Wearable Sensor Framework for Rehabilitation Monitoring Following

Knee-Related Conditions

Can Tekdemir

, Yusuf Ziya Hayirlioglu

, Olgar Birsel

2 a

and Beren Semiz

1 b

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

Department of Orthopaedics and Traumatology, School of Medicine, Koc¸ University, Istanbul, Turkey

{ctekdemir21, yhayirlioglu17, olgarbirsel, besemiz}@ku.edu.tr

Keywords:

Knee Health, Acoustical Emissions, Electrical Bioimpedance, Inertial Activity, Rehabilitation Monitoring.

Abstract:

The knee, being the largest and one of the most complex joints in the body, is highly vulnerable to injury due to

its intricate structure and exposure to multi-directional forces. Surgical interventions are among the most fre-

quently utilized treatments for advanced degenerative joint disease. However during the rehabilitation process,

physician assessments can be subjective, which may lead to inconsistent evaluations, particularly in subtle or

complex cases. To enhance accuracy and reduce variability, objective methodologies like data-driven tools and

standardized protocols are necessary. Hence, we propose a multi-modal wearable sensor framework leveraging

electrical bioimpedance, acoustical emissions, inertial activity and temperature measurements simultaneously

to achieve objective and quantiﬁable information that can complement clinical judgement, ensuring more re-

liable and reproducible outcomes in patient care. The system was validated through active (ﬂexion-extension)

and inactive (sedentary) measurements, and proven successful in capturing knee-related signatures and assess-

ing the knee joint health. Such a system could facilitate the establishment of a direct relationship between

signal characteristics and key knee health parameters, enabling more informed decisions regarding disease or

injury status and treatment progress.

1 INTRODUCTION

The knee is the largest and one of the most com-

plex joints in the human body, making it highly sus-

ceptible to stress from multi-directional forces during

physical activity (Austermuehle, 2001). Its intricate

structure, comprising two distinct joints, accommo-

dates loads from various directions while relying on

both static and dynamic soft tissues to maintain stabil-

ity. This complexity, along with age-related wear and

tear, makes the knee particularly vulnerable to injury

and a leading cause of rehabilitation among muscu-

loskeletal disorders. Globally, an estimated 1.71 bil-

lion people suffer from musculoskeletal conditions,

a number projected to rise with an aging population

(WHO, 2022).

Surgical interventions are one of the most com-

monly used treatment approaches for advanced de-

generative joint disease. Indeed, the number of to-

tal knee arthroplasties performed each year in the

United States alone is estimated to reach 3.5 million

https://orcid.org/0000-0002-2137-1164

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

in 2040 (Carmichael et al., 2022). Following surgery,

the main goal in the early postoperative period is to

maintain proper healing of the surgical wound. This

process is monitored by simple observation and phys-

ical examination ﬁndings. Healing, which begins as

soon as the surgical incision is made, should follow

a decrease in soft tissue edema, redness and temper-

ature, ﬁnally leading to restoration of skin integrity

(George Broughton et al., 2006). The second im-

portant point is to restore the range of motion of the

knee joint, which is calculated through angular val-

ues while the physician is passively moving the knee

joint or asking the patient to actively bend the knee.

The physician can then conclude whether the wound

is healing smoothly in the early postoperative period

and the inﬂammatory process has regressed as ex-

pected – or can make necessary interventions and pro-

vide warnings by identifying disruptions in the natural

course of recovery.

One problem with the physician assessments is

their subjectivity as they often rely on personal ex-

perience, intuition and observation, which can vary

between individuals and lead to inconsistent evalu-

ations, especially in subtle or complex cases. To

Tekdemir, C., Hayirlioglu, Y. Z., Birsel, O. and Semiz, B.

Wearable Sensor Framework for Rehabilitation Monitoring Following Knee-Related Conditions.

DOI: 10.5220/0013144000003911

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

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 1, pages 45-52

ISBN: 978-989-758-731-3; ISSN: 2184-4305

Figure 1: Most important circuit components.

reduce variability and improve accuracy, objective

methodologies, such as data-driven tools and stan-

dardized protocols, are required. In the literature, sev-

eral attempts have been made to assess knee health

non-invasively which are expected to provide pa-

tients and physicians with more readily available

quantitative data regarding health or treatment sta-

tus. Most of these attempts have focused on the

design of wearable sensing modalities leveraging

the measurement of joint acoustics, edema or activ-

ity range through miniature microphones, electrical

bioimpedance (EBI) systems and inertial measure-

ment units (IMUs), respectively. For example, joint

acoustics (vibrations emitted from the mid-patella

during active movements) have been used to distin-

guish between subjects with osteoarthritis and healthy

subjects (Sarillee et al., 2014), to track treatment

and medication effectiveness in children with arthri-

tis (Semiz et al., 2018) and to monitor rehabilitation

advancements in athletes with an acute injury (Hersek

et al., 2017). On the other hand, several studies have

leveraged the use of EBI in joint edema (swelling)

detection and monitoring (Hersek et al., 2016); and

IMUs for knee joint stability (Kianifar et al., 2017)

and range of motion assessment (Seel et al., 2014).

In this work, we propose a multi-modal sensor

system prototype leveraging electrical bioimpedance,

acoustical emissions, inertial activity and tempera-

ture measurements simultaneously to achieve objec-

tive and quantiﬁable information that can complement

clinical judgement, ensuring more reliable and repro-

ducible outcomes in patient care. Such a multi-modal

system could potentially facilitate the investigation of

one-to-one correspondence between signal character-

istics and important knee health parameters, enabling

more informed decisions regarding the disease/injury

state or treatment progression.

2 METHODS

2.1 Hardware Setup

In the system, a Raspberry Pi Pico based on RP2040

microcontroller is used as the development kit. The

circuit visuals are provided in Figure 1 and the sen-

sor locations are visualized in Figure 2. The system

has have two modes: an active mode and an inac-

tive (passive) mode. In the active mode, acoustical

emissions and inertial activity measurements are em-

ployed, whereas during inactive mode, the electrical

bioimpedance and temperature values are measured.

Details regarding electrical bioimpedance, tempera-

ture, acoustical emissions and inertial activity are pre-

sented in the upcoming sub-sections.

2.1.1 Electrical Bioimpedance Measurement

System

The bioimpedance system uses an AD9833 signal

generator module to generate a 60 kHz sine wave with

an amplitude and DC offset of 325 mV each. Subse-

quently, this generated signal is fed into a DC off-

set remover and voltage buffer circuit (which uses an

LM358 op-amp) to be turned into a pure AC signal.

The output from the LM358 circuit is then fed into

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

Figure 2: Sensors included in the system.

Figure 3: Experimental protocol including active and inac-

tive (passive) measurements.

a NE5532 op-amp based VCCS (Voltage Controlled

Current Source) circuit. The VCCS circuit sends out

a constant 2 mA peak-to-peak current from the op-

amp’s output pin to its inverting input pin. The leg

electrodes are connected to these pins to send a con-

stant AC current through the knee joint. Additionally,

the VCCS circuit uses a potentiometer to ﬁne-tune the

magnitude of the AC current.

The voltage across the knee joint is measured with

an NE5532 op-amp based circuit. The circuit con-

sists of a simple inverting summing ampliﬁer (which

uses the NE5532) and a low-pass ﬁlter. The top-leg

electrode’s potential acts as the input signal, while the

bottom-leg’s is connected to the ADC ground (which

makes it act as a reference ground for the system).

The circuit adds the input voltage and a voltage source

of -1.5151 volts (which is half of the reference volt-

age of the RP2040’s ADC) together, inverts the signal

(so that it is positively shifted for the ADC), and then

feeds it into an anti-aliasing (low pass) ﬁlter, which

then gets connected to one of the ADC pins on the

Pico. It should be noted that the output signal is

not too large to cause any clipping or go beyond the

ADC’s range, which is why this level of DC offset is

perfectly adequate for the Pico’s ADC.

The negative voltage producing circuit consists of

a NE555 precision timer-based square wave gener-

ator that drives a negative charge-pump circuit and

a LM7905 negative voltage regulator. The system

functions by continuously driving a negative charge-

pump circuit at 60.33 kHz. The negative charge-pump

circuit consists of a 680 uH inductor, a 1N914 fast

switching diode, a 680 uF capacitor, and a 2N3906

PNP BJT transistor to act as a switch. At 60.33

kHz, the charge-pump circuit produces a stable -11.5

volts while being able to deliver a current around -

6.8 mA. The output from the charge-pump circuit is

then fed into a LM7905 negative voltage regulator,

which reduces and stabilizes the negative voltage cir-

cuit’s output to -5 volts. The negative voltage produc-

ing circuit’s output is primarily used in powering all

of the op-amps found in the system. It is also used

as a constant offset voltage for the aforementioned

bioimpedance voltage measurement system.

2.1.2 Temperature Measurement System

The temperature measurement circuit is based on a

voltage divider that consists of a thermistor and a re-

sistor. A 10 kOhm negative-thermal-coefﬁcient ther-

mistor (model: TTC05103JSY) is connected in se-

ries with a 10 kOhm resistor. As the thermistor’s

temperature increases, its resistance decreases, which

causes the voltage across the 10 kOhm resistor to in-

crease. The output voltage on the resistor is fed into

a NE5532 op-amp based voltage buffer and the out-

put of the voltage buffer is fed into a low pass ﬁl-

ter with a frequency cutoff of 15.92 Hz. It should be

noted that the temperature on the thermistor changes

very slowly due to thermodynamic physical phenom-

ena and that the DC component of the output signal is

signiﬁcantly more important than the AC component.

Therefore, the low-pass ﬁlter is implemented to func-

tion as a pseudo-AC ﬁlter that also blocks off power-

line interference and other unneeded higher frequen-

cies. The NE5532 is used as a buffer to preserve the

voltage so that the voltage divider’s output is not af-

fected due to the low pass ﬁlter. The output of the ﬁl-

ter is fed into the Pico’s second ADC pin to calculate

the thermistor’s temperature. It should also be noted

that the voltage-to-temperature calculation is done in-

ternally in the Pico itself.

2.1.3 Acoustical Emissions System

Two miniature Knowles BU-23173 analog contact

microphones were used in the acoustic emission sys-

tem. Both microphones were ampliﬁed through the

TDA2822m audio ampliﬁer integrated with 39 dB

gain. After ampliﬁcation, the signals are passed

through an anti-aliasing ﬁlter. Since the sampling rate

used in the microphones was 20 kHz, this results in a

Nyquist frequency of 10 kHz; hence, two ﬁrst-order

low-pass ﬁlters with a cut-off frequency of 9042.89

Hz are used to avoid aliasing during the recordings.

Wearable Sensor Framework for Rehabilitation Monitoring Following Knee-Related Conditions

Figure 4: Raw IMU measurements showing accelerometer and gyroscope recordings individually.

2.1.4 Inertial Activity System

The inertial activity system measures joint motion and

angular changes along with knee acoustical emissions

during active movements. Accordingly, 2 IMU sen-

sors are included in the system, one ﬁxed to the thigh

(upper) and the other to the calf (lower). A 6-axis

(3-axis accelerometer and 3-axis gyroscope) digital

BMI160 sensor is selected as the IMU. Each of the

IMUs receives data with a sampling rate of 100 Hz.

The raw accelerometer and gyroscope data are then

used for direction/angle calculation via quaternions.

2.1.5 Final System Design

In addition to the individual systems described above,

various peripheral elements are also used in the de-

sign.

The system switches between “active” and “inac-

tive” measurements via a DPDT (double pole, double

throw) switch that is controlled externally. When the

switch is in the “active” measurement position, the

output signals from the inertial activity system and

the acoustical emissions system are fed into the Pico’s

ADC pins. When the switch is set into the “passive”

position, the outputs from the electrical bioimpedance

system and the temperature measurement system are

fed into the ADC pins instead. This way, only one set

of measurements are taken at a time.

When the switch is in “active” mode, three dif-

ferent buttons are used to toggle between the “Start”,

“Hold” and “Stop” commands. When the recording

starts, a ﬁle is created on the microSD card for the ac-

tive mode; where the acoustic emission signals and

IMU data are recorded into this ﬁle. The ﬁles are

named using the date and time information captured

via the DS3231 RTC module. A ﬂag is also added to

the ﬁle name to indicate that the record is an “active”

record.

When switching to “inactive” mode using the

switch, a second start button is used to create a

separate ﬁle with a different ﬂag (note that the ﬁle

name still uses the date and time information from

the DS3231). Although a second “Start” button is

used for this mode, it shares the same “Hold” and

“Stop” buttons. This mode measures only 2 signals;

namely the electrical bioimpedance and the tempera-

ture, which are both analog signals. The voltage-to-

temperature measurement is done internally (as men-

tioned before), and the measured skin voltage is saved

directly into the ﬁle. All of the recorded measurement

data is saved unto an SD card through a Adafruit mi-

croSD card module.

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

Figure 5: IMU-derived angle and acoustical emission sig-

nals shown together.

Additionally, there are 3 LEDs in the system that

indicate whether the system is functioning properly.

The green LED lighting up indicates that the record-

ing is running smoothly (for the “active” mode), while

a red LED indicates that there is a problem with the

system (e.g. SD card non-contact). For the “inactive”

mode, both the green and the yellow LED light up to

indicate that the system is recording.

2.2 Data Collection and Experimental

Protocol

The protocol for the measurements basically consists

of two main parts: (i) the sedentary recording part

during rest and (ii) the active recording part includ-

ing active tasks, as seen in Figure 3. Data was col-

lected from a healthy male participant in laboratory

settings. First, bioimpedance and temperature val-

ues were measured during 20 seconds of sedentary

measurement, followed by the active period. Dur-

ing the active period, 10 ﬂexion-extension exercises

were performed, during which the corresponding in-

ertial activity and acoustic emission measurements

were taken. The protocol was terminated with a sec-

ond 20-second sedentary period.

2.3 Analysis Framework

2.3.1 Electrical Bioimpedance Analysis

The recorded bioimpedance signal was analyzed with

a simple signal processing algorithm implemented in

Matlab. First, the signal was passed through a very

tight band-pass ﬁlter ([59.5, 60.5] kHz) and then mul-

tiplied by 41.1392. This scaling value was obtained

experimentally: The output of the function generator

was fed as an input to the bioimpedance measurement

circuit and then the amount of drift and noise was ana-

lyzed. It was observed that the use of a bandpass ﬁlter

and this scaling factor resulted in a signal very close

to the signal seen on the oscilloscope. This algorithm

prevents any external noise from passing through and

also makes the calculated knee impedance more ac-

curate.

2.3.2 Temperature Analysis

For the temperature measurement, the Pico was pro-

grammed to shift the incoming data by 13 bits before

any value was calculated. The main reason behind

this is to avoid noise interference and to compensate

for the voltage difference between the output of the

ﬁlter and the output of the voltage divider. In ad-

dition, a simple compression algorithm was used to

reduce the temperature range to [24.5, 50.0] degrees

Celsius. Following the compression algorithm, the

temperature was shifted down to eliminate the added

bits and any DC noise.

2.3.3 Inertial Activity Analysis

The raw accelerometer and gyroscope data recorded

on the SD card were ﬁrst ﬁltered between 0.1 Hz and

48 Hz using a ﬁnite impulse response (FIR) Kaiser

window band-pass ﬁlter. The triaxial accelerometer

and gyroscope signals recorded from the calf (lower)

and thigh (upper) regions for each IMU of a sample

subject are visualized in Figure 4.

The raw accelerometer and gyroscope data were

then evaluated for direction/angle calculation using

quaternions (Seel et al., 2014). Accordingly, an ex-

tended Kalman ﬁlter was used to estimate an orien-

tation represented as a quaternion. The algorithm

ﬁrst estimates the new state (the most recent ori-

entation) for the calf and thigh using the instanta-

neous measurements of the gyroscopes, then makes

the necessary corrections using the measurements of

the accelerometers. Next, quaternions are extracted

from the calculated orientations for the calf and thigh

and rotation matrices are calculated for both regions.

These two rotation matrices are multiplied to obtain

Wearable Sensor Framework for Rehabilitation Monitoring Following Knee-Related Conditions

Table 1: The calculated inertial activity characteristics.

Flexion Time Flexion Range Flexion Slope Extension Time Extension Range Extension Slope

Mean 0.417 4.966 0.121 0.382 4.959 0.133

Std 0.051 0.267 0.015 0.064 0.381 0.021

Max 0.510 5.345 0.138 0.540 5.517 0.172

Min 0.360 4.543 0.091 0.320 4.261 0.099

the relative rotation. Finally, from this relative rota-

tion matrix, relative angle values in x, y and z direc-

tions are obtained through Euler transformation.

The relative angles x, y and z represent the rela-

tive roll, pitch and yaw signals respectively. For an-

gle tracking and the extraction of relevant features,

relative pitch graph was used in accordance with the

mechanical orientations. On the relative pitch graph,

several features were calculated. First, the peak am-

plitudes, durations and slopes were extracted for each

ﬂexion or extension movement. Then, the mean, stan-

dard deviation, minimum, and maximum values of

the peak, duration, and slope of all cycles were cal-

culated to evaluate the irregularity of the exercises, as

well as the maximum and minimum angles reached. It

should be noted that the features were extracted from

the non-normalized signal. This would potentially en-

able the assessment of the maximum/minimum angle

differences between the participants in future studies.

2.3.4 Acoustical Emissions Analysis

The .bin ﬁles saved on the SD card were ﬁrst con-

verted to the appropriate format using Matlab. It

should be noted that the following pre-processing and

feature extraction steps were repeated for both the

left and right microphones. First, hardware noise was

suppressed using narrow band-stop ﬁlters ([185, 205]

and [940, 1050] Hz). The remaining pre-processing

and feature extraction steps were then employed in

Python (version 3.11).

In order to extract the clinically relevant part of the

acoustic emission data from the microphones, a ﬁnite

impulse response (FIR) Kaiser window bandpass ﬁl-

ter between 250 Hz and 5 kHz was used. An example

signal is presented in Figure 5. In order to visualize

that the system successfully captures the felt clicks, a

signal containing clicks was speciﬁcally selected.

The signal was then divided into 500 ms windows

in parallel with the studies in the literature and 250 ms

overlap was used between the subsequent windows

(50% overlap). For the two different window types

(with and without clicks), we reported the distinctive

differences in the features selected from 3 different

groups: temporal (peak-to-peak), statistical (kurtosis)

and spectral (roll-off).

3 RESULTS AND DISCUSSION

3.1 Inertial Activity and Acoustical

Emissions Results

As mentioned in Section 2.2, the participant per-

formed 10 ﬂexion-extension exercises during the pilot

data acquisition. Figure 5 clearly shows the angular

change. The dots on the graph represent the moments

of ﬂexion, while the peaks represent the moments of

maximum extension. Although the preliminary analy-

sis focused on angular features, direct analysis of raw

accelerometer and gyroscope data would also be pos-

sible in the long term due to their high signal qual-

ity. In addition, the peak amplitudes, durations and

slopes were extracted for each ﬂexion or extension

movement. Then, the mean, standard deviation, mini-

mum, and maximum values of the peak, duration, and

slope of all cycles were calculated to evaluate the ir-

regularity of the exercises, as well as the maximum

and minimum angles reached. The derived values are

presented in Table 1.

Figure 6 shows a superimposition of 500 ms

(10000 samples)-long segments, and examples of a

silent window and a window with clicks. As ex-

plained in Section 2.3.4, three different features were

extracted from a sample silent window and a win-

dow with clicks. As seen, the peak-to-peak values

were 0.081 and 0.023; whereas the kurtosis values

were 26.863 and 0.628, respectively. As higher kur-

tosis refers to the presence of more extreme values or

outliers (i.e. tailedness) compared to a normal dis-

tribution, it was indeed aligning with the morphol-

ogy differences between a window with click and a

silent window. Similarly, the spectral roll-off val-

ues were calculated to be 0.988 and 0.628, respec-

tively. Higher spectral roll-off in signal analysis indi-

cates that a greater proportion of the signal’s energy

is concentrated in the higher frequency components.

The joint clicks are characterized by high energy and

short duration, typically lasting between 10 and 20

milliseconds. Furthermore, these emissions contain

high-frequency components, with bandwidths extend-

ing up to 20 kHz, a range commonly expected for

acoustic emissions (Teague et al., 2016). Hence, hav-

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

Figure 6: (a) Superimposition of acoustical emission frames. (b) Examples of a frame with a click and a silent frame.

Figure 7: Temperature measurement.

ing a higher spectral roll-of value indeed aligns with

the literature.

3.2 Temperature and Bioimpedance

Results

The temperature and impedance measurement analy-

ses were explained in Sections 2.3.2 and 2.3.1, respec-

tively. Accordingly, the resulting temperature mea-

surement for the participant is presented in Figure 7.

Similarly, the impedance value was calculated to be

97.9246 Ohm. This value was indeed in line with the

impedance values of healthy knees in studies in the

literature (Hersek et al., 2015).

4 CONCLUSION

In this work, we presented a multi-modal wear-

able sensor framework leveraging electrical

bioimpedance, acoustical emissions, inertial ac-

tivity and temperature measurements simultaneously

to obtain objective and measurable information that

can support clinical decision-making. The system

was validated using active (ﬂexion-extension) and

inactive (sedentary) measurements and proven to be

successful in measuring the knee-related parameters.

The main limitation of the current design is the

extensive number of components. Hence, future work

will focus on transforming the current design into a

printed-circuit-board (PCB) and validating it in actual

clinical settings more easily. The ﬁrst potential use

area will be to monitor the rehabilitation period of

the patients undergone total knee arthroplasty. Such

a system could help establishing direct relationships

between signal characteristics and key knee health pa-

rameters, allowing for more informed decisions about

the disease or injury status and treatment progress.

ACKNOWLEDGEMENTS

This work was supported by the Health Institutes of

Turkiye (TUSEB) under grant number 22554.

REFERENCES

Austermuehle, P. D. (2001). Common knee injuries in pri-

mary care. The Nurse Practitioner, 26(10):26–32.

Carmichael, J., Dennis, D., Jennings, J., Stevens-Lapsley,

J., and Bade, M. (2022). Feasibility and initial efﬁcacy

of a multimodal swelling intervention after total knee

arthroplasty: A prospective pilot study with historical

controls. The Knee, 35:25–33.

George Broughton, I., Janis, J. E., and Attinger, C. E.

(2006). Wound healing: an overview. Plastic and

reconstructive surgery, 117(7S):1e–S.

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.

Hersek, S., T

oreyin, H., and Inan, O. T. (2015). A robust

system for longitudinal knee joint edema and blood

ﬂow assessment based on vector bioimpedance mea-

surements. IEEE Transactions on biomedical circuits

and systems, 10(3):545–555.

Hersek, S., T

oreyin, H., Teague, C. N., Millard-Stafford,

M. L., Jeong, H.-K., Bavare, M. M., Wolkoff, P.,

Sawka, M. N., and Inan, O. T. (2016). Wearable vec-

tor electrical bioimpedance system to assess knee joint

health. IEEE Transactions on Biomedical Engineer-

ing, 64(10):2353–2360.

Kianifar, R., Lee, A., Raina, S., and Kuli

c, D. (2017). Auto-

mated assessment of dynamic knee valgus and risk of

knee injury during the single leg squat. IEEE journal

Wearable Sensor Framework for Rehabilitation Monitoring Following Knee-Related Conditions

of translational engineering in health and medicine,

5:1–13.

Sarillee, M., Hariharan, M., Anas, M., Omar, M., Aishah,

M., and Oung, Q. W. (2014). Assessment of knee joint

abnormality using acoustic emission sensors. In 2014

IEEE International Conference on Control System,

Computing and Engineering (ICCSCE 2014), pages

378–383. IEEE.

Seel, T., Raisch, J., and Schauer, T. (2014). Imu-based

joint angle measurement for gait analysis. Sensors,

14(4):6891–6909.

Semiz, B., Hersek, S., Whittingslow, D. C., Ponder, L. A.,

Prahalad, S., and Inan, O. T. (2018). Using knee

acoustical emissions for sensing joint health in pa-

tients with juvenile idiopathic arthritis: A pilot study.

IEEE sensors journal, 18(22):9128–9136.

Teague, C. N., Hersek, S., T

oreyin, H., Millard-Stafford,

M. L., Jones, M. L., Kogler, G. F., Sawka, M. N., and

Inan, O. T. (2016). Novel methods for sensing acous-

tical emissions from the knee for wearable joint health

assessment. IEEE Transactions on Biomedical Engi-

neering, 63(8):1581–1590.

WHO (2022). Musculoskeletal health — who.int. World

Health Organization. [Accessed 06-10-2024].

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices