Journal of Abnormal Psychology 1978, Vol. 87, No. 5, 491-496
Smooth-Pursuit Eye Movements: A Comparison of Two Measurement Techniques for Studying Schizophrenia Philip S. Holzman Harvard University
Delwin T. Lindsey University of Chicago
Shelby Haberman and Nicholas J. Yasillo University of Chicago Simultaneous recording of smooth-pursuit eye movements by electrooculographic (EOG) and infrared reflection techniques showed good correspondence between the two methods. The parameter of pursuit arrests, previously used to quantify smooth-pursuit performance, was not well correlated in the two methods. The natural logarithm of the signal:noise ratio obtained from harmonic regression of digitized and standardized eye movement data provides a valid quantitative assessment of smooth pursuit and suggests that such scoring of EOG records is effective and generally free of artifacts.
Research performed in this laboratory and elsewhere over the past 4 years has established a possible link between schizophrenia and impairment of smooth-pursuit eye movements (SPEM). Between 65% and 80% of schizophrenic patients and about 45% of their first-degree relatives have shown disordered SPEM. In contrast, the prevalence among normals has been about 6%. Moreover, neuroleptic drugs administered to schizophrenic patients have not accounted for SPEM impairment and single doses of chlorpromazine hydrochloride or diazepam have not disrupted SPEM in normals (Holzman, Levy, Uhlenhuth, Proctor, & Freedman, 1975; Holzman, Proctor, & Hughes, 1973; Holzman, Proctor, Levy, Yasillo, Meltzer, & Hurt, 1974; Shagass, Amadeo, & Overton, 1974). Although the weight of the evidence appears to validate the fundamental conclusion The authors are indebted to John Trimble, Joel Pokorny, and David Wallace for their indispensable technical help and to Betteanne Lindsey and Denise Hansen for their needed assistance. This study was supported by funds from Public Health Service Grant #MH-3I1340 to Philip Holzman. Requests for reprints should be sent to Philip S. Holzman, Department of Psychology, Harvard University, 33 Kirkland Street, Cambridge, Massachusetts 02138.
of an association between vulnerability to schizophrenia and disordered SPEM drawn from these studies, criticism has been directed toward the technical aspects of measuring SPEM (Troost, Daroff, & Dell'Osso, 1974). In the reports described above, visual tracking was assessed on the basis of an electrooculographic (EOG) method recorded while subjects followed the motion of a slowly oscillating pendulum. The raw EOG record contains information about eye position and also bioelectric noise. It has been argued that the impaired SPEM may reflect leakage of these bioelectric artifacts into the filtered EOG record. This interpretation implies that either (a) differences seen between schizophrenic patients and normals are methodological artifact, or (b) the differences are real but may or may not imply differences in actual eye movements. We believe that the first possibility is not reasonable considering the general stability of intrasubject EOG scores in very large samples over both short and long intertest intervals and the fact that the basic phenomenon has been replicated in several independent studies (e.g., Kuechenmeister, Linton, Mueller, & White, 1977; Shagass et al., 1974). We cannot, however, rule out the second possibility, and therefore we require a more systematic
Copyright 1978 by the American Psychological Association, Inc. 0021-843X/78/8705-0491$00.75
491
492
LINDSEY, HOLZMAN, HABERMAN, AND YASILLO
analysis of the contribution of EOG noise to EOG scores. In the study described below, we obtained EOG scores from persons with normal and impaired SPEM. In addition, we recorded eye position simultaneously by the method of infrared reflection (IR). We then assessed the validity of previously employed scoring procedures applied to both EOG and IR and compared them to a new statistical method of objectively scoring eye movement records. Method Subjects The subjects were five psychiatric patients and five normal controls, chosen to give a wider distribution of good and poor pursuit than would be found in the normal population by chance. The subject patients were recently admitted to the psychiatric unit at the University of Chicago. This group was composed of three males and two females, with the following tentatively diagnosed clinical conditions: psychotic depression (Subject 6), organic brain syndrome (Subject 7), and schizophrenia (Subjects 8, 9, 10). Only Subject 10 was receiving neuroleptic medication. The control subjects were recruited from the university community. None of the subjects had previously participated in an eye movement study and none were familiar with the objectives of this study.
The IR technique used in this study was similar to that first described by Stark and Sandberg (Note 1) and later by Young (1963). Briefly, the technique is based on the differential reflectivity of the iris and sclera to uniform infrared illumination. Photodetectors placed at the iris-sclera border on each side of the vertical meridian respond differentially depending on the proportions of radiation received from the iris and sclera. The output of each photodetector then varies with position of the irissclera borders with respect to each detector. The difference in output between the detectors has been found to vary approximately linearly with eye position. In our study, uniform illumination of the eye was provided by a small tungsten filament lamp mounted in an opaque tube located above the eye and directed toward it at about a 45-degree angle. A removable gelatin filter (Wratten #87C) which blocked radiation shorter than 850 nm was fitted over the front of the tube. The photodetectors consisted of two photodiodes (TI12N2175) mounted to an XYZ mechanical stage. The stage was fixed rigidly to the apparatus table. In this way, the position of the photodiode pair could be adjusted easily for each subject. The photodiode pair was placed below the subject's eye and aimed toward it, thus affording the subject an unobstructed view of the stimulus field. The output of each photodiode was amplified and fed into a differential amplifier, and the result fed into one channel of the tape recorder. The IR device had a frequency response flat to 10 kHz. A mouthbite of dental impression wax was provided to minimize head movements during recording.
Procedure Apparatus The stimulus consisted of a red circular spot of light subtending 6 min. of visual arc. Its position was modulated sinusoidally in the horizontal plane, with an excursion of ± 10 degrees of visual arc. The stimulus was produced by reflecting a laser beam from the surface of a mirror onto the surface of a white projection screen placed 1 m from the subject. The mirror was mounted to the shaft of a reotilinearly corrected servomotor. In this way, a sine wave input to the servomotor produced sinusoidal angular oscillation of the mirror and therefore a sinusoidal deflection of the light beam on the projection screen. The nominal frequency of oscillation was .4 Hz, but actual frequency varied slightly from subject to subject. For measurement of the EOG, silver - silver chloride electrodes were applied to the outer canthi of both -eyes, and a ground electrode was applied to the middle of the forehead. Electrical potentials were differentially amplified by a preamplifier (high frequency cutoff = 300 Hz; low frequency cutoff = .1 Hz) and fed into one channel of an instrumentation reel-to-reel Ml tape recorder.
Each session began by familiarizing the subject with the nature of the task. A mouthbite was then made and the EOG electrodes applied. The subject was seated in the apparatus and the IR monitor adjusted until saccades made to ± 10 degrees yielded outputs symmetrical about the 0-degree output. Although this procedure does not insure linearity of the IR output over the full range, data obtained from controls (see below) suggest that this calibration procedure was sufficient. Recording was done in a dimly lighted, electrically shielded room. Each subject was required to follow the stimulus for two SO-sec periods, separated by a 1-min rest during which the subject was permitted to release himself/herself from the mouthbite. The IR monitor was calibrated before the start of the second recording period. To insure that the subjects maintained their tracking performance for the entire 50-sec period, a realerting instruction was issued halfway through each period ("Keep your eyes right on the moving dot. Follow it as carefully as you can"). Two experimenters were used during the recording session. One was located outside the recording room and monitored EOG and IR output on an
EYE MOVEMENTS AND SCHIZOPHRENIA oscilloscope. The other experimenter was located inside the room to insure that the subject was performing the task and to help the subject patient feel more at ease during the recording session.
Scoring The quality of SPEM for the 10 subjects for the two methods of measurement was assessed by two scoring procedures. The first was identical to that employed previously by Holzman et al. (1974) and included both qualitative rating of the individual eye-tracking records and a quantitative scoring procedure of counting the number of times the derivative of the direct eye-tracking record went to zero. Since the derivative is proportional to the instantaneous velocity of the eye during tracking, zero crossings of the derivation record are called pursuit arrests. The second scoring procedure assessed the quality of SPEMs by the method of harmonic regression of the eye-tracking data and computation of a statistic referred to below as the natural logarithm signal: noise ratio (ln[S/N]).
Qualitative Scoring and Pursuit Arrest Scoring The recorded eye-tracking data were displayed on two channels of a chart dynograph at a paper speed of 10 mm/sec. Channel 1 of the dynograph displayed the eye movement data directly, and Channel 2 displayed the derivative of the data. The 3-db. points of Channels 1 and 2 were 10 Hz and 6 Hz, respectively. Each S'PEM record was classified qualitatively as normal or deviant independently by two of the authors. The IR and EOG records were evaluated blindly. The criterion for a normal tracing was that it appear reasonably sinusoidal; tracings departing from this were scored as deviant. All of the tracings obtained in this study were easily classified. Pursuit arrests were determined from the derivative tracing. We scored as pursuit arrests all deflections within 2 mm or less of the zero line, which corresponds to 4 degrees/ sec or less. Since normal tracking requires two pursuit arrests per cycle, these obligate pursuit arrests were not counted. The pursuit arrest score reflected arrests per tracking cycle.
493
The IR data were corrected for eye-blink artifacts. Blink artifacts appear as off-scale reflections on the IR record during the blink. If the standardized IR record exceeded 2 in magnitude, it was replaced by the last preceding value not exceeding 2 in magnitude. Harmonic regression (Bloomfield, 1976) was performed on the digitized and standardized eye movement data obtained by both IR and EOG methods. The sums of squares of the real plus imaginary portions of the fit were taken as the estimate of signal (S) present in the eye movement record in the proportion (in the frequency domain) of the record which was correlated with the stimulus. The fast fourier transform of the residual was computed and the estimate of noise (N) present was obtained by summing the squares of the real and imaginary parts of the transformed residual from .8 Hz to 8 Hz. The low cutoff frequency of .8 Hz insured minimal contribution of EOG baseline drift to the estimate of noise. We chose 8 Hz as the cutoff frequency for two reasons: (a) This frequency approaches the limit of resolution of the dynograph, and (b) 8 Hz limits leakage of noise uncorrelated with eye movements into the calculation of the signal: noise ratio for EOG records. Specifically, contribution from the electroencephalographic alpha rhythm and from muscle contraction are minimized. An alternative method of determining the signal: noise ratio would have been to transform the data first and then to compare the regression on the fast fourier transformed data and residual. We chose to regress before transformation for two reasons: (a) The fundamental in our data is wellknown (a .4-Hz sine wave), making regression on the data easy. Transformation before regression is useful only when little is known about the signal, (b) Transformation before regression can introduce systematic error into the signal: noise calculations if the fundamental is not a multiple of the sampling rate. We determined that aliasing was not a problem since inspection of the power spectra shows a gradual rolling off at 8 Hz with insignificant power above 40 Hz. Visual inspection of the fast fourier transformed data indicated that the natural logarithm of the signal: noise ratio yielded a quantitative statement in agreement with a qualitative evaluation of the power spectra. We therefore report values in natural logarithmic units of the signal: noise ratio: ln(S/N).
Computation of Signal: Noise Ratio All computations were performed on a DEC PD'P 8 digital computer. These were based on a 12.5-sec segment of each record. The recorded data of these records were first digitized using conventional techniques. Conversion rate was 180 points/sec for a total of 2,250 data points. The data points were standardized by subtracting the sample mean from each data point and dividing by the sample standard deviation.
Results Samples of stimulus tracking measured simultaneously by EOG and IR techniques are shown for four subjects in Figure 1. The EOG sample appears at the top of each panel and is labeled A. Pen deflections from top to bottom correspond to eye movements from
494
LINDSEY, HOLZMAN, HABERMAN, AND YASILLO DIRECT EYE MOVEMENT
DERIVATIVE
SUBJECT 9 DIRECT EYE MOVEMENT
DERIVATIVE
.
SUBJECT 7
SUBJECT 8 SUBJECT 4
A A Figure 1. Samples of electrooculographic (A) and infrared reflection (B) records of four subjects. (Both were recorded simultaneously.) left to right. Inspection reveals that although the IR traces are generally smoother than the EOG, there is excellent correspondence between tracings for each subject. Table 1 shows that qualitative assessment of SPEM
across method is in agreement for all subjects. If the distortions in the eye movement records were unrelated to deviant SPEM, it is unlikely that the two methods of recording
Table 1 Qualitative Scores and Quantitative Assessments, Including Pursuit Arrests and for Two Methods of Recording Smooth-Pursuit Eye Movements
Infrared reflection
Electrooculogram
Subjects
1
Quality of tracking0
1 2 3 4 S
A A A A A
6 7 8 9 10
A B B B B
ln(S/N)
Quality of tracking"
Pursuit arrests per cycle
4.19 4.24 7.57 8.49 4.94
Normals 4.34 2.58 4.31 2.89 5.19
A A A A A
3.59 1.13 1.07 0.77 1.10
4.34 4.69 4.99 4.88 4.81
3.92 5.21 7.00 7.79 13.82
Patients 5.69 -3.88 -2.89 -2.90 3.25
A B B B B
0.51 2.11 1.74 4.90 3.32
5.46 -3.90 -2.79 -2.62 3.93
Pursuit arrests per cycle
A = good pursuit and B = poor pursuit.
ln(S/N)
ln(S/N)
EYE MOVEMENTS AND SCHIZOPHRENIA
would appear so well correlated since the major sources of measurement error occupy different frequency bands. We therefore conclude that the deviant SPEM measured for these subjects reflects truly dysfunctional smooth-pursuit eye movements. In the present study, the dysfunction is easily revealed by either EOG or IR methods. Deviant SPEM in schizophrenia is thus not a methodological artifact. Table 1 also contains the pursuit arrest scores for each subject. The scores represent pursuit arrests per cycle averaged over the two tracking periods. Scores obtained from the IR records are uniformly lower than those obtained from the EOG and may be assumed to reflect the qualitatively smoother appearance of the IR data. It is noteworthy, moreover, that the distribution of pursuit arrest scores is rather narrow compared with previous populations we have tested. The Pearson product-moment correlation between the pursuit arrest score for EOG and IR is not significantly different from zero. Subject 4, in particular, showed a high number of pursuit arrests in the EOG recording relative to those in the IR record. It is unlikely that EOG tracings reflect very fine movements not resolved by the IR monitor. The ripple riding on the fundamental is within the pass band of the IR monitor and of sufficient amplitude to be resolved by it. We interpret the differences between the tracings for Subject 4 as reflecting a larger than normal leakage of noise into the EOG trace, probably because of faulty electrode placement. Thus, the present study weakens the use of the pursuit arrest score as an index of SPEM performance. Calculations in ln(S/N) are also contained in Table 1. The Pearson correlation coefficient between EOG and IR for these calculations is .96 (p < .01). The point biserial correlation between the qualitative assessment of pursuit (A or B) and the ln(S/N) is .82 (p < .01) for EOG and .86 (p < .01) for the IR method. Thus, although the sample size obtained in this study is small, the differences in concordance using the pursuit arrest and harmonic regression strongly recommends the use of ln(S/N) as a significant improvement in the measurement of SPEM performance. In
495
addition, we wish to stress the ease with which this index can be implemented with conventional analog circuitry. This combined with the EOG method of eye movement measurement affords the possibility of quick and accurate assessment of SPEM in the clinical setting. Work is presently under way in this laboratory to develop the necessary electronic device. Discussion The IR technique (Stark, 1971), and more recently the Purkinje eye tracker (e.g., Cornsweet & Crane, 1973), are considered to be optimal for quantitative studies of eye movements. The IR technique has good linearity and resolution and introduces few artifacts. Unfortunately, both the IR and the Purkinje eye tracker are somewhat less adaptable for field studies, in which the investigator conducts the measurement in subjects' homes or on the wards of mental hospitals. EOG is the method of choice for that latter situation. A principal issue concerns the comparability of the two techniques of measurement. Qualitative comparability is high. The comparability of the pursuit arrest score is, however, poor and suggests that although this score in the EOG method is rather stable for a specific person, it may not be a particularly robust estimation of SPEM performance. It is noteworthy that a replication of the Holzman et al. (1973, 1974) studies using IR technology indicated that the observation of SPEM dysfunction in schizophrenic patients and their relatives is valid (Kuechenmeister et al., 1977). Pursuit arrests are probably confounded to some degree by biologic noise and therefore this measure is no longer recommended. The ln(S/N) shows a very high correlation between the IR and EOG methods and recommends itself as a good measure of SPEM performance. Reference Note 1. Stark, L., & Sandberg, A. A simple instrument for measuring eye movements. In, Quarterly Progress Report (No. 62). Boston: Massachusetts Institute of Technology, Research Laboratory of Electronics, 1961.
496
LINDSEY, HOLZMAN, HABERMAN, AND YASILLO
References Bloomfield, P. Fourier analysis oj time series: An introduction. New York: Wiley, 1976. Cornsweet, T, N., & Crane, H. D. Accurate twodimensional eye tracker using first and fourth Purkinje images. Journal of the Optical Society of America, 1973, 63, 921-930. Holzman, P. S., Levy, D. L., Uhlenhuth, E. H., Proctor, L. R., & Preedman, D. X. Smooth-pursuit eye movements and diazepam, CPZ, and secobarbital. Psychopharmacologia, 1975, 44, 111-115. Holzman, P. S., Proctor, L. R., & Hughes, D. W. Eye tracking patterns in schizophrenia. Science, H973,181, 179-181. Holzman, P. S., Proctor, L. R., Levy, D. L., Yasillo, N. J., Meltzer, H. Y., & Hurt, S. W. Eye-tracking dysfunctions in schizophrenic patients and their
relatives. Archives oj General Psychiatry, 1974, 31, 143-151. Kuechenmeister, C. A., Linton, P. H., Mueller, T. V., & White, H. B. Eye tracking in relation to age, sex, and illness. Archives of General Psychiatry, 1977, 34, 578-579. Shagass, C., Amadeo, M., & Overton, D. A. Eye tracking performance in psychiatric patients. Biological Psychiatry, 1974, 9, 245-260. Stark, L. The control system for versional eye movements. In P. Bach-y-Rita, C. C. Collins, & J. E. Hyde (Eds.), The control oj eye movements. New York: Academic Press, 1971. Troost, B. T., Daroff, R. B., Dell'Osso, L. R. Technical comments. Science, 1974, 184, 1202. Young, L. R. Measuring eye movements. The Medical Journal of Medical Electronics, Oct.-Dec., 1963. Received March 2, 1978 •
Smooth-Pursuit Eye Movements: A Comparison of Two Measurement Techniques for Studying Schizophrenia Philip S. Holzman Harvard University
Delwin T. Lindsey University of Chicago
Shelby Haberman and Nicholas J. Yasillo University of Chicago Simultaneous recording of smooth-pursuit eye movements by electrooculographic (EOG) and infrared reflection techniques showed good correspondence between the two methods. The parameter of pursuit arrests, previously used to quantify smooth-pursuit performance, was not well correlated in the two methods. The natural logarithm of the signal:noise ratio obtained from harmonic regression of digitized and standardized eye movement data provides a valid quantitative assessment of smooth pursuit and suggests that such scoring of EOG records is effective and generally free of artifacts.
Research performed in this laboratory and elsewhere over the past 4 years has established a possible link between schizophrenia and impairment of smooth-pursuit eye movements (SPEM). Between 65% and 80% of schizophrenic patients and about 45% of their first-degree relatives have shown disordered SPEM. In contrast, the prevalence among normals has been about 6%. Moreover, neuroleptic drugs administered to schizophrenic patients have not accounted for SPEM impairment and single doses of chlorpromazine hydrochloride or diazepam have not disrupted SPEM in normals (Holzman, Levy, Uhlenhuth, Proctor, & Freedman, 1975; Holzman, Proctor, & Hughes, 1973; Holzman, Proctor, Levy, Yasillo, Meltzer, & Hurt, 1974; Shagass, Amadeo, & Overton, 1974). Although the weight of the evidence appears to validate the fundamental conclusion The authors are indebted to John Trimble, Joel Pokorny, and David Wallace for their indispensable technical help and to Betteanne Lindsey and Denise Hansen for their needed assistance. This study was supported by funds from Public Health Service Grant #MH-3I1340 to Philip Holzman. Requests for reprints should be sent to Philip S. Holzman, Department of Psychology, Harvard University, 33 Kirkland Street, Cambridge, Massachusetts 02138.
of an association between vulnerability to schizophrenia and disordered SPEM drawn from these studies, criticism has been directed toward the technical aspects of measuring SPEM (Troost, Daroff, & Dell'Osso, 1974). In the reports described above, visual tracking was assessed on the basis of an electrooculographic (EOG) method recorded while subjects followed the motion of a slowly oscillating pendulum. The raw EOG record contains information about eye position and also bioelectric noise. It has been argued that the impaired SPEM may reflect leakage of these bioelectric artifacts into the filtered EOG record. This interpretation implies that either (a) differences seen between schizophrenic patients and normals are methodological artifact, or (b) the differences are real but may or may not imply differences in actual eye movements. We believe that the first possibility is not reasonable considering the general stability of intrasubject EOG scores in very large samples over both short and long intertest intervals and the fact that the basic phenomenon has been replicated in several independent studies (e.g., Kuechenmeister, Linton, Mueller, & White, 1977; Shagass et al., 1974). We cannot, however, rule out the second possibility, and therefore we require a more systematic
Copyright 1978 by the American Psychological Association, Inc. 0021-843X/78/8705-0491$00.75
491
492
LINDSEY, HOLZMAN, HABERMAN, AND YASILLO
analysis of the contribution of EOG noise to EOG scores. In the study described below, we obtained EOG scores from persons with normal and impaired SPEM. In addition, we recorded eye position simultaneously by the method of infrared reflection (IR). We then assessed the validity of previously employed scoring procedures applied to both EOG and IR and compared them to a new statistical method of objectively scoring eye movement records. Method Subjects The subjects were five psychiatric patients and five normal controls, chosen to give a wider distribution of good and poor pursuit than would be found in the normal population by chance. The subject patients were recently admitted to the psychiatric unit at the University of Chicago. This group was composed of three males and two females, with the following tentatively diagnosed clinical conditions: psychotic depression (Subject 6), organic brain syndrome (Subject 7), and schizophrenia (Subjects 8, 9, 10). Only Subject 10 was receiving neuroleptic medication. The control subjects were recruited from the university community. None of the subjects had previously participated in an eye movement study and none were familiar with the objectives of this study.
The IR technique used in this study was similar to that first described by Stark and Sandberg (Note 1) and later by Young (1963). Briefly, the technique is based on the differential reflectivity of the iris and sclera to uniform infrared illumination. Photodetectors placed at the iris-sclera border on each side of the vertical meridian respond differentially depending on the proportions of radiation received from the iris and sclera. The output of each photodetector then varies with position of the irissclera borders with respect to each detector. The difference in output between the detectors has been found to vary approximately linearly with eye position. In our study, uniform illumination of the eye was provided by a small tungsten filament lamp mounted in an opaque tube located above the eye and directed toward it at about a 45-degree angle. A removable gelatin filter (Wratten #87C) which blocked radiation shorter than 850 nm was fitted over the front of the tube. The photodetectors consisted of two photodiodes (TI12N2175) mounted to an XYZ mechanical stage. The stage was fixed rigidly to the apparatus table. In this way, the position of the photodiode pair could be adjusted easily for each subject. The photodiode pair was placed below the subject's eye and aimed toward it, thus affording the subject an unobstructed view of the stimulus field. The output of each photodiode was amplified and fed into a differential amplifier, and the result fed into one channel of the tape recorder. The IR device had a frequency response flat to 10 kHz. A mouthbite of dental impression wax was provided to minimize head movements during recording.
Procedure Apparatus The stimulus consisted of a red circular spot of light subtending 6 min. of visual arc. Its position was modulated sinusoidally in the horizontal plane, with an excursion of ± 10 degrees of visual arc. The stimulus was produced by reflecting a laser beam from the surface of a mirror onto the surface of a white projection screen placed 1 m from the subject. The mirror was mounted to the shaft of a reotilinearly corrected servomotor. In this way, a sine wave input to the servomotor produced sinusoidal angular oscillation of the mirror and therefore a sinusoidal deflection of the light beam on the projection screen. The nominal frequency of oscillation was .4 Hz, but actual frequency varied slightly from subject to subject. For measurement of the EOG, silver - silver chloride electrodes were applied to the outer canthi of both -eyes, and a ground electrode was applied to the middle of the forehead. Electrical potentials were differentially amplified by a preamplifier (high frequency cutoff = 300 Hz; low frequency cutoff = .1 Hz) and fed into one channel of an instrumentation reel-to-reel Ml tape recorder.
Each session began by familiarizing the subject with the nature of the task. A mouthbite was then made and the EOG electrodes applied. The subject was seated in the apparatus and the IR monitor adjusted until saccades made to ± 10 degrees yielded outputs symmetrical about the 0-degree output. Although this procedure does not insure linearity of the IR output over the full range, data obtained from controls (see below) suggest that this calibration procedure was sufficient. Recording was done in a dimly lighted, electrically shielded room. Each subject was required to follow the stimulus for two SO-sec periods, separated by a 1-min rest during which the subject was permitted to release himself/herself from the mouthbite. The IR monitor was calibrated before the start of the second recording period. To insure that the subjects maintained their tracking performance for the entire 50-sec period, a realerting instruction was issued halfway through each period ("Keep your eyes right on the moving dot. Follow it as carefully as you can"). Two experimenters were used during the recording session. One was located outside the recording room and monitored EOG and IR output on an
EYE MOVEMENTS AND SCHIZOPHRENIA oscilloscope. The other experimenter was located inside the room to insure that the subject was performing the task and to help the subject patient feel more at ease during the recording session.
Scoring The quality of SPEM for the 10 subjects for the two methods of measurement was assessed by two scoring procedures. The first was identical to that employed previously by Holzman et al. (1974) and included both qualitative rating of the individual eye-tracking records and a quantitative scoring procedure of counting the number of times the derivative of the direct eye-tracking record went to zero. Since the derivative is proportional to the instantaneous velocity of the eye during tracking, zero crossings of the derivation record are called pursuit arrests. The second scoring procedure assessed the quality of SPEMs by the method of harmonic regression of the eye-tracking data and computation of a statistic referred to below as the natural logarithm signal: noise ratio (ln[S/N]).
Qualitative Scoring and Pursuit Arrest Scoring The recorded eye-tracking data were displayed on two channels of a chart dynograph at a paper speed of 10 mm/sec. Channel 1 of the dynograph displayed the eye movement data directly, and Channel 2 displayed the derivative of the data. The 3-db. points of Channels 1 and 2 were 10 Hz and 6 Hz, respectively. Each S'PEM record was classified qualitatively as normal or deviant independently by two of the authors. The IR and EOG records were evaluated blindly. The criterion for a normal tracing was that it appear reasonably sinusoidal; tracings departing from this were scored as deviant. All of the tracings obtained in this study were easily classified. Pursuit arrests were determined from the derivative tracing. We scored as pursuit arrests all deflections within 2 mm or less of the zero line, which corresponds to 4 degrees/ sec or less. Since normal tracking requires two pursuit arrests per cycle, these obligate pursuit arrests were not counted. The pursuit arrest score reflected arrests per tracking cycle.
493
The IR data were corrected for eye-blink artifacts. Blink artifacts appear as off-scale reflections on the IR record during the blink. If the standardized IR record exceeded 2 in magnitude, it was replaced by the last preceding value not exceeding 2 in magnitude. Harmonic regression (Bloomfield, 1976) was performed on the digitized and standardized eye movement data obtained by both IR and EOG methods. The sums of squares of the real plus imaginary portions of the fit were taken as the estimate of signal (S) present in the eye movement record in the proportion (in the frequency domain) of the record which was correlated with the stimulus. The fast fourier transform of the residual was computed and the estimate of noise (N) present was obtained by summing the squares of the real and imaginary parts of the transformed residual from .8 Hz to 8 Hz. The low cutoff frequency of .8 Hz insured minimal contribution of EOG baseline drift to the estimate of noise. We chose 8 Hz as the cutoff frequency for two reasons: (a) This frequency approaches the limit of resolution of the dynograph, and (b) 8 Hz limits leakage of noise uncorrelated with eye movements into the calculation of the signal: noise ratio for EOG records. Specifically, contribution from the electroencephalographic alpha rhythm and from muscle contraction are minimized. An alternative method of determining the signal: noise ratio would have been to transform the data first and then to compare the regression on the fast fourier transformed data and residual. We chose to regress before transformation for two reasons: (a) The fundamental in our data is wellknown (a .4-Hz sine wave), making regression on the data easy. Transformation before regression is useful only when little is known about the signal, (b) Transformation before regression can introduce systematic error into the signal: noise calculations if the fundamental is not a multiple of the sampling rate. We determined that aliasing was not a problem since inspection of the power spectra shows a gradual rolling off at 8 Hz with insignificant power above 40 Hz. Visual inspection of the fast fourier transformed data indicated that the natural logarithm of the signal: noise ratio yielded a quantitative statement in agreement with a qualitative evaluation of the power spectra. We therefore report values in natural logarithmic units of the signal: noise ratio: ln(S/N).
Computation of Signal: Noise Ratio All computations were performed on a DEC PD'P 8 digital computer. These were based on a 12.5-sec segment of each record. The recorded data of these records were first digitized using conventional techniques. Conversion rate was 180 points/sec for a total of 2,250 data points. The data points were standardized by subtracting the sample mean from each data point and dividing by the sample standard deviation.
Results Samples of stimulus tracking measured simultaneously by EOG and IR techniques are shown for four subjects in Figure 1. The EOG sample appears at the top of each panel and is labeled A. Pen deflections from top to bottom correspond to eye movements from
494
LINDSEY, HOLZMAN, HABERMAN, AND YASILLO DIRECT EYE MOVEMENT
DERIVATIVE
SUBJECT 9 DIRECT EYE MOVEMENT
DERIVATIVE
.
SUBJECT 7
SUBJECT 8 SUBJECT 4
A A Figure 1. Samples of electrooculographic (A) and infrared reflection (B) records of four subjects. (Both were recorded simultaneously.) left to right. Inspection reveals that although the IR traces are generally smoother than the EOG, there is excellent correspondence between tracings for each subject. Table 1 shows that qualitative assessment of SPEM
across method is in agreement for all subjects. If the distortions in the eye movement records were unrelated to deviant SPEM, it is unlikely that the two methods of recording
Table 1 Qualitative Scores and Quantitative Assessments, Including Pursuit Arrests and for Two Methods of Recording Smooth-Pursuit Eye Movements
Infrared reflection
Electrooculogram
Subjects
1
Quality of tracking0
1 2 3 4 S
A A A A A
6 7 8 9 10
A B B B B
ln(S/N)
Quality of tracking"
Pursuit arrests per cycle
4.19 4.24 7.57 8.49 4.94
Normals 4.34 2.58 4.31 2.89 5.19
A A A A A
3.59 1.13 1.07 0.77 1.10
4.34 4.69 4.99 4.88 4.81
3.92 5.21 7.00 7.79 13.82
Patients 5.69 -3.88 -2.89 -2.90 3.25
A B B B B
0.51 2.11 1.74 4.90 3.32
5.46 -3.90 -2.79 -2.62 3.93
Pursuit arrests per cycle
A = good pursuit and B = poor pursuit.
ln(S/N)
ln(S/N)
EYE MOVEMENTS AND SCHIZOPHRENIA
would appear so well correlated since the major sources of measurement error occupy different frequency bands. We therefore conclude that the deviant SPEM measured for these subjects reflects truly dysfunctional smooth-pursuit eye movements. In the present study, the dysfunction is easily revealed by either EOG or IR methods. Deviant SPEM in schizophrenia is thus not a methodological artifact. Table 1 also contains the pursuit arrest scores for each subject. The scores represent pursuit arrests per cycle averaged over the two tracking periods. Scores obtained from the IR records are uniformly lower than those obtained from the EOG and may be assumed to reflect the qualitatively smoother appearance of the IR data. It is noteworthy, moreover, that the distribution of pursuit arrest scores is rather narrow compared with previous populations we have tested. The Pearson product-moment correlation between the pursuit arrest score for EOG and IR is not significantly different from zero. Subject 4, in particular, showed a high number of pursuit arrests in the EOG recording relative to those in the IR record. It is unlikely that EOG tracings reflect very fine movements not resolved by the IR monitor. The ripple riding on the fundamental is within the pass band of the IR monitor and of sufficient amplitude to be resolved by it. We interpret the differences between the tracings for Subject 4 as reflecting a larger than normal leakage of noise into the EOG trace, probably because of faulty electrode placement. Thus, the present study weakens the use of the pursuit arrest score as an index of SPEM performance. Calculations in ln(S/N) are also contained in Table 1. The Pearson correlation coefficient between EOG and IR for these calculations is .96 (p < .01). The point biserial correlation between the qualitative assessment of pursuit (A or B) and the ln(S/N) is .82 (p < .01) for EOG and .86 (p < .01) for the IR method. Thus, although the sample size obtained in this study is small, the differences in concordance using the pursuit arrest and harmonic regression strongly recommends the use of ln(S/N) as a significant improvement in the measurement of SPEM performance. In
495
addition, we wish to stress the ease with which this index can be implemented with conventional analog circuitry. This combined with the EOG method of eye movement measurement affords the possibility of quick and accurate assessment of SPEM in the clinical setting. Work is presently under way in this laboratory to develop the necessary electronic device. Discussion The IR technique (Stark, 1971), and more recently the Purkinje eye tracker (e.g., Cornsweet & Crane, 1973), are considered to be optimal for quantitative studies of eye movements. The IR technique has good linearity and resolution and introduces few artifacts. Unfortunately, both the IR and the Purkinje eye tracker are somewhat less adaptable for field studies, in which the investigator conducts the measurement in subjects' homes or on the wards of mental hospitals. EOG is the method of choice for that latter situation. A principal issue concerns the comparability of the two techniques of measurement. Qualitative comparability is high. The comparability of the pursuit arrest score is, however, poor and suggests that although this score in the EOG method is rather stable for a specific person, it may not be a particularly robust estimation of SPEM performance. It is noteworthy that a replication of the Holzman et al. (1973, 1974) studies using IR technology indicated that the observation of SPEM dysfunction in schizophrenic patients and their relatives is valid (Kuechenmeister et al., 1977). Pursuit arrests are probably confounded to some degree by biologic noise and therefore this measure is no longer recommended. The ln(S/N) shows a very high correlation between the IR and EOG methods and recommends itself as a good measure of SPEM performance. Reference Note 1. Stark, L., & Sandberg, A. A simple instrument for measuring eye movements. In, Quarterly Progress Report (No. 62). Boston: Massachusetts Institute of Technology, Research Laboratory of Electronics, 1961.
496
LINDSEY, HOLZMAN, HABERMAN, AND YASILLO
References Bloomfield, P. Fourier analysis oj time series: An introduction. New York: Wiley, 1976. Cornsweet, T, N., & Crane, H. D. Accurate twodimensional eye tracker using first and fourth Purkinje images. Journal of the Optical Society of America, 1973, 63, 921-930. Holzman, P. S., Levy, D. L., Uhlenhuth, E. H., Proctor, L. R., & Preedman, D. X. Smooth-pursuit eye movements and diazepam, CPZ, and secobarbital. Psychopharmacologia, 1975, 44, 111-115. Holzman, P. S., Proctor, L. R., & Hughes, D. W. Eye tracking patterns in schizophrenia. Science, H973,181, 179-181. Holzman, P. S., Proctor, L. R., Levy, D. L., Yasillo, N. J., Meltzer, H. Y., & Hurt, S. W. Eye-tracking dysfunctions in schizophrenic patients and their
relatives. Archives oj General Psychiatry, 1974, 31, 143-151. Kuechenmeister, C. A., Linton, P. H., Mueller, T. V., & White, H. B. Eye tracking in relation to age, sex, and illness. Archives of General Psychiatry, 1977, 34, 578-579. Shagass, C., Amadeo, M., & Overton, D. A. Eye tracking performance in psychiatric patients. Biological Psychiatry, 1974, 9, 245-260. Stark, L. The control system for versional eye movements. In P. Bach-y-Rita, C. C. Collins, & J. E. Hyde (Eds.), The control oj eye movements. New York: Academic Press, 1971. Troost, B. T., Daroff, R. B., Dell'Osso, L. R. Technical comments. Science, 1974, 184, 1202. Young, L. R. Measuring eye movements. The Medical Journal of Medical Electronics, Oct.-Dec., 1963. Received March 2, 1978 •
