Med. & Biol. Eng. & Comput., 1977, 15, 184-188
Some applications of human.operator research to the assessment of disability in stroke P. A. Lynn W. R, Parker
G . A . L . Reed
R. Langton H e w e r Avon Stroke Unit, Frenchay Hospital, Bristol, England
Department of Electrical & Electronic Engineering, University of Bristol, England
A b s t r a c t - - T h e present assessment of partially-paralysed victims of stroke is complicated by a lack of objective measurement, A tracking task of the type widely used to measure the performance of skilled opelators in a manual control task is used here to encourage controlled movement in the semiparalysed arm. Quantitative measures derived from such tests provide sensitive indication of trends in performance during rehabilitation, which cannot be detected by routine clinical examination. Keywords--Stroke research, Rehabilitation
Introduction EVrRX" year about 100 000 people in England and Wales suffer a stroke, which is a rather general term used to describe the effects of a disturbance in the blood supply to the brain. In approximate terms, one third of this total die within the first three weeks, and of those who survive about half are left with serious disability--usually a hemiplegia, involving paralysis of the arm and leg on one side of the body. The hemiplegia may be accompanied by a wide variety of other problems, including partial loss of vision (hemianopia), speech disturbance (dysphasia), loss of spatial orientation, and loss of sensation in the affected limbs: and there are very often severe psychological and social problems b o t h for the patient and for his or her family. Although 70 ~ of strokes affect people aged 65 or over, a significant proportion occur in people who are in their 40s and 50s, and who therefore tend to have demanding responsibilities to their families or to their jobs. The precise disabilities suffered by the victim of stroke depend upon the site and extent of brain damage. A right hemiplegia (i.e. paralysis of the right arm and leg) is caused by left-sided brain damage and is often accompanied by speech difficulties, whereas right-sided brain damage often gives rise not only to left-sided paralysis, but also to much more subtle difficulties of comprehension and spatial orientation which may be even more handicapping. Hemiplegia is generally characFirst received 8th March and in final form 7th April 1976
184
terised in its early stages by flaccidity in the affected limbs, giving way after several weeks to a spastic phase in which stiffening of the limbs tends to mask any residual voluntary control in the neuromuscular system. Although about 7 0 ~ of survivors ultimately learn to walk again with or without the aid of a stick or tripod, only about I0 ~ recover normal function in the affected arm and hand. Effective rehabilitation of the surviving victims of stroke is obviously an extremely important problem, from the point of view of both the individual patient and the community. One of the biggest problems is to decide on the basis for physical treatment. For example, a majority of hemiplegic patients receive physiotherapy, but various techniques are currently in use and these are not easy to assess objectively. Such a situation can only be clarified by reliable measurement of the effects of different types of treatment, so that the problem of rehabilitation of the patient is inevitably bound up with the need for objective assessment of his condition.
aemand . ~ display
Fig. I A pursuit tracking task
Medical & Biological Engineering & Computing
March 1977
Human-operator research Our initial interest has centred on previous engineering studies of the skilled human subject in a feedback-control environment, and their implications for stroke rehabilitation. A common experimental technique in this field [see, for example, the reports of the NASA-University annual conferences on manual control (1965 onwards)] is the tracking task, one variation of which is illustrated in Fig. 1. In 'pursuit' tracking, the operator controls the position of a symbol (e.g. a cross) on a cathoderay tube, using a suitable transducer. A second symbol (e.g. a circle) is moved automatically, and the operator attempts to track it so as to minimise the positional error between the two symbols. The 'controlled dynamics' may be adjusted to simulate the properties of the system being controlled. When the demand symbol is moved in a simple repetitive m a n n e r - - f o r example, sinusoidally-there is evidence that the operator first recognises the pattern and then generates a response of similar amplitude which he attempts to synchronise with the demand (PEw et al., 1966). The ability to recognise simple patterns may be counteracted by use of random demand signals, and studies such as that by TAYLOR (1967) represent attempts to model performance using established system identification techniques. Authors such as WASIKO et al. (1966) and NIELSEN (1972) have attempted more explicitly physiological interpretations of experimental results, whereas POULXON (1967), amongst others, has sought to overcome the difficulties of continuous, linear, representations of human performance by an approach based upon intermittent (sampling) behaviour. In seeking to apply such techniques to the assessment of stroke victims, there are perhaps two main dangers. The first is that experiments designed to test skilled operators may prove quite unsuitable for the disabled; and secondly, certain attributes of skilled performance--for example, consistency between successive tests--may be largely irrelevant to a study of recovery in stroke patients.
arm. About 2 min is allowed between each test for relaxation, during which time the screen is blanked. Deflection of the circular symbol (the demand) is achieved using a maximal-length pseudorandom binary sequence (p.r.b.s.) of 255 characters, clocked at 4~Hz, and lowpass filtered by a 4th-order Butterworth filter having an adjustable cutoff frequency to give different degrees of task difficulty. In the initial version of the experiment, symbol generation and deflection was produced by analogue circuits, and the demand and response waveforms were recorded on a f.m. tape recorder prior to off-line analysis using the University of Bristol's main computer. By early 1975, a PDP 11 digital computer had taken over the task of routine acquisition and analysis of experimental data, and is now being used also for symbol deflection and experimental control.
demand
;
skJl~d sebject
patent A
patJent B m~ent C
patient D
patient E ~
The experiment Our initial experiment is of the pursuit tracking type already illustrated in Fig. 1. The patient sits in front of a display oscilloscope with the arm supported in a horizontal arm rest, freely pivoted at the elbow. A circle on the screen is deflected horizontally in an apparently random way, and the patient attempts to track it with a second symbol (a cross), controlled by movement at the elbow. The duration of the test is about one minute, and a score representing time-on-target is displayed to the patient after each attempt. In the great majority of cases, a session has consisted of an initial test with the patient's undamaged arm, followed by three attempts with the hemiplegic Medical & Biological Engineering & Computing
m~xx~s
'- ~
I
I
0
30
patient F
,
-I
60 secs.
Fig. 2 Typical tracking performances of one skilled subject and six hemiplegic patients, in response to a band-limited demand waveform (0 to 1 tad/s)
Initial evaluation of the suitability of the task for a wide range of hemiplegic subjects suggested the following additional parameters: mean angle of patient's elbow joint = 100~ distance of face from screen = 1-6m; symbol size = 0.06m; maxim u m deflection o f circular symbol on the screen March 1977
185
= + O. lOm, corresponding to an elbow movement of + 20~ and a filter cutoff frequency of 1 rad/s (although 2 or 4 rad/s proved suitable for some less-disabled patients).
Results Nearly 40 hemiplegic patients were tested on the tracking apparatus during the early stages of the project, in order to assess the suitability of this type of experiment for a wide range of patients with varying degrees of disability. The patients were selected as suitable for tracking during clinical examination by a neurologist, on the basis that they would be able to comprehend the task and seemed able to exert some degree--however small--of controlled movement in the hemiplegic elbow. In most cases, these patients were seen for between one and three sessions, but not followed up in the longer term. Typical performances achieved by six of these patients and one skilled, normal, subject are shown in Fig. 2. It is at once apparent that the degree of controlled movement varies very widely indeed: and, perhaps more interestingly, there appear to be marked differences in the type of control strategy used, to which we will refer later. Having established a suitable set of experimental parameters, we then tested a small number of patients over a much longer period, with the aim of developing quantitative measures for following trends in performance. Although space does not permit a full account of these tests, which are reported elsewhere (PARKER, 1976), some of the measures we have investigated may be discussed by reference to results from just two patients. Patients A and B, whose initial tracking performances have
left arm 77 wps ~.
I
right arm
,
,
-15 =
0
i
I
*15 ~
-15 ~
~
,15 =
Fig. 3 Histograms of instantaneous tracking error for patient A, obtained at 77, 80 and 93 weeks post stroke (w.p.s.)
already been illustrated in Fig. 2, were both female, aged 55, and had their strokes about one year prior to testing. Both patients had suffered a right hemiplegia, and although they could walk with the aid of a stick, neither had regained significant functional use of the damaged arm. They attempted the 186
tracking task at approximately monthly intervals over a period of about a year and a half. Fig. 3 shows typical histograms of instantaneous error between demand and response for patient A (sampled 10 times per second over the 1 min task), relevant to early, intermediate and later stages of the test programme. These results suggest a modest improvement with time for the left (nominally undamaged) arm, and a much more dramatic improvement for the right arm.
t A
5
-
f/Left
I
5
10
15
ll0
115
session
no.
arm
session
no.
Fig. 4 Trends in mean-square tracking error for two patients with right hemiplegia
Whereas patient A tracked a demand waveform band limited to 0 to 1 rad/s, as already shown in Fig. 2, patient B was soon able to tackle a more difficult task with a bandwidth from 0 to 2 rad/s. However, equivalent histograms for patient B suggest hardly any improvement in performance over the test period. Trends in performance cannot be conveniently stored in the form of a histogram for each tracking test. In an attempt to reduce the data, we next computed the mean square error--the second moment of the error histogram--and averaged it for left and right arm over any one tracking session (each session consisting of one or two attempts with the left arm, followed by three attempts with the damaged right arm). Session averages for both patients are shown in Fig. 4, and clearly suggest both differences between tracking ability with left and right arms, and the substantial initial improvement in performance achieved by patient A with her hemiplegic arm, compared with patient B. The difficulty of trying to represent the tracking performance of such patients by a quasilinear, time-invariant, transfer function is illustrated by Fig. 5, which shows typical frequency responses relevant to a single attempt at tracking by each patient. In spite of the relatively skilled performance of these particular patients, the short duration and very limited bandwidth of the demand waveform gives results which are so statistically unreliable
Medical & Biological Engineering & Computing
March 1977
as to make a description by, say, a 2nd- or 3rd-order transfer function quite unrealistic. We have therefore investigated a simpler approach--that of a characterisation by just two parameters, average gain and time delay. Although these parameters could be estimated by fitting straightline approximations to gain and phase data, we have used a more convenient method based upon computing the crosscorrelation function between demand and response waveforms. A n essentially similar approach has been used by CASSELL (1973) in a study of tracking performance by patients suffering from Parkinson's disease. Fig. 6 illustrates trends in gain and time-delay parameters for both our patients, obtained by the above method after an initial tapering off of the gain
patient A
1.0 ~
~
.
-
~
~
A
lb
15
session no. - - ~
10
15
session no.
~9
(dB)
9
-
o .o
o
v9[
.2~I _o
patient B
Fig. 6 Trends in gain and time delay. (s)
:o%
o
*
o
" 9 ~'~%%*'-'_ o o OTp o
hase
o oo
o 05_60,
I 0.01
I
0.1
5
ase
o
gain
responses contain a rather large amount of highfrequency jitter, which cannot be accounted for by a quasilinear model; and secondly, the use of a
I
1.0
frequency (Hz)
Fig. 5 Frequency-response estimates
first and last 10% of the raw data records with window functions of half-Hanning form (raised cosine bell). I n spite of short-term fluctuations, the results once again suggest fairly dramatic improvement in tracking performance by patient A with her hemiplegic arm, and clearly indicate the differences between damaged and undamaged arms of both patients at all stages of the test programme. Discussion
The limited selection of results reported here illustrate some of the technical possibilities and difficulties in quantifying dynamic control in the hemiplegic arm. It is clear that relatively simple measures such as mean square error, or gain and time delay, can indicate trends in tracking performance which are undetectable by normal clinical examination (which failed to reveal any of the trends seen in Figs. 4 and 6). On the other hand, conventional technical measures of a more comprehensive kind, such as the frequency response, are problematical. The reasons for this seem straightforward: firstly, Fig. 2 shows that most patients'
Medical & Biological Engineering & Computing
1 min tracking task, rather than the 2 or even 4 m i n tasks which skilled subjects can tackle, makes the estimation of comprehensive measures subject to large statistical errors. Different approaches to the characterisation of tracking performance may, of course, prove more effective. For example, the discontinuous nature of some of the responses in Fig. 2 has prompted us to start investigating the use of sampled-data models, with promising initial results. Fig. 2 also raises the problem of control strategy. If tracking performance is characterised by a simple measure (such as mean-square error), a given level of performance can clearly be achieved with a variety of strategies. For example, Fig. 7, in which the mean-square error arising from various combinations of gain and time shift has been computed for the demand waveform of Fig. 2, shows that a patient with a very limited range of movement (small gain) coupled with the ability to predict the future time course of the demand waveform (small, .10
1.5
."~ gain -0.5
0-5 ~) *0.5 time shift (see)
1'5[ 0.5
time shift (sec,)
Fig. 7 Contour and 3-dimensional plots o f meansquare tracking error as a function o f gain and time shift
March 1977
187
or even negative time delay) could achieve the same error as a patient with a fuller range o f movement who responded rather slowly. The variety o f patients' tracking responses m a y reflect differences in such factors as visual perception or muscle spasticity, rather than differences in conscious strategy, and i t would be useful to quantify the various factors contributing to overall performance. This, however, would require performance models more explicitly related to the special problems of the stroke patient.
Acknowledgments--The authors wish to acknowledge the generous provision by the Medical Research Council of an online digital-computer system for data acquisition and analysis, and tke postgraduate research studentship awarded by the Science Research Council to W. R. Parker.
References CASS~LL, K. J. (1973) The usefulness of a temporal correlation technique in the assessment of human motor performance on a tracking device. Med. & BioL Eng. 11, 755
NASA-UNWERSITYannual conferences on manual control NASA special publications (1965 onwards) NIELSON, P. D. (1972) Speed of response or bandwidth of voluntary system controlling elbow position in intact man. Med. & Biol. Eng. 10, 450 PARKER, W. R. (1976) Aspects of electronic and control engineering applied to a study of recovery in hemiplegic stroke patients. Ph.D. thesis, University of Bristol PEW, R. W., DUrrENBACK,J. C. and FENSCH, L. K. (1966) A summary of sine wave tracking studies. NASAUniversity conference on manual control. No. 2, SP128, paper 1 POULTON, E. C. (1967) A quantal model for human tracking. NASA-University conference on manual control no. 3, SP144, paper 18
TAYLOR, L. W. (1967) A comparison of human performance modelling in the time and frequency domains. NASA-University conference on manual control no. 3, SP144, paper 9 WASIKO, R. J., McRuER, D. T. and MA6BALENO, R. E. (1966) Human pilot dynamic response in single loop systems with compensatory and pursuit displays. Air Force Flight Dynamics Laboratory technical report no. AFFDL-TR-66-137
Quelques-unes des applications de la recherche sur les op6rateurs humains das le domaine de I'evaluations des incapacit6s entrain6es par les attaques Sommaire---A pr6sent, le domaine de l'~valuation des victimes partiellement paralys~es par les attaques, est compliqu6, faute de mesurage objectif. Ici, on se sert d'une 'tgtche de piste', du type beaucoup utilis6 pour mesurer la performance d'op6rateurs sp6cialis6s, qui font une tO.che de contr61e manuelle pour encourager ]e mouvement contr616 dans le bras demi-paralys6. Des mesures quantitatives obtenus par de telles 6preuves, donnent une indication sensible des tendances de performance pendant la r66ducation, qu'on ne peut pas d6tecter a rexamen clinique courant.
188
Medical & Biological Engineering & Computing
March 1977
Some applications of human.operator research to the assessment of disability in stroke P. A. Lynn W. R, Parker
G . A . L . Reed
R. Langton H e w e r Avon Stroke Unit, Frenchay Hospital, Bristol, England
Department of Electrical & Electronic Engineering, University of Bristol, England
A b s t r a c t - - T h e present assessment of partially-paralysed victims of stroke is complicated by a lack of objective measurement, A tracking task of the type widely used to measure the performance of skilled opelators in a manual control task is used here to encourage controlled movement in the semiparalysed arm. Quantitative measures derived from such tests provide sensitive indication of trends in performance during rehabilitation, which cannot be detected by routine clinical examination. Keywords--Stroke research, Rehabilitation
Introduction EVrRX" year about 100 000 people in England and Wales suffer a stroke, which is a rather general term used to describe the effects of a disturbance in the blood supply to the brain. In approximate terms, one third of this total die within the first three weeks, and of those who survive about half are left with serious disability--usually a hemiplegia, involving paralysis of the arm and leg on one side of the body. The hemiplegia may be accompanied by a wide variety of other problems, including partial loss of vision (hemianopia), speech disturbance (dysphasia), loss of spatial orientation, and loss of sensation in the affected limbs: and there are very often severe psychological and social problems b o t h for the patient and for his or her family. Although 70 ~ of strokes affect people aged 65 or over, a significant proportion occur in people who are in their 40s and 50s, and who therefore tend to have demanding responsibilities to their families or to their jobs. The precise disabilities suffered by the victim of stroke depend upon the site and extent of brain damage. A right hemiplegia (i.e. paralysis of the right arm and leg) is caused by left-sided brain damage and is often accompanied by speech difficulties, whereas right-sided brain damage often gives rise not only to left-sided paralysis, but also to much more subtle difficulties of comprehension and spatial orientation which may be even more handicapping. Hemiplegia is generally characFirst received 8th March and in final form 7th April 1976
184
terised in its early stages by flaccidity in the affected limbs, giving way after several weeks to a spastic phase in which stiffening of the limbs tends to mask any residual voluntary control in the neuromuscular system. Although about 7 0 ~ of survivors ultimately learn to walk again with or without the aid of a stick or tripod, only about I0 ~ recover normal function in the affected arm and hand. Effective rehabilitation of the surviving victims of stroke is obviously an extremely important problem, from the point of view of both the individual patient and the community. One of the biggest problems is to decide on the basis for physical treatment. For example, a majority of hemiplegic patients receive physiotherapy, but various techniques are currently in use and these are not easy to assess objectively. Such a situation can only be clarified by reliable measurement of the effects of different types of treatment, so that the problem of rehabilitation of the patient is inevitably bound up with the need for objective assessment of his condition.
aemand . ~ display
Fig. I A pursuit tracking task
Medical & Biological Engineering & Computing
March 1977
Human-operator research Our initial interest has centred on previous engineering studies of the skilled human subject in a feedback-control environment, and their implications for stroke rehabilitation. A common experimental technique in this field [see, for example, the reports of the NASA-University annual conferences on manual control (1965 onwards)] is the tracking task, one variation of which is illustrated in Fig. 1. In 'pursuit' tracking, the operator controls the position of a symbol (e.g. a cross) on a cathoderay tube, using a suitable transducer. A second symbol (e.g. a circle) is moved automatically, and the operator attempts to track it so as to minimise the positional error between the two symbols. The 'controlled dynamics' may be adjusted to simulate the properties of the system being controlled. When the demand symbol is moved in a simple repetitive m a n n e r - - f o r example, sinusoidally-there is evidence that the operator first recognises the pattern and then generates a response of similar amplitude which he attempts to synchronise with the demand (PEw et al., 1966). The ability to recognise simple patterns may be counteracted by use of random demand signals, and studies such as that by TAYLOR (1967) represent attempts to model performance using established system identification techniques. Authors such as WASIKO et al. (1966) and NIELSEN (1972) have attempted more explicitly physiological interpretations of experimental results, whereas POULXON (1967), amongst others, has sought to overcome the difficulties of continuous, linear, representations of human performance by an approach based upon intermittent (sampling) behaviour. In seeking to apply such techniques to the assessment of stroke victims, there are perhaps two main dangers. The first is that experiments designed to test skilled operators may prove quite unsuitable for the disabled; and secondly, certain attributes of skilled performance--for example, consistency between successive tests--may be largely irrelevant to a study of recovery in stroke patients.
arm. About 2 min is allowed between each test for relaxation, during which time the screen is blanked. Deflection of the circular symbol (the demand) is achieved using a maximal-length pseudorandom binary sequence (p.r.b.s.) of 255 characters, clocked at 4~Hz, and lowpass filtered by a 4th-order Butterworth filter having an adjustable cutoff frequency to give different degrees of task difficulty. In the initial version of the experiment, symbol generation and deflection was produced by analogue circuits, and the demand and response waveforms were recorded on a f.m. tape recorder prior to off-line analysis using the University of Bristol's main computer. By early 1975, a PDP 11 digital computer had taken over the task of routine acquisition and analysis of experimental data, and is now being used also for symbol deflection and experimental control.
demand
;
skJl~d sebject
patent A
patJent B m~ent C
patient D
patient E ~
The experiment Our initial experiment is of the pursuit tracking type already illustrated in Fig. 1. The patient sits in front of a display oscilloscope with the arm supported in a horizontal arm rest, freely pivoted at the elbow. A circle on the screen is deflected horizontally in an apparently random way, and the patient attempts to track it with a second symbol (a cross), controlled by movement at the elbow. The duration of the test is about one minute, and a score representing time-on-target is displayed to the patient after each attempt. In the great majority of cases, a session has consisted of an initial test with the patient's undamaged arm, followed by three attempts with the hemiplegic Medical & Biological Engineering & Computing
m~xx~s
'- ~
I
I
0
30
patient F
,
-I
60 secs.
Fig. 2 Typical tracking performances of one skilled subject and six hemiplegic patients, in response to a band-limited demand waveform (0 to 1 tad/s)
Initial evaluation of the suitability of the task for a wide range of hemiplegic subjects suggested the following additional parameters: mean angle of patient's elbow joint = 100~ distance of face from screen = 1-6m; symbol size = 0.06m; maxim u m deflection o f circular symbol on the screen March 1977
185
= + O. lOm, corresponding to an elbow movement of + 20~ and a filter cutoff frequency of 1 rad/s (although 2 or 4 rad/s proved suitable for some less-disabled patients).
Results Nearly 40 hemiplegic patients were tested on the tracking apparatus during the early stages of the project, in order to assess the suitability of this type of experiment for a wide range of patients with varying degrees of disability. The patients were selected as suitable for tracking during clinical examination by a neurologist, on the basis that they would be able to comprehend the task and seemed able to exert some degree--however small--of controlled movement in the hemiplegic elbow. In most cases, these patients were seen for between one and three sessions, but not followed up in the longer term. Typical performances achieved by six of these patients and one skilled, normal, subject are shown in Fig. 2. It is at once apparent that the degree of controlled movement varies very widely indeed: and, perhaps more interestingly, there appear to be marked differences in the type of control strategy used, to which we will refer later. Having established a suitable set of experimental parameters, we then tested a small number of patients over a much longer period, with the aim of developing quantitative measures for following trends in performance. Although space does not permit a full account of these tests, which are reported elsewhere (PARKER, 1976), some of the measures we have investigated may be discussed by reference to results from just two patients. Patients A and B, whose initial tracking performances have
left arm 77 wps ~.
I
right arm
,
,
-15 =
0
i
I
*15 ~
-15 ~
~
,15 =
Fig. 3 Histograms of instantaneous tracking error for patient A, obtained at 77, 80 and 93 weeks post stroke (w.p.s.)
already been illustrated in Fig. 2, were both female, aged 55, and had their strokes about one year prior to testing. Both patients had suffered a right hemiplegia, and although they could walk with the aid of a stick, neither had regained significant functional use of the damaged arm. They attempted the 186
tracking task at approximately monthly intervals over a period of about a year and a half. Fig. 3 shows typical histograms of instantaneous error between demand and response for patient A (sampled 10 times per second over the 1 min task), relevant to early, intermediate and later stages of the test programme. These results suggest a modest improvement with time for the left (nominally undamaged) arm, and a much more dramatic improvement for the right arm.
t A
5
-
f/Left
I
5
10
15
ll0
115
session
no.
arm
session
no.
Fig. 4 Trends in mean-square tracking error for two patients with right hemiplegia
Whereas patient A tracked a demand waveform band limited to 0 to 1 rad/s, as already shown in Fig. 2, patient B was soon able to tackle a more difficult task with a bandwidth from 0 to 2 rad/s. However, equivalent histograms for patient B suggest hardly any improvement in performance over the test period. Trends in performance cannot be conveniently stored in the form of a histogram for each tracking test. In an attempt to reduce the data, we next computed the mean square error--the second moment of the error histogram--and averaged it for left and right arm over any one tracking session (each session consisting of one or two attempts with the left arm, followed by three attempts with the damaged right arm). Session averages for both patients are shown in Fig. 4, and clearly suggest both differences between tracking ability with left and right arms, and the substantial initial improvement in performance achieved by patient A with her hemiplegic arm, compared with patient B. The difficulty of trying to represent the tracking performance of such patients by a quasilinear, time-invariant, transfer function is illustrated by Fig. 5, which shows typical frequency responses relevant to a single attempt at tracking by each patient. In spite of the relatively skilled performance of these particular patients, the short duration and very limited bandwidth of the demand waveform gives results which are so statistically unreliable
Medical & Biological Engineering & Computing
March 1977
as to make a description by, say, a 2nd- or 3rd-order transfer function quite unrealistic. We have therefore investigated a simpler approach--that of a characterisation by just two parameters, average gain and time delay. Although these parameters could be estimated by fitting straightline approximations to gain and phase data, we have used a more convenient method based upon computing the crosscorrelation function between demand and response waveforms. A n essentially similar approach has been used by CASSELL (1973) in a study of tracking performance by patients suffering from Parkinson's disease. Fig. 6 illustrates trends in gain and time-delay parameters for both our patients, obtained by the above method after an initial tapering off of the gain
patient A
1.0 ~
~
.
-
~
~
A
lb
15
session no. - - ~
10
15
session no.
~9
(dB)
9
-
o .o
o
v9[
.2~I _o
patient B
Fig. 6 Trends in gain and time delay. (s)
:o%
o
*
o
" 9 ~'~%%*'-'_ o o OTp o
hase
o oo
o 05_60,
I 0.01
I
0.1
5
ase
o
gain
responses contain a rather large amount of highfrequency jitter, which cannot be accounted for by a quasilinear model; and secondly, the use of a
I
1.0
frequency (Hz)
Fig. 5 Frequency-response estimates
first and last 10% of the raw data records with window functions of half-Hanning form (raised cosine bell). I n spite of short-term fluctuations, the results once again suggest fairly dramatic improvement in tracking performance by patient A with her hemiplegic arm, and clearly indicate the differences between damaged and undamaged arms of both patients at all stages of the test programme. Discussion
The limited selection of results reported here illustrate some of the technical possibilities and difficulties in quantifying dynamic control in the hemiplegic arm. It is clear that relatively simple measures such as mean square error, or gain and time delay, can indicate trends in tracking performance which are undetectable by normal clinical examination (which failed to reveal any of the trends seen in Figs. 4 and 6). On the other hand, conventional technical measures of a more comprehensive kind, such as the frequency response, are problematical. The reasons for this seem straightforward: firstly, Fig. 2 shows that most patients'
Medical & Biological Engineering & Computing
1 min tracking task, rather than the 2 or even 4 m i n tasks which skilled subjects can tackle, makes the estimation of comprehensive measures subject to large statistical errors. Different approaches to the characterisation of tracking performance may, of course, prove more effective. For example, the discontinuous nature of some of the responses in Fig. 2 has prompted us to start investigating the use of sampled-data models, with promising initial results. Fig. 2 also raises the problem of control strategy. If tracking performance is characterised by a simple measure (such as mean-square error), a given level of performance can clearly be achieved with a variety of strategies. For example, Fig. 7, in which the mean-square error arising from various combinations of gain and time shift has been computed for the demand waveform of Fig. 2, shows that a patient with a very limited range of movement (small gain) coupled with the ability to predict the future time course of the demand waveform (small, .10
1.5
."~ gain -0.5
0-5 ~) *0.5 time shift (see)
1'5[ 0.5
time shift (sec,)
Fig. 7 Contour and 3-dimensional plots o f meansquare tracking error as a function o f gain and time shift
March 1977
187
or even negative time delay) could achieve the same error as a patient with a fuller range o f movement who responded rather slowly. The variety o f patients' tracking responses m a y reflect differences in such factors as visual perception or muscle spasticity, rather than differences in conscious strategy, and i t would be useful to quantify the various factors contributing to overall performance. This, however, would require performance models more explicitly related to the special problems of the stroke patient.
Acknowledgments--The authors wish to acknowledge the generous provision by the Medical Research Council of an online digital-computer system for data acquisition and analysis, and tke postgraduate research studentship awarded by the Science Research Council to W. R. Parker.
References CASS~LL, K. J. (1973) The usefulness of a temporal correlation technique in the assessment of human motor performance on a tracking device. Med. & BioL Eng. 11, 755
NASA-UNWERSITYannual conferences on manual control NASA special publications (1965 onwards) NIELSON, P. D. (1972) Speed of response or bandwidth of voluntary system controlling elbow position in intact man. Med. & Biol. Eng. 10, 450 PARKER, W. R. (1976) Aspects of electronic and control engineering applied to a study of recovery in hemiplegic stroke patients. Ph.D. thesis, University of Bristol PEW, R. W., DUrrENBACK,J. C. and FENSCH, L. K. (1966) A summary of sine wave tracking studies. NASAUniversity conference on manual control. No. 2, SP128, paper 1 POULTON, E. C. (1967) A quantal model for human tracking. NASA-University conference on manual control no. 3, SP144, paper 18
TAYLOR, L. W. (1967) A comparison of human performance modelling in the time and frequency domains. NASA-University conference on manual control no. 3, SP144, paper 9 WASIKO, R. J., McRuER, D. T. and MA6BALENO, R. E. (1966) Human pilot dynamic response in single loop systems with compensatory and pursuit displays. Air Force Flight Dynamics Laboratory technical report no. AFFDL-TR-66-137
Quelques-unes des applications de la recherche sur les op6rateurs humains das le domaine de I'evaluations des incapacit6s entrain6es par les attaques Sommaire---A pr6sent, le domaine de l'~valuation des victimes partiellement paralys~es par les attaques, est compliqu6, faute de mesurage objectif. Ici, on se sert d'une 'tgtche de piste', du type beaucoup utilis6 pour mesurer la performance d'op6rateurs sp6cialis6s, qui font une tO.che de contr61e manuelle pour encourager ]e mouvement contr616 dans le bras demi-paralys6. Des mesures quantitatives obtenus par de telles 6preuves, donnent une indication sensible des tendances de performance pendant la r66ducation, qu'on ne peut pas d6tecter a rexamen clinique courant.
188
Medical & Biological Engineering & Computing
March 1977
