Estimation of Fluid Intake Volume from Surface Electromyography

Signals: A Comparative Study of Seven Regression Techniques

Iman A. Ismail

and Ernest N. Kamavuako

Department of Engineering, King’s College London, London WC2R 2LS, U.K.

Keywords:

Dehydration, Electromyography Sensors, Fluid Intake, Volume Quantifying, Swallowing.

Abstract:

Insufﬁcient ﬂuid intake in older adults, in particular, is a worrying problem and an actual concern that warrants

scrutiny. Monitoring ﬂuid intake is essential to avoid dehydration and overhydration problems. This paper

presents an investigation to estimate the ﬂuid intake volume using surface Electromyographic (sEMG) sensors.

Eleven subjects participated in the experiment, and sEMG recordings of swallows from cups, bottles, and

straws were collected. Four features were extracted from the EMG signals. Seven regression algorithms were

implemented for quantifying the volume of swallowed ﬂuid: Random Forest (RF), Support Vector Regressor,

K-nearest neighbour (KNN), Linear Regressor (LR), Decision Tree (DT), Lasso and Ridge. The mean sip

volume across subjects was 14.85 ± 5.05 ml. Results showed that using Random Forest, the root mean

square (RMSE) for estimating ﬂuid intake volume using one the Mean Absolute Value feature gave 1.37 ±

1.1 ml. These results indicate a step forward in estimating ﬂuid intake volume based on sEMG for hydration

monitoring.

1 INTRODUCTION

The terms ”hydration” and ”healthy” can be used in-

terchangeably because water is necessary for every

organ, cell, and tissue in the body to function nor-

mally. In other words, being hydrated is essential to

good health because it aids in processes like lubricat-

ing joints, avoiding infections, feeding cells nutrients,

and preserving the general health of the body’s or-

gans. Yet, older adults experience considerable hy-

dration concerns since their bodies contain 10 − 15%

less water. This can be a signiﬁcant contributing fac-

tor to the majority of health problems that older adults

experience. Studies show that most older individuals

are more susceptible to renal issues and electrolyte

abnormalities due to medications that lead to dehy-

dration, making them more susceptible to changes in

conditions and illnesses (El-Sharkawy, 2021). De-

mentia, Alzheimer’s, diabetes, and poor mobility are

just a few of the health problems that older adults may

experience that cause their ability to feel thirst de-

creases, making them less conscious of their body’s

need for water. Dehydration is an issue for many older

persons as a result. It also increases the risk of death

in seniors relative to the general population and is one

https://orcid.org/ 0000-0001-9846-6674

https://orcid.org/ 0000-0001-6846-2090

of the most common reasons for hospital admissions.

Dehydration occurs when the human body uses or

loses more ﬂuid and minerals, such as sodium and

potassium than it takes in. The body cannot perform

everyday tasks due to the lack of water and other liq-

uids. Thus, if the body cannot replace the lost ﬂuids,

it will become dehydrated, which is very dangerous.

Therefore, monitoring the quantity of intaken ﬂuid is

essential to decrease the risk of dehydration. Fluid

charts are one of the vital clinical methods used to

monitor patients’ ﬂuid intake and output throughout

the day in hospitals and care facilities where nurses

stay with the patients to try to keep an eye on their

consumption of meals and liquids. The medical team

uses the data to make later clinical decisions, such

as whether to perform surgery or prescribe medica-

tion. It is crucial to fully complete the ﬂuid balancing

charts to detect changes in the ﬂuid’s input or output.

Any ﬂuid intake should be precisely measured and its

type recorded on a ﬂuid balance chart. That technique

can be used to estimate and record any output ﬂuid,

including urine, loose stools, and vomiting. These

ﬂuid charts may only be used to monitor ﬂuid levels;

however, they are not always accurate because doctors

or nurses sometimes fail to note a patient’s input or

output (Malvuccio and Kamavuako 2021). According

to the study, only 25% of the ﬂuid charts at Kettering

General Hospital had precise measurements, and only

118

Ismail, I. and Kamavuako, E.

