Cerebral hemispheric specialization for spatial attention: spatial distribution of search-related eye fixations in the absence of neglect

Neuropsychologia 41 (2003) 1396–1409

Cerebral hemispheric specialization for spatial attention: spatial distribution of search-related eye fixations in the absence of neglect Mark Mapstone a,c,1 , Sandra Weintraub a,b,c,∗ , Caralynn Nowinski b , Gülüstu Kaptanoglu b , Darren R. Gitelman b,c , M.-Marsel Mesulam a,b,c a c

Department of Psychiatry and Behavioral Sciences, Northwestern University Feinberg School of Medicine, Chicago, IL, USA b Department of Neurology, Northwestern University Feinberg School of Medicine, Chicago, IL, USA Cognitive Neurology and Alzheimer’s Disease Center, Northwestern University Feinberg School of Medicine, Chicago, IL, USA Received 20 September 2001; received in revised form 9 January 2003; accepted 9 January 2003

Abstract The “specialization” of the right hemisphere for spatial attention is widely accepted but poorly understood. While several theories have been supported by studies of patients with acute hemispatial neglect, generalizability beyond this population remains unclear. In this study, we compared the predictions of two attention models [Brain 119 (1996) 841; Trans. Am. Neurol. Assoc. 95 (1970) 143] when applied to data obtained from subjects with unilateral right- or left-cerebral lesions, but without clinical evidence of neglect during a visual search task. Both Left Lesion and Right Lesion subjects detected fewer targets in the contralesional hemispace. However, the Right Lesion subjects also made fewer visual fixations and longer saccades in the contralesional hemispace, suggesting a fundamental alteration in the architecture of visual search. The spatial distribution of fixations made by Right Lesion subjects more closely fits the prediction of a “salience” model than of the strict interpretation of a linear “gradient” model. These data support the long-standing notion of right hemisphere dominance for spatial attention, especially for the top–down processes entailed in self-directed visual search, and extend this to lesion patients without clinically evident neglect. A theoretical model based on the salience of extrapersonal space appears useful for understanding alterations of attentional allocation, particularly after recovery from stroke. © 2003 Elsevier Science Ltd. All rights reserved. Keywords: Cerebral dominance; Visual search; Eye movements; Neuropsychology

1. Introduction The dramatic manifestations of the hemispatial inattention (neglect) syndrome have stimulated vigorous investigation into the behavioral and neural mechanisms of spatial attention. One central question focuses on the nature of hemispheric specialization for this important behavioral domain. Early studies of patients with focal cerebrovascular lesions suggested that contralateral hemispatial neglect is more frequent and severe following right-sided lesions than following unilateral left-cerebral damage (Bisiach, Cornacchia, Sterzi, & Vallar, 1984; Fullerton, McSherry, & Stout, 1986; ∗ Corresponding author at: Cognitive Neurology and Alzheimer’s Disease Center, Northwestern University Feinberg School of Medicine, 320 East Superior, Searle 11-467, Chicago, IL, 60611, USA. Tel.: +1-312-908-9023; fax: +1-312-908-8789. E-mail address: [email protected] (S. Weintraub). 1 Present address: Department of Neurology, University of Rochester Medical Center, Rochester, NY.

Mosidze, Mkheidze, & Makashvili, 1994; Weintraub & Mesulam, 1987). This conclusion was reiterated in a recent review that revisited studies directly comparing the occurrence of unilateral spatial neglect following right- or left-sided unilateral cerebral lesions (Bowen, McKenna, & Tallis, 1999). Results of recent functional neuroimaging studies in young, non-brain damaged individuals further have suggested that the right cerebral hemisphere may be relatively more important than the left in modulating directed spatial attention (Coull & Nobre, 1998; Nobre, Coull, & Frith, 1999). According to one model of right cerebral dominance for directed spatial attention, neural units in the right cerebral hemisphere modulate attention within both the contralateral and ipsilateral hemispaces, while neural units in the left are only directed at the contralesional right hemispace (Mesulam, 1981, 1985, 1990, 1998, 1999). Several clinical observations support this model. First, the study of patients undergoing the intracarotid sodium amytal procedure prior

0028-3932/03/$ – see front matter © 2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0028-3932(03)00043-5

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

to surgical intervention for epilepsy has shown that anesthetization of the right cerebral hemisphere results in marked contralateral hemispatial neglect while injection of the left hemisphere does not produce the same effect (Spiers et al., 1990). Second, it has been demonstrated that right-sided lesions that cause contralesional neglect are associated with a milder degree of inattention within the ipsilesional right hemispace (Weintraub & Mesulam, 1987). Finally, functional imaging studies also have provided support for this model by showing greater activation of right hemispheric regions during performance of naturalistic tasks during which attention is equally allocated to the left and right hemispaces (Gitelman et al., 1996; Nobre et al., 1997). A number of models addressing the control of spatial attention have been proposed to account for neglect behavior. An early “gradient” model (Kinsbourne, 1970a, 1970b, 1987) was proposed to characterize the shift of attention to the ipsilesional space. This model is based on the notion that the cerebral hemispheres direct attention toward the contralateral visual hemispace in an increasing linear gradient. Furthermore, it hypothesizes that moment-to-moment shifts of attention can be influenced by the nature and location of external stimuli. Experiments in normal individuals have demonstrated that such a gradient exists within each hemispace as well, and that under certain circumstances such as orientation conflict, a rightward bias of attention can be demonstrated (Reuter-Lorenz, Kinsbourne, & Moscovitch, 1990). In the simplest interpretation of this model, the distribution of attention following a right cerebral lesion adopts a linear gradient from a nadir in the leftmost side of space to its peak in the rightmost side of space. More recently, studying the phenomenon of line bisection in patients with hemispatial neglect, Anderson (1996) proposed a different theoretical model to account for the spatial distribution of attention. This model hypothesizes that the spatial distribution of attention can be predicted by determining the salience of each point within the extrapersonal space. According to Anderson, the salience of a point in space is “. . . a function of its spatial location along the linear dimension of left to right and is something akin to the weight or attraction a point has as a result of a subject’s attention” (Anderson, 1996, p. 843). According to this model, each cerebral hemisphere contributes to attentional processing over the extrapersonal space in a bell-shaped “salience” curve. The right hemisphere curve normally is broad and encompasses the entire extrapersonal space, while the left hemisphere curve is narrower and is centered in the right hemispace. When these curves are summed, they provide a bimodal distribution of the total salience of horizontal space. Following a right hemisphere lesion, salience is shifted toward the ipsilesional side of space whereas relatively little change occurs after a left hemisphere lesion. Both the gradient and salience models depict an ipsilesional shift of attention following right hemisphere damage but the shape of the predicted distribution differs considerably.

1397

Although eye movements are often used to study underlying attentional processes, the relationship between occulo-motor programming and attention is complex. Most eye movements produce overt shifts of attention to the location of the saccadic target, however attention can be shifted covertly without eye movements (Shepard, Findlay, & Hockey, 1986). Numerous studies have explored this relationship and in general, provide evidence that networks involved in shifts of attention and eye movement generation overlap substantially (e.g. Gitelman et al., 1996; Moore & Fallah, 2001; Sheliga, Riggio, Craighero, & Rizzolatti, 1995). This appears to be especially true when both the termination of the saccade and the focus of attention are in the same location or are the same object (Deubel & Scheider, 1996). There is recent evidence suggesting that eye movements are altered in hemispatial neglect in a way that implies a shift of egocentric space to the ipsilesional side (e.g. Barton, Behrmann, & Black, 1998; Chedru, Leblanc, & Lhermitte, 1973; Ishiai, Koyama, Seki, & Nakayama, 1998; Karnath & Fetter, 1995). Barton et al. (1998) analyzed the spatial distribution of eye movements during the performance of a line bisection task by patients with hemispatial neglect. Alterations in the distribution were attributed to a “re-centering” of attention slightly to the right of the subject’s midline and were cited as support for the gradient model. This re-centering has been characterized both as a shift in the egocentric frame of reference to the ipsilesional side of space (Karnath & Fetter, 1995) and also as an intensification of a rightward gradient covering the global work space (Behrmann, Watt, Black, & Barton, 1997). In another recent study by Karnath and Fetter (1995), neglect patients were asked to search a dark room for a non-existent target amid an array of distractors. Eye movements recorded during search revealed that exploration of space was shifted to the right of the objective mid-sagittal plane. However, when the eye movement data were analyzed with the subjects’ self-reported mid-sagittal plane, the eye movements were distributed symmetrically around a rightward-shifted subjective mid-sagittal plane. The authors suggest that the essential component in hemispatial neglect is a systematic ipsilesional error in the computation of the egocentric coordinates of spatial reference. One paradoxical limitation of neuropsychological investigations in the study of hemispheric specialization for spatial attention is the use of subjects with clinical neglect, who, by definition, fail to perform the relevant task in the neglected hemispace. This prevents a direct comparison of the putatively differential attentional strategies linked to each hemisphere. In this investigation, we compared the impact of unilateral lesions on visual search strategies of patients without neglect who displayed accurate target detection in both hemispaces. We also used sensitive eye tracking measurements to reveal more complex hemispheric asymmetries in the neural encoding of search behaviors.

1398

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

2. Methods 2.1. Subjects All subjects in the current experiment were selected from a larger group (n = 106) participating in a study of visual attention at the Cognitive Neurology and Alzheimer’s Disease Center at Northwestern University Feinberg School of Medicine. Subjects were drawn from the Northwestern Memorial Hospital Neurology Service, the Rehabilitation Institute of Chicago, the control subject registry at the Buehler Center on Aging at Northwestern University Feinberg School of Medicine, and the subject pool of healthy older control participants in the Northwestern Alzheimer’s Disease Center. The Institutional Review Board of Northwestern University approved the study. 2.2. Inclusion and exclusion criteria A total of 40 stroke subjects with single unilateral strokes were initially identified from the larger subject group. These subjects were then screened for the following inclusion criteria: (1) absence of hemianopia on confrontation testing at the time of experiment; (2) absence of clinically salient hemispatial inattention on traditional bedside tasks of neglect (e.g. line bisection, visual extinction, manual exploration, and pencil and paper cancellation test); and (3) absence of verbal comprehension deficits (e.g. aphasia) that might interfere with understanding task instructions. A total of 11 patients with right-sided lesions and 12 with left-sided lesions met criteria. All stroke patients were tested at least 2 months after stroke, and in many instances, much later. Subject demographics appear in Table 1. A total of 19 community-dwelling, neurologically normal subjects of approximately the same age (range from 31 to 74 years) and level of education (range from 10 to 20 years) as the stroke subjects were also identified. All control subjects were administered research neurological examinations and a research MRI or CT brain scan to rule out abnormalities. Patients underwent a full clinical neurological evaluation and MRI or CT brain scans to confirm the presence and location of the lesion. The size of the lesion was determined from analysis of research MRI scans in 20 subjects. Three subjects (two Left Lesion and one Right Lesion) had clinical imaging studies and did not wish to undergo a research MRI scan. Two subjects in each lesion group had only sagit-

tal images that were analyzed on a Macintosh computer using the Scion Image program, as modified from NIH Image for the Macintosh by Scion Corporation and available on the internet at http://www.scioncorp.com. In Scion Image, the lesion volume was calculated by measuring the area of the lesion as it appears on consecutive slices and then multiplying by the slice thickness. All the remaining subjects had axial images and lesion size was analyzed in these subjects using the Stereo Investigator analysis program (MicroBrightField, Colchester, VT). Stereo Investigator uses the Cavalieri method to estimate lesion size from axial slices sampled every 10 mm. In both programs, total brain volume was estimated by measurements taken every 10 mm. The percentage of total brain volume occupied by lesion and the interval from stroke to test for each stroke subject can be found in Table 2. The two lesion groups did not differ with respect to percentage of whole brain volume occupied by the lesion. However, the Left Lesion group had a significantly longer interval from stroke to test (P < 0.01) (Left Lesion mean = 51.3 months, S.D. = 40.5; Right Lesion mean = 12.6 months, S.D. = 12.8). This can be attributed in large part to the presence of two subjects in this group who were tested nearly 9 years after stroke (Table 2). 2.3. Procedure Subjects were tested at the Cognitive Neurology Laboratory at Northwestern University Feinberg School of Medicine. The experimental protocol was explained to all subjects in advance and signed consent was obtained. Subjects underwent neuropsychological testing and were administered clinical tests of neglect and the experimental search task. 2.4. Neuropsychological and clinical neglect measures A version of the American National Adult Reading Test (ANART) (Schwartz & Saffran, 1987) and the Complex Ideational Material Subtest of the Boston Diagnostic Aphasia Examination (Goodglass & Kaplan, 1983) were administered as measures of estimated premorbid general intelligence and auditory language comprehension, respectively. Subjects were tested for evidence of clinical neglect with four tasks. The first was a test of visual extinction using

Table 1 Subject demographics and neuropsychological test data Group

N (M/F)

Mean Age (years)

Mean education (years)

ANART EIQ

BDAE

Left Lesion Right Lesion Control

12 (6/6) 11 (7/4) 19 (7/12)

58.9 (10.9) 57.3 (11.3) 55.5 (12.4)

14.3 (3.2) 14.0 (2.4) 15.8 (2.7)

114.8 (12.2) 118.4 (10.2) 122.7 (7.6)

10.5 (2.3) 11.3 (0.9) 11.2 (0.7)

ANART EIQ: American National Adult Reading Test, estimated full scale IQ; BDAE: Complex Ideational Subtest score from the Boston Diagnostic Aphasia Examination (maximum score = 12); numbers in parentheses are the standard deviations, with the exception of column 1.

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

1399

Table 2 Characteristics of subjects with strokes Lesion volumea

Lesion Site

Left Lesion group 1 56 2 17 3 39 4 4 5 2 6 103 7 36 8 20 9 51 10 119 11 111 12 57

n/a 0.80b 0.80b 1.08c 0.12c 1.00b 1.00b 1.10b 3.40b n/a 21.10b 5.60b

L central sulcus region and precentral and post-central gyri L external capsule, lateral basal ganglia and thalamus L basal ganglia L posterior thalamus and medial temporal lobe L inferior frontal gyrus L caudate nucleus and putamen Mid-L frontal lobe and middle and inferior frontal gyri L inferior frontal gyrus and internal capsule L frontal and parietal lobes and insular cortex L inferior frontoparietal region L MCA distribution L MCA distribution to temporal, posterior frontal and anterior parietal lobes

Right Lesion group 1 24 2 7 3 5 4 9 5 3 6 14 7 27 8 1 9 41 10 6 11 2

12.10b 1.30b 1.66c 1.50b 0.60b 7.00c 2.10b 0.20b 0.10b 0.80b n/a

R R R R R R R R R R R

Subject

Months since stroke

MCA distribution to frontal, temporal and parietal lobes lateral basal ganglia, insular cortex and body of caudate nucleus MCA distribution to insular cortex, basal ganglia and internal capsule insular cortex and lateral basal ganglia thalamus, basal ganglia, putamen and head of caudate nucleus frontal lobe and basal ganglia parietal lobe internal capsule thalamus external capsule posterior parietal lobe

a

Lesion volume is expressed as percentage of total brain volume. Lesion size estimated by Stereo Investigator. c Lesion size estimated by NIH Image. b

routine clinical methods of bilateral simultaneous stimulation. There were a total of 18 counterbalanced trials (6 left stimulation only, 6 right only, and 6 bilateral). Neglect was defined as present if the subject made 30% more errors on bilateral trials than on unilateral trials. The second was a line bisection task in which subjects were asked to place a mark bisecting a 26 cm horizontal line presented on an 8.5 in. × 11 in. sheet of paper. The average magnitude (in mm) and direction (left, right) of the deviation of the mark from true center was calculated over four trials for each subject and compared to the mean deviation of the normal control group. Neglect was defined as greater than 2.5 standard deviations from true center. Exploratory neglect was tested by blindfolding subjects and asking them to search by palpation with the preferred hand (unless hemiplegic) for 20 Velcro targets arranged in a non-linear fashion on a 62 cm × 46 cm Plexiglas board. Neglect was deemed present if 30% more targets were undetected in one hemispace than the other. Finally, visual target cancellation was measured with a paper-and-pencil test in which subjects were required to locate 60 targets (30 in each hemispace) within an array of over 300 non-linear distractors (Weintraub, 2000; Weintraub & Mesulam, 1985). Neglect was judged present on this test if 30% fewer targets were detected in one hemispace than in the other.

2.5. Experimental computerized visual search task Subjects were seated 40 in. in front of a 21-in., high-resolution display monitor subtending an angle of 23◦ horizontally and 17◦ vertically. The subject’s eye position was sampled during the experimental task by an ISCAN® RK-426PC (ISCAN, Burlington, MA) Pupil/Corneal Reflection Tracking System at a rate of 60 Hz. The position of the eye was automatically recorded in two-dimensional space and saved to a Macintosh computer for analysis using ILAB© (Gitelman, 2002), custom-designed software operating in the MATLAB® software environment (Mathworks, Natick, MA). All subjects in this study had reliable eye position data defined as the presence of valid eye position coordinates in at least 75% of each subject’s sample. Very few subjects had more than 10% missing data. Missing eye data was usually due to blinking during a trial. The experiment began with a standardized nine-point calibration procedure. Subjects maintained gaze on a central fixation point in the center of the display monitor. The central fixation point disappeared and was replaced by a non-linear array of 80 stimuli (20 per visual quadrant, consisting of 1 target number and 79 letter distractors). On each of 40 trials, one of four target numbers (2, 4, 6 or 7) appeared at one of the test locations (10 per visual quadrant). The target numbers were selected on the basis of pilot testing to

1400

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

determine which numbers were most easily discriminated from the distracter letters. Trial order was randomized so that a target could not appear in the same quadrant on more than three consecutive trials. Subjects were asked to search for “a number” within the letter distractors and to press a response key centered on a table in front of them. Subjects were not told which numbers would serve as targets. Subjects were also required to call out the number after pressing the response key to verify accuracy of response. Most stroke subjects used the preferred hand to respond. However, six Left Lesion subjects had significant right hemiplegia at the time of test and thus used the non-preferred left hand to respond. In order to minimize any bias introduced by using the non-preferred hand to respond, six Control subjects were also tested using the non-preferred hand (left in all cases). The target and distracter stimuli remained on the monitor until the subject responded or until 10 s had elapsed. 2.6. Data analysis Because we were interested in examining spatial asymmetries in visual search we operationally defined two visual hemispaces (i.e. two adjacent bins each 320 horizontal pixels wide and 480 pixels high) in the 640 × 480 pixel video display. This was done after data collection and the hemispace designation was not visible to the subjects during the search task. From the eye movement data collected during the search task we extracted five dependent measures for analysis and computed left and right hemispace values for each dependent measure described below. Accuracy of target detection was defined as a correct verbal response in addition to a button press within the 10-s time limit for each trial and was expressed as a percentage of the 40 trials. The total area covered during search, was defined as the number of display monitor pixels covered by the eye during all 40 trials, regardless of whether or not the target was found. The ISCAN system represents point of gaze as a single pixel in the display monitor. This does not imply that the system is able to provide one pixel spatial resolution. Nor does this imply that subject gaze was limited to this single pixel. Passing over a single pixel twice in a trial resulted in that pixel being counted twice. The total number of eye fixations made during search across all 40 trials constituted a third measure. An eye fixation was defined as stable eye position within any 6 pixel horizontal by 4 pixel vertical area on the display for at least 100 ms (Zihl & Hebel, 1997). For most subjects, the first fixation of each trial was usually of very long duration (often three times longer than the average fixation duration for that subject). This initial fixation was thought to reflect a delay in initiating active search following disappearance of the central fixation point and was excluded from analysis. Mean duration of eye fixations for each subject was also computed. Finally, we computed the mean distance of saccades made by each subject. A saccade was defined as a continuous eye move-

ment between two successive fixations. Saccadic distance was measured in screen pixels. 2.7. Magnitude of search asymmetry In order to facilitate between-group comparisons of search asymmetry, we computed an asymmetry composite for each of the five dependent measures. We used the following formula to derive these measures for the Left Lesion group: (left hemispace − right hemispace) (left hemispace + right hemispace) For the Right Lesion group, the asymmetry measures were computed using the following formula: (right hemispace − left hemispace) (left hemispace + right hemispace) These formulae were based on the a priori assumption that subjects with stroke would have larger values in the ipsilesional hemispace and were used to maximize the likelihood of positive values for the comparison of absolute magnitudes of asymmetry. A single mixed within- and between-subjects multivariate analyses of variance (MANOVA) compared the three subject groups on the five variables. The within-subjects factor was visual hemispace (left or right) and the between-subjects factor was group (Control, Left Lesion, Right Lesion). Significant effects and interactions were explored using paired t-tests with Bonferonni adjustment of alpha levels. Statistical significance was thus P < 0.01. A single multivariate analysis of variance (MANOVA) was used to compare the three groups on magnitude of search asymmetry derived as above for accuracy, area covered, total number of eye fixations, mean fixation duration, and mean saccade distance. The between-subjects factor was group (Control, Left Lesion, Right Lesion). Post hoc comparisons were run using the Bonferonni procedure. 2.8. Mapping spatial distribution of eye fixations within the global work space Linear and non-linear regression methods were used to compare the obtained distribution of eye fixations with predicted distributions based on the gradient and salience models. In addition, the distribution of fixations within the global work space for the Right Lesion group was fit to the salience model equation by best-fit equation modeling using the parameters specified by Anderson (1996). The distribution of eye fixations across the workspace during search was used as a measure of the spatial distribution of attention to evaluate the predictions of the two models. The display monitor, subtending approximately 23◦ of visual angle, was operationally partitioned into 23 separate vertical bins, each 1◦ of visual angle. Data from all 40 trials of the experimental task were used, regardless of accuracy of

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

1401

Fig. 1. Predictions of the gradient and salience models for the spatial distribution of attention in Normal and Right Lesion subjects. Each bar represents ∼1◦ on the display monitor which subtends a total of 23◦ of visual angle. The distribution of eye fixations predicted by the gradient model (A) follows a linear increase from left to right. The predicted distributions of eye fixations based on the salience model for Control (B) and Right Lesion (C) subjects are based on the sums of two bell-shaped salience curves and were determined using the algorithm and parameters proposed by Anderson (1996). The method for determining the numbers of fixations expected at each bin for each of the models is described in detail in the text.

detection, to calculate the average number of eye fixations within each bin for the Control and Right Lesion groups. The raw number of fixations in each bin was converted into percentages of total fixations for direct comparison to each other and to the model predictions. The distributions predicted by the gradient and salience models (Fig. 1A and C) were compared to the obtained distributions for the Right Lesion and Control groups only (Fig. 2). The Control group data were included to test the prediction of the salience model for a normal distribution of attention and to provide

a reference point with which to compare the Right Lesion data. The Left Lesion distribution was not modeled because this group did not demonstrate a significant asymmetry on this measure and because performance on this measure was almost identical to that of the Control group. The distribution of eye fixations following a right hemisphere lesion as predicted by the gradient model was obtained by fitting a straight line to the Right Lesion group data obtained in this experiment. We desired only to impose a linear fit to the data and not constrain the model

1402

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

Fig. 2. Obtained eye fixation distributions for Normal Control and Right Lesion groups. Both Control (A) and Right Lesion (B) distributions are tri-modal with the largest proportion of fixations occurring in the middle third of the defined workspace. The distribution of the Right Lesion group is shifted to the right of center.

with regard to the starting or ending point of the line. Thus, the y-intercept was not specified. This fitting resulted in the equation y = 0.1523x + 2.5209. The predicted number of fixations for each of the 23 bins was derived from this equation and the root-mean-squared (RMS) difference between the model and the data obtained from the Right Lesion group was calculated. The RMS difference is a method for evaluating regression models and represents the average distance from the model to each data point. It is calculated by squaring the differences between the data and the model at each point then taking the square root of the average of these squared residuals. A lower RMS value indicates a better fit between the model and the data.

The predicted distribution of fixations for a non-braininjured individual and for a patient with a right hemisphere lesion according to the salience model was computed using the algorithm and parameter values described by Anderson (1996). According to this model, three parameters define each of two bell-shaped distributions provided by each cerebral hemisphere. The six parameters define the breadth, height, and location of the two curves on the horizontal plane (see Table 3). The relevant predictions of the salience model were compared to the Control and Right Lesion group distributions obtained in the present study and the RMS difference was calculated for both comparisons.

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

1403

Table 3 Evaluation of gradient and salience models Model

Formula and parameter values

RMS of residuals

Gradient (Right Lesion) Salience (Normal Control)

y = 0.1523x + 2.5209 y = 100 arctan((x − 750)/100)0.6 + 250 arctan((x − 475)/250)0.6 + 100 arctan((−x + 750)/100)0.6 + 250 arctan((−x + 475)/250)0.6 y = 100 arctan((x − 750)/100)0.6 + 75 arctan((x − 480)/75)0.6 + 100 arctan((−x + 750)/100)0.6 + 75 arctan((−x + 480)/75)0.6 y = 87 arctan((x − 780)/87)5 + 70 arctan((x−470)/70)6.7 + 87 arctan((−x + 780)/87)5 + 70 arctan((−x + 470)/70)6.7 y = 97.8 arctan((x − 842.4)/97.8)5.2 + 181.6 arctan((x − 532.8)/181.6)6.78 + 97.8 arctan((−x + 842.4)/97.8)5.2 + 181.6 arctan((−x + 532.8)/181.6)6.78

1.90 1.90

Salience (Right Lesion) Salience best-fit starting values (Right Lesion) Salience best fit (Right Lesion)

1.75 na 0.99

Formula variables: x = location (bin) on horizontal axis (1–23), y = predicted number of fixations at x. Salience algorithm (Anderson, 1996): salience = S.D.L arctan((XR − ML )/S.D.L )SFL + S.D.R arctan((XR − MR )/S.D.R )SFR + S.D.L arctan((−XL + ML )/S.D.L )SFL + S.D.R arctan((−XL + MR )/S.D.R )SFR . In this formula, X represents any point along the horizontal dimension, M is the point on the horizontal axis under the peak of each curve, S.D. is the standard deviation, or breadth of each curve, and SF is a scaling factor for the height of each curve. The L and R subscripts are used to denote the contributions from the left and right attentional systems, respectively.

In a final analysis, the salience model algorithm was used to provide the best fit for the fixation distribution of the Right Lesion group obtained in this experiment. Algorithm starting values for this best-fit analysis can be found in Table 3.

3. Results 3.1. Neuropsychological and neglect screening measures All subjects had estimated premorbid verbal IQ’s within the above-average to superior range and both lesion groups had normal language comprehension (see Table 1). None of the subjects demonstrated neglect on any of the clinical neglect screening measures. 3.2. Computerized visual search task Both MANOVA’s produced significant omnibus F statistics at P < 0.01 each. The following sections describe the univariate tests for each of the five dependent measures. 3.2.1. Accuracy of target detection The three groups differed with respect to the accuracy of target detection in the two hemispaces (F (2, 39) = 25.21, P < 0.001). Both the Right and Left Lesion groups detected significantly fewer targets in the contralesional hemispace (P < 0.01 and P < 0.001, respectively), while control subjects did not differ in this regard. The three groups differed in the magnitude of this asymmetry (F (10, 70) = 3.56, P < 0.001) with the Right Lesion group demonstrating greater asymmetry compared to the Control group (P < 0.001), but not when compared to the Left Lesion group (P = 0.09). The Left Lesion group did not differ from the Control group in asymmetry of target detection. Fig. 3 shows the accuracy for the three groups by hemispace.

3.2.2. Area covered during search The three groups differed with respect to the area searched in the two hemispaces (F (2, 39) = 11.39, P < 0.001). Although both Right and Left Lesion groups covered less area while searching the contralesional space, these differences (P = 0.02 each) failed to reach statistical significance with adjusted P-value criterion of 0.01. Interestingly, the Control group also showed a nearly significant bias toward searching more area in the left than the right hemispace. The three groups did not differ in the magnitude of the asymmetry described above (Fig. 3). 3.2.3. Total eye fixations The groups differed with respect to the total number of eye fixations made in the two hemispaces during search (F (2, 39) = 9.45, P < 0.001). This effect was supported primarily by the Right Lesion group who made significantly fewer fixations in the contralesional compared to the ipsilesional hemispace (P < 0.01). The Left Lesion and Control groups did not show significant differences as a function of hemispace on this measure. Accordingly, the magnitude of the Right Lesion group’s asymmetry on this measure was greater than both the Left Lesion group (P < 0.01) and the Control group (P < 0.001) (Fig. 3). 3.2.4. Mean fixation duration The three groups did not differ in the average duration of fixations made in the two hemispaces, nor did they differ in the magnitude of the asymmetry on this measure. 3.2.5. Mean saccade distance The three groups did not differ significantly in the average length of saccades made during search of the two hemispaces. However, the Right Lesion group did make significantly longer saccades in the left hemispace compared to the right (P < 0.001) and this asymmetry was greater than that

1404

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

Fig. 3. Performance on dependent measures; Control, Left Lesion, and Right Lesion group performance with respect to accuracy of target detection, area covered, total number of fixations, mean fixation duration, and mean saccade distance during search for the targets (A, B, C, D, and E respectively). Significant within-group hemispheric asymmetries (P < 0.01) are indicated by asterisks.

of the other two groups who essentially showed no asymmetry at all (P < 0.01 each) (Fig. 3). 3.2.6. Spatial distribution of eye fixations The distributions of eye fixations across the horizontal plane for all groups are best characterized as trimodal, with a

major percentage occurring in the central region (Fig. 2). The Right Lesion group’s obtained distribution was shifted to the right of midline and, consistent with the eye fixation asymmetry results, most fixations occurred in the right hemispace. Correspondence of the gradient model to the obtained data was lowest in the central region and at the rightmost edge of

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

1405

Fig. 4. Evaluation of model predictions to obtained distributions. The distribution of eye fixations in the Right Lesion group is compared with the predicted distributions from the gradient and salience models. The salience model resulted in a lower root-mean-squared (RMS) residual value, which implies that it is a better fit than the gradient model with the actual data.

Fig. 5. Salience model for Right Lesion using best fit parameters; depiction of the best fit of the salience model to the Right Lesion group data. Starting parameter estimates and RMS residual value can be found in Table 3.

1406

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

the visual workspace. The root mean square difference for this comparison was 1.90 (Fig. 4A). The salience model’s prediction of the Right Lesion group’s distribution was better with a RMS value of 1.75 (Fig. 4B). The distribution of fixations made by the Left Lesion group was nearly identical to that made by the Control group. Indeed, the percentage of fixations made by the Left Lesion group in the majority of the 23 bins did not differ from the Control group by separate Mann–Whitney U-tests. This is expected given the lack of group differences in the eye-movement-dependent behavioral measures (especially in the number of fixations made). Because the Left Lesion and Control groups did not differ in this regard and because the gradient and salience models do not make explicit predictions about alterations in attention following left-cerebral lesions, we chose to evaluate only the Right Lesion and Control group’s distributions with the two models. The distribution of fixations for the Control group was adequately described by the salience model’s prediction. The RMS difference for this comparison was 1.90. Finally, the best-fit estimate of the six-parameter model proposed by the salience theory produced a RMS value of 0.988 (Fig. 5). The parameter values of the salience model and the best fit for the current data can be found in Table 3. In summary, this study demonstrates that: (1) Contralesional deficits in accuracy of target detection are present following either left or right cerebral lesions, but contralesional deficits in the number of eye fixations and the average length of saccades occur only in subjects with right cerebral lesions. (2) Right unilateral cerebral lesions alter the architecture of visuomotor search in the contralesional hemispace, even in the absence of clinically observable neglect. (3) Differences in the programming of fixations may constitute a major mechanism of hemispheric asymmetry in spatial attention. (4) The salience model Anderson (1996) provides a better characterization of the distribution of attention following right-sided damage without hemispatial neglect than does a simple gradient model (Kinsbourne, 1970a, 1970b, 1987).

4. Discussion The present study sought to explore attentional asymmetries reflected in visual search following unilateral strokes that did not produce clinically observable neglect. Control subjects explored slightly more area in the left visual hemispace when compared to the right. However, this did not reach statistical significance. This finding may reflect the impact of the organizational structure imposed by reading the English language. Several early studies (e.g. Kugelmass, Lieblich, & Ehrlich, 1972) have demonstrated a bias in the initiation of perceptual search based on the direction from which text in the native language is read (left to right in En-

glish readers, right to left in Hebrew, etc.). The linguistic nature of our stimuli may have reinforced the adoption of this organizational search strategy. The Left Lesion group demonstrated significant asymmetry in target detection with fewer targets detected in the contralesional right visual hemispace than the left. This pattern closely resembled that seen in the Control group and did not differ in magnitude from the Control group. On the other measures, the Left Lesion group did not show significant asymmetries. In general, the pattern of performance seen in the Left Lesion group was the same as the Control group’s (i.e. L > R), but usually of slightly greater magnitude. Thus, the Left Lesion group’s performance appears to be a mild exaggeration of the Control group’s pattern. One possible reason for this mild exaggeration of search-related asymmetry may be the use of linguistic stimuli in the search task. Although all Left Lesion subjects performed within normal limits on a test of language comprehension, it is possible that the lesions could have also interfered with the ability to rapidly process the linguistic targets in this task. This, coupled with mild disruption of the left attentional network may have resulted in an isolated contralesional deficit in target detection. The use of non-linguistic stimuli in future experiments would be useful in determining the roles of linguistic and attentional processes in explaining this result. The Right Lesion group also demonstrated a significant asymmetry in target detection. However the pattern was opposite (i.e. L < R) that of the Control and Left Lesion groups. This finding underscores the notion that visual search performance in the Right Lesion group is qualitatively different from the pattern seen in normal control subjects and Left Lesion subjects. In addition to an asymmetry in target detection, the Right Lesion group made fewer visual fixations and longer saccades while searching the left visual hemispace. These dependent measures may more directly tap the fundamental components of the attentional network than target detection accuracy, which may require additional cognitive networks such as language. For the purposes of this study, we defined a saccade as a continuous eye movement between two fixations. Thus, the number of fixations and the number of saccades are highly correlated. Theoretically, this fact does not constrain the length of any individual saccade. However, in our subjects the average length of saccades was also highly correlated with the number of fixations in both the left (r 2 = −0.54, P = 0.000) and right (r 2 = −0.63, P = 0.000) hemispaces. Thus, fewer fixations resulted in longer saccades and although fixations and saccade length were treated as independent measures, they are highly interdependent and may rely on a common process. The reason for longer saccades (or fewer fixations) is unclear and deserves further study. In contrast to the isolated asymmetry in target detection seen in the Left Lesion group, the asymmetries seen in the Right Lesion group suggest an elementary disruption of the

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

architecture of visual search revealed by eye movement measures. Furthermore, these results suggest that the fundamental components of search revealed by eye movement monitoring largely are supported by right cerebral hemisphere structures. The finding of significant contralesional hemifield asymmetries in the Right Lesion group is in contrast to the results of early work on the activation-orienting model. For example, Reuter-Lorenz et al. (1990) reported data suggesting that hemisphere-specific attentional resources of normal subjects can be manipulated by presentation of stimuli in the contralateral visual hemifield. In their study, young normal subjects viewed tachistoscopically-presented bisected lines of varying length in several locations in the right and left visual hemifields. Subjects judged the accuracy of the bisection point. The results showed that bisection judgments were affected depending on which cerebral hemisphere was activated by contralateral stimulation. The authors interpreted this as strong evidence for the activation-orienting hypothesis which would appear to be evidence against a right hemisphere dominance for spatial attention. In attempting to reconcile these findings with the findings reported in this paper, it is clear that differences in tasks must be considered. The task used by Reuter-Lorenz et al. is based on externally generated stimulation which may utilize more automatic or perceptually driven processes. In fact, the authors touch on this point briefly in their paper. The visual search task used in the present study, in contrast, ostensibly requires a great deal of internally generated, top–down, cognitive processing to construct an effective search strategy. Under this condition, there is a clear preference for a distribution of attention that is best described by the Salience model proposed by Anderson. Indeed, Anderson’s model was originally proposed to account for line bisection behavior in subjects, whereas in the Reuter-Lorenz study the lines presented to subjects were already bisected. Our results also suggest that although traditional tests of neglect may have many advantages for assessing acute patients or those with marked visual neglect, they are not sensitive enough to detect subtle alterations of visual attention that remain after the acute phase of stroke. The failure of traditional bedside pencil and paper tasks to register mild deficits in spatial attention may ultimately result in an under-representation of the true frequency of hemispatial inattention in the population of patients with stroke. With regard to the spatial distribution of visual fixations made during visual search, all groups demonstrated a trimodal distribution with peaks of fixations centered in the left and right hemispaces and in the center of the visual space. For all three groups, the largest percentage of fixations occurred in the center of the defined workspace. The first fixation of each trial was excluded from the analysis to account for fixations resulting from latency in disengaging from the pre-trial central fixation point. In a recent study using the same visual search task to study the impact of aging and Alzheimer’s disease on attention (Rosler et al., 2000), young

1407

normal subjects made several small fixations around the center of the global workspace in the first several seconds prior to making subsequent larger saccades. This observation suggests that these subjects may have been “planning a strategy” before initiating a visual search. This preparation might consist of an immediate deployment of covert attention allowing for an initial scan of the global workspace (or portions of it) (Greenwood, Parasuraman, & Alexander, 1997; Treisman & Gelade, 1980; Zelinsky, Rao, Hayhoe, & Ballard, 1997). This initial scan may have been used to identify likely areas for deployment of focal attention and more detailed search. This notion is similar to a recent model (Guided Search 2.0) proposