Volume 9, Issue 8, August – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24AUG995

Combining Gyroscope and Electromyogram

Analysis for the Detection of Resting Tremor and
Muscle Activity in Parkinson's Disease
Arnab Biswas Shamayita Mukherjee Arnab Maji Ayan Manna
Applied Electronics & Applied Electronics & Applied Electronics & Applied Electronics &
Instrumentation Engineering Instrumentation Engineering Instrumentation Engineering Instrumentation ENgineering
Heritage Institute of Heritage Institute of Heritage Institute of Heritage Institute of
Technology(MAKAUT) Technology(MAKAUT) Technology(MAKAUT) Technology(MAKAUT)
Kolkata, India Kolkata, India Kolkata, India Kolkata, India

Shouvik Sarkar
Electronics and Communication Engineering
Assam University,Silchar

Abstract:- In this paper, we focus on designing and As with other tremors, there are fundamental differences
implementing a wearable device for detecting Parkinson's between resting tremors associated with Parkinson's disease
disease (PD) symptoms by analyzing resting tremors and and other tremors, such as their characteristic frequency, their
abnormal muscle activity which contribute to PD progression and pathophysiology. However , there are
combining gyroscope and electromyogram(EMG) similarities in their appearance , occurrences which leads to
analysis. Using advanced sensor technology, real-time data misdiagnosis of resting tremor associated with PD and other
about movement and muscle activity is captured by the tremors . Misdiagnosis occur regarding the symptoms that can
device. Here, we outline a hardware framework for mimic PD such as essential tremors , dystonic tremors or other
optimizing data acquisition by identifying sensors to be neurological disorders[5] . Thus , due to overlapping of such
used, their placement and integration strategies. In order symptoms ,a reliable and accurate diagnosis is needed .
to analyze data, machine learning algorithms are used to Moreover, misdiagnosis can lead to improper medications ,
distinguish between tremors and muscle activity that are ineffective treatement , progression of the disease to a more
specific to Parkinson's disease and normal movements complicated stage. Despite these challenges , the accurate
using classification technique. By enabling proactive diagnosis is highly dependent on clinical expertise , specialist
healthcare interventions and customized patient consultation , and also includes previous history and physical
management strategies, the proposed device represents a examination . Neuroimaging and DAT scans have been a
promising tool for the detection of early-stage potential option to diagnose PD[6,13].However, it takes a huge
Parkinson's disease. time for the whole procedure,which can pose serious threats to
patients.
Keywords:- Parkinson’s Disease(PD), Resting Tremor,
Muscle Activity, Gyroscope, Electromyogram(EMG). Methods combining gyroscope and electromyography
(EMG) electrodes[8,9,10] is a subject of growing interests as
I. INTRODUCTION they are readily available , non-invasive , cost-effective
diagnostic tools . Combination of these provides a satisfactory
A tremor is a rhythmic, involuntary , shaking movements accuracy and diagnosis of resting tremor and abnormal muscle
of the body . Resting tremor is a type of tremor that occurs activity associated with PD . However , accuracy is not enough
mainly at rest and not specifically engaged in any movement. for validation of the disease . It requires validation from
Resting tremor is a very important symptom of physician or a clinical expert . In this work , we proposed of
PD[1,4].Worldwide, an estimated 7–10 million people are creating a smart wearable device by detecting tremor , muscle
living with PD and its occurrence increases with age, being rare activity associated with PD using gyroscope and EMG
before age 50 and more common in men than in women[2,3]. analysis. Recently, the application of machine learning
The occurrence of PD increases with age and PD affects 1% of techniques has increased in healthcare monitoring systems.
the population above 60 years of age. It typically involves Here , we have utilized SVM – based classifier to classify
movement of limbs , wrist or other parts of body. In most cases, resting tremor associated with PD and abnormal muscle
it operates at a frequency between 2 and 6 Hz. activities. This can be used to accurately predict the occurance
of the disease and can take actions accordingly.

IJISRT24AUG995 www.ijisrt.com 1436

Volume 9, Issue 8, August – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24AUG995

Fig. 1. Project Running Condition ( Initally with Scattered Components)

II. METHODOLOGY

Our work comprises of the following steps : (1)hardware circuitry , (2)data acquisition , (3)data preprocessing ,(4)feature
extraction , (5)dimensionality reduction and (6)classification . The schematic diagrams are provided below.

