A System for Age-related Changes Focusing on EMG Co-contraction in Upper Limb for Geriatric Applications
Abstract— The paper\’s aim is to design a system and focusing on exploring EMG measurements, specifically co-contraction between biceps and triceps of participants\’ upper limb that may correspond with a decline in physiological reserves. This could serve as a surrogate marker of frailty in the elderly that may be clinically easy to obtain. First, we design a system consisting of various sensors which could be used for detecting frailty. Then we propose to recruit a cross-sectional sample and assess upper limb EMG co-contraction and dexterity across 2 population groups: 37 young participants and 38 elderly participants living in the community. We obtain participants\’ EMG co-contraction using electromyography (EMG) sensors in addition to motion recording using IMU sensors and clinical dexterity assessment tools such as the “Box and Block” test. Patients will interact with a computer program while wearing these sensors. The main focus of this paper is on EMG co-contraction parameters across the population groups to explore the feasibility of using EMG parameter metrics that may indicate frailty in the elderly.
The number of elderly people is increasing at an unprecedented pace worldwide. With regard to this global aging phenomenon, how to maintain or improve the autonomy and wellbeing of the elderly has become a crucial research topic. Generally speaking, healthy old people tend to have preserved activities of daily living, good cognitive and social status. In the literature, some studies have also identified several factors related to being an autonomous old and healthy person, such as good visual acuity, regular exercise, spontaneous awakening in the morning, preserved mastication, no history of drinking alcohol, no severe falls, frequent protein intake, living at home, etc. To the contrast, some opposite factors, especially a low level of exercise, a tendency to fall, and low protein intake are some composite factors that could hinder successful ageing.
The clinical characteristics of older people may exhibit anorexia, sarcopenia, osteoporosis, fatigue, risk of falls, and poor physical health. Elderly people also tend to be highly vulnerable to adverse health outcomes, such as disability, dependency, need for long-term care and even death in extreme cases. Aging may reflect declines in the molecular, cellular and physiological systems of the aged body. Elderly people also tend to have reduced stress tolerance because of decreased physiological reserves in the muscles, bones, circulation, hormone and immune systems. The reasons for such physiological degradation include genetic and acquired factors. The natural course of aging is progressive rather than abrupt, increasing the risk of comorbidity and disability over time. The increasing frequency of obesity in elderly people further complicates the clinical diagnosis. Two issues are important clinically with regard to aging. First, identification of the causes of aging and its association with chronic inflammation and vascular disease should be performed. Secondly, the possibilities for prevention and their effectiveness need to be established. For effective prevention and subsequent treatment of aging in elderly people, some parameters must be recognized and interventions need to start early.
Aging will also tend to cause frailty. Frailty has been identified as a clinical syndrome in the elderly that is distinctly different from comorbidity and disability. Frail elderly tend to have an increased vulnerability of adverse outcomes to minor insults which could result in disproportionate deconditioning [1, 2]. Several models have been proposed in the literature to define frailty. The category of physical strength such as grip strength has been used as an important measure . It is also agreed that targeting pre-frail elderly for successful intervention is crucial since frailty is potentially reversible [3, 4]. Current approaches usually involve questionnaires  which are subjective metrics. In addition, objective measurements have also been studied such as gait speed and step width variability. Such objective measurements have emerged as a promising tool for aging and frailty prediction . However, gait measurements generally require an elaborate and complicated set-up and are not a practical screening tool from the viewpoint of clinical practice. Therefore, easy and convenient detection will be of great help and importance.
As a result, exploring physiological parameters using convenient schemes for old people should be performed for the study of aging issue. Since aging is also reflected in the reduced muscle force exertion, EMG signal changing and dexterity degradation are some phenomena of which aging could possibly exhibit and hence requires more study. In this paper, we first propose a system consisting of various sensors which could be used for detecting frailty. Then we recruit a cross-sectional sample and assess upper limb EMG co-contraction, dexterity and other parameters across two population groups, namely the young and the elderly. Although the whole system is illustrated subsequently in the paper, the analysis is mainly on EMG part since it is the focus of this paper and also one of our targeted metrics for detecting aging or even frailty in clinical trials. The organization of the paper is as follows. Introduction is given in Section I. The overall system and design is illustrated in Section II. EMG signal and co-contraction is introduced in Section III. Section IV explains the trial setup and Section V provides experiment details and the corresponding analysis. Conclusion is given in Section VI.
