signals. In both sub-datasets, there were 36

patients with stenosis and 44 patients with regurgita-

tion present either of the valves. It should be noted

that these signals were simultaneously acquired from

the same subjects and the classiﬁcation and regression

tasks were employed on the same samples. Thus the

data was comparable. However, some GCG signals

were shorter in duration.

In the datasets, some recordings had a sampling

rate of 256 Hz while the others had a rate of 512 Hz.

To ensure standardization, all sampling rates were set

to 256 Hz. Each signal was then divided into 10-

second long segments to have higher number of in-

stances for model training. For the GCG dataset, this

resulted in 1070 samples for stenosis and 2017 sam-

ples for regurgitation; whereas for the SCG dataset,

there were 1521 samples for stenosis, and 2017 for

regurgitation. No other pre-processing step was ap-

plied on the signals.

Performance Comparison of Gyrocardiogram and Seismocardiogram Signals in Valvular Heart Disease Assessment

787

2.3 Feature Extraction

Considering that the stenosis and regurgitation im-

pairments exhibit different blood ﬂow characteristics,

it was hypothesized that there would be differences in

the information encoded in the spectral content of the

acquired signals. In addition, previous studies showed

that VHD assessment can highly beneﬁt from spec-

tral analysis of the SCG signals through spectrogram,

wavelet, chromagram and tempogram features (Erin

and Semiz, 2023). In this work, a similar approach

was inherited, however instead of using a high num-

ber of features (and to prevent the possibility of over-

ﬁtting), only ﬁve features (one from the time domain,

four from the spectral domain) were extracted from

each of the 10-second-long X, Y, Z segments.

• Entropy: Entropy quantiﬁes the abrupt energy

changes in the time domain of the signal. When

the signal exhibits sudden changes, it is expected

to observe relatively reduced entropy at the onset

of these changes (Hersek et al., 2017).

• Spectral Entropy: The interpretation behind

spectral entropy is similar to the energy entropy.

However computation is carried out in the fre-

quency domain. The resulting value represents

how complex the spectrum is, i.e., the larger the

complexity, the higher the spectral entropy value

(Hersek et al., 2017).

• Spectral Rolloff: It measures the frequency at

which a speciﬁc portion of the signal’s energy is

concentrated below. Based on the literature, this

ratio is usually chosen as 90% (Giannakopoulos

and Pikrakis, 2014).

• Spectral Centroid: It computes the center of

mass of the spectrum. If the signal mostly in-

cludes high frequencies, centroid is expected to be

relatively higher (Giannakopoulos and Pikrakis,

2014).

• Spectral Spread: Spread corresponds to the dis-

tribution of the frequencies around the centroid.

If the frequencies are tightly gathered around the

center frequency, a lower spectral spread value is

expected (Giannakopoulos and Pikrakis, 2014).

As there were 5 different features extracted from

each of the 10-second-long segments in the X, Y, Z di-

rections, the dataframe was consisting of 15 features,

i.e. columns. Segments corresponding to stenosis and

regurgitation were labeled as 0 and 1, respectively.

2.4 Model Training and Validation

2.4.1 Stenosis and Regurgitation Classiﬁcation

Model Selection and Training: In the ﬁrst task, the

aim was to compare the performances of the GCG-

and SCG-based stenosis and regurgitation classiﬁca-

tion models. As the classiﬁcation model, extreme gra-

dient boosting trees (XGBoost) was chosen. Rather

than relying on a single estimator, XGBoost involves

employing multiple estimators concurrently. During

training, multiple decision trees are trained in an iter-

ative manner, enabling the prediction and reﬁnement

of residual errors from the preceding iteration as the

training progresses (Dietterich et al., 2002).

For GCG- and SCG-based classiﬁcation, two dif-

ferent XGBoost models were trained. Before split-

ting each dataframe into train and test sets, the fea-

tures were scaled using standard scaler. Following

that, the datasets were split into training (80%) and

testing (20%) portions. As the dataset had some in-

balance regarding the number of samples available in

each group, leave-one-out methodology could not be

used. During training, the objective function was de-

termined as binary:logistic. On the other hand, the de-

fault values were used for the remaining parameters.

The performance of the model was assessed using ac-

curacy, precision, recall, and f1-score metrics. These

equations are presented in Equations 1, 2, 3 and 4, re-

spectively (TP: true positives, FP: false positives, TN:

true negatives and FN: false negatives). In addition,

the area under the receiver operating characteristics

curve (ROC AUC) was computed.

Accuracy =

T P + T N

T P + T N + FP + FN

(1)

Precision =

T P

T P + FP

(2)

Recall =

T P

T P + FN

(3)

score = 2 ∗

precision ∗ recall

precision + recall

(4)

Investigating the Performance of the Individual

Axes: In addition to investigating the performance

of SCG and GCG signals in distinguishing between

stenosis and regurgitation cases, the performances of

the individual GCG and SCG axes (X, Y and Z) were

assessed. To that aim, similar feature extraction and

model training steps were implemented. Performance

assessment was again employed through accuracy,

precision, recall, f1-score metrics.

