Closed-loop system for assisted strength exercising Tomislav Pozaić*, Matija Varga*, Dominik Džaja* i Sara Žulj* *
University of Zagreb, Faculty of Electrical Engineering and Computing / Dept. of Electronic Systems and Information Processing, Zagreb, Croatia
[email protected]
Summary - Accurate assessment of the qualitative and quantitative aspects of physical activity is considered very important in order to achieve better results in rehabilitation or training. We have developed Wireless Body Area Network for continuous and personalized health monitoring of basic physiological parameters (ECG, breathing rate and body temperature) and added the functionality of mobility measurement and closed-loop assisted strength exercising. The results has shown that assisted exercising using on-line feedback is performed closer to prescribed instructions in time and intensity while self-assessment and exercising without on-line feedback vary much more showing a higher deviation from the prescribed instructions. Our system enables sophisticated tracking of the rehabilitation/exercise performances of patients with closed-loop quantitative and qualitative results in real-time.
I.
INTRODUCTION
Broad range of therapy approaches are currently practiced in clinics. These therapies differ significantly due to type of the disease and severity of the damage the acute disease caused to the patient. Therapists and physicians are confronted with difficult task of selecting optimal at home rehabilitation programs for their patients. Impossibility to do more frequent health monitoring is particularly difficult during rehabilitation. When the rehabilitation is not successful medical specialist can not explain what exactly has happened because they usually see patients when major complications have already occured. Moreover, dramatically aging population makes it absolutely necessary to develop systems that keep people out of hospitals, but with an appropriate medical care. From previous systematic reviews and meta-analysis there is a consensus about positive effect of exercise training on health-related quality of life, mortality and hospital admissions [1]. However, the effects of exercise within self-management programs remain unclear or gave negative feedback. According to the impending problems in healthcare, most important challenge for every system that offers assistance in rehabilitation is to give an accurate feedback on continuous physical activity for patients and their therapists [2]. Closed-loop exercising enables efficient continuous recording of physiological parameters and real-time assessment of the quantity and quality aspects of physical activity. Efficient closed-loop assessment is
considered very important in order to achieve full potential of rehabilitation or training [3]. Questionnaires, surveys and diaries of self-control have disadvantages such as inaccurate perception and recall of information by the subject, unsuitable design and delay [4]. Important limitations for wider acceptance of the existing systems for continuos monitoring are: lack of system integration of invidual sensors, interference on a wireless communication channel shared by multiple devices, nonexistent support for massive data collection and lack of closed-loop support [5, 6]. Miniature, lighweight, ultra-low power, intelligent monitoring devices enabled continuous, accurate and closed-loop monitoring of physiological parameters of all kinds of patient without influencing their usual daily activities. On the other hand, sensor integration and well-structured patients information allows professional staff more flexible treatment and more efficient data processing. Wireless Body Area Network (WBAN) for continuous and personalized health monitoring of basic physiological parameters (ECG, breathing rate and body temperature) was developed at the Department of Electronic Systems and Information Processing at the Faculty of Electrical Engineering and Computing, University of Zagreb and we added the funcionality of mobility measurement and assisted exercising [7]. In this paper a part of this system which provides closed-loop assisted strength exercising is described. The goal of this system is to enhance the effects of rehabilitation and training in patients that suffer from different types of chronic diseases. It can also support early stage detection of abnormal conditions and thus prevent its serious consequences. Additionally, real-time critical health monitoring implemented in this system can provide professionals with life-important information and reduce the risk of further complications. System integrates the knowledge of experts in kinesiology, medicine and engineering. II.