Estimation of Fluid Intake Volume from Surface Electromyography Signals: A Comparative Study of Seven Regression Techniques.

DOI: 10.5220/0011795600003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 4: BIOSIGNALS, pages 118-124

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

14% had thorough records of all intakes and outputs

(Asogan, 2021).

Many technologies use machine learning and

other techniques to monitor ﬂuid intake in older per-

sons, according to a survey of literature reviews on

methods of monitoring dehydration. One of these

technologies is wearable technology, including ac-

celerometers, inertial sensors, smartwatches, cell-

phones, acoustic sensors, and psychological sensors.

These items are widely available on the commercial

market and have helped detect drinking activities (Co-

hen, 2021). Still, wearables cannot reliably quan-

tify consumption volume, despite research showing

that they can detect drinking events with an accuracy

of ¿ 80%. Furthermore, some senior citizens dislike

these devices and do not wish to wear them (Wellnitz,

2019). For textile techniques to be useful in daily

life, they must be connected to the clothing and se-

curely laundered in the washing machine. Respiratory

Inductance Plethysmography (RIP), for example, has

produced positive results for swallowing detection but

hasn’t calculated the volume of ﬂuid consumed. Ac-

cordingly, none of these methods has been used to

quantify the volume of ﬂuid intake in the clinic, de-

spite the encouraging results of these methods for de-

tecting swallowing and drinking events (Dong, 2017,

Cheng, 2010 - Tatulli, 2020).

Another approach to measure ﬂuid intake is sur-

faces with embedded sensors. These surfaces require

the users to lift the containers used for drinking and

place them on the surface every time they drink to de-

termine the drinking actions and to record the amount

of the drink. Any additional object placed on the

surface will give inaccurate information, leading to

erroneous detections. Further, some vision and en-

vironmental approaches, like wearable cameras and

radar, have concentrated on intake detection. Still,

the detection accuracy depends on the camera resolu-

tion and the surrounding environment (lighting, pro-

cessing power, and data storage), most of which have

not operated in real-time. Although these techniques

can recognise drinking events with a 90% accuracy

using deep learning techniques, they cannot calculate

the volume of ﬂuid consumed (Cohen, 2021).

The use of smart containers paired with Inertial

Measurements Units (IMU) placed outside the bot-

tle to estimate the sip volume according to the event

orientation and duration is another effective method

that has been used to quantify ﬂuid intake. Though,

the usability of these containers remains low. Numer-

ous research has examined ultrasonic sensors to cal-

culate the volume of ﬂuid being absorbed using the

container, but they have not evaluated how accurate

the sensors are. Nevertheless, these techniques have

not found their way to the clinics routinely.

Methods based on physiological signals to moni-

tor ﬂuid intake include sEMG and acoustic sensors,

such as microphones. Nakafuji et al. (2015) ob-

tained a classiﬁcation accuracy of 84% using micro-

phones to record swallowing noises to discriminate

between discrete ﬂuid volumes (Nakafuji, 2014). Us-

ing the frequency and amplitude characteristics of the

recorded signals, Kobayashi et al. (2014) captured

swallowing sound using a throat microphone to ac-

curately detect ﬂuid intake with 95% accuracy us-

ing the cross-validation of SVM and estimate how

much the individual was drinking with a 3.33 ml

RMSE using the amplitude characteristic of swallow-

ing sound (Kobayashi, 2014). Malvuccio and Ka-

mavuako (2021) have also applied sEMG recordings

of individual and continuous swallows to distinguish

between liquid and saliva swallows using Fine KNN

with an accuracy of 86.69 ± 5.52% to classify be-

tween the noise and swallows using Fine Gaussian

SVM with an accuracy of 99±1.31 (Malvuccio Ka-

mavuako, 2021). Surface electromyography (sEMG)

and microphones have been used for continuous mon-

itoring of the swallowing events by Amft and Tr

oster

to discriminate between solid and liquid meals in a

single participant (Amft, 2006).

