S152 Published posters / Gait & Posture 24S (2006) S98–S289
PP-037
Detecting post-operative change in gait function using principal component analysis in subjects with cerebral palsy
Kjell- ˚Ake Nilsson
School of Health Sciences, J¨onk¨oping University, J¨onk¨oping, Sweden
1. Summary/conclusions
Principal components analysis is a multivariate statisti-cal method that has been used in gait analysis. One example of use of the method is the production of The Gillette Gait Index (GGI). This index, indicating normality in gait func-tion, has been presented and validated by previous authors. According to suggestions made by these authors, the index could potentially be used to evaluate change in gait function after surgical intervention in subjects with cerebral palsy. The GGI was calculated using principal components analysis for nineteen individuals with cerebral palsy in a retrospective study. The change in index value per individual from pre-to post-operative situations was compared pre-to the evaluation of change made by an experienced clinician. Agreement was evaluated using Cohen’s kappa (κ), resulting in a value of
κ = 0.406, which is usually considered to be a fair to moderate
level of agreement. Reasons were found to question the pre-viously published lower bound for detection of change using the GGI. Although the method seems promising, there is not yet enough evidence to justify the introduction of the GGI as a daily evaluation tool in clinical gait analysis (Table 1).
2. Introduction
Movement laboratories all over the world collect vast amounts of data for single subjects, in most cases to eval-uate certain aspects of the gait of those subjects for clinical purposes. Comparisons are difficult due to the number of gait variables and data points collected[1]. Clinical use of mul-tivariate methods to reduce the multidimensional gait data set and to facilitate the comparison process could be of great help to clinicians. Principal components analysis (PCA) has been used to produce a single-number index that indicates
Table 1
Table shows the kappa calculation matrix, where GGI represents index change and Clin represents the outcome evaluation made by the clinician
GGIw GGI0 GGIb
Clinw 2 0 0 2 Clin0 0 0 2 2 Clinb 1 1 13 15 3 1 15 19
Indices w, 0 and b represent worse, no change, or better.
closeness to normality for subjects with cerebral palsy (CP) by performing calculations on 16 selected gait variables. This index has been named the Gillette Gait Index (GGI)[2]. It was suggested by the original authors that the GGI could potentially be used to detect functional change in individual subjects from pre- to post-operative situations.
3. Statement of clinical significance
Evaluations of gait function normality are usually made to identify the need for and type of surgical intervention. A prevalent view is that a change towards normality is beneficial for the patient. It is obvious that efforts to quantify a change towards normality would be helpful in the post-surgical eval-uation made by the clinician.
4. Methods
This study was approved by the regional ethical review board. Informed written consent was obtained from all sub-jects. A total of nineteen subjects, all of whom had undergone both pre- and post-operative gait analysis and clinical evalua-tion, were used for this study (5 hemiplegics, 13 diplegics and 1 quadraplegic). There were 10 male and 9 female subjects (mean age 16 years, range 10–31 years), all ages identified at the time of 1-year follow-up and post-operative gait anal-ysis. No subjects were younger than 8 years at the time of pre-operative gait analysis. The PCA method to calculate the GGI, described by previous author, was applied to the data sets both pre- and post-operatively to observe the result-ing index change. The change in index value was compared to the statement regarding outcome made by the evaluating clinician during post-operative analysis. Three levels were defined both for index change and outcome evaluation; bet-ter, no change, or worse. This was done for each subject. Agreement between index and perceived change was evalu-ated using Cohen’s kappa statistic.
5. Results
After applying the calculation sequence described by pre-vious authors, three principal components were identified and used for calculation of final GGI values for all subjects. Mean GGI values for subjects grouped by CP involvement were obtained. Both the mean and range GGI values decreased from pre- to post-operative situations in all groups. The means and ranges presented in this study also seem to concur with the results presented in previous studies, however there were individual changes showing an index increase. Kappa calcu-lations gave κ = 0.406, showing a fair to moderate level of agreement between GGI and clinician.