In this section, we present a system which facilitates our subsequent trial protocol. The block diagram of the system setup is conceptualized in Fig. 1, in which the user wears various sensors and interacts with the computer. Motion sensors are worn on the wrist of the dominant arm of the subject, whilst EMG sensors are worn on the biceps and triceps of the upper arm. EMG sensor is connected wirelessly to the computer via Bluetooth while the motion sensor is connected to the computer visa a proprietary 2.4 GHz wireless protocol. Both the touch pad sensors and the dynamometer, which measures the exerted force, are connected to the computer via wires.
Fig. 1. Conceptual block diagram of system setup
Fig.2 shows the view of real scenario. The subject sits on the chair and wears the motion sensor and the EMG sensors on the right wrist and the right upper arm respectively. The touch pad is put in front him on the desk where a notebook PC is also stationed.
Fig.2 Real view of system setup
The program running on the notebook will collect the information transmitted by the sensors and also provide the visual aid to the subject.
A. Touch Pad Sensor
The touch pad consisting of two buttons, which in this scenario are in red and green color respectively, can be also spotted in Fig. 2. When the hand touches and presses the buttons, the interrupts will be generated and sent to PC which records the precise timing of pressing the buttons . The two buttons are used in order to facilitate recording the timing of movement of hands. The subject moves the hand according to the metronome provided in the software program running on the computer.
B. Motion Sensor
The motion sensor is also termed as a mote in the wireless network which can perform some processing, obtain sensor data and communicate with the gateway node. They use flash memory and onboard memory of the microcontroller. The onboard memory is used for programming the device whilst the FLASH memory or the user memory is used for application related data. As for the MEMs motion chip, Invensense MPU-9250 is used as the 9-axis inertial motion unit (IMU) because of low power consumption and high performance requirements of wearable sensors. It contains a 3-axis accelerometer, a 3-axis gyroscope and a 3-axis compass. There is also one Digital Motion Processor (DMP) built in, which can output the quaternion representation of the rotation movement. The DMP is responsible for converting rotation angles to quaternion representation and refining the accuracies using acceleration information. The refinement is achieved through recursive filtering.
C. Wireless Transceiver
As for the wireless communication of motion sensor, nRF24L01 Transceiver is chosen as the wireless transmission interface in this study due to its low cost and low-power radiofrequency transmission. It is a highly integrated, ultra low power (ULP) 2Mbps RF transceiver IC for the 2.4GHz ISM (Industrial, Scientific and Medical) band. Thus, it can be used in the biomedical field for functions such as healthcare data transmission. The chip is with peak RX/TX currents lower than 14mA, a sub μA power down mode, advanced power management, and a 1.9 to 3.6V supply range. The Enhanced ShockBurst hardware protocol accelerator additionally offloads time critical protocol functions from the application microcontroller. Such characteristics enable the implementation of advanced and robust wireless connectivity. This facilities the powerful wireless transmission scheme in terms of the requirements of our in-house environment. The gateway attached to computer receives the wireless-transmitted data and relays it to the host computer through UART. A Visual Basic program is designed and coded to collect the data packets from the gateway.
D. EMG Sensor and Amplifier
Such surface EMG sensors and instrumentation are used extensively in the fields of ergonomics, sports science and medical research. The SX230 brand we used is the precision bipolar, differential EMG sensor having integral electrodes with a fixed electrode distance of 20mm, which can provide sufficient quality of signal and ease of use. Since the input impedance is great than 10,000,000 M Ohms, it indicates that little or no skin preparation is required in practices. In addition, conducting gels are not needed due to this characteristic. The electrode can be just applied to the muscle using the die cut medical grade double sided adhesive tape. The EMG sensor is connected to the portable DataLOG system, which is a general purpose data amplifier and may collect signals from a wide range of sensors simultaneously with EMG. DataLOG, captures and transfers data in real time to the computer via Bluetooth for storage.
With regard to waveforms, EMG signals are fundamentally bursts of random signals with certain characteristics that are able to indicate timing, force and fatigue. After acquiring a waveform, we remove unwanted sections of the trace. Then, the waveforms are converted to RMS signal to obtain envelope of the signal for the purpose of subsequent processing. Specifically, each sample of data is first squared and then a moving average is taken of these squares. The output is the square root of each average calculated.
Dynamometer is used for measuring the grip strength . The subject uses the hand of the dominant arm to hold the device as hard as he or she can. The force exerted can be then measured and stored in the computer. Strong people usually are expected to have stronger muscles and thus more force exerted and measured.
Base on the hardware set-up of the system, we design a trial protocol, which includes a well-established active-daily-life(ADL) hand function test, the Box and Block test. It is similar to gait analysis in that the movement is repetitive and requires coordination and cognition. Upper limb movement is also evaluated in the Box and Block test. Besides, in the test the subjects are invited to participate in a research study examining the use of IMU and EMG sensors. The motion data is used to capture the upper limb, specifically the speed, angles of the upper arm and lower arm movement. EMG sensors are used as non-invasive body sensors to record electrical activity of the muscles. In addition to IMU and EMG, touch pad sensors are also used, as illustrated in Fig. 2. All these sensors are used for the purpose of screening for people who may be frail and at risk for health decline. In other words, the aim is to explore whether we can infer some information related to aging by using these devices and if yes to what extent the information can be inferred. In subsequent sections, the focus is on analyzing EMG co-contraction of the subjects when pressing touch pad sensors. Clinically, it is a more convenient approach to gait analysis. The EMG information is collected and analyzed to explore if some patterns exist for different group of subjects. In particular, what interests us is to explore whether this information reflects whether the subject is aging/frail or not. In summary, by designing this test and adding technological adjuncts such as touch sensors, we hope to propose a compact and sensitive method for predicting frailty.
EMG Signal and Co-Contraction
A. Raw EMG Signal
EMG, abbreviation of electromyography, is used to analyze muscle strength. It is commonly used in determining neuromuscular diseases in people, such as stroke and auto-immune disorders. It can also be applied in other applications, which include physiotherapy, and strength training. Raw EMG signal shows positive and negative fluctuations with regard to time.
The EMG signal helps to observe the muscle’s state and movement, by measuring the muscular performance of an individual. Usually when a doctor wants to decide the procedure before and after some surgery, they can rely on EMG to help them for decision. It also helps the doctor to document treatment and training programs for their patients. The patients can then understand their current muscle condition and take measures to train on their strength. With regard to the subjects in our trial, the EMG sensors will help us to gauge their muscle performance.
B. EMG Signal Rectification
Since the raw EMG signal has both positive and negative components, the signal has to be rectified in order for further processing to calculate energy spent. Full wave rectification is performed, by eliminating the signal’s negative portion, through converting the value to positive.
Since the random environmental noise can interfere with the raw EMG data, a moving average digital filter can be applied.
y[n]=1/M ∑_(k=0)^(M-1)▒〖x[n-k]〗 (2)
Where x[ ] is the input signal, y[ ] is the output signal, and M is number of average points. For example if M is 5, an equation of a five point moving average filter is given as
The moving average filter is a FIR (finite impulse response) low pass filter, which is effective in reducing random noise and other high frequencies and maintaining a smooth step response.
D. Root Mean Square
Other schemes can also be used to remove additive random noise from the EMG signal. The aim is to obtain a smooth linear envelope of the EMG signal in the time domain. Root Mean Square of a signal is used to find envelope of the random EMG signal, to make the signal values more precise and meaningful in our case. Sliding Window is used to time step across the signal, one time step at a time, to remove the Gaussian noise while retaining the original signal values. The graph below shows how the shape of EMG signal envelope is obtained by RMS scheme.
Co-contraction refers to the tendency of various muscle groups to become activated simultaneously. It is commonly expressed as percentage for normalized SEMG. This co-activation is called a co-contraction. In the literature, there are a number of approaches measuring co-contraction. E.g, in , maximum isometric tests were performed on a customized dynamometer consisting of a strength chair and a force transducer attached on the dynamometer lever-arm. The EMG raw data were full-wave rectified and low pass filtered at 6 Hz, yielding the linear envelopes of each muscle EMG. The EMG of each muscle was then expressed as a percentage of the EMG value during the MVC. Co-contraction index was calculated by finding the overlap between the agonist and antagonist curves.
One way to compute co-contraction is by calculating the value of the integral of the enveloped EMG signal of each of the muscles and as follows. The envelope is calculated using root mean square (RMS). Co-contraction is calculated as the percentage of co-contraction between the two antagonistic muscles. As illustrated in Fig. 3, area A is the area under the enveloped EMG A curve. Area B is the area under the enveloped EMG B curve. The common area is the common area of activity between two antagonistic muscles. And co-contraction is calculated as the follows:
cocon=2(common area of A and B)/(area A+area B) (4)
where cocon is the percentage of co-contraction between the two different muscles. In subsequent sections, the co-contraction is calculated using this approach.