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

788

Investigating the Importance of the Spectral Fea-

tures: To investigate which features contribute the

most in regurgitation and stenosis classiﬁcation, XG-

Boost feature importance ranking was leveraged. In

this approach, for each split point, the signiﬁcance of

each attribute in improving the performance measure

is computed. The importance scores are then aver-

aged across all decision trees available in the XG-

Boost model. The resulting scores are used to rank

the features, which represent the importance of the

features. Following that, the minimum number of

features sufﬁcient was determined to be used in the

regression tasks and the redundant features were ex-

cluded from the feature set.

2.4.2 Investigating the Cardiovascular

Parameter Prediction Performances

After determining the top ﬁve most important features

in classiﬁcation of regurgitation and stenosis, those

features were used to predict the ejection fraction, left

ventricular end diastolic dimension and left ventric-

ular posterior wall thickness values, which were in-

cluded in the referred database. The reason why only

top ﬁve features were used was that adding additional

features was not increasing the model performances

any further. Below, the deﬁnition of each cardiovas-

cular parameter and their relationship with the steno-

sis and regurgitation conditions are detailed.

• Ejection fraction: It represents how well the heart

pumps blood and is deﬁned as the ratio of the

stroke volume and end diastolic left ventricular

volume. In case of stenosis and regurgitation, the

primary compensatory mechanism required to up-

hold a normal effective stroke volume is an eleva-

tion in left ventricular end-diastolic volume. An

increase in end-diastolic volume, results in a de-

crease in ejection fraction (Maurer, 2006; Cham-

bers, 2006).

• Left Ventricular Posterior Wall Thickness

(LVPW): Speciﬁcally under aortic stenosis con-

dition, an increase in wall thickness typically

occurs to compensate the elevated intracavitary

pressure (Chambers, 2006; Mehrotra et al., 2015;

Borow et al., 1985).

• Left Ventricular End Diastolic Dimension

(LVEDD): Similar to LVPW, an increase is

observed in LVEDD in regurgitation (Maurer,

2006), however studies found no signiﬁcant rela-

tionship between LVEDD and stenosis (Mehrotra

et al., 2015; Borow et al., 1985).

Using the top ﬁve features, separate SCG- and

GCG-based XGBoost regression models were trained

Table 1: Performance Metrics for SCG and GCG.

Signal Precision Recall Accuracy AUC

SCG 94.7 94.5 94.5 0.99

GCG 96.0 95.9 95.9 0.99

Table 2: Performance Metrics for the individual GCG Axes

in Classiﬁcation Task.

Axis Precision Recall Accuracy f1-score

X 87.5 87.5 87.5 87.5

Y 87.3 87.4 87.4 87.4

Z 72.3 73.0 73.0 72.1

for each cardiovascular parameter. First, similar to the

previous task, the SCG and GCG datasets were split

into training (80%) and testing (20%) portions. Dur-

ing training, the objective function was determined as

reg:squarederror and the remaining parameters were

kept as the default values. The performance of the

models was assessed using the root mean squared

error (RMSE) and mean absolute percentage error

(MAPE) metrics. The corresponding equations are

presented in Equations 5 and 6, respectively ( ˆy

: pre-

dicted, y

: actual).

RMSE(y, ˆy) =

∑

N−1

i=0

− ˆy

)

(5)

MAPE(y, ˆy) =

100%

N−1

∑

i=0

− ˆy

. (6)

3 RESULTS AND DISCUSSION

3.1 Stenosis and Regurgitation

Classiﬁcation

Classiﬁcation Performance: To compare the perfor-

mance of SCG and GCG in stenosis and regurgitation

assessment, two different XGBoost models were

trained. The performance results and confusion

matrices for both models are presented in Table 1

and Fig. 2. As expected, the GCG-based model

resulted in slightly better performance compared

to the SCG-based model. Overall, the precision,

recall and accuracy values were 94.7, 94.5, and 94.5

for the SCG, and 96.0, 95.9 and 95.9 for the GCG,

respectively. On the other hand, both models resulted

in an ROC AUC of 0.99.

Assessment of Axes Contributions: Additionally,

the performance of the individual GCG and SCG

axes in the classiﬁcation task was investigated. The

Performance Comparison of Gyrocardiogram and Seismocardiogram Signals in Valvular Heart Disease Assessment

789

Figure 2: Confusion matrices for the classiﬁcation models.

Table 3: Performance Metrics for the individual SCG Axes

in Classiﬁcation Task.

Axis Precision Recall Accuracy f1-score

X 86.9 86.9 87.2 86.8

Y 86.7 86.7 86.9 86.7

Z 81.6 81.6 81.7 81.6

performance results are presented in Tables 2 and 3

for the GCG- and SCG-based models, respectively.

The accuracy values of the models trained with

the GCG signals were 87.5 and 87.4 for the X

(lateral) and Y (head-to-foot) axes, whereas for the

SCG-based models, these values were 87.2 and

86.9, respectively. These ﬁndings were indeed in

parallel with the previous study (Shandhi et al.,

2019). The model trained with the Z (dorso-ventral)

axis of the GCG signals resulted in a relatively lower

performance compared to the one trained with the

SCG signals, with accuracy and f1-score being 73.0

and 72.1, respectively. On the other hand, the one

with the SCG signals resulted in an accuracy score

of 81.7 and f1 score of 81.6, which is in parallel

with the results presented in (Erin and Semiz, 2023).

These ﬁndings could be attributed to the fact that the

ﬂow characteristics of blood can be captured better in

the lateral and head-to-foot axes by the GCG as they

represent the ejection direction and path; whereas the

beating of the heart results in a linear acceleration

in the dorso-ventral direction, hence could be better

captured by the SCG.

Importance Ranking of Spectral Features: In ad-

dition to assessing the performance of the individual

axes, importance ranking of the features was also in-

vestigated. The top ﬁve features for both GCG-based

model and SCG-based model are presented in Fig. 3.

Indeed, both signal types resulted in the same top ﬁve

features. The spectral centroid in the lateral and head-

to-foot directions appeared as the top two features,

showing that the center frequency for blood ﬂow has

a distinguishable effect in stenosis and regurgitation

classiﬁcation. In addition, entropy, which is a time

domain feature, did not appear among the top fea-

tures. Another important observation was that fea-

tures from the dorso-ventral axis have relatively lower

importance compared to the features from the lateral

and head-to-foot directions.

Figure 3: Feature importance scores for the SCG and GCG

models.

Table 4: Performance Metrics for SCG Regression Task.

SCG Ejection Fraction

LVEDD

LVPW

RMSE 11.5 10.6 1.71

MAPE 19.4 15.5 12.0

Table 5: Performance Metrics for GCG Regression Task.

GCG Ejection Fraction

LVEDD

LVPW

RMSE 11.9 10.0 1.85

MAPE 20.6 14.5 13.1

3.2 Cardiovascular Parameter

Prediction Results

Using the top ﬁve features, separate SCG- and GCG-

based XGBoost regression models were trained for

each of ejection fraction (EF), left ventricular pos-

terior wall thickness (LVPW) and left ventricular

end diastolic dimension (LVEDD). The RMSE and

MAPE values for the SCG- and GCG-based models

are presented in Table 4 and 5, respectively. The re-

sults show that the SCG and GCG have comparable

performance in the estimation of all three parameters.

While the (RMSE, MAPE) values were (11.5, 19.4),

(10.6, 15.5), (1.71, 12.0) for the SCG-based models,

these values were (11.9, 20.6), (10.0, 14.5), (1.85,

13.1) for the GCG-based models for ejection fraction,

LVEDD and LVPW, respectively. SCG-based models

performed slightly better than the GCG ones in the

estimation of ejection fraction and LVPW, whereas

LVEDD estimation resulted in relatively lower error

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

790

when GCG was used. Overall, it can be deduced that

both SCG and GCG can potentially be leveraged in

the estimation of cardiovascular parameters.

4 CONCLUSION

In this work, the performances of the SCG- and GCG-

based models in stenosis and regurgitation assessment

were investigated. Additionally, the predictive per-

formances of SCG- and GCG-based models on the

cardiovascular parameters (ejection fraction, left ven-

tricular end diastolic dimension and left ventricle pos-

terior wall thickness) under stenosis and regurgita-

tion conditions were studied. Overall, it was found

that the GCG-based model performs slight better than

the SCG-based model in distinguishing between the

stenosis and regurgitation cases, most probably as

the GCG could capture the angular characteristics of

the blood ﬂow better than the SCG. Additionally, the

best performing axes were found to be the lateral and

head-to-foot axes.

For the regression tasks, the SCG and GCG had

comparable performance in the estimation of ejec-

tion fraction, left ventricular posterior wall thickness

and left ventricular end diastolic dimension. Models

based on SCG demonstrated slightly higher perfor-

mance compared to those based on GCG in estimat-

ing ejection fraction and LVPW. On the other hand,

the estimation of LVEDD showed a relatively lower

error when GCG-based model was used. In conclu-

sion, it can be inferred that both SCG and GCG can

potentially be used in estimating various cardiovascu-

lar parameters.