SYSTEM ARCHITECTURE
OZIMS (Osobni Zdrastveni Informacijski i Mobilni Sustav), a personal healthcare information and mobile system, is a wearable systems with tremendous potential in proactive healthcare. System integration, wireless communication and sensor development enabled lightweight, easy-to-use, unobustrive, flexible, non-invasive, personalized health systems. Our system has the ability to communicate wirelessly with a personal
server and subsequently through the Internet with a emergency services or medical and research database. OZIMS is a combination of non-invasive sensing and contextual features such as ECG, EMG, mobility, body temperature, voice and visual features. The whole system can be divided into four layers:
sensor layer (data collection hardware with sensors for continuous measurement of mobility and physical activity)
communication layer (communication hardware for data stream to a PC, tablet or cloud)
processing layer (algorithms for real-time processing of phsyological and movement data)
display layer (graphical user interface (GUI) for display of clinically-relevant information and assisted strength exercising on PC or tablet)
A. Sensor Layer Very important role for sensor applications in the field of rehabilitation are advances in development of microelectromechanical systems (MEMS). MEMS sensors are ultra-low power and lightweight sensors that have been used in motor activity and other health status monitoring systems [8]. Sensor node has three inertial sensor for trunk position measurement, mobility and assessment of physical activity (accelerometer, gyroscope, magnetometer). It also has sensor for body temperature measurement, ECG and EMG amplifier, voice interface and communication module. The node is powered by one rechargeable lithium battery Nokia BL-5B (890 mAh) It can supply power to the node for up to 13 days without recharging. One or more sensor nodes form a Sensor Layer for every patient who is being monitored during rehabilitation or training. For data acquisition closed-loop controlled strength exercising, sensor node is using embedded three axis accelerometer (ADXL345) and three axis gyroscope (L3G4200D). The axes of the coordinate system of inertial sensors are not related to the coordinate system of the Earth but rather, they match with coordinate system with centre in sensor case. The sensitivity of the accelerometer is set to 4 mg/LSB. Although the frequency of recorded signals depends on sensor placement, it has been shown that in 95% cases frequency of human movement is under 10 Hz and amplitude is less than 6g (1g = 9.81 m/s2) [9]. To capture all important signal frequency components and to fulfil ultra-low power demands, the sampling frequency of a 10-bit ADC is set to 100 Hz per channel.
Figure 1. Sensor node block diagram
reliability and broad range of different types of communication packets (23) which allow flexibility and efficiency. Central node in WBAN is communicating with the closest access point in the network (hospital, elderly home, working space, gym etc.) The node has also the option to record the data to MicroSD card in cases when the patient is out of reach of access point and GSM network. Communication module is intended for local networks with low power consumption bellow 5 µA/byte. Module works in the ISM frequency band (2.4 GHz). To significantly reduce the power consumption, node sends only processed and carefully extracted data. In the case when power consumption is not important, it is possible to stream raw sensor's signal to the computer and than make further processing. To allow fully closed-loop exercising, two-way communication is enabled. FlexyNet packets that are exchanged between sensor node and application are shown in Figure 2. At the beginning of exercising PC or tablet application sends packet to the sensor node that have information about patients ID, number of exercises and necessary information about every exercise (exercise ID, pause duration, dominant axis and number of repetition in each series). At the end of every exercise, the node sends packet to application with information about patient and
Microcontroller LPC1347 controls all functions of the sensor node. FreeRTOS, a real time operating system, is implemented on the microcontroller. Whole algorithm is divided into tasks and low power modes are used to reduce the consumption. Detailed sensor node block diagram is shown in Figure 1. B. Communication Layer The main protocol used for communication between nodes and PC is flexyNet protocol based on TCP/IP protocol [10]. Its main characteristics are high speed,
Figure 2. Communication packets between application and sensor node
exercise that is performed (exercise ID, quantitative index, repetition duratio, pause duratio, amplitude ratio, fluency of movements and described angles). C. Processing Layer The algorithm for processing of data collected from sensors was implemented on the microcontroller using cascaded integrator-comb filtering, derivating, signal thresholding and peak detecting. The algorithm is performed in real-time and no search-back is used. The result of processing is a feature vector with characteristic values for each type exercise that application user has to perform. Feature vector is a 5-dimensional vector of numeric features of exercise (minimum and maximum amplitude of signal, average duration of one repetition, average duration of pause and described angle during exercising). Feature vector is stored in database and it is used to describe qualitative and quantitative parameters of every strength exercise. Data that was gathered during processing is then transferred to the Display Layer. Algorithm used in processing of signals obtained during exercising can be divided into three major blocks:
Time Management,
Quantity Management,
Quality Management.
For every patient and for each exercise personalized parameters are entered into the exercise plan. Personalized parameters are stored in the database and depend on patient's age, therapy, health conditions etc. Once the patient has entered the assisted exercise program and accepted the exercise plan, at the beginning of each exercise personalized parameters are sent to patient's central sensor node using flexyNet protocol. Time Management block is used for controlling the time aspects of exercising. It is using timeouts to determine whether the patient is exercising as instructed, i.e. according to recommended timing. Thus, outputs of the Time Managment block participate in quality estimation of physical activity. Quantity Management block is used for repetition counting. Its main part is peak detection algorithm which detects each single repetition. Quality Management according to the parameters previously stored in database, then determines the quality of every repetition and it gives the final average score of whole physical activity during specific therapy. D. Display Layer The results of processing are sent to PC application where they are displayed in an intuitive and effective application. The application can, on demand, also analyze raw signal data in real-time from any combination of sensors, acting as a central sensor unit. Along with the display functions, application also provides key information that apply to patient's health and wellness such as: qualitative and quantitative score of physical activity, efficiency and intensity of physical activity, calories burnt, activity schedule and therapy information. It is developed in Microsoft Visual C# 2010 Express as an object oriented user interface so additional upgrades or integrations are easy to perform.
Application's Graphical User Interface (GUI) is particularly adapted for professional and non-professional users (athletes, patients). On user's PC, depending on the status of registered user, GUI is shown as a separate node which is communicating over the Central Server with the database and other nodes. Providing information about physical activity and therapy on any PC and whenever needed, it is very easy to further enhance rehabilitation and training output. III.
METHODS
A. Anthropometric measurement In this experiment anthropometric measurements were conducted for better qualitative and quantitative signal analysis and to achieve the desired level of diagnostic certainty and resolution. Anthropometric measurements are important for better definition of sensor placement on the body, more accurate description and performance of given physical activity and more credible and repeatable statistic score of result of exercising. For every examinee, body mass index was calculated. Optimal position for placing the nodes was defined by analysis of previous papers [11]. Sensor node was placed on the wrist of dominant arm. Results of anthropometric measurement of examinees in this project are shown in Table I. Each individual value is shown along with mean and standard deviation rounded to one decimal place. B. Experiment setup One experiment was set for the investigation of user friendliness, intuitiveness, accuracy, assessment and usefulness of the closed-loop strength exercising application is implemented. A group of ten healthy examinees was formed, comprising of 8 male and 2
TABLE I. ANTHROPOMETRIC MEASUREMENTS Male
Examinee
weight/kg
hight/cm
age/years
1
72
174
23
2
80
182
26
3
70
175
23
4
87
185
23
5
82
180
23
6
75
183
23
7
86
184
24
8
86
181
22
average
79,6
180,5
23,4
SD
6,7
4,0
1,2
Female
Examinee
weight/kg
hight/cm
age/years
1
64
172
26
2
62
168
21
average
63,0
170,0
23,5
SD
1,4
2,8
3,5
female examinees. Before exercising, every examinee signed the Informed Consent for Physical Activity. Experiment was divided into two steps. In the first step, examinees were exercising following written instructions and without the assistance of our application. Signals were recorded on MicroSD card so further processing and feature extraction in OZIMS application would be possible. Every examinee got a plan of exercising in writing with detailed description of every exercise (number of repetition, number of series, recommended duration of each repetition, recommended duration of each pause, initial and ending extremity and/or body position). They had the assignment to count repetitions of prescribed movements and to estimate the time, effort and efficiency they have spent exercising. The aim was to be as close as possible to the default repetition and pause duration and to do exact number of prescribed repetitions and series. Examinees filled the form which was used for their self-assessment and which was compared with the recorded signals. In the second step examinees had auditory and visual assistance during exercising and all information about every exercise have been displayed in application in real-time. Signal from the sensor node was streamed to application where it was further processed and where feature extraction was performed so examinee had real-time estimation of physical activity during exercising. Quantitative and qualitative parameters of recorded signals from first and second step were compared. After finishing the prescribed exercising program, the examinees overall satisfaction with application was tested by filling a short questionnaire. Results are shown in the next chapter of the paper. In both phases of the experiment, all examinees were performing two different dumbbell exercises (Figure 3) which were chosen based on professional advice of medical staff [12]. Every exercise had three series of repetition with pause of 15 seconds after each series. In the first series the examinees had to perform six repetitions, in second they had to perform five repetitions and in the third four repetitions. The duration of each single repetition was set to 2 seconds. Exercises were performed with dumbbells having a mass of 1 kg. The target group of muscles were arm muscles. We hypothesize that assisted exercising using on-line feedback through OZIMS application will be performed closer to prescribed instructions in time and intensity while exercising performance results inter and intra-examinees without on-line feedback as well as the