Fig. 3 Co-contraction of two EMG signals
The aim of the research targets on differences in EMG co-contraction of biceps and triceps of upper arm movements between young and elderly people. Namely, the subjects who participate in the whole trial process are grouped into two general categories, namely Young and Elderly. A total number of 75 people have been recruited in the trial. The category of Young includes 37 male/female subjects aging between 21 and 40, and they are mostly junior staff or university/polytechnic students. The category of Elderly consists of a total number of 38 people who are over 65 years old. Participants with a history of stroke or other known disease that could affect hand motor function are excluded. Professional musicians are also excluded. In addition, the following requirements are applied, such as being able to understand and follow instructions, no previous known illness causing weakness of the arms, not admitted to hospital in the past year, able to travel on his or her own and take care of self dressing, bathing and hygiene. They are mostly recruited from the community or recommended through acquaintances. The participants are also required to read the form “Participant Information Sheet” and sign to be accepted as “Consent By Research Subject”. The subject can also choose to opt out, which however may not be encouraged for the purpose of research of this topic. With regard to ethical issues, the study is not invasive and does not include the obtaining of blood or tissue samples. Two EMG sensors are applied to the subject’s dominant upper arm, specifically, one on biceps and the other on triceps. The IMU sensor which could detect movement of the upper arm is worn like a watch on the wrist of the dominant arm.
The assessment protocol involved three parts, questionnaires, strength test and tapping task. The questionnaires included a Music Ability Assessment, the Fried Phenotype , and the Vulnerable Elders Survey . The strength tests consisted of grip strength measurements and the Box and Block test. The Precision Dynanomometer G2000 was used for the grip strength test at the standard second handle position. For the grip strength test, the dominant arm held the grip dynamometer perpendicular to the ground with the elbow flexed at 90 degree, and the shoulder abducted at 10 degree. The participants were instructed to grip as fast as they could and as hard as they could for 5 seconds. This was repeated three times with a mandatory rest time of 1 minute between each test. The box and block test was administered using the standard established protocol . As for the tapping task, participants were seated at a desk in front of the tapping apparatus. The tapping apparatus consisted of two buttons spaced 43.4 cm apart (measured from the center of the buttons) which were separated by a vertical divider as shown in Fig. 2. The buttons were 5cm in diameter. A custom made program produced the sound metronome and collected the response from the buttons which could generate temporal resolution of 20ms.
Participants were tested individually with the experimenter beside them. Demographic information was obtained before administrating tapping test. Then participants were instructed to anticipate and synchronize their hand taps on the buttons as precisely as possible with the sound stimuli, alternating buttons with each tap by bringing their dominant hand over the vertical divider. The divider was centralized to the body of the participant. The distance between the apparatus and the body of the participant was determined by aligning the wrist of the extended dominant arm to the ipsilateral button. This distance produced a slight flexion of the elbow when the patient tapped the button using their fingers.
The tapping tests consisted of tests isochronously at various beats per minute (BPM). Each test duration was 30 seconds, with a mandatory rest of at least a minute between each test. The order of the tests was such that they were administered at 50, 65 and then 80 BPM. After the trial, the participants were given a mandatory rest time where they completed the questionnaires. Before introducing a new test, participants watched a demonstration and performed a trial run.
The trials collected various types of data. And our analysis focuses on EMG data, which are recorded in different plain text files. For ease of organization, all the files related to one subject are stored in the corresponding directory. To make the analysis, programs have been coded in Matlab language which recursively search the relevant files under the test data directory to process the collected data. Co-contraction of the EMG is calculated whilst the mean and standard deviation are the measurement for the analysis. The results are visualized in Matlab figures.
Fig. 4 Co-contraction results for elderly, all subjects combined
Fig. 4 shows the co-contraction of Elderly at 50, 65 and 80 BPM respectively. The X-axis shows the subject number and Y-Axis shows the co-contraction between biceps and triceps in terms of percentage. From the chart, it shows that there exist big differences of co-contraction among various elderly subjects, which indicate the co-contraction could range between 19.08% and 72.28% for 50 BPM, between 20.43% and 73.85% for 65 BPM, between 20.06% and 74.39% for 80 BPM respectively. From the chart, it shows that there are small variations for the same subject at different speeds. It should be noted that the data related to subject 16 is recorded incorrectly and set to null, which could be attributed to the loose contact of EMG sensor.
Fig. 5 Co-contraction results for the young, all subjects combined
Fig. 5 shows the co-contraction of Young at 50, 65 and 80 BPM respectively. The X-axis shows the subject number and Y-Axis shows the co-contraction between biceps and triceps in terms of percentage. Similar patterns as Elderly group could also be observed. The chart shows that there also exist big differences of co-contraction among various young subjects, which indicate the co-contraction could range between 23.41% and 88.73% for 50 BPM, between 28.09% and 80.06% for 65 BPM, between 31.07% and 84.83% for 80 BPM respectively. Similarly, it demonstrates that small variations exist for the same subject at different speeds.
Fig.6 Elderly vs. Young (averaging all subjects)
Fig. 6 and Table 1 shows side-by-side comparison between elderly and young on three speeds. It shows that elderly group has lower co-contraction mean than Young group. It may indicate that the bicep and triceps of elderly have degraded co-contraction performance due to aging. The standard deviations of the two groups are comparable. At 50 BPM, the contraction has the highest mean among the three speeds for both groups, while the contractions at 65 BPM and 80 BPM have similar means. One explanation could be that the slower speed movement makes the co-ordination of biceps and triceps function at a better degree than higher speeds.
The comparison is also analyzed using ANOVA as shown in Table 2. The ANOVA table shows the between-groups variation and within-groups variation. SS is the sum of squares, and df is the degrees of freedom. The total degree of freedom is total number of observations minus one, which is 12 – 1 = 11. The between-groups degree of freedom is number of groups minus one, which is 2 – 1 = 1. The within-groups degree of freedom is total degrees of freedom minus the between groups degrees of freedom, which is 11 – 1 = 10. MS is the mean squared error, which is SS/df for each source of variation. The F-statistic is the ratio of the mean squared errors (75.4065/3.239). The p-value is the probability that the test statistic can take a value greater than or equal to the value of the test statistic, i.e., P(F > 23.2807). The small p-value of 0.00069698 indicates that differences between the group means are significant.
Source SS df MS F Prob>F
Columns 75.4065 1 75.4065 23.2807 0.00069698
Error 32.3902 10 3.239
Total 107.7967 11
Table 2. ANOVA Table
Generally, the Young group has higher average co-contraction level than Elderly group at the 3 speeds. The co-contraction level varies significantly for each individual subject at the same speed (max difference is 50%). For each subject, the co-contraction difference over various speed is between 1% and 20%. When speed increases, the co-contraction for a subject may increase or decrease.
We propose a system which could help decide aging status and for that purpose the subjects are invited to participate in a research study examining the use of Inertial Measurement Units (IMU) and Electromyography (EMG) sensors to screen for people who may be frail and at risk for health decline. The aim is to explore whether we can indicate frailty using the timing differences corresponding to various metronome. We propose to recruit a cross-sectional sample and assess upper limb EMG co-contraction and dexterity across 2 population groups: 37 young participants and 38 elderly participants living in the community. We obtain participants\’ EMG co-contraction using electromyography (EMG) sensors in addition to motion recording using IMU sensors and clinical dexterity assessment tools such as the “Box and Block” test. The statistical analysis demonstrates that elder group and young group exhibit different co-contraction characteristics which help us explore the feasibility of frailty indication in the elderly.
The authors would like to thank comments and advice from fellow colleagues who make the improvement of this document possible.
Clegg A., et al., “Frailty in elderly people,” The Lancet, 2013. 381(9868): p. 752-762.
Morley J.E., et al., “Frailty consensus: a call to action,” J Am Med Dir Assoc, 2013. 14(6): p. 392-7.
Ruiz M., Cefalu C., and Reske T., “Frailty syndrome in geriatric medicine,” The American Journal of the Medical Sciences, 2012. 344(5): p. 395-398 10.1097.
Chen X., Mao G., and Leng S.X., “Frailty syndrome: an overview,” Clin Interv Aging, 2014. 9: p. 433-41.
Schwenk M., et al., “Frailty and technology: a systematic review of gait analysis in those with frailty,” Gerontology, 2014. 60(1): p. 79-89.
Usbbuttons, USB Button Applications, http://usbbuttons.com/applications/, accessed Dec. 10, 2015.
InvenSense, MPU-9250 Datasheet, http://www.invensense.com/products/motion-tracking/9-axis/mpu-9250/, accessed Dec. 20, 2015.
Nordic Semiconductor, nRF24L01 Ultra low power 2.4GHz RF Transceiver IC, http://www.nordicsemi.com/eng/Products/2.4GHz-RF/nRF24L01, accessed Dec. 21, 2015.
Biometrics, Surface EMG, http://www.biometricsltd.com/emg-measurements.html , accessed Dec. 23, 2015.
Precision, Dynamometer G2000, http://www.biometricsltd.com/myogrippinch.htm, accessed Dec. 28, 2015.
Kellis E., Arabatzi F. and Papadopoulos C., “Muscle co-activation around the knee in drop jumping using the co-contraction index, ”Journal of Electromyography and Kinesiology 13 (2003), 229–238.
...(download the rest of the essay above)