C/in. Biomech.
1994; 9: 21-27
Spectral signature of forces to discriminate perturbations in standing posture B A McClenaghan L Thorn bs PhD4
PED’,
H Williams
PW, J Dickerson
PhD3,
’Motor Rehabilitation
Laboratory; ‘Motor Control/Development Laboratory, Department of Exercise Science, School of Public Health; 3Department of Civil Engineering, College of Engineering; ‘Department of Statistics, College of Science and Mathematics, University of South Carolina, Columbia, USA
Summary This study proposes a methodology for the collection and analysis of the spectral characteristics of human movement patterns. Specifically, the purpose of this study was to determine the reliability of the spectral signature obtained from postural forces and the usefulness of the technique in identifying perturbations in standing posture. Data collected included trials of the experimental protocol under normal standing conditions and under three experimental conditions designed to perturb stability. Results of this investigation indicated that spectral signatures created from ground forces using the methodology proposed in this study were highly reliable within individuals and across different testing sessions. These data further indicate that spectral signatures obtained from ground reaction forces during standing provide a sensitive indicator of an individual’s postural stability.
Relevance Interest in the preclinical assessment of individuals exhibiting neurological impairments has resulted in the development of new techniques to evaluate motor function. This investigation suggests the use of spectral analysis as a technique to evaluate postural stability. Key words: Spectral signature, standing posture
Introduction
The belief that the progression of many neurological impairments may be slowed by modern rehabilitative techniques has stimulated an interest in the development of assessment techniques which can be used to identify ‘high risk’ individuals during the preclinical period. It has been suggested that a number of motor manifestations of neuromuscular disease are evident prior to the time clinical features become evident. This need and recent advances in computerized data acquisition and the reduced cost of Received: 23 April 1992 Accepted: 16 December 1992 Corrkpondence and reprint requests to: Bruce
A McClenaghan, Professor and Director. Motor Rehabilitation Laboratorv. Department of Exercise Science. University of South Carolina, Columbia. South Carolina 29208, USA 0 1994 Butterworth-Heinemann 026%0033/941010021-07
Ltd
data processing have resulted in new ways of addressing clinical questions not previously possible. One such in the field of mechanical approach prominent engineering is the identification of a spectral signature. The term signature is used to describe signal patterns that characterize the state or dynamic properties of systems from which they are acquired. In other words, signatures are used to profile the dynamic properties and/or behaviour of mechanical systems’. Monitoring the spectral signature of mechanical systems periodically has allowed industrial engineers to obtain the earliest possible warning signs of system failure, thus providing for early diagnosis, and/or prediction, of impending failure. This technique involves transforming vibrations into analogue signals, decomposing these signals into their frequency components, describing distinctive characteristics of the frequency spectra, and identifying changes in these spectral characteristics that indicate impaired functioning.
22
C/in. Biomech. 1994; 9: No 1
In a typical application the vibrational characteristics of a piece of equipment is monitored on a regular and systematic basis and then compared with previously established spectral characteristics. If subsequent monitoring of vibrational characteristics of the equipment indicates that the spectral signature is ‘close enough’ to the original signature the equipment is considered to be mechanically sound. If, however, there are significant differences between the two spectral signatures, specific features of the most recent spectra are analysed in order to diagnose where potential failures may lie (e.g. bearing fault or shaft imbalance). Numerous other applications of this technique have been documented, ranging from discriminating between seismic records originating from nuclear explosions and those originating from earthquakes to classifying an individual as being affected or unaffected by a certain neurological disease on the basis of brain waves extracted from the electroencephalogram2. Additional examples of this technique can be found in audiology, psychology, and other disciplines3.
Application of the spectral signature to the human motor control system
Application of spectral signature theory to the human motor system has had only limited study. There are, however, considerable data on motor systems of animals and humans that suggest that such a phenomenon exists. The most compelling evidence is from research on pattern generators. Marsden suggested that the basal ganglia are responsible for the automatic execution of learned motor plans. There appears to exist in animals an elaborate neural circuitry in the spinal cord which acts to generate and maintain basic locomotor patterns. These neural circuits or pattern generators, when activated, produce an identifiable pattern of inter-limb coordination which is used in a variety of locomotor patterns. Different locomotor patterns are produced by modifying time and force characteristics of the pattern generator output. If the concept of a pattern generator can be compared to a spectral signature, then it would appear that the human motor system has a definite signature associated with producing motor patterns. There is also indirect evidence that there may be a similar signature in the postural control system. This is seen in well-documented research on postural synergies. Woolacott et al.” have shown that when a person is standing in the upright position, and that position is perturbed unexpectedly, the motor system initiates a set of automatic postural responses to correct for perturbation of balance (sway). The general time and force characteristics of these postural synergies are similar in normal adults. However, such parameters are often modified in individuals with selected CNS dysfunction and in the aged. Analysis of frequency characteristics has been applied to the study and verification of other motor
patterns, including handwriting and vocalization. For example, several studies have used a three-axis force-sensitive pen to transform the dynamics of human handwriting into electrical signals that can be analysed 6,7. In the Lam and Kamins’ study6 a fast Fourier transform (FIT) was used to convert normalized handwritten signatures into frequency spectra that were used to determine whether or not future signatures were produced by the same individual. In this way positive identification of personnel for security reasons could be quickly and accurately accomplished. In another study investigators were able to classify infants’ cries as indicating pain or hunger based on frequency characteristics of the vocal patterns’. Classification in this study was made by comparing the spectral characteristics of an unknown cry with the known spectral signature derived from data previously collected during periods of known hunger and pain. A recent approach to studying postural stability is that of analysing spectral or frequency characteristics of sway patterns. Spectral analysis allows one to identify the frequency characteristics of sway patterns and to determine the distribution of energy across a range of harmonics. A variety of physical impairments have been associated with different energy distributions’. Other work has indicated that selected frequency characteristics of postural sway patterns may reflect the influence of different sensory or neural control systems”. This investigation was designed as an initial step in determining if the spectral signature concept is applicable to the human motor control system. This study proposes a methodology for the collection and analysis of the spectral characteristics of postural data. Specifically, the purpose of this study was to determine the reliability of the spectral signature obtained from postural forces and the usefulness of the technique in predicting perturbations in standing posture. The overall goal of this research is to develop a ‘marker’ that can be used to screen large populations of individuals for the purpose of identifying those at high risk of falling prior to such an event.
Relationship of postural sway and instability
Since standing is a dynamic event, a small amount of postural sway is a normal phenomenon”. During standing the centre of gravity is placed above a relatively small base of support. Standing posture has been described as a constant struggle to remain erect against the force of gravity12. Campbell13 noted that it was a common mistake to associate the idealized erect posture with a static concept of standing, instead of considering constant, continuous posture as adaptation. Hellebrandt14 similarly stated that standing is in reality movement upon a stationary base, with postural sway being inseparable from man’s upright stance. In more recent work, BrantonI defined postural behaviour as spontaneous attempts at
McClenaghan
attaining relative stability of the body’s segmented structure. Postural adjustments underlying the maintenance of equilibrium result from the integration of afferent information (proprioceptive, vestibular, and visual) into motor responses that produce postural sway and maintain the centre of gravity over its base of support. Traditionally, postural sway has been used extensively as a clinical measure of patient stability. Sheldon16 conducted the first investigation into the changing pattern of instability with age. He demonstrated that children have more postural sway than adults and the amount of sway tends to decrease during later adolescence and is maintained until about the age of 60 years. Hasselkus and Shambes” confirmed the results of this study with women and noted that the elderly sway more than young adults. Overstall et al.” similarly found that postural sway increased with age and was greater in women at all ages. Results of this study indicated there were no differences between subjects with no history of falls and those who fell due to tripping. In both sexes, postural sway was significantly greater in subjects who had experienced a fall caused by loss of stability. Overstall et al.‘” reported a relationship between sway and the number of past falls experienced by the individual. Sheldon” suggested that the inability to control sway in advancing years played an important part in the tendency of old people to fall. It was suggested that the change in sway with age was due to a change in the ability to control random movements that originated centrally. The history of evaluating postural sway is extensive and with modern techniques it is possible to get a reliabie and accurate picture of postural sway patternsiY.‘“. Using postural sway during normal standing to identify individuals at risk of future falls has been suggested by several investigations’8.‘9”-2j. Although the relationship between age and increased sway and falls has been recognized, little effort has been directed toward developing and quantifying measures of sway to predict the risk of falling in the elderly2”. It has been suggested that the limited clinical use of sway in predicting risk of falling is due to the expense of the equipment and the inability to obtain
Table 1. Experimental
conditions
et al.: Spectral characteristics
of postural forces
23
readily usable resultsis. Methods Subjects for the study were undergraduates (n = 1.5) enrolled in upper division courses in the Department of Exercise Science, University of South Carolina. All subjects had normal vision and no history of impaired stability. Subjects came to the laboratory on three occasions at similar times during the day and were tested at l-week intervals. Data collected during sessions 1 and 2 included nine trials of the experimental protocol under normal standing conditions. During session 3, postural data were collected under three experimental conditions designed to perturb stability. All subjects were exposed to each of the following conditions: dark environment, visual conflicting environment and vestibular disturbance (Table 1). Five trials of postural data were collected in each condition. Conditions were randomly presented with sufficient time between to allow for complete recovery. A trial consisted of standing on the force platform for 30 s. Foot placement was standardized across trials. Subjects were seated between trials and body weight was subtracted from the measurement platform prior to data collection. Multiple channels of force data were collected and used to create a spectral signature characteristic of the subject. The creation of the spectral signature is completed in four steps including (1) collection of the time series force vectors, (2) conversion of the force data using a spectral analysis and smoothing the resulting spectral data to highlight spectral characteristics of the data, (3) extracting from the data selected dependent measures that are used to construct the spectral signature, and (4) statistical analysis of the spectral signature to address specific research questions.
Collection of force vectors
Analogue signals were obtained from a Kistler measurement platform. This instrument produces three analogue signals that are amplified and input (20 Hz)
used to perturb postural stability
‘I. Full visual environment The subject fixates on a black circle placed at eye-height and directly in front of the subject (-2 lighted with no objects in the line of sight.
m). The room is fully
2. Dark environment After the subject is in position on the force platform, lights are turned off. The laboratory is totally dark. 3. Conflicting visual feedback A large white, opaque spherical dome with black vertical stripes (spaced at 25 mm intervals) is secured onto the subject’s head. The subject fixates on an ‘X’ placed on the centre of the dome. Movement of the dome is in concert with the subject’s sway and give cues that indicate ‘no sway’or movement is occurring. 4. Vestibular conflict Subject’s are spun in a chair prior to each trial. The quantity and intensity of the spin was controlled.
24
Clin. Biomech.
1994; 9: No 1
through a 1Zbit A/D converter (resolution = 0.2 N) to a laboratory computer. As the subject stood stationary on the force platform, the three analogue signals of the represented a time-history generated instantaneous three-dimensional forces applied to the surface of the platform. Software, written in ASYST (Kiethley-Metrabyte, Taunton, MA, USA) was developed for real-time data collection, analysis and storage of the digitized force vectors.
6.0 z s b LL
0.0
-6.0
I
-12.0 -
I Spectral analysis and smoothing the spectrum
A fast fourier transform (FIT) algorithm was used to obtain the spectral characteristics of the time series data. Typically the spectrum has an erratic and fluctuating form which often makes interpretation with formal statistical procedures difficult. It was reasonable to believe that the power spectrum, being a distribution of energy, is a smooth function, which suggests that any reasonable estimator should share this property. For this reason a Bartlett lag window (M = 5) was used to smooth the spectrum (Figure 1).
3.0
1 9.0
a
I
I
I
15.0
21.0
27.0
7.0
9.0
3.50
4.50
Time (5)
36.0 1, ;
28.0
I "z
20.0
k ;
12.0
4.0
1.o
Feature extraction
The spectral data created from each force vector (lateral, anteroposterior, and vertical) contained a set of unique features that were extracted and used to create the spectral signature. The primary objective of this phase of the study was to identify those bandwidths and force directions that most discriminated between two sets of spectral data which represented stable and unstable postures. For this purpose spectral data (O-5 Hz) were reduced by integrating the spectrum at selected bandwidth intervals of 0.2 Hz. The spectral signature (three forces combined) thus consisted of a set of 75 values.
Once the spectral signature was created, these data were reduced to a subset of features that could be more easily analysed (i.e. working with a large set of features presents computational difficulties since matrix inversions are required). A principal components analysis was used to reduce the data set and to identify those factors (i.e. principal components) that explained the maximum variation in the data with as few factors as possible. A new data set was created using the factor coefficients obtained from the principal components analysis and the raw data using the SCORE procedure (SAS Institute, Cary, NC). This statistical procedure multiplies values from two data sets, one containing factor coefficients and the other containing raw data. The result of this multiplicationis a data set containing linear combinations of the coefficients and the raw data values. The resulting factor scores were used in subsequent analysis. The reliability of factor scores generated from the principal components analysis was estimated using a
5.0 Frequency
18.0
(Hz)
Fn
2.0 0.50
C
Statistical analysis of spectral signature
3.0
b
1.50
2.50 Frequency
(Hz)
Figure 1. Spectral analysis and smoothing of the spectrum. Steps in creating the spectral signature included (a) obtaining the time-series force data, (b) frequency analysis of the data, and (c) smoothing of the power spectrum.
2-way ANOVA and intra-class correlation technique. The intra-class correlation coefficient is widely recognized as the most appropriate method for estimating reliability since it is sensitive to more sources of measurement error than other methods26,27. Discriminant analyses were used to determine if those factors identified through the principal components analysis were sensitive enough to identify and correctly classify perturbations in postural stability. Results
Spectral profiles were averaged across all subjects for each experimental condition. The resulting profiles in the lateral and anteroposterior directions are presented
McClenaghan
40
. 30
et al.: Spectral characteristics
of postural forces
25
variance in the data and was characterized by forces exerted in the anteroposterior direction between 0.6 and 3.1 Hz. Factors 2 and 3 were best characterized by forces exerted in the mediolateral and vertical directions respectively and accounted for 9.83% and 4.47% of the sample variance. Factor 4 accounted for less than 5% of the variance in the data and was characterized by forces in the anteroposterior plane (~3.1 Hz).
20
Reliability
Data collected under full visual conditions were used to estimate the reliability of the spectral signature. A 2-way ANOVA (subject x trial) was used to estimate reliability of factor scores. Results of this analysis indicated that there were significant (P < 0.05) differences between subjects for all factors. There were no significant differences within individuals across trials. Intra-class reliability coefficients (R) were calculated for each factor as a function of trials and sessions (Table 2). Within-subject across-trials intra-class correlations were high (R > 0.90) for factors 1, 2, and 4. Factor 3, spectral data in the vertical direction, was less reliable across trials (R = 0.66). Reliability of the spectral data across testing sessions (l-week intervals) followed a similar trend; coefficients were high (R > 0.90) for factors 1, 2, and 4. Reliability for factor 3 was lower (R = 0.77).
10
';; I N
z
i a"
estimates
0 40
30
20
10
Discriminant 0
Figure 2. Averaged smoothed power spectra. Profiles created across all subjects in the (a) lateral and (b) anteroposterior direction for each experimental condition. Full vision; ---dark; ----visual conflict; .. .... vestibular.
in Figure 2. The greatest similarity was observed between spectral profiles obtained during full and visual conflict conditions. Vestibular conflict resulted in greater total energy distributed over the widest bandwidth. Elimination of visual feedback (dark) tended to increase the energy of the spectrum and result in a slightly lower mean frequency.
Principal
Discriminant analysis was used to determine if the spectral signature was a sensitive indicator of postural stability. Because significant differences were found between subjects, a discriminant analysis was conducted for each subject. Classifications for each perturbation were based on criteria developed from trials collected under that condition. Results from these analyses were tabulated and summarized in Table 3. Perturbations were appropriately classified according to postural sway/standing conditions with 87-98% accuracy.
Table 2. Subject
inter-trial
and inter-session Inter-trial
reliability
(17) Inter-session
Factor
Description
1
Anteroposterior spectral data
0.90
0.94
Lateral spectral data
0.91
0.94
Vertical spectral data
0.66
0.71
Anteroposterior spectral data > 3.13 Hz
0.92
0.97
components
Spectral data were initially analysed using the principal components technique followed by varimax rotation. A total of ten factors were derived from the principal components analysis, accounting for 56.9% of the variance in the data. Four of the extracted factors, representing 44.34% variance, were used in subsequent analysis. The initial factor accounted for 25.93% of the
analysis
2 3 4
(R)
26
Clin. Biomech.