To the best of our knowledge, there is a limited

number of studies using the sEMG to estimate ﬂuid

intake volume effectively. Among the above three

cited studies, only one attempted to use a continuous

estimation approach using artiﬁcial neural networks.

The challenge is not to classify discrete values but

to estimate continuous volume. Therefore, this study

aims to compare the capability of different regressors

in estimating ﬂuid intake volume using sEMG. Novel

contributions of this paper include (1) investigating

different machine learning regressors to ﬁnd the opti-

mum regressor in estimating ﬂuid volume (2) unravel-

ling the dependency between the choice of regressor

and features; and (3) proposing optimum placement

of sEMG electrodes with minimum error.

2 METHODOLOGY

2.1 Dataset

This study uses a previously recorded dataset; details

can be found in (Malvuccio Kamavuako, 2021). In

brief, three females and eight men, ranging in age

from 20 to 67 years, participated in this study. Two of

the Delsys Tringo sensors were placed on the belly of

the suprahyoid muscles, and two were placed on the

belly of the infrahyoid muscles. Drinking events oc-

Estimation of Fluid Intake Volume from Surface Electromyography Signals: A Comparative Study of Seven Regression Techniques

119

curred through various classes (drinking using a cup,

straw, bottle, and scale). After data checks, two sub-

jects had poor EMG data and were removed from the

investigation.

2.2 Experimental Procedure

Subjects consumed water using: a bottle, a cup, and

a straw, referred to as containers for simplicity. Each

subject had to swallow water ﬁve times for each con-

tainer while taking regular sips. We used a digital

scale to weigh the container before and after drink-

ing to quantify the true sip volume. A sip is a drink,

taking only a small amount at a time. For the ﬁnal

group of drinks, 5 ml were added to the highest cup

volume that could be computed, and this assignment

was only done once.

Figure 1: A graphical representation of the experimental

approach and data analysis pipeline.

2.3 Data Analysis

On Google Collab, we carried out data analysis using

Python 3.8 and preprocessing consisted of bandpass

ﬁltering between 6–400 Hz. As shown in Figure 1, the

EMG signals were rectiﬁed, and the signal envelope

was computed to detect the highest peak where the

swallowing event occurred. The EMG burst was then

extracted using the peak position. From that burst,

features of the Mel frequency cepstral coefﬁcients

(MFCCs), Mean Absolute Value (MAV), Waveform

Length (WL), and Willison Amplitude (WAMP) were

calculated on the raw data of the EMG. These features

had positive outcomes when applied to EMG signals

in earlier investigations.

• Mean Absolute Value (MAV): It is a method for

identifying and evaluating the intensity of muscu-

lar contractions. It can be represented as the mov-

ing average of the full wave rectiﬁed EMG signal,

as shown in equation 1[49].

MAV =

∑

i=1

| (1)

While N is the length of the segment, i is the seg-

ment increment, and Xi is the signal amplitude

value.

• Mel-frequency cepstral coefﬁcients (MFCCs):

MFCCs are coefﬁcients that form the Mel-

frequency cepstrum (MFC) based on a linear co-

sine transform of a log power spectrum on a non-

linear Mel scale of frequency. It works by seg-

menting the signal to a number of windows, then

applying the Discrete Fourier Transform (DFT)

and taking the log of the magnitude. Then, it

makes wrapping the frequencies on a Mel scale

and, in the end, applies the inverse Discrete Co-

sine Transform (DCT).

• Willison Amplitude (WAMP): The WAMP fea-

ture counts the number of changes in the ampli-

tude of the EMG signal that surpass a speciﬁc

threshold, as shown in equation 2 (Negi, 2016).

WAMP =

∑

i=1

[ f (|X

− X

i+1

|)];

f (x) =

(

1, x >= threshold

0, otherwise

(2)

• Waveform Length (WL): It is the total length

of the waveform for the segment. The results

obtained from the WL computation indicate the

waveform’s amplitude, frequency, and duration as

shown in equation 3 (Spiewak, 2018).