Published posters / Gait & Posture 24S (2006) S98–S289 S153
6. Discussion
The values produced in this study, using existing data col-lected with another system in another gait laboratory, are shown to be of approximately the same magnitude as those previously reported. Somewhat different GGI values should be anticipated. Reasons for this may include day-to-day dif-ference in subject spasticity, actual subject change or marker placement error. While the GGI seems to be a promising tool in clinical gait analysis, it is not without problems. In general, one must realise that producing a single-value global index of this sort can actually conceal important data when used on its own. While it seems efficient to reduce the multivariate gait data into a single number, there is a risk that one uses this index to make decisions that are possibly life-changing for the subject. It must be stated that an index of this sort should only be seen as a rough guideline that might support other interpretations of the data, and that it should only be used as a tool for the experts. Although there seems to be potential for use of the GGI, much verification work using different aspects is needed before it can and/or will be trusted as a reliable tool in the gait laboratories of today regarding post-surgical outcome evaluations.
References
[1] Simon S. Quantification of human motion: gait analysis—benefits and limitations to its application to clinical problems. J Biomech 2004;37(12):1869–80.
[2] Thompson N, Harrington M, vol. 22, p. 377.
doi:10.1016/j.gaitpost.2006.11.105 PP-038
Variability of SEMG of five stump muscles during stance phase of gait in TF amputees with osseointegrated pros-theses
Annette Pantalla,∗, Sally Durhamb, David Ewinsa,b
aCentre for Biomedical Engineering, University of Surrey, Guildford, UK bGait Laboratory, Queen Mary’s Hospital, Roehampton, UK
1. Summary/conclusions
Surface electromyograms (SEMG) were recorded from 5 transfemoral (TF) amputees with osseointegrated prostheses and from 10 normal subjects during walking on a level sur-face. The smoothed and rectified signals were divided into stance and swing phases and the coefficient of multiple cor-relations (CMC) calculated for successive stances for each muscle. Results showed that the amputee group had consid-erable differences in the values of the CMC. Three of the amputees had good reproducibility for at least one muscle with gluteus maximus having the least variability. The
find-ings suggest that in certain amputees it may be possible to use SEMG as a natural sensor for control of a micro-processor controlled knee joint.
2. Introduction
A number of investigations have been undertaken on the variability of SEMG during locomotion with different statis-tical techniques used to assess repeatability. Factors affecting constancy of SEMG include varying levels of noise, differ-ent placemdiffer-ent of the electrodes and changing neural input
[1]. It has been reported that SEMG has a poor repeatability compared to other parameters measured during locomotion
[2]. Explanations for this include firstly that the measured EMG is shorter than the time period during which the force develops within a muscle, thereby making the SEMG profiles more variable[3], and secondly that there are an indetermi-nate number of ways in which the muscles can achieve the desired movement[4].
3. Statement of clinical significance
The clinical significance is to determine whether the repeatability of the SEMG is sufficiently high to make the deployment of SEMG as a natural sensor for the control of an intelligent knee in lower limb prosthesis a feasible option. Myoprocessors have been successfully incorporated in upper limb prostheses, but not to date in lower limb prostheses.
4. Methods
Two groups of subjects were selected. Group A consisted of 5 male TF amputees with osseointegrated prostheses and Group B of 10 normal male subjects. SEMG was measured from hip muscles of the stump side in Group A and the left lower limb in Group B. The muscles selected were gluteus maximus (GMX), gluteus medius (GMD), rectus femoris (RF), adductor magnus (AM) and biceps femoris (BF). The signal was collected using the Biometrics DataLINK DLK800 system with SX230 surface pre amplifiers at a sam-pling rate of 1000Hz. The bandwidth was 20–450 Hz and the electrode impedance 106M. A 3.3 m dual walkway with eight transducers sampling at 2000 Hz was used to collect force data. The electrode sites selected for all the muscles in Group B and GMX and GMD in Group A were the locations recommended by SENIAM[5]. The electrode placement for RF, AM and BF for Group A was determined by palpation and resistive testing. Each subject was asked to walk at their normal pace 10 times along the force walkway. The data were analysed in MATLAB (The MathWorks, Inc.). The signal was divided into stance and swing sections and full wave recti-fied and smoothed. Each stance signal was normalised to the