Gait Analysis in Lower-Limb Prosthesis Design, State of the Art
(Alt Title: Walk this Way)
While humans have been replacing amputated lower limbs for thousands of years, behind the deceptive simplicity of normal walking is a complicated network of kinematic parameters defined by unseen joint interactions in the body. Prosthetic devices must sufficiently mimic the complex nature of this movement in order to be successful. For the most part, the goodness of fit for these devices has been assessed subjectively, either by the user assessing how it feels to walk with the device or by an external observer eyeballing the user and assessing how “normal” they look while walking. While it has always been acknowledged that a robust objective methodology needs to be introduced into the device evaluation process, only in recent years has instrumented gait analysis technology begun to bring such desires into reality . In particular, instrumented gait analysis has been used to take the “eyeball” test and quantize certain specific parameters of asymmetry between the prosthetic leg and the sound leg. Three specific applications of this technology include (i) developing a priori models that define the parameters of motion, (ii) direct comparison of competing devices, and (iii) supplementing rehabilitation efforts with patients who already have a device.
For a device to adequately replace a limb requires it to not only replicate the motion of the missing limb but also fit within the physical boundaries established by the body . For example, if a device replicates the swing phase accurately but requires a mechanism that weighs 200 lbs., it will not be feasible for the patient to use. Therefore, a sophisticated process is required to optimize the multiple properties of the replacement device. Since experimental methods were found to be tedious, it became apparent mathematical models would be much more efficient . Pejhan, et al. (2008), using kinematic data collected from gait analysis, created such a model to optimize the design of above-knee prosthesis. Pejhan focused on analyzing the dynamics of the device during the entire gait cycle, including swing and stance phases. This was in contrast to previous modeling attempts which were primarily focused on either modeling the healthy leg or modeling only a selected section of the prosthesis gait. Following conventional definitions, the transfemoral design was defined with standard three rigid sections (thigh, shank, and foot) connected at two joints (knee and ankle). Swing phase of the knee motion was assumed to be controlled by a hydraulic controller and the stance phase motion was assumed to be controlled by an elastic controller. The flexibility and energy storing functions in the ankle joint were assumed to be provided by a torsional spring and damper. Each of these components were modeled by Lagrangian equations. Next, the biomechanical data collected from gait analysis of a normal human walking provided the controller parameters and initial conditions. The optimized values for spring stiffness and damping coefficients were found for the knee and ankle components (Table 1).
Table 1. Optimized parameters in the Pejhan above-knee prosthetic leg model.
|Elastic controller stiffness (knee)||1980 N/m|
|Hydraulic damping coeff. (knee)||0.7 Kg/s|
|Torsional stiffness (ankle)||5.35 Kg/s|
|Torsional damping coeff. (ankle)||10.5 N*cm/rad|
When compared to the collected data, the Pejhan model predictions for knee-flexion angle differed by 3.3 degrees on average (3.3 SD) and predictions for ankle-plantar flexion differed by 3.4 degrees on average (2.9 SD). Thus, successfully providing a full description of prosthesis behavior from stance phase through swing phase within acceptable error boundaries.
In a related effort, Awad, et al (2016) estimated parameters for actuation systems used in the knee and ankle joints . While the Pejhan model started with pre-defined components and worked forwards to predict behavior, the Awad model started with the physiological system and established the parameters that any given device would have to fit within. The actuation mechanism is defined along five variables: maximum peak torque (Tmax), rated continuous, torque (Tr), maximum angular velocity required from the actuator (ωmax), maximum angular position (θmax) allowed by the actuator mechanism, and the inertia of the mechanical components of the system. Using instrumented gait analysis, the Awad model derived the physiologically parallel parameter values for both the stance and swing phase (Table 2).
Table 2. Selected knee parameters for level ground walking at normal speed. Reproduced from 
|Parameter||Stance phase||Swing phase|
Both of these models, Pejhan and Awad, are derived for normal walking along a flat walking area. Some inherent limitations to the models include the fact that the joint torque parameters are only the average generated torque from the entire joint and not broken up into agonist and antagonist components. It is likely the case that the physiological muscle pairings in vitro affect the movement control of the particular joint. Thus, matching the joint velocity and joint torque in the device may not be sufficient to actually model the precise movement of the leg.
Instrumented gait analysis has also been used to compare different prosthetics head-to-head in their direct performance with specific patients. These studies will typically compare the influence of individual components of the device on the gait of the patient. Examples include: foot design, range of motion, energy storing, and knee components . The knee components of trans-femoral designs are most interesting because of the swing-phase control elements; as the leg is raised up, the lower segment must swing through to the next heel strike. Different mechanisms exist to facilitate this motion from simple, purely mechanical designs to complex computerized feedback systems. It is often hypothesized that the computerized system with a more physiologically similar swing phase reduces the asymmetry of the patient since it moves more naturally with the walking speed. Instrumented gait analysis has been immensely helpful in quantifiably answering this question. Schaarschmidt, et al (2012) and Segal, et al (2006) both tackled this issue using instrumented gait analysis to evaluate the computerized C-Leg with non-computerized models.