A. Hardware Circuitry
In this work , we used the following components : (1)arduino uno which acted as central processing unit and control interface
, (2)EMG sensor and electrodes which helped in capturing muscle activity data , (3)Gyroscope sensor(GYROSCOPE) which helped
detecting resting tremor and abnormal movements associated with PD , (4)buzzer which helped for real time monitoring feedback
based on symptom detection ,(5)LCD which was used to display real-time data and system status. And finally used machine learning
model on arduino to classify and analyze sensor data for detecting PD . User interacts with the device through a push button to start
the monitoring process and upon pressing the button the device starts the data acquisition from EMG sensor and gyroscope .

Fig. 2.Basic Block Diagram

IJISRT24AUG995 www.ijisrt.com 1437

Volume 9, Issue 8, August – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24AUG995

Fig. 3. Block Diagram of our Classification System

B. Data Acquisition D. Feature Extraction

We have taken data from 81 patients among whom (1)36 We used autoencoder technique for feature extraction
normal patients who was not diagnosed with any resting which extracts relevant features from the sensor data . We have
tremor or abnormal muscle activity , (2)28 patients who were considered many layers for encoding and decoding . Mostly
diagnosed with abnormal muscle activities but not tremor and we have use rectifier linear unit(ReLU) as the activation
finally(3)17 patients who were diagnosed with both resting functions rather than sigmoid for autoencoder as it gives better
tremor and abnormal muscle activities . For (3) many were classification results . In the majority of tremor related
detected with early-stage PD and for (2) some were diagnosed patients, one side is more affected by the disease than the full
with PD not essentially having tremor . Data was collected body . Feature extraction is performed based on the most
through EMG sensor and gyroscope . EMG sensor affected side of the patients in order to obtain more descriptive
continuously read EMG signals from muscles and also features.
converted analog signals to digital for processing . Through
gyroscope we collected data which detected resting tremor E. Dimensionality Reduction
associated with PD. Dimensionality reduction was performed to reduce the
dimensionality of the feature vector to avoid any interrupts in
The sensors were placed on different parts of the body classification step . This is important as we have higher
particularly on wrist , limbs . The subjects were kept at resting number of features and samples than the number of subjects .
position . Prior to sensor placement , the wrists and limbs were We achieved this by using principal component
cleaned with ethanol and cotton . Three 30-seconds tests were analysis(PCA)[14]. PCA actually projects the feature space
performed for each patient. To discard artifacts and transitions into principal components in the direction of maximum
from the recorded data, we considered only the signal from variance. These new components make a feature space with
10 to 25 seconds. reduced dimensionality. In our work, the number of
components was optimized empirically in order to achieve a
C. Preprocessing and Splitting high classification performance.
Raw sensor data went for a preprocessing to remove
artifacts and high frequency components using low pass F. Classification
butterworth filter . Signals were filtered and normalised to For this work we have trained binary support vector
enhance the symptom detection and model accuracy. We have machines (SVM)[15,20] classifier for the 2 classes : resting
split the data into test and train labels according to the sensor tremor and muscle activity . We used two kernels : radial basis
data and symptoms we were measuring for PD . function(RBF) and linear for the classification task . SVM
consists of a cost parameter, which controls number of
misclassifications of training examples. The RBF kernel has

IJISRT24AUG995 www.ijisrt.com 1438

Volume 9, Issue 8, August – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24AUG995

an additional parameter known as gamma, which controls how individual classification using SVM. To evaluate the
far the influence of a single training example reaches. We did performance we found out the accuracy[7] , and the
a grid-search as a method of model selection to adjust the SVM performance metrics precision, recall , F1 score as in the
parameters. The range of gamma was { 1, 0.1, 0.01, 0.001}. classification report .
For better understanding of our device we also employed

Table 1 Patient Characteristics

Details Normal RT and MA MA
Number 36 17 28
Age(range) 30-70 50-80 35-65
Gender M-18 ; F-18 M-12 ; F-5 M-16 ; F-12
RT – Resting tremor , MA – Muscle activity, M – Male,
F- Female

III. RESULT
IV. FUTURE ENHANCEMENTS AND
Accuracy was computed to check the performance of our CONCLUSION
device. As mentioned earlier we computed both individual and
overall accuracy of our method. For individual computations , This work provides potential benefits by early detection
the best result came for muscle activity detection from EMG of PD enabling timely intervention . This helps the clinical
sensor of about 98% and for resting tremor detection we got professionals to track the disease progression and improve
accuracy approximately about 80% . The overall accuracy of treatment efficiency. This device can be enhanced by
our method to detect PD is approximately about 80%. Talking incorporating more advanced sensors such as heart monitoring
about the performance metrics (in individual cases) is given in sensors , temperature sensors to detect more symptoms related
TABLE II. We conducted tests under diverse environmental to PD that can provide more details about this disease and can
conditions and also measured response time from symptom be more helpful for medical experts to diagnose PD[11,12] .
onset to device alert activation that is of buzzer and LCD update More enhancements can be done by refining of machine
displaying the results. During the work we encountered with learning (ML) algorithms by exploring more complex
various challenges , more particularly EMG sensor readings symptom recognition and classification[16,17] . Future
were initially affected by electrical noise and required adequate developments and research could focus on remote monitoring
filtering techniques to enhance signal-to-noise ratio , and for and telemedicine integration by securely transmitting data to
gyroscope sensor which measures the angular rate that providers of healthcare . This also can provide personalized
indicates changes in the orientation or rotation speed of the treatment adjustments through real-time data thereby helping
device that can further detect tremor activity and can be in dosage management and required medications. Clinical
quantified. For this data, due to drift over time , it needed trials and research could be done by standardizing tremor
periodic calibrations to maintain the accuracy. We have taken related metrics and providing high-resolution data for
the gyroscope data in terms of degrees/seconds which we evaluating and introducing new therapies .
converted further into hertz by using the formula :
Currently , wearable technology is very important in
(𝑎𝑛𝑔𝑢𝑙𝑎𝑟 𝑣𝑒𝑙𝑜𝑐𝑖𝑡𝑦) healthcare monitoring particularly for chronic disease like PD
𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦(𝐻𝑧) = [12,18]. Early detection of PD and timely monitoring is helpful
360
for both patients and clinical professionals for improvement
We have taken the EMG sensor data in terms of and treatment efficiency . This can potentially improve patient
microvolts(µV). This represents the amplitude of the electrical outcomes and quality of life. Further , enhancing the device’s
potential produced by the muscle contraction. We also utility in both clinical practice and research can help paving
performed model optimizations by tuning machine learning the way for more effective management of PD.
hyperparameters[19] to enhance model performance and
reduce computational load on Arduino Uno.

TABLES

TABLE 2 DATA ACQUISITION T ABLE

gx gy gz EMG value Resting tremor detected Muscle Activity
detected
504 328 90 71 0 0
495 203 87 779 0 1
497 172 85 670 0 1
1713 497 -596 716 1 1
607 -1670 -17 723 1 1

IJISRT24AUG995 www.ijisrt.com 1439

Volume 9, Issue 8, August – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24AUG995

478 165 107 564 0 1

295 292 78 91 0 0
-418 314 -69 58 0 0
511 138 105 81 0 0
493 181 88 778 0 1
1817 -1217 283 716 1 1
533 210 112 733 0 1
510 205 111 780 0 1
-657 -69 566 573 0 1
-627 -1707 456 657 1 1
496 227 84 352 0 0
519 184 89 128 0 0
628 182 114 438 0 0
-987 1564 1648 586 1 1
‘0’ – Not Detected ; ‘1’ – Detected ; gx , gy , gz : Gyroscope data

TABLE 3 EVALUATION TABLE

Classification Accuracy Precision Recall F1 score
Resting tremor 0.80 0.80 0.95 0.82
Muscle Activity 0.98 0.98 1.00 0.99

ACKNOWLEDGMENT [8]. Palmes P., Ang W.T., Widjaja F., Tan L.C., Au W.-L.
Pattern Mining of Multichannel sEMG for Tremor
We would like to thank all participants of this study as Classification. IEEE Trans. Biomed. Eng.
well as the staff helping with data collection. 2010;57:2795–2805. doi:
10.1109/TBME.2010.2076810.
REFERENCES [9]. Hssayeni M.D., Jimenez-Shahed J., Burack M.A.,
Ghoraani B. Wearable Sensors for Estimation of
[1]. Prasad R., Babu S., Siddaiah N., Rao K. A review on Parkinsonian Tremor Severity during Free Body
techniques for diagnosing and monitoring patients with Movements. Sensors. 2019;19:4215. doi:
parkinson’s disease. J. Biosens. Bioelectron. 2016;7:2. 10.3390/s19194215.
[2]. Reeve A., Simcox E., Turnbull D. Ageing and [10]. Mailankody P., Thennarasu K., Nagaraju B.C., Yadav R.,
Parkinson’s disease: Why is advancing age the biggest Pal P.K. Re-emergent tremor in Parkinson’s disease: A
risk factor? Ageing Res. Rev. 2014;14:19–30. doi: clinical and electromyographic study. J. Neurol. Sci.
10.1016/j.arr.2014.01.004. 2016;366:33–36. doi: 10.1016/j.jns.2016.04.041.
[3]. Wirdefeldt K., Adami H.-O., Cole P., Trichopoulos D., [11]. Rovini E., Maremmani C., Cavallo F. How Wearable
Mandel J. Epidemiology and etiology of Parkinson’s Sensors Can Support Parkinson’s Disease Diagnosis and
disease: A review of the evidence. Eur. J. Epidemiol. Treatment: A Systematic Review. Front. Neurosci.
2011;26:1–58. doi: 10.1007/s10654-011-9581-6. 2017;11:555. doi: 10.3389/fnins.2017.00555.
[4]. Sveinbjornsdottir S. The clinical symptoms of [12]. López-Blanco R., Velasco M.A., Méndez-Guerrero A.,
Parkinson’s disease. J. Neurochem. 2016;139:318–324. Romero J.P., Del Castillo M.D., Serrano J.I., Rocon E.,
doi: 10.1111/jnc.13691. Benito-León J. Smartwatch for the analysis of rest tremor
[5]. D. Hirtz, D. J. Thurman , K. Gwinn-Hardy, M. in patients with Parkinson’s disease. J. Neurol. Sci.
Mohamed, A. R. Chaudhuri, and R. Zalutsky, "How 2019;401:37–42. doi: 10.1016/j.jns.2019.04.011.
common are the "common" neurologic disorders?," [13]. K. Badiavas, E. Molyvda, I. Iakovou, M. Tsolaki, K.
Neurology, vol. 68, no. 5, pp. 326-337, 2007. Psarrakos, and N. Karatzas, "SPECT imaging evaluation
[6]. K. Badiavas, E. Molyvda, I. Iakovou, M. Tsolaki, K. in movement disorders: far beyond visual assessment,"
Psarrakos, and N. Karatzas, "SPECT imaging evaluation vol. 38, no. 4, pp. 764-773, 2011.
in movement disorders: far beyond visual assessment," [14]. S. Wold, K. Esbensen, and P. Geladi, "Principal
vol. 38, no. 4, pp. 764-773, 2011. component analysis," Chemometrics and Intelligent
[7]. A. J. Hughes, S. E. Daniel, L. Kilford, and A. J. Lees, Laboratory Systems, vol. 2, no. 1-3, pp. 37-52, 1987.
"Accuracy of clinical diagnosis of idiopathic Parkinson's [15]. M. A. Hearst, S. T. Dumais, E. Osman, J. Platt, and B.
disease: a clinico pathological study of 100 cases," Scholkopf, "Support vector machines," IEEE Intelligent
Journal of Neurology, Neurosurgery, and Psychiatry, Systems and their Applications, vol. 13, no. 4, pp. 18 -
vol. 55, no. 3, pp. 181–184, 1992. 28, 1998

IJISRT24AUG995 www.ijisrt.com 1440

Volume 9, Issue 8, August – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24AUG995

[16]. Kubota K.J., Chen J.A., Little M.A. Machine learning for
large-scale wearable sensor data in Parkinson’s disease:
Concepts, promises, pitfalls, and futures. Mov. Disord.
2016;31:1314–1326. doi: 10.1002/mds.26693.
[17]. Maetzler W. A clinical view on the development of
technology-based tools in managing Parkinson’s disease.
Mov. Disord. 2016;31:1263–1271. doi:
10.1002/mds.26673.
[18]. Matias R., Paixão V., Bouça R., Ferreira J.J. A
Perspective on Wearable Sensor Measurements and Data
Science for Parkinson’s Disease. Front. Neurol.
2017;8:8. doi: 10.3389/fneur.2017.00677.
[19]. Li L., Jamieson K., Desalvo G., Rostamizadeh A.,
Talwalkar A. Hyperband: A Novel Bandit-Based
Approach to Hyperparameter Optimization. arXiv.
20181603.06560
[20]. Cortes C., Vapnik V. Support-vector networks. Mach.
Learn. 1995;20:273–297. doi: 10.1007/BF00994018.

IJISRT24AUG995 www.ijisrt.com 1441