This is an Author's Original Manuscript of an article published by Elsevier in Gait and Posture Published online 6 June 2015 and available at: doi:10.1016/j.gaitpost.2015.05.021 Foot side detection from lower lumbar spine acceleration Khaireddine BEN MANSOUR, Nasser REZZOUG, Philippe GORCE HandiBio - EA4322 - Université de Toulon, Toulon Var, 83957 La Garde cedex, France Abstract The purpose of this paper is to present a reliable algorithm to discriminate between left/right foot contact using an accelerometer located over the lower lumbar spine. With the given accelerometer frame orientation, the side detection algorithm, based on the sign of the derivative of the sinusoidal shape obtained from the filtered mediolateral (ML) acceleration, showed 100% correct side detection for all subjects at all walking velocities. From the obtained results, it is concluded that in healthy subjects, the side of subsequent foot contact can be reliably obtained from the ML acceleration pattern of the lower lumbar spine. Keywords: Accelerometer; Gait; Side detection; Lumbar spine
1. Introduction Accelerometers have been found reliable for the assessment of gait parameters from different body locations [1,2]. To minimize the number of sensors, lower lumbar spine placement is often recommended for gait assessment [3 6]. Unlike accelerometers located at thighs, shanks and feet, parameters assessed from lumbar accelerometer cannot be directly attributed to the left/right side. During walking, differences between left/right lower extremities have been mentioned frequently. In the case of asymmetrical gait, the determination of the affected side is often an advocated goal for physical therapy. Only a few studies attempted to discriminate left/right foot contacts despite the importance of this identification [3 6]. In some projects, gyroscopes were introduced as a tool of side detection [3]. However, the gathered information are not sufficient to quantify gait parameters. To cover these shortages, new tendencies propose the fusion between gyroscope and accelerometer for gait assessment [3]. While the results were encouraging, some other studies suggest the use of a single accelerometer. Zijlstra and Hof [4] developed a method based on the analysis of the sign of foot contact for the phase of the first harmonic from lumbar accelerometer. Their results showed that six out of fifteen subjects presented 12% of 756 foot contacts labeled falsely. Köse et al [5] presented a wavelet-based approach to detect the gait events and identify the foot side contact simultaneously from right pelvic acceleration. However, some studies found that it can generate extra initial contacts (IC) [7]. Capela et al [6] described a method based on the filtered ML acceleration gathered from smartphone located at the lower lumbar spine. Then, the tangent at one artibrarily chosen point (a=0.25 of step duration) was computed. A right step was identified if the y-coordinate of two points on the tangent with x-coordinates equal to (a+locking period) and (a-locking period) were above the filtered ML acceleration. The appropriate locking period was defined by a dedicated algorithm. The exposed condition depends on the orientation of the ML axis of the accelerometer and the authors did not account for it. Furthermore, the authors reported that some foot steps were not identified without giving further information about the issue. In this framework, the aim of this study is to present a simple yet robust algorithm to discriminate between left/right foot contact using a single accelerometer located over the lower lumbar spine. 2. Methods 2.1 Subjects Ten able bodied subjects (29 (6) years, 1.79 (0.08) m and 79 (9) kg), with no clinical history of injury that could interfere with their walking pattern, participated in the study. Each participant provided written informed consent before starting. The experiment was approved by the local ethics committee.
2.2 Equipment and data acquisition Acceleration and ground reaction forces (GRF) were measured by a triaxial capacitive accelerometer (MMA8453Q, Free scale Semiconductor, Austin, Texas USA, 200Hz, range: ±4g, resolution: 7.8mg) and a treadmill with dual integrated force-plates (ADAL3D-F TECHMACHINE, Andrezieux, France, 200Hz), respectively. For this study, the accelerometer was fixed firmly over the lower lumbar spine (L3-L4) by a waist belt and double-sided tape. 2.3 Procedures The capacity of the accelerometer located at the lower lumbar spine to discriminate between left/right step was evaluated during treadmill walking. After a 30s warm up, measurements were recorded for a period of 30 s at the following treadmill walking speeds (2.7kmh -1, 3.6kmh -1, 4.5kmh -1, 5.4kmh -1 and 6.3kmh -1 ). Gait is a succession of foot contact with the floor during a set of subsequent left/right steps. Each IC marks the beginning of the stance phase with different subphases. The first one corresponds to the double limb support which starts with an IC and extends until the transfer of the body weight from the back to the front leg. This transfer is characterized by a ML acceleration of the body center of mass toward the front leg [8]. To discriminate between left/right step, an algorithm based on the study of the sign of the derivative of the filtered ML acceleration was proposed. The following steps for the side detection algorithm (SDA) were realized: 1. Apply the mathematical algorithm proposed by Moe-Nilssen to transform the dynamic acceleration into a vertical-horizontal coordinate system [9]. 2. Detect the moment of IC from the anteroposterior acceleration [4] which were verified against treadmill data. 3. Filter the ML acceleration component with fourth-order zero-lag Butterworth filter at 1 Hz (Low-pass filter). A sinusoidal shape (SinS) was obtained. 4. Study the sign of the derivative of the SinS at the moment of IC. As presented in the SDA at the second step, IC were detected from anteroposterior acceleration and left-right normalized vertical GRF (Fig 2, top-panel) measured by the instrumented treadmill. Square and circles represent the instants of left/right contact, respectively. These IC were transcribed on the filtered ML lumbar acceleration signal. With the given accelerometer frame orientation following the ISB standard, foot contact on the right side (circles) coincides with a descending part of the SinS. Thereby, the derivative of SinS at each right IC is negative. Likewise, the IC of left foot (squares) were located on the ascending part of the
SinS and its derivative is positive.to validate the detection of the side of IC, the instrumented treadmill was set as the gold standard. Based on the vertical component of the GRF, the IC was detected as the first value greater than 20N [10]. Fig 1 Flowchart for step side detection
3. Results Compared to the treadmill, the SDA showed 100% success in detection of the side of foot contact for the set of subjects across all walking velocities (Table 1). Table 1 Percentage of success in detection of the side of foot contact for the set of subjects across all walking velocities Gait velocity Step number Percentage of step labeled correctly 2.7 kmh -1 466 100% 3.6 kmh -1 509 100% 4.5 kmh -1 562 100% 5.4 kmh -1 604 100% 6.3 kmh -1 650 100% Fig 2 Data for subject 3 at 2.7kmh -1. Top panel: left (black) and right (gray) normalized vertical GRF. Bottom panel: recorded ML acceleration (grey) and SinS (black). For both, the instant of IC was presented in square and circles for the left and right side 4. Discussion Through the availability of left/right IC detected from the vertical GRF, reliable indications were obtained to validate the capacity of ML acceleration to discriminate between left/right foot contacts. Compared to the gold standard, SDA showed 100% success for all subjects at all walking velocities, contrary to the method of Zijlstra and Hof [4]. Despite similarities between SDA and the tangent method proposed by Capela et al [6] some differences exist. The tangent was computed at an arbitrary point, while, for the SDA method, the choice of the instant to compute the derivative of filtered ML acceleration is based on biomechanical gait characteristics. The process of side detection of the "tangent method" works only if the chosen arbitary (a) value occurs on the same side of the filtered
ML inflection point. In contrast, the SDA method integrates the fact that IC marks the start of the transfer of the body center of mass characterized by a ML acceleration toward the front leg. Finally, SDA method was tested during five walking speed including low, normal and fast speed, while the tangent method was used only during freely chosen speed. Compared to the presented methods SDA was found simpler and more robust. 5. Conclusion The ability to accurately discriminate between left/right gait events is usually a prerequisite for detection of the pathological lower limb side in order to prescribe a physical or surgical treatment. This study shows that left/right contact can be reliably identified through the use of the ML acceleration recorded at the lower lumbar spine.
Acknowledgments This study was financially supported as part of doctoral scholarship from the French region Provence Alpes Cote d Azur, the FEDER European fund and the ESPHI society. Conflict of interest The authors report that there are no conflicts of interests. Reference [1] Lee J-A, Cho S-H, Lee Y-J, Yang H-K, Lee J-W. Portable Activity Monitoring System for Temporal Parameters of Gait Cycles. J Med Syst 2010;34:959 66. doi:10.1007/s10916-009-9311-8. [2] Selles RW, Formanoy MA, Bussmann J, Janssens PJ, Stam HJ. Automated estimation of initial and terminal contact timing using accelerometers; development and validation in transtibial amputees and controls. IEEE Trans Neural Syst Rehabil Eng 2005;13:81 8. doi:10.1109/tnsre.2004.843176. [3] McCamley J, Donati M, Grimpampi E, Mazzà C. An enhanced estimate of initial contact and final contact instants of time using lower trunk inertial sensor data. Gait Posture 2012;36:316 8. doi:10.1016/j.gaitpost.2012.02.019. [4] Zijlstra W, Hof AL. Assessment of spatio-temporal gait parameters from trunk accelerations during human walking. Gait Posture 2003;18:1 10. [5] Köse A, Cereatti A, Croce UD. Bilateral step length estimation using a single inertial measurement unit attached to the pelvis. J NeuroEngineering Rehabil 2012;9:9. doi:10.1186/1743-0003-9-9. [6] Capela NA, Lemaire ED, Baddour N. Novel algorithm for a smartphone-based 6-minute walk test application: algorithm, application development, and evaluation. J NeuroEngineering Rehabil 2015;12:19. doi:10.1186/s12984-015-0013-9. [7] Trojaniello D, Cereatti A, Della Croce U. Accuracy, sensitivity and robustness of five different methods for the estimation of gait temporal parameters using a single inertial sensor mounted on the lower trunk. Gait Posture 2014;40:487 92. doi:10.1016/j.gaitpost.2014.07.007. [8] John CT, Seth A, Schwartz MH, Delp SL. Contributions of muscles to mediolateral ground reaction force over a range of walking speeds. J Biomech 2012;45:2438 43. doi:10.1016/j.jbiomech.2012.06.037. [9] Moe-Nilssen R. A new method for evaluating motor control in gait under real-life environmental conditions. Part 1: The instrument. Clin Biomech 1998;13:320 7. doi:10.1016/s0268-0033(98)00089-8. [10] Zeni Jr JA, Richards JG, Higginson JS. Two simple methods for determining gait events during treadmill and overground walking using kinematic data. Gait Posture 2008;27:710 4.