The Trendelenburg Gait

A Personalised Monitoring and Recommendation Framework for Kinetic Dysfunctions: The Trendelenburg Gait

Nikolaos Michalopoulos George E. Raptis Christina Katsini Moment Kinetics Moment Kinetics Moment Kinetics nmichalopoulos@ graptis@ ckatsini@ momentkinetics.com momentkinetics.com momentkinetics.com Theoﬁlos Chrysikos Michail Mamalakis Andrew Vigotsky ECE Department ECE Department Biomedical Engineering Dept. University of Patras University of Patras Northwestern University [email protected] [email protected] [email protected]

ABSTRACT bilitative perspective, it is possible to identify the risk of Kinetics and kinematics have been the focus of a lot of re-injury on an individual basis through the application of research throughout the years, and technological advances biomechanical kinematics and kinetics. The identification have contributed towards advanced monitoring and analysing of causality often requires precise analysis, as patients fre- of motion. Focusing on Trendelenburg Gait, an abnormal quently present a plethora of compensatory movement pat- gait, a personalisation framework for monitoring and eval- terns. Such compensatory deficits and impairments are often uating the movement is proposed in this paper based on a more noticeable than the underlying cause [15], making the number of factors, such as individuals’ body composition, diagnosis process even more challenging. However, quantify- motion mechanisms, capturing process, analysis tools and ing the mobility state of a medical disorder and determining information presentation techniques. A formal description the neuromuscular-skeletal contributions to that state helps of such factors allow for a better understanding and deeper in prescribing treatment and assessing the outcome with analysis of the critical points, enabling the framework to greater confidence. The most accurate systems for captur- provide improvement recommendations, applied in interdis- ing gait patterns are camera-based, which require an expen- ciplinary contexts such as rehabilitation, medical applica- sive experimental setup and a complicated calibration phase tions, strength and conditioning, and sport performance. [1, 15]. Such systems calculate and report reduced internal hip abduction moments and external knee adduction moments following intramuscular hypertonic saline injections CCS Concepts [10]. However, research regarding gait analysis on the Tren- •Hardware → Sensor devices and platforms; •Social and delenburg gait pattern is limited. Examination of this gait professional topics → Medical technologies; pattern arises from the Trendelenburg test, in which the individual is seen ”standing on the treated (affected) leg and Keywords raising the buttock of the other side up to or above the hori- zontal line” [27]. Failure of the test implies being unable to Trendelenburg gait; kinetic dysfunction; personalisation and stand on this position [27]. recommendation framework; ultrasonic sensors Advances in computer technology, simulation models, and artificial intelligence have increased the potential of design- 1. INTRODUCTION ing and developing personalisation, both on-demand and in- time, mechanisms. Utilising computational techniques to The general aim of clinical movement analysis is to iden- model patients’ locomotion offers a personalised experience, tify and understand mechanopathology and pathomechan- improving individuals’ life quality. However, the design of ics. With the application of biomechanical analysis, injury such effective and efficient personalised platforms is an elu- causality could be identified and the kinematics associated sive task, as it is related to an interdisciplinary background, with an injury can effectively be treated, leading to improve- such as biomechanics and computer engineering, taking into ments in quality of life and performance. From a reha- account various physical, human, technological, and applica- Permission to make digital or hard copies of all or part of this work for personal or tion design factors. The interrelation and interdependences classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation among the aforementioned factors are defined by the charac- on the first page. Copyrights for components of this work owned by others than the teristics of the designed application, such as anthropometric author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or data, human locomotion kinetics and kinematics, data trans- republish, to post on servers or to redistribute to lists, requires prior specific permission mission mechanisms, and tools of data visualisation. In this and/or a fee. Request permissions from [email protected]. context, this paper contributes to the design for personalised PCI ’16, November 10 - 12, 2016, Patras, Greece experience by proposing a factor-based framework that can c 2016 Copyright held by the owner/author(s). Publication rights licensed to ACM. ISBN 978-1-4503-4789-1/16/11. . . $15.00 be used to monitor, evaluate, and recommend improvements DOI: http://dx.doi.org/10.1145/3003733.3003786 on the treatment process of Trendelenburg gait. 2. RELATED WORK 4. DATA MINING Trendelenburg Gait has not received a lot of interdisci- Several tracking systems have been used to capture the plinary attention. To the authors’ knowledge, only two critical points of the Trendelenburg gait pattern; however, teams [1, 10] have formerly researched this abnormal gait they are mainly based on sport performance, tracking exer- pattern either implicitly or explicitly. Existing monitoring cises like barbell squats. Taking into consideration that sim- systems [12, 18] are not specified for Trendelenburg gait ilar kinetics are used for Trendelenburg gait pattern, funda- analysis, but they could be applied in this domain. How- mental mechanisms of the proposed tracking systems could ever, these systems model the human body based on the be applied to our case study. Research reports that camera- average human metrics, thus disregarding the characteris- based tracking systems are the most accurate, nonetheless tics of each particular person. Nonetheless, personalisation, the sensitivity of such systems in terms of lighting, shadow recommendation and adaptation have received a lot of re- and clutter question the system effectiveness [3]. Attempts search in other contexts than kinetic dysfunction domain, to use microelectromechanical systems and inertial sensors with promising results in terms of enhanced user experi- have also been made, but they have proven to be prone to ences and enriched provided services [8, 24]. Given that, error due to sensor bias and measurement noise [12, 18]. In in the context of medical applications and kinetic dysfunc- the present paper, we propose an alternative method which tion, patients’ locomotion and kinetics have a great impact makes use of the ultrasound technique. on the treatment process, a personalised monitoring and recommending system could add value to the existing capturing 4.1 Ultrasonic sensors for positioning systems and boost the supported procedures. This requires measurement in depth understanding of the patients’ locomotion and ki- Advances in the research and experimental study of sen- netics combined with exploitation of existing wireless wear- sors have allowed for various protocols functioning in the able tracking systems and high-level system adaptation. infrared-radio ultra wide band (IR-UWB) channel [14, 21]. Methods due to innovative techniques in microelectromechanical system technology such as accelerometer and gyro- 3. TRENDELENBURG GAIT scope have been investigated in [12, 13, 18]. Human tracking The Trendelenburg gait pattern, named after German sur- is feasible both in tri-axial and dual-axial reference systems geon Dr Friedrich Tredelenburg, is an abnormal gait caused and user movement phenomena have been mitigated with by weakness of the abductor muscles of the lower limb, glu- the application of the Kalman filter [17]. The transmis- teus medius and gluteus minimus [25]. When a person walks sion of the acquired empirical data is robust and accurate, with such an abnormal gait, the hip joint is subjected to however the increased complexity and cost of implemen- stresses in areas not normally stressed during gait. This re- tation and maintenance renders this method inappropriate sults in the development of other pathologies of the bones for on-the-fly solutions regarding positioning measurements. and cartilage in the hip and knee, such as arthritis or prema- In addition, path loss and multipath propagation phenom- ture wear in the cartilage hip joints [26, 28]. Trendelenburg ena need to be accounted for. Our application, designed gait is also associated with the development of pathology for sporting areas and gymnastic halls, as well as medical at the knee or ankles over a period of years [25], e.g. val- centers and treatment rooms, requires the investigation of gus of the knee is caused by the laterally displaced weight distance-dependent free space attenuation as well as large- over the hip [15]. It is characterized by a trunk shift over scale (shadow) fading which contribute to the overall losses the affected hip, due to weakness of the hip abductors, and of the signal strength and the stochastic variations around is best visualized from behind or in front of the patient. its local mean [5]. Observation of the patients’ gait from the side enables the As an alternate to the complex UWB method, we have examiner to detect stride and step length deficiencies; mo- chosen to apply the ultrasonic technique. This solution tion of the trunk and lower extremity in the sagittal plane, avoids all the intrinsic channel propagation characteristics including the extensor or gluteus maximus lurch, in which and can be implemented independent of the nature of the the patient extends the trunk posteriorly to compensate for topology as well as noise and interference issues. In addition, weak hip extensors. It also enables detection of ankle dorsi- this scheme is preferred due to its low cost, safety, simplic- flexor weakness and foot drop, leading to the inability of the ity and high temporal resolution for low range measurement foot to clear the ground, which is compensated for by ex- [20]. As already shown [22], the distance measurement of cessive lower extremity flexion to facilitate foot clearance of such an ultrasonic sensor is the returned distance reflected the ground. The integrity of the hip abductor muscle func- from the ground/surface, and the orientation of the sensor tion is determined by clinical and X-ray assessments and is not taken into consideration. The proposed scheme inves- by the Trendelenburg test. While no treatment modalities tigates a triaxial tracking system for the Trendelenburg gait currently exist, medical management attempt to deal with pattern. The degrees of freedom and the angle of movement underlying causes, e.g. pelvic support osteotomies lead to a are also considered, from a mechanical point of view. significant improvement in posture, gait and walking toler- ance in patients who have untreated congenital dislocations 4.2 Sensors positioning [6, 7]. Research has shown the importance of strengthening Having analysed the technical details of the monitoring hip abductors, quadriceps and hamstrings, resulting in re- system, it is worth mentioning that the accuracy of any ducing the degree of Trendelenburg gait [16]. The purpose of human model reconstruction is strongly dependant on the the treatment is to increase the otherwise abnormal range of extracted human silhouette [2]. We propose a 3D approach motion in the hip and trunk. Visual feedback collected from instead of 2D, so that kinematic data could be extracted the patients’ gait is used for adjusting the physiotherapeutic with minimal error. A musculoskeletal model for gait sim- treatment [9]. ulation was created, inducing 80 muscle-tendon units, con- sisting of 41 retro-reflective markers, built of 22 articulating validated against real life data derived from credible and rigid bodies. Degrees of freedom were 20 in the lower body valid tools used nowadays, such as markers, EMGs, ECGs (six on the pelvis and seven per leg) and 17 in the torso and and MRI-CT. The optimal treatment procedure an indi- upper body (three on the lumbar joint and seven per arm). vidual should adopt is recommended by the personalisation This model has met the fidelity criteria by comparing (i) the tools of the proposed framework based on the individuals’ musculoskeletal geometry to experimental data, (ii) the sim- user and group context models. Therefore, the proposed ulated muscle-generated joint moments to inverse dynamics personalisation approach consists of four primary sections: joint moments, and (iii) the simulated muscle activity to electromyography (EMG) data. It also met the speed crite- • Scenarios through rule-based and content-based ria by computing the time required to generate a simulation mechanisms: Capturing and formulating anthropo- of a single gait cycle. It is ensured that this model was as metric, kinetic and kinematic data along with technol- fast or faster than other frequently used models, while cor- ogy attributes will be used in developing and simulat- recting some drawbacks that existed in previous gait models ing scenarios, which will be then used by our frame- and it was implemented in the open source software platform work to collect and analyse data aiming to optimise the OpenSim [23]. adaptation, monitoring and recommendation process. • Context-aware critical situations: In ubiquitous 5. PERSONALISATION and pervasive computing the context model is modified and updated dynamically in real time. Our framework Taking into consideration the factors discussed in the pre- should adapt the changes made through context-aware vious sections, user models can be developed to collect and reasoning and process information in real time, provid- utilize individuals’ information either implicitly or explicitly ing the users, e.g. patients and physiotherapists, with regarding the Trendelenburg gait pattern. The adaptation valuable information to complete their objectives. types applied on the system content, behaviour and func- • Reproduction of personalised healthy movement: tionality, are applied through adaptation algorithms, follow- Based on personalised patients’ anthropometric and ing various approaches such as rule-based and content-based muscle-anatomical parameters, a patient specified model mechanisms [4]. The process of producing high-level infor- is produced. Blending the input data and the data de- mation from a set of low-level metrics, e.g. foot position or rived from motion and EMG analysis of a healthy neu- knee joint angle, is controlled by rule-based mechanisms ei- romuscular model, a hybrid forward-inverse dynamic ther implicitly or explicitly, statically or dynamically. Such model is implemented, reproducing the patients’ per- rules can initialise the system and adjust it in real time by sonalised healthy movement of walking gait. The cap- comparative and predictive models exploiting the data gath- tured movement is iteratively compared with the spec- ered and analysed for individuals of similar characteristics ified model, providing recommendations by combining and gait. Content-based mechanisms suggest positioning, the mechanism of the gait pattern collector tool, rec- measurement and analysis of body sensors placed on individ- ommendation engine and training process. uals while walking. Such mechanisms are primarily based on • System iterative training process: User interac- the anthropometric data for each individual gathered explic- tion and cooperation with our framework should be ex- itly. Factors such as body stance, straddle and walking pace ploited to understand the user behavioural and struc- should also be considered. Other mechanisms which could tural patterns. Therefore, the data forage, acquisi- enhance the adaptation process include collaborative and tion, aggregation, monitoring, reasoning and recom- group data analysis, e.g. people with similar anthropomet- mendation procedures should follow an iterative pro- ric characteristics suffering from Trendelenburg syndrome, cess which is used for system training based on its and sophisticated data mining mechanisms, including pat- users’ individual characteristics and patterns. tern discovery, rule investigation, clustering, classification, etc., implemented using Markov and probabilistic models focusing on adaptation and personalisation [19]. 6. CONCLUSION AND FUTURE WORK In this paper we proposed a personalised framework for 5.1 Personalisation of Kinetic Dysfunction monitoring, evaluating and recommending improvements to The primary aim of our system is to determine the differ- the Trendelenburg gait treatment process. The main chal- ences between patients’ motion and healthy motion. Thus, lenge is to personalise the recommendations based on pa- it is necessary to model and produce this healthy motion in tients’ locomotion, kinetics and anthropometrics, since they each case. This can be done with a neuromuscular simula- introduce important differences and limitations. The pro- tion model based on specific muscular and anatomical pa- posed framework suggests that the non-technical users i.e. rameters, which will reproduce each patient’s personalised patients, physiotherapists, rehabilitation experts etc. are healthy motion. Considering the difficulty to estimate these allowed to configure the system, providing a personalised muscle and anatomical parameters, evaluating their impor- experience. The machine learning allows for system train- tance for each specific movement with a sensitivity analy- ing and enhances user experience by providing more suit- sis is needed. Finally, validation of the simulated model’s able recommendations. The proposed approach certainly motion needs to be executed before it provides any consid- requires practical validation in several application contexts eration for the patient’s dysfunction and Hicks et al. [11] in order to identify difficulties in the practical process and present some ways for providing efficient solutions to the better understand the evaluation and recommendation pro- aforementioned issue. This procedure involves matching the cedure. User studies could contribute towards refining and system’s anthropometric data, joint loads, muscle geome- improving the proposed framework by identifying and op- tries and moment arms with a sensitivity analysis. Every timising the interplay among human mechanics, technology part of the neuromuscular-tendon model must be separately and information presentation through real life case studies. 7. REFERENCES IEEE Transactions on Instrumentation and [1] M. Archdeacon, K. R. Ford, J. Wyrick, M. V. Measurement, 62(5):1073–1083, 2013. Paterno, S. Hampton, M. B. Ludwig, and T. E. [15] J. Perry and J. M. Burnfield. Gait analysis: normal Hewett. A prospective functional outcome and motion and pathological function. 1992. analysis evaluation of the hip abductors after femur [16] J. Petrofsky. The use of electromyogram biofeedback fracture and antegrade nailing. Journal of orthopaedic to reduce trendelenburg gait. European journal of trauma, 22(1):3–9, 2008. applied physiology, 85(5):491–495, 2001. [2] B. Bhanu and J. Han. Model-based human recognition [17] A. J. Petruska and S. G. Meek. Non-drifting limb – 2d and 3d gait. In Human Recognition at a Distance angle measurement relative to the gravitational vector in Video, pages 65–94. Springer, 2010. during dynamic motions using accelerometers and rate [3] V. Bonnet, C. Mazza, P. Fraisse, and A. Cappozzo. A gyros. In 2011 IEEE/RSJ International Conference on least-squares identification algorithm for estimating Intelligent Robots and Systems, pages 3632–3637. squat exercise mechanics using a single inertial IEEE, 2011. measurement unit. Journal of Biomechanics, [18] P. Picerno, A. Cereatti, and A. Cappozzo. Joint 45(8):1472 – 1477, 2012. kinematics estimate using wearable inertial and [4] P. Brusilovski, A. Kobsa, and W. Nejdl. The adaptive magnetic sensing modules. Gait & posture, web: methods and strategies of web personalization, 28(4):588–595, 2008. volume 4321. Springer Science & Business Media, [19] D. Pierrakos, G. Paliouras, C. Papatheodorou, and 2007. C. D. Spyropoulos. Web usage mining as a tool for [5] T. Chrysikos and S. Kotsopoulos. Characterization of personalization: A survey. User modeling and large-scale fading for the 2.4 ghz channel in user-adapted interaction, 13(4):311–372, 2003. obstacle-dense indoor propagation topologies. In [20] Y. Qi, C. B. Soh, E. Gunawan, and K.-S. Low. Vehicular Technology Conference (VTC Fall), 2012 Ambulatory measurement of foot kinematics using IEEE, pages 1–5. IEEE, 2012. wearable ultrasonic sensors. In 2014 36th Annual [6] A. M. DiGioia, B. Jaramaz, A. Y. Plakseychuk, J. E. International Conference of the IEEE Engineering in Moody, C. Nikou, R. S. LaBarca, T. J. Levison, and Medicine and Biology Society, pages 6282–6285. IEEE, F. Picard. Comparison of a mechanical acetabular 2014. alignment guide with computer placement of the [21] Y. Qi, C. B. Soh, E. Gunawan, K.-S. Low, and socket. The Journal of arthroplasty, 17(3):359–364, A. Maskooki. Using wearable uwb radios to measure 2002. foot clearance during walking. In 2013 35th Annual [7] R. Froelich. De l’ostéotomie sous-trochantérienne. International Conference of the IEEE Engineering in bifurcation des allemands. traitement de la luxation Medicine and Biology Society (EMBC), pages congénitale de la hanche. Rev. Orthop, 32:309, 1925. 5199–5202. IEEE, 2013. [8] M. Gao, K. Liu, and Z. Wu. Personalisation in web [22] Y. Qi, C. B. Soh, E. Gunawan, K.-S. Low, and computing and informatics: Theories, techniques, R. Thomas. Lower extremity joint angle tracking with applications, and future research. Information Systems wireless ultrasonic sensors during a squat exercise. Frontiers, 12(5):607–629, 2010. Sensors, 15(5):9610–9627, 2015. [9] D. Hamacher, D. Bertram, C. Fölsch, and L. Schega. [23] A. Rajagopal, C. L. Dembia, M. S. DeMers, D. D. Evaluation of a visual feedback system in gait Delp, J. L. Hicks, and S. L. Delp. Full-body retraining: A pilot study. Gait & posture, musculoskeletal model for muscle-driven simulation of 36(2):182–186, 2012. human gait. IEEE Transactions on Biomedical [10] M. Henriksen, J. Aaboe, E. B. Simonsen, T. Alkjær, Engineering, 63(10):2068–2079, 2016. and H. Bliddal. Experimentally reduced hip abductor [24] J. Relan, P. Chinchapatnam, M. Sermesant, function during walking: implications for knee joint K. Rhode, M. Ginks, H. Delingette, C. A. Rinaldi, loads. Journal of biomechanics, 42(9):1236–1240, 2009. R. Razavi, and N. Ayache. Coupled personalization of [11] J. L. Hicks, T. K. Uchida, A. Seth, A. Rajagopal, and cardiac electrophysiology models for prediction of S. L. Delp. Is my model good enough? best practices ischaemic ventricular tachycardia. Interface Focus, for verification and validation of musculoskeletal page rsfs20100041, 2011. models and simulations of movement. Journal of [25] C. Romano, C. Frigo, G. Randelli, and A. Pedotti. biomechanical engineering, 137(2):020905, 2015. Analysis of the gait of adults who had residua of [12] G. X. Lee and K.-S. Low. A factorized quaternion congenital dysplasia of the hip. J Bone Joint Surg approach to determine the arm motions using triaxial Am, 78(10):1468–79, 1996. accelerometers with anatomical and sensor [26] A. Sutherland, S. D’Arcy, D. Smart, and G. Ashcroft. constraints. IEEE Transactions on Instrumentation Abductor weakness and stresses around acetabular and Measurement, 61(6):1793–1802, 2012. components of total hip arthroplasty: a finite element [13] J. McCamley, M. Donati, E. Grimpampi, and analysis. International orthopaedics, 23(5):275–278, C. Mazzà. An enhanced estimate of initial contact and 1999. final contact instants of time using lower trunk inertial [27] F. Trendelenburg. Trendelenburg’s test: 1895. Clinical sensor data. Gait & posture, 36(2):316–318, 2012. orthopaedics and related research, (355):3–7, 1998. [14] G. Panahandeh, N. Mohammadiha, A. Leijon, and [28] P. Vasudevan, K. Vaidyalingam, and P. B. Nair. Can P. Händel. Continuous hidden markov model for trendelenburg’s sign be positive if the hip is normal? pedestrian activity classification and gait analysis. Bone & Joint Journal, 79(3):462–466, 1997.