Future work will focus on improving the current

pipelines further to enable real-time monitoring of

VHDs and validating these pipelines in larger datasets

to ensure generalizability.

REFERENCES

Borow, K. M., Colan, S., and Neumann, A. (1985). Al-

tered left ventricular mechanics in patients with valvu-

lar aortic stenosis and coarction of the aorta: effects on

systolic performance and late outcome. Circulation,

72(3):515–522.

Chambers, J. (2006). The left ventricle in aortic steno-

sis: evidence for the use of ace inhibitors. Heart,

92(3):420–423.

Dietterich, T. G. et al. (2002). Ensemble learning.

The handbook of brain theory and neural networks,

2(1):110–125.

Erin, E. and Semiz, B. (2023). Spectral analysis of cardio-

genic vibrations to distinguish between valvular heart

diseases.

Faisal, I. A., Purboyo, T. W., and Ansori, A. S. R. (2019). A

review of accelerometer sensor and gyroscope sensor

in imu sensors on motion capture. J. Eng. Appl. Sci,

15(3):826–829.

Giannakopoulos, T. and Pikrakis, A. (2014). Introduction

to audio analysis: a MATLAB® approach. Academic

Press.

Go, A. S., Mozaffarian, D., Roger, V. L., Benjamin, E. J.,

Berry, J. D., Borden, W. B., Bravata, D. M., Dai, S.,

Ford, E. S., Fox, C. S., et al. (2013). Heart disease

and stroke statistics—2013 update: a report from the

american heart association. Circulation, 127(1):e6–

e245.

Hersek, S., Pouyan, M. B., Teague, C. N., Sawka, M. N.,

Millard-Stafford, M. L., Kogler, G. F., Wolkoff, P.,

and Inan, O. T. (2017). Acoustical emission analysis

by unsupervised graph mining: A novel biomarker of

knee health status. IEEE Transactions on Biomedical

Engineering, 65(6):1291–1300.

Inan, O. T., Migeotte, P.-F., Park, K.-S., Etemadi, M.,

Tavakolian, K., Casanella, R., Zanetti, J., Tank, J.,

Funtova, I., Prisk, G. K., et al. (2014). Ballistocardio-

graphy and seismocardiography: A review of recent

advances. IEEE journal of biomedical and health in-

formatics, 19(4):1414–1427.

Jafari Tadi, M., Lehtonen, E., Saraste, A., Tuominen, J.,

Koskinen, J., Ter

as, M., Airaksinen, J., P

ank

a, M.,

and Koivisto, T. (2017). Gyrocardiography: A new

non-invasive monitoring method for the assessment

of cardiac mechanics and the estimation of hemody-

namic variables. Scientiﬁc reports, 7(1):1–11.

Klabunde, R. (2011). Cardiovascular physiology concepts.

Lippincott Williams & Wilkins.

Maurer, G. (2006). Aortic regurgitation. Heart, 92(7):994–

1000.

Mehrotra, P., Flynn, A. W., Jansen, K., Tan, T. C., Mak, G.,

Julien, H. M., Zeng, X., Picard, M. H., Passeri, J. J.,

and Hung, J. (2015). Differential left ventricular out-

ﬂow tract remodeling and dynamics in aortic stenosis.

Journal of the American Society of Echocardiography,

28(11):1259–1266.

Shandhi, M. M. H., Semiz, B., Hersek, S., Goller, N.,

Ayazi, F., and Inan, O. T. (2019). Performance

analysis of gyroscope and accelerometer sensors for

seismocardiography-based wearable pre-ejection pe-

riod estimation. IEEE journal of biomedical and

health informatics, 23(6):2365–2374.

Svensson, L. (2008). Aortic valve stenosis and regurgita-

tion: an overview of management. Journal of Cardio-

vascular Surgery, 49(2):297.

WHO (2020). The top 10 causes of death. Geneva: World

Health Organization. https://www.who.int/news- r

oom/fact-sheets/detail/the-top-10-causes-of-death

(visited: 2022-09).

Yang, C., Fan, F., Aranoff, N., Green, P., Li, Y., Liu, C., and

Tavassolian, N. (2021). An open-access database for

the evaluation of cardio-mechanical signals from pa-

Performance Comparison of Gyrocardiogram and Seismocardiogram Signals in Valvular Heart Disease Assessment

791

tients with valvular heart diseases. Frontiers in Phys-

iology, 12.

Yang, C., Tang, S., and Tavassolian, N. (2017). Utilizing

gyroscopes towards the automatic annotation of seis-

mocardiograms. IEEE Sensors Journal, 17(7):2129–

2136.

Yang, C. and Tavassolian, N. (2017). Combined seismo-and

gyro-cardiography: A more comprehensive evaluation

of heart-induced chest vibrations. IEEE journal of

biomedical and health informatics, 22(5):1466–1475.

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

792