1994; 9: No 1
Table 3. Results of discriminant
analysis Into group
From group
Full vision (%)
Dark (%)
Visual conflict (%)
Vestibular (%I
Trials (n)
87.1
1.8
10.1
1.0
270
Dark
7.2
90.4
1.2
1.2
75
Visual conflict
2.3
0
97.7
0
75
Vestibular
2.4
0
1.2
96.4
75
Full vision
Discussion
This investigation used spectral and multivariate analysis techniques to determine the reliability of spectral signature obtained from postural forces and its ability to discriminate between normal standing and three experimental conditions designed to perturb standing stability. The initial question addressed by this study was: is the spectral signature technique reliable? Results of this investigation indicated that spectral signatures created from ground forces using the methodology proposed in this study were highly reliable within individuals and across different testing sessions. Significant individual differences were observed in the data, indicating that the spectral signature obtained from ground reaction forces is subject dependent. Murry et al.” reported similar results for movement of the centre of pressure of subjects during standing. They reported large standard deviations for movement of centre of pressure across subjects; much smaller standard deviations were observed across repeated tests within individual subjects. Other investigations also have demonstrated that various measures of postural sway are reliable”~20~28. Significant individual differences in the spectral data support the assumption that a unique spectral signature can be identified for individuals and that this signature is different for individuals. Individual differences in postural sway as manifested in differences in individual spectral signatures may reflect differences in such characteristics as body size, body proportions, and physiological, and/or sensory, neuromuscular functioning. The second thrust of this study was to determine whether or not the derived spectral signature of the individual could be used to distinguish between stable and unstable postures. Discriminant analyses using spectral data resulted in high classification accuracy for data obtained during different standing/postural sway conditions. These data indicate that spectral signatures, obtained from ground reaction forces during standing, change when the postural system is perturbed and provide a sensitive indicator of an individual’s postural stability. The sensitivity of spectral signature to changes in postural stability has important implications for its
use in identification of impaired postural control. For example, to predict or identify individuals who are at
high risk of falling (i.e. elderly), one has to be able to detect changes in postural control that reflect instability. Data from the present study suggest that the spectral signature technique potentially can be used for such detection. If the spectral signature technique is successfully used in this application, it is possible that it can also be used to provide a ‘marker’ that can be used to identify other impending impairments, that is to identify the risk of an impairment in motor control before its actual onset. Tetrud2’ suggested that handwriting and speech are commonly affected prior to clear diagnosis of the onset of Parkinson’s disease. Additional studies need to be conducted before the potential for this type of analysis can be explored. Studies conducted on a variety of subject populations must confirm the reliability of the spectral signature and investigate its use in the study of additional motor patterns.
References Braun S. Mechanical signature analysis. Proceedings of the Ninth Biennial Conference on Mechanical Vibration and Noise of the Design and Production Engineering Technical Conferences. 1983: ll- 14 Alagon J. Spectral discrimination for two groups of time series. J Time Series Anal 1989; 10: 203- 14 Shumway RH. Discriminant analysis for time series. In: Krishnaiah PR, Kanal LN, ed. Handbook of Statistics, Vol. 2 Amsterdam: North-Holland, 1982: l-46 Marsden CD. The mysterious motor function of the basal ganglia. Neurology 1982; 32: 514-39 Woolacott M, Shumway-Cook A, Nashner L. Aging and postural control: changes in sensory organization and muscular coordination. Int J Aging Hum Dev 1986; 23: 97-114
Lam CF, Kamins D. Signature recognition through spectral analysis. Pattern Recognition 1989; 22: 39-44 Liu CN, Herbst NM, Anthony NJ. Automatic signature verification: system description and field test results. IEEE Trans Systems, Man Cybern 1979; 9: 35-8
Cohen A, Zmora E. Automatic classification of infants’ hunger and pain cry. In: Cappellini V, Constantinides A, ed. Digital Signal Processing - 84. New York: NorthHolland, 1984 Mauritz KH, Dichgans J, Hufschmidt A. Quantitative analysis of stance in late cortical cerebellar atrophy of the anterior lobe and other forms of cerebellar ataxia. Brain 1979; 102: 461-82
McClenaghan
10 Grantchev G, Popov V. Quantitative evaluation of induced body oscillation in man. Aggressologie 1973; 14c: 91-4
11 Murry MP, Siereg AA, Sepic SB. Normal postural stability and steadiness: quantitative assessment. J Bone Joint Surg 1975: 57-A: 510-16 12 Wagenhauser FJ. Epidemiology of postural disorders in young people. In: Fehr K, Huskisson EC, Wilhelmi E, ed. Rheumatological Research and the Fight Against Rheumatic Diseases in Switzerland. Basel, Eular, 1978 13 Campbell DG. Posture: a gesture toward life. Physiother Rev 193.5; 15: 43-7 14 Hellebrandt FA. Standing as a geotropic reflex. Am J Physioll938; 121: 194-202
15 Branton P. Behavior, body mechanics and discomfort. In: Grandjean E, ed. Proceedings of the Symposium on Sitting Posture. London: Taylor & Francis, 1969: 202-13 16 Sheldon JH. The effect of age on the control of sway.