Figure 3. Dumbbell exercises
results of the following self-assessment will vary much more showing a higher deviation from the prescribed instructions. C. Signal processing In both experiment stages signals where processed within a desktop application. In the first, self-assessment stage of this experiment, signals were recorded on the MicroSD card and later fed to desktop application with the aid of data generator form. The form was used to simulate real-time signal. In the second stage signals were processed in real-time within the desktop application. Signal processing is applied on one axis off acceleration signal with the highest amplitudes (dominant axis). The first stage of signal processing was filtering and decimation. The most convenient solution to this problem was designing cascaded integrator-comb filter. We have implemented 8 sections of CIC decimator with the decimation factor of 5, thus sampling frequency at the output was 20 Hz. The second stage of signal processing involved edge detection. Each exercise has predefined parameters that are loaded from the database. One of the parameters is derivation thresholds, as well as signal thresholds at which derivations should occur. Two pairs of derivation and signal thresholds are used, one for the repetition start and one for the repetition end. To eliminate false positive edge detections an approach similar to debouncing was used. If initial edge was detected, the algorithm starts searching for local extreme until the final edge is detected. The initial and the final edge denote single repetition duration. Another local extreme is searched between the final edge and the initial edge of the next repetition. Local extrema are used to qualitatively describe the dominant axis waveform. One repetition is characterized by:
the two local extrema,
repetition duration,
2 pairs of derivation and signal thresholds.
Depending on the workload assigned to the patient and his physical status, different tolerances are applied on the signal and derivation thresholds. This feature makes an algorithm less rigid and more suitable for different types of users. For example, if one person is at the beginning of his rehabilitation and can not follow the prescribed pace of exercising, his fitness level will be saved to database as low. As a consequence, high tolerances will be applied in the second stage of signal processing. At the beginning of one's rehabilitation, even finishing the whole exercise plan is satisfactory rather than following strict rules of exercising. On the other hand, if a person is at the end of his/her rehabilitation, his/her fitness level are expected to have reached a medium or high level, hence the tolerances will be narrowed. Repetitions that are not within the tolerances will be marked as a poor performance.
IV.
A detailed questionnaire regarding exercising with the assistance of the developed application consisted of 3 major questions:
V.
RESULTS
usefulness of the OZIMS system for the assisted strength exercising, user friendliness and intuitive aspect of the OZIMS system, satisfaction with the provided instructions for exercising.
Responses to each question were on 6-point scale where a score 0 indicated that examinee could not finish the exercise, 1 indicated the least satisfaction, while 5 indicated the total satisfaction with the application. The information obtained from these questionnaires is shown in Figure 4 as an average value of all scores rounded to one decimal place. Results of the questionnaire have shown average score of 4.4 for the usefulness of the OZIMS systems for the assisted strength exercising, average score of 4.3 for the user friendliness and intuitive aspect of the OZIMS system and 4.7 for the satisfaction with the provided instructions for exercising. Number of repetitions for first exercise in self-assessment was estimated to (16.2 ± 3.1), number of repetitions for second exercise was estimated to (15.5 ± 1.6), duration of each repetition in self-assessment was estimated to (2.0 ± 0.8) seconds and duration of each pause was estimated to (13.5 ± 2.1) seconds. Using the developed algorithm in application, we processed signals and compared average repetition duration of each type of exercise with and without the support of application for assisted exercising. Repetition duration in first phase of experiment for first exercise was (1.6 ± 0.9) seconds, while for second exercise was (1.5 ± 1.0) seconds. In the second phase of experiment, average first exercise repetition duration was (1.9 ± 0.3) seconds, while average second exercise repetition duration was (1.9 ± 0.2) seconds. Quality of each repetition was calculated within C#. It was measured on scale from 0 to 10, where 0 designates the case when the quality mark could not be given. Grade 10 was given if both local extrema are within tolerances. Average quality of each repetition in first phase of experiment was (6.0 ± 1.6), while average quality of each repetition in second phase of experiment was (8.6 ± 0.5).
The results of self-assessment have shown that examinee's quantitative assessment of exercising vary from prescribed exercising. Self-assessment depends on various factors such as motivation, concentration and trained level of examinee, understanding of given instructions, successfulness in taking starting position and performing the exercise to the instructed ending position. Furthermore, examinees in the first phase of experiment showed greater deviation of repetition and qualitative performance of exercises than in the second phase. With the results of questionnaires, we have shown that our system is well accepted within the group of examinees and since it ensures quantitative and qualitative assistance during strength exercising it might have the positive effect on rehabilitation and training in larger groups as well. Streaming information between sensor node and desktop or mobile application, real-time feature extraction and context classification using OZIMS system that integrates commodity hardware and attached application for assisted exercising enables the real-time classification of performances with feedback without the need for other infrastructure, invasive methods and obstruction of patient's daily life. Since in the rehabilitation is not only important to simply exercise, but to exercise exactly as prescribed and as much as prescribed, it is obvious that OZIMS application can further enhance the process of rehabilitation. Streaming processed data from the sensor node to Central Server where they are easy accessible to professional and end users for further review, we introduced closed loop for continuous monitoring of physical activity of patients in real-time and enabled sophisticated tracking the rehabilitation/exercise performance to patients and their therapists. ACKNOWLEDGMENT The authors would like to acknowledge the contribution of L. Celić in the design and development of Wireless Body Area Network and flexyNet protocol. The authors would also like to thank to prof. Ratko Magjarević for supervising the development of the OZIMS application. LITERATURE [1]
[2]
[3]
[4]
[5] Figure 4. Average grades on questions about OZIMS system
CONCLUSION
G. Ninot et al., "Cost-saving effect of supervised exercise associated to COPD self-management education program", Respiratory Medicine, vol. 3, pp. 377-385, 2011. H. Viswanathan, C. Baozhi, D. Pompili, "Research challenges in computation, communication, and context awareness for ubiquitous healthcare", IEEE Communication Magazine, vol. 50, pp. 92-99, 2012. B. Najafi, K. Aminian, A. Paraschiv-Ionescu, F. Loew, C. J. Bula, P. Robert, "Ambulatory system for human motion analysis using a kinematic sensor: Monitoring of Daily physical activity in the elderly", IEEE Transaction on Biomedical Engineering, vol. 50, pp. 711-723, 2003. F. Pitta, T. Troosters, V. S. Probst, M. A. Spruit, M. Decramer, R. Gosselnik, "Quantifying physical activity in daily life with questionnaires and motion sensors in COPD", European Respiratory Journal, vol 27, pp. 1040-1055, 2006. E. Jovanov, A. Milenkovic, C. Otto, P. de Groen, "A wireless body area network of intelligent motion sensors for computer
[6]
[7]
[8]
assisted physical rehabilitation", Journal of NeuroEngineering and Rehabilitation, vol. 2, pp. 6-16, 2005. M. Sung et al., "Wearable feedback systems for rehabilitation", Journal of NeuroEngineering and Rehabilitation, vol. 2, pp. 17-29, 2005. L. Celic, M. Varga, T. Pozaic, D. Dzaja, S. Zulj, R. Magjarevic, "WBAN for physical activity monitoring in health care and wellness", World Congress on Medical Physics and Biomedical Engineering, Beijing, China, 2012. S. Patel, H. Park, P. Bonato, L. Chan, M. Rodgers, "A review of wearable sensors and systems with application in rehabilitation", Journal of NeuroEngineering and Rehabilitation, vol. 9, pp. 21-38, 2012.
[9]
A. G. Bonomi, "Physical activity recognition using a wearable accelerometer: New perspectives for energy expenditure assessment and health promotion", Eindhoven: Datawyse / Universitaire Pers Maastricht, 2010. [10] L. Celic, D. Trogrlic, I. Paladin, "Integration of measurement devices supporting diabetic patients into a remote care system", IFMBE Proc., vol. 37, pp. 39-42, 2011. [11] S. E. Crouter, K. G. Clowers, D. R. Bassett, "A nove method for using accelerometer data to predict energy expenditure", Journal of Applied Physiology, vol. 100, pp. 1324-1331, 2006. [12] P. Manocchia, "Anatomy of Exercise: a trainer's guide to your workout", 1st ed., Mozaik knjiga, 2011.
