Technology in Rehabilitation: Comparing Personalised and Global Classification Methodologies in Evaluating the Squat Exercise with Wearable IMUs

Summary

Background: The barbell squat is a popularly used lower limb rehabilitation exercise. It is also an integral exercise in injury risk screening protocols. To date athlete/patient technique has been assessed using expensive laboratory equipment or subjective clinical judgement; both of which are not without shortcomings. Inertial measurement units (IMUs) may offer a low cost solution for the objective evaluation of athlete/patient technique. However, it is not yet known if global classification techniques are effective in identifying naturally occurring, minor deviations in barbell squat technique.

Objectives: The aims of this study were to: (a) determine if in combination or in isolation, IMUs positioned on the lumbar spine, thigh and shank are capable of distinguishing between acceptable and aberrant barbell squat technique; (b) determine the capabilities of an IMU system at identifying specific natural deviations from acceptable barbell squat technique; and (c) compare a personalised (N=1) classifier to a global classifier in identifying the above.

Methods: Fifty-five healthy volunteers (37 males, 18 females, age = 24.21 +/- 5.25 years, height = 1.75 +/- 0.1 m, body mass = 75.09 +/- 13.56 kg) participated in the study. All participants performed a barbell squat 3-repeti- tion maximum max strength test. IMUs were positioned on participants’ lumbar spine, both shanks and both thighs; these were utilized to record tri-axial accelerometer, gyroscope and magnetometer data during all repetitions of the barbell squat exercise. Technique was assessed and labelled by a Chartered Physiotherapist using an evaluation framework. Features were extracted from the labelled IMU data. These features were used to train and evaluate both global and personalised random forests classifiers.

Results: Global classification techniques produced poor accuracy (AC), sensitivity (SE) and specificity (SP) scores in binary classification even with a 5 IMU set-up in both binary (AC: 64%, SE: 70%, SP: 28%) and multi- class classification (AC: 59%, SE: 24%, SP: 84%). However, utilising personalised classification techniques even with a single IMU positioned on the left thigh produced good binary classification scores (AC: 81%, SE: 81%, SP: 84%) and moderate-to-good multi- class scores (AC: 69%, SE: 70%, SP: 89%).

Conclusions: There are a number of challenges in developing global classification exercise technique evaluation systems for rehabilitation exercises such as the barbell squat. Building large, balanced data sets to train such systems is difficult and time intensive. Minor, naturally occurring deviations may not be detected utilising global classification approaches. Personalised classification approaches allow for higher accuracy and greater system efficiency for end-users in detecting naturally occurring barbell squat technique deviations. Applying this approach also allows for a single-IMU set up to achieve similar accuracy to a multi-IMU setup, which reduces total system cost and maximises system usability.

Keywords

Exercise therapy - biomedical technology - lower extremity - physical therapy speciality - inertial measurement units

^* These authors contributed equally to this work

• References

• 1 Cook G, Burton L, Hoogenboom B. Pre-participation screening: the use of fundamental movements as an assessment of function – part 1. North American Journal of Sports Physical Therapy 2006; 1 (02) 62-72.
• 2 Hall M, Nielsen JH, Holsgaard-Larsen A, Nielsen DB, Creaby MW, Thorlund JB. Forward lunge knee biomechanics before and after partial menis- cectomy. The Knee 2015; 22 (06) 506-509.
• 3 Ahmadi A, Mitchell E, Destelle F, Gowing M, O’Connor NE, Richter C. et al. Automatic Activity Classification and Movement Assessment During a Sports Training Session Using Wearable Inertial Sensors. Proceedings of the 11th International Conference on Wearable and Implantable Body Sensor Networks (BSN). IEEE; 2014: 98-103.
• 4 Bonnechere B, Jansen B, Salvia P, Bouzahouene H, Omelina L, Moiseev F. et al. Validity and reliability of the Kinect within functional assessment activities: Comparison with standard stereophotogrammetry. Gait & Posture 2014; 39 (01) 593-598.
• 5 Bonnet V, Mazza C, Fraisse P, Cappozzo A. Realtime estimate of body kinematics during a planar squat task using a single inertial measurement unit. IEEE Transactions on Biomedical Engineering 2013; 60 (07) 1920-1926.
• 6 Whiteside D, Deneweth JM, Pohorence MA, Sandoval B, Russell JR, McLean SG. et al. Grading the functional movement screen: A comparison of manual (real-time) and objective methods. The Journal of Strength & Conditioning Research 2016; 30 (04) 924-933.
• 7 McGrath D, Greene BR, O’Donovan KJ, Caulfield B. Gyroscope-based assessment of temporal gait parameters during treadmill walking and running. Sports Engineering 2012; 15 (04) 207-213.
• 8 Morris D, Saponas TS, Guillory A, Kelner I. Reco-Fit: using a wearable sensor to find, recognize, and count repetitive exercises. Proceedings of the SIG- CHI Conference on Human Factors in Computing Systems. ACM; 2014: 3225-3234.
• 9 Leardini A, Lullini G, Giannini S, Berti L, Ortolani M, Caravaggi P. Validation of the angular measurements of a new inertial-measurement-unit based rehabilitation system: comparison with state-of-the-art gait analysis. Journal of Neuroengineering and Rehabilitation 2014; 11 (01) 1-7.
• 10 Tang Z, Sekine M, Tamura T, Tanaka N, Yoshida M, Chen W. Measurement and Estimation of 3D Orientation using Magnetic and Inertial Sensors. Advanced Biomedical Engineering 2015; 4: 135-143.
• 11 Muehlbauer M, Bahle G, Lukowicz P. What can an arm holster worn smart phone do for activity recognition?. Proceedings of the International Symposium on Wearable Computers (ISWC). IEEE; 2011: 79-82.
• 12 Chang K-H, Chen MY, Canny J. Tracking free- weight exercises. Krumm J, Abowd GD, Seneviratne A, Strang T. UbiComp 2007: Ubiquitous Computing.. Berlin, Heidelberg: Springer; 2007: 19-37.
• 13 Seeger C, Buchmann A, Van Laerhoven K. my-HealthAssistant: a phone-based body sensor network that captures the wearer’s exercises throughout the day. Proceedings of the 6th International Conference on Body Area Networks. ICST; 2011: 1-7.
• 14 Pernek I, Hummel KA, Kokol P. Exercise repetition detection for resistance training based on smartphones. Personal and Ubiquitous Computing 2013; 17 (04) 771-782.
• 15 Pernek I, Kurillo G, Stiglic G, Bajcsy R. Recognizing the intensity of strength training exercises with wearable sensors. Journal of Biomedical Informatics 2015; 58: 145-155.
• 16 Taylor PE, Almeida GJ, Hodgins JK, Kanade T. Multi-label classification for the analysis of human motion quality. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE; 2012: 2214-2218.
• 17 Melzi S, Borsani L, Cesana M. The virtual trainer: supervising movements through a wearable wireless sensor network. 6th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks Workshops. 2009. IEEE; 2009: 1-3.
• 18 Velloso E, Bulling A, Gellersen H, Ugulino W, Fuks H. Qualitative activity recognition of weight lifting exercises. Proceedings of the 4th Augmented Human International Conference. ACM; 2013: 116-123.
• 19 Giggins OM, Sweeney KT, Caulfield B. Rehabilitation exercise assessment using inertial sensors: a cross-sectional analytical study. Journal of Neuroengineering and Rehabilitation 2014; 11 (01) 158-168.
• 20 O’Reilly M, Whelan D, Chanialidis C, Friel N, Delahunt E, Ward T. et al. Evaluating squat performance with a single inertial measurement unit. IEEE 12th International Conference on Wearable and Implantable Body Sensor Networks (BSN). IEEE; 2015: 1-6.
• 21 Whelan D, O’Reilly M, Ward T, Delahunt E, Caulfield B. Evaluating Performance of the Single Leg Squat Exercise with a Single Inertial Measurement Unit. Proceedings of the 3rd Workshop on ICTs for Improving Patients Rehabilitation Research Techniques. ACM; 2015: 144-147.
• 22 Whelan D, O’Reilly M, Ward T, Delahunt E, Caulfield B. Evaluating Performance of the Lunge Exercise with Multiple and Individual Inertial Measurement Units. Pervasive Health 10th EAI International Conference on Pervasive Computing Technologies for Healthcare. 2016: 101-108.
• 23 Whelan D, O’Reilly M, Huang B, Giggins O, Kechadi T, Caulfield B. Leveraging IMU data for accurate exercise performance classification and musculoskeletal injury risk screening. IEEE 38th Annual International Conference of the Engineering in Medicine and Biology Society (EMBC). 2016: 659-662.
• 24 Chawla NV. Data mining for imbalanced datasets: An overview. Maimon O, Rokach L. Data mining and knowledge discovery handbook.. New York: Springer; 2005: 853-867.
• 25 He H, Garcia EA. Learning from imbalanced data. IEEE Transactions on Knowledge and Data Engineering 2009; 21 (09) 1263-1284.
• 26 Kotsiantis SB, Zaharakis I, Pintelas P. Supervised machine learning: A review of classification techniques. Maglogiannis I, Karpouzis K, Wallace M, Soldatos J. Emerging Artificial Intelligence Applications in Computer Engineering: Real Word AI Systems with Applications in EHealth, HCI, Information Retrieval and Pervasive Technologies.. Amsterdam: IOS Press; 2007: 3-25.
• 27 Baechle TR, Earle RW. Resistance Training Exercise Techniques. Baechle TR, Earle RW. NSCA’s Essentials of Personal Training.. 1st ed. Champaign, IL: Human Kinetics; 2004: 295-332.
• 28 Jerri AJ. The Shannon sampling theorem – Its various extensions and applications: A tutorial review. Proceedings of the IEEE 1977; 65 (11) 1565-1596.
• 29 Shimmer 9DOF calibration [cited 2017 March 13]. Available from: http://www.shimmersensing.com/shop/shimmer-9dof-calibration.
• 30 Madgwick SOH, Harrison AJL, Vaidyanathan R. Estimation of IMU and MARG orientation using a gradient descent algorithm. Proceedings of the IEEE International Conference on Rehabilitation Robotics (ICORR). IEEE; 2011: 1-7.
• 31 Katz MJ, George EB. Fractals and the analysis of growth paths. Bulletin of Mathematical Biology 1985; 47 (02) 273-286.
• 32 Single-level discrete 1-D wavelet transform [cited 2017 March 13]. Available from: http://uk.math-works.com/help/wavelet/ref/dwt.html.
• 33 Breiman L. Random forests. Machine Learning 2001; 45 (01) 5-32.
• 34 Mitchell E, Ahmadi A, O’Connor NE, Richter C, Farrell E, Kavanagh J. et al. Automatically detecting asymmetric running using time and frequency domain features. Proceedings of the 12th International Conference on Wearable and Implantable Body Sensor Networks (BSN). IEEE; 2015: 1-6.
• 35 Fushiki T. Estimation of prediction error by using K-fold cross-validation. Statistics and Computing 2011; 21 (02) 137-146.
• 36 Whelan D, O’Reilly M, Ward T, Delahunt E, Caulfield B. Technology in Rehabilitation: Evaluating the Single Leg Squat Exercise with Wearable Iner- tial Measurement Units. Methods of Information in Medicine 2016; 55 (06) 88-94.
• 37 Bonnet V, Mazza C, Fraisse P, Cappozzo A. A least-squares identification algorithm for estimating squat exercise mechanics using a single inertial measurement unit. Journal of Biomechanics 2012; 45 (08) 1472-1477.