Falls in the elderly related to postural imbalance. Br Med J 1977; 1: 261 19 Nasher L, Woolacott M. The organization of rapid postural adjustments of standing humans: An experimental-conceptual model. In: Talbott RE. Humphrey DR, ed. Posture and Movement. New York, Raven Press, 1979
of postural forces
1982; 49: 169-77 21 Brocklehurst JC, Robertson
22 23
24 25 26
27
D, James-Groom P. Clinical correlates of sway in old age: sensory modalities. AgeAging 1982; 11: l-10 Era P, Heikkinen E. Postural sway during standing and unexpected disturbance of balance in random samples of men of different ages. J Gerontol1985; 8: 287-95 Fernie GR, Gryfe CI, Holliday PJ, Llewellyn A. The relationship of postural sway in standing to the incidence of falls in geriatric subjects. Age Ageing 1982; 11: 11-16 Wootton R, Bryson E, Eisasser U et al. Risk factors for fractured neck in the elderly. Age Aging 1982; 11: 160-8 Kirshen AJ, Cape RDT, Hayes KC, Spencer JD. Postural sway and cardiovascular parameters associated with falls in the elderly. J Clin Exp Gerontoll984; 6: 291-307 Kroll W. A note on the coefficient of intra-class correlation as an estimate of reliability. Res Q 1962; 33: 313-6 Baumgartner TA. (1989) Norm-referenced measurement reliability. In: Safrit MJ, Wood TM, ed. Measurement Concepts in Physical Education and Exercise Science.
Human Kinetics, Champaign, Illinois 28 Lord SR, Clark RD, Webster IW. Postural stability and associated physiological factors in a population of aged persons. J Gerontoll991; 46: M69-76 29 Tetrud JW. Preclinical detection of motor and nonmotor manifestations. Geriatrics 1991; 46 (Suppl 11: 43-6
International Conference on Biomedical Engineering
Hong Kong
27
20 Soames RW, Atha J. The spectral characteristics of postural sway behavior. Eur J Appl Physiol Occup Physiol
Gerontol Clin 1963; 5: 129-38 17 Hasselkus BR, Shambes GM. Aging and postural sway in women. J Gerontol1975; 30: 661 18 Overstall PM, Exton-Smith A, Imms FJ, Johnson AL.
et al.: Spectral characteristics
Technology for Health
Hong Kong Convention & Exhibition Centre 7-9 April 1994 BME’94 will bring together engineers, technologists, physicists, physicians and other professionals from around the world, to share and report their work in the area of biomedical engineering. This conference will also be of interest to manufacturers of biomedical engineering and health technology products. The programme of free paper presentations will be augmented by two plenary keynote lectures. A satellite symposium on rehabilitation technology will be held on 10th April. For further information on BME’94 please contact: Conference Secretary, BME’94, c/o Rehabilitation Engineering Centre, Hong Kong Polytechnic, Hunghom, Kowloon, Hong Kong. Tel: (852) 766 7683 Fax: (852) 362 4365 Telex: 38964 Polyx Hx Email:
[email protected] Organized by The Hong Kong Institution of Engineers - Biomedical Engineering Division