Segal compared the C-Leg with the Mauch SNS and selected a wide variety of measures of symmetrical walking. As expected, the C-Leg had a statistically significantly (p < 0.005) lower peak knee flexion angle (55.2º ± 7) as compared to the Mauch SNS (64.4° ± 6) when the subject was under controlled walking speed conditions. The prosthetic limb step length for the C-Leg was lower than for the Mauch SNS device (0.66 m ± 0.04 and 0.70 m ± 0.06 respectively) when at controlled walking speed. The C-Leg thus showed a much closer fit to the intact limb step length (0.64 m ± 0.06) and can be said to have increased the symmetry. However, at self-selected speeds, there was no statistically significant difference in the step-length between the two devices . In sum, while the C-Leg appears to increase symmetry in some parameters, it is unclear if it increases symmetry across the board.
Schaarschmidt compared the same C-Leg to a non-computerized 3R80 knee component with more ambiguous results, reporting that “enhanced stance phase security and swing phase control [by the] C-Leg did not affect the asymmetry between the intact and prosthetic leg” . In particular, both devices showed the contact time for the prosthetic limb was substantially shorter than the contact time for the intact limb. At 1.1 m/s walking speed, the C-Leg contact times were 0.68 sec ± 0.04 and 0.79 sec ± 0.06 for the prosthetic and intact limb respectively. The 3R80 contact times were nearly identical at 0.69 sec ± 0.04 and 0.78 sec ± 0.05 for the prosthetic and intact limb respectively .
Even after the patient has his or her device, instrumented gait analysis has proven to be a helpful tool in rehabilitating the patient and getting them to recover closer to normal walking motion. One of the particularly common challenges in this stage of the process is with alignment of the socket component of the device as it strongly determines the gait motion. Esquenazi et al (2014) demonstrated the effectiveness of instrumented gait analysis in rehabilitation of transtibial patients. Initial baseline kinematic data were collected using motion capture as well as instrumented treadmills. After baselines were established, the socket was realigned by a prosthetist and the same data collected. All defined gait characteristics either significantly improved or were statistically insignificant in their change. One unique measure of balance was the trunk lean – the degree to which the patient leaned over as determined by reflective markers; a decrease of approximately 10° was observed post- alignment adjustment (Figure 1).
Figure 1. Trunk lean of transtibial patients before (left) and after (right) socket adjustment, horizontal axis is % of gait, vertical axis is degrees. Reproduced from 
Because of the quantifiable and predictive nature of the parameters defining these devices, attempts have been made to incorporate machine learning into the process and possible remove the human element from processes such as rehabilitation. In one case in particular, an instrument gait analysis system integrated with a machine learning algorithm was used to distinguish between passive and active tibial devices. Because the two devices produce slightly different ground reaction forces, a force plate was used to collect the data. The particular system was able to distinguish between the temporal properties of the Solid Ankle Cushioned Heel (SACH) and the iWalk BiOM powered prosthesis 100% of the time . This proof of concept study will pave the way for future, non-human analysis of prosthetic limbs based purely on biomechanical data from instrumented gait analysis.
In summary, instrumented gait analysis has been used highly successfully in lower-limb prosthetic design. Three such areas of success include the development of a priori mathematical models describing lower limb movement, device to device comparison demonstrating that swing-phase control is not sufficient to maintain symmetrical gait, and gait improvements from socket alignment assessed through instrumented gait analysis. Cutting-edge efforts are now pushing for incorporating machine learning into the process and potentially removing humans from the analysis process.
- Rietman, J.S., K. Postema, and J.H. Geertzen, Gait analysis in prosthetics: opinions, ideas and conclusions. Prosthet Orthot Int, 2002. 26(1): p. 50-7.
- Pitkin, M., What can normal gait biomechanics teach a designer of lower limb prostheses? Acta Bioeng Biomech, 2013. 15(1): p. 3-10.
- Pejhan, S., F. Farahmand, and M. Parnianpour, Design optimization of an above-knee prosthesis based on the kinematics of gait. Conf Proc IEEE Eng Med Biol Soc, 2008. 2008: p. 4274-7.
- Awad, M., et al. Estimation of actuation system parameters for lower limb prostheses. in Mechatronics (MECATRONICS)/17th International Conference on Research and Education in Mechatronics (REM), 2016 11th France-Japan & 9th Europe-Asia Congress on. 2016. IEEE.
- Segal, A.D., et al., Kinematic and kinetic comparisons of transfemoral amputee gait using C-Leg and Mauch SNS prosthetic knees. J Rehabil Res Dev, 2006. 43(7): p. 857-70.
- Schaarschmidt, M., et al., Functional gait asymmetry of unilateral transfemoral amputees. Hum Mov Sci, 2012. 31(4): p. 907-17.
- Esquenazi, A., Gait analysis in lower-limb amputation and prosthetic rehabilitation. Phys Med Rehabil Clin N Am, 2014. 25(1): p. 153-67.
- LeMoyne, R., et al., Implementation of machine learning for classifying prosthesis type through conventional gait analysis. Conf Proc IEEE Eng Med Biol Soc, 2015. 2015: p. 202-5.