W L =

N−1

∑

i=1

i+1

− X

| (3)

We used a regression approach to estimate drink-

ing volume from sEMG. Our analysis included the

following techniques: Random Forest (RF), Sup-

port Vector Regressor (SVR), K-Nearest Neighbor

Regressor (KNN), Linear Regression (LR), Decision

Tree (DT), Lasso, and Ridge regressors. For each

subject, we had 16 observations; thus, a leave-one-

sample-out was used employed with permutation,

with the Root Mean Square Error (RMSE) as the per-

formance metric. In the ﬁrst part of data analysis,

we investigated the impact of using all four channels

versus the two lower (infrahyoid muscles) and upper

(suprahyoid muscles) channels using all regressors,

single features and all features together. We used

a three-way repeated measures analysis of variance

(3-ANOVA) with factors (Channels, Regressors and

features) to test for statistical differences between the

factors and interactions. In the second part of the data

analysis, we selected the three regressors and three

features with the lowest RMSE to investigate the ef-

fect of using single and mixed channels (one upper

and one lower) on performance. Similarly, we used

3-ANOVA to test for statistical differences. Results

are expressed as mean ± standard error.

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

120

3 RESULTS

The overall grand mean of the sip volume

across subjects and estimation RMSE across

channels, regressors and features were 13.91

± 1.27 ml and 1.37 ± 0.39 ml, respectively.

The Effect of Regressors and Features: The

RMSE of all four channels was 2.65 ± 0.32 ml, not

signiﬁcantly lower than the infrahyoid (2.90 ± 0.35

ml) and suprahyoid (2.76 ± 0.33 ml) pairs. There

was a statistical difference between regressors (P =

0.003), with RF (2.25 ± 0.25 ml), SVR (2.29 ± 0.30

ml) and KNN (2.41 ± 0.37 ml) performing better than

the others, also summarised in Figure 2 There was

no interaction between channels and regressors (P =

0.379).

Figure 2: An error bar plot that summarizes the RMSE and

the STD for the seven regressors.

The mean performance of each feature in ascend-

ing order, was 1.37 ± 0.39 ml for MAV, 1.82 ± 0.45 ml

for MFCC, 1.82 ± 0.45 ml for all features combined,

1.99 ± 0.60 ml for WAMP and 2.12 ± 0.44 ml for

WL, not statistically different (P = 0.132) from each

other. There was an interaction (P = 0.01) between

regressors and features, meaning that the choice of

features affects the performance of the regressors, as

summarised in Table 1. Figure 3 depicts the relation-

ship between regressors and features when using the

two suprahyoid muscles, ﬁgure 4 depicts the relation-

ship between regressors and features when using the

two infrahyoid muscles, and ﬁgure 5 depicts the rela-

tionship between regressors and features when using

the two suprahyoid and the two infrahyoid muscles.

Table 2 indicates that performance can be max-

imised using suprahyoid muscles with Random for-

est as regressor with the MAV feature. Nevertheless,

SVR and KNN are good regressor candidates with the

MFCC feature.

Table 1: Association between features and best regressor

for different channels.

Features Four

channels

Infrahyoid

muscles

Suprahyoid

muscles

MFCC RF

2.26±0.27ml

SVR

2.27±0.41ml

KNN

1.82±0.45ml

MAV RF

1.44±0.25ml

Lasso / Ridge

2.19±0.42ml

1.37±0.39ml

WAMP KNN

1.99±0.60ml

KNN

2.09±0.57ml

KNN

2.03±0.57ml

WL SVR

2.24±0.36ml

SVR

2.15±0.44ml

2.12±0.44ml

ALL SVR

2.06±0.60ml

SVR

1.97±0.36ml

KNN

1.82±0.45ml

Figure 3: The bar plot with the root mean square error for

the seven regressors with the features using the Suprahyoid.

Single channel investigation showed no difference

between channels nor their combinations (infrahyoid

and suprahyoid). It is worth noting that using the

left infrahyoid or left suprahyoid channel alone with

the RF regressor with either MAV or MFCC pro-

vided RMSE values close to 1.6 ml. The combined

left supra and right infra channels performed down

to 1.5 ml using RF and MAV. Combination of MAV

and MFCC using the two suprahyoid channels with

the RF has not improved the RMSE results. Table

V demonstrates the average sip volume and the Root

Mean Square error for each subject using the upper

two suprahyoid muscles with RF.

4 DISCUSSION

This study aimed to compare the power of various

regressor techniques in estimating ﬂuid intake vol-

ume using surface Electromyographic (sEMG) sen-

sors. The study’s regression ﬁndings strongly suggest

that estimation of ﬂuid intake volume is feasible us-

ing surface EMG. This study demonstrated how re-

gression performance differed depending on whether

signals were coming from the upper two muscles

Estimation of Fluid Intake Volume from Surface Electromyography Signals: A Comparative Study of Seven Regression Techniques

121

Figure 4: The bar plot with the root mean square error for

the seven regressors with the features using the Infrahyoid

Muscles.

Figure 5: The bar plot with the root mean square error for

the seven regressors with the features using the Infrahyoid

Muscles and the Suprahyoid muslecs.

(Suprahyoid muscles), the lower two muscles (In-

frahyoid muscles), or all the muscles. We found that

solely employing suprahyoid muscles did not produce

signiﬁcant superior results to those of the infrahyoid

muscles. The seven regression models were run on

each feature individually and then on all four features

collectively to select the best features.

This demonstrated that no single regressor works

best for all features and that the regressor depends on

the feature. There was a signﬁcant difference between

the regressors. For example, Random Forest regres-

sor performs best using the Mean Absolute Value fea-

ture. Except for the Willison amplitude and MFCC

features, statistical analysis did reveal signiﬁcant vari-

ation in how different regressors performed with var-

ious features. Therefore, utilising a single character-

istic can be advantageous since it will decrease com-

puting costs and time, particularly for online jobs. Us-

ing single channels of the infrahyoid or the suprahy-

oid muscles or combining a single channel of each

to estimate the volume has not improved the results

of RMSE. However, the error difference was not too

high, indicating that single EMG channels may be

used to record the intake data. As a result, in a sub-

sequent study, single channels will be used to record

the intake data and will be investigated if the ﬂuid es-

timation performance will be improved or not.

Table 2: Final Results for the Best Regression Model (RF)

using Mean Absolute Volume feature.

Subjects Average Sip

Volume (ml)

RMSE (ml) Percentage

error (%)

S1 19.42 ± 5.00 1.07 ± 1.67 5

S2 8.72 ± 2.95 1.42 ± 0.98 16

S3 12.18 ± 4.19 0.82 ± 1.4 6

S4 11.4 ± 3.59 2.99 ± 1.19 26

S5 18.71 ± 5.94 0.13 ± 1.98 1

S6 12.72 ± 3.85 1.85 ± 1.28 14

S7 21.32 ± 8.47 0.39 ± 2.82 2

S8 13.66 ± 3.15 0.29± 1.05 2

S9 7.14 ± 2.86 3.35 ± 0.95 47

Average 13.91 ± 1.27 1.37 ± 0.39 13.22

The number of studies aiming at ﬂuid volume es-

timation from sEMG is very limited. Kobayashi et al.

attempted to measure the amount of liquids consumed

using a throat microphone with an RMSE value of

3.33 ml (Kobayashi, 2014). Malvuccio also estimated

the amount of ﬂuid consumed using sEMG record-

ings of both individual and continuous swallows, but

her work had an RMSE than ours (Malvuccio, 2021).

Despite the similar performance, decreasing the error

further will be beneﬁcial, and thus our future study

should include a larger sample size with advanced

techniques.

Although the sEMG performance in estimating

ﬂuid intake volume has shown encouraging results

compared to other approaches, further validation of

these data is necessary. We aim to increase the sample

size to improve the outcomes and model performance.

Additionally, the test volunteers must be older adults

since their swallowing habits change as people age,

which may impact the system’s functionality. Find-

ing out if the performance would be affected by age

and by increasing the number of subjects is required

because this study only used a small number of partic-

ipants. Additionally, this study did not consider other

variables that could impact the sip volume, such as

the liquid temperature and composition. These fac-

tors may cause the sip volumes to differ from sub-

ject to subject, compromising the ﬂuid intake volume

technique’s estimation ability.

5 CONCLUSIONS

We have compared for the ﬁrst time the capability of

various regressors to estimate ﬂuid intake from sEMG

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

122

signals recorded during swallowing events. The re-

sults of utilizing only the suprahyoid or infrahyoid

muscles did not differ statistically; however, there

were statistical differences between the various re-

gressors. RF, then SVM regressors were the best ones

using the Mean Absolute Value feature in estimating

the ﬂuid volume with the lowest error. Furthermore,

there is an indication that regressor performance is

feature dependent. This outcome is a step forward

in using sEMG for hydration monitoring. Further re-

search is needed to investigate the use of single EMG

channels to record and estimate the ﬂuid data and

whether two channels work better for regression and

the other are better for classiﬁcation.

REFERENCES

El-Sharkawy, A. M., Sahota, O., Maughan, R. J., & Lobo,

D. N. (2014). The pathophysiology of ﬂuid and elec-

trolyte balance in the older adult surgical patient. Clin-

ical Nutrition, 33(1), 6-13.

Dehydration risks for seniors. Cleveland Clinic. (2021,

November 10). Re- trieved December 30, 2021,

from https://health.clevel and clinic.org/drink- up-

dehydration- is-an-often-overlooked-health-risk-for-

seniors/

Malvuccio, C., & Kamavuako, E. N. (2021, March). De-

tection of Swallowing Events and Fluid Intake Vol-

ume Estimation from Surface Electromyography Sig-

nals. In 2020 IEEE-EMBS Conference on Biomedical

Engineering and Sciences (IECBES) (pp. 245-250).

IEEE.

Asogan, H., & Raoof, A. (2021). Education and training as

key drivers for improving the quality of ﬂuid balance

charts: ﬁndings from a quality improvement project.

BMJ Open Quality, 10(3), e001137.

Cohen, R., Fernie, G., & Roshan Fekr, A. (2021). Fluid in-

take monitoring systems for the elderly: a review of

the literature. Nutrients, 13(6), 2092.

Wellnitz, A., Wolff, J. P., Haubelt, C., & Kirste, T. (2019,

September). Fluid intake recognition using inertial

sensors. In Proceedings of the 6th international Work-

shop on Sensor-based Activity Recognition and Inter-

action (pp. 1-7).

Dong, B., & Biswas, S. (2017). Meal-time and duration

monitoring using wearable sensors. Biomedical Sig-

nal Processing and Control, 32, 97-109.

Cheng, J., Amft, O., & Lukowicz, P. (2010, May). Ac-

tive capacitive sensing: Exploring a new wearable

sensing modality for activity recognition. In Interna-

tional conference on pervasive computing (pp. 319-

336). Springer, Berlin, Heidelberg.

Cheng, J., Zhou, B., Kunze, K., Rheinl

ander, C. C., Wille,

S., Wehn, N., ... & Lukowicz, P. (2013, September).

Activity recognition and nutrition monitoring in ev-

ery day situations with a textile capacitive neckband.

In Proceedings of the 2013 ACM conference on Per-

vasive and ubiquitous computing adjunct publication

(pp. 155-158).

Zhang, R., Freund, M., Amft, O., Cheng, J., Zhou, B.,

Lukowicz, P., ... & Chabrecek, P. (2016, September).

A generic sensor fabric for multi-modal swallowing

sensing in regular upper-body shirts. In Proceedings

of the 2016 ACM International Symposium on Wear-

able Computers (pp. 46-47).

Amft, O., & Troster, G. (2006, November). Methods for de-

tection and classiﬁcation of normal swallowing from

muscle activation and sound. In 2006 Pervasive Health

Conference and Workshops (pp. 1-10). IEEE.

Amft, O., & Tr

oster, G. (2008). Recognition of dietary ac-

tivity events using on-body sensors. Artiﬁcial intelli-

gence in medicine, 42(2), 121-136.

Moreau–Gaudry, A., Sabil, A., Benchetrit, G., & Franco,

A. (2005). Use of respiratory inductance plethysmog-

raphy for the detection of swallowing in the elderly.

Dysphagia, 20(4), 297-302.

Dong, B., & Biswas, S. (2012, August). Swallow moni-

toring through apnea detection in breathing signal. In

2012 Annual International Conference of the IEEE

Engineering in Medicine and Biology Society (pp.

6341-6344). IEEE.

Dong, B., & Biswas, S. (2013, May). Liquid intake monitor-

ing through breathing signal using machine learning.

In Sensing Technologies for Global Health, Military

Medicine, and Environmental Monitoring III (Vol.

8723, pp. 141-148). SPIE.

Dong, B., & Biswas, S. (2016). Analyzing breathing signals

and swallow sequence locality for solid food intake

monitoring. Journal of medical and biological engi-

neering, 36(6), 765-775.

Tatulli, E., Fontecave-Jallon, J., Calabrese, P., & Gumery,

P. Y. (2020). Respiratory Inductance Plethysmogra-

phy for Automated Swallowing Detection. Interna-

tional Journal of E-Health and Medical Communica-

tions (IJEHMC), 11(2), 64-77.

Nakafuji, H., Imura, M., Uranishi, Y., Yoshimoto, S., & Os-

hiro, O. (2014). Estimation of amount of swallowed

water by analysis of swallowing sounds. Biomedical

Engineering, 52(Supplement), O-11.

Kobayashi, Y., & Mineno, H. (2014, October). Fluid in-

take recognition for nursing care support by leverag-

ing swallowing sound. In 2014 IEEE 3rd Global Con-

ference on Consumer Electronics (GCCE) (pp. 620-

621). IEEE.

Amft, O., & Troster, G. (2006, November). Methods for de-

tection and classiﬁcation of normal swallowing from

muscle activation and sound. In 2006 Pervasive Health

Conference and Workshops (pp. 1-10). IEEE.

Negi, S., Kumar, Y., & Mishra, V. M. (2016, September).

Feature extraction and classiﬁcation for EMG signals

using linear discriminant analysis. In 2016 2nd In-

ternational Conference on Advances in Computing,

Communication, & Automation (ICACCA)(Fall) (pp.

1-6). IEEE.

Spiewak, C., Islam, M., Zaman, A., & Rahman, M.

H. (2018). A comprehensive study on EMG feature

extraction and classiﬁers. Open Access Journal of

Biomedical Engineering and Biosciences, 1(1), 1-10.

Estimation of Fluid Intake Volume from Surface Electromyography Signals: A Comparative Study of Seven Regression Techniques

123

Hara, K., Tohara, H., Kobayashi, K., Yamaguchi,

K., Yoshimi, K., Nakane, A., & Minakuchi, S.

(2018). Age-related declines in the swallowing mus-

cle strength of men and women aged 20–89 years:

A cross-sectional study on tongue pressure and jaw-

opening force in 980 subjects. Archives of Gerontol-

ogy and Geriatrics, 78, 64-70. Chicago

What factors affect EMG signal quality? – Signal Qual-

ity Mon- itor. Delsys. (2020, June 24). Retrieved June

23, 2022, from https://delsys.com/emgworks/signal-

quality-monitor/factors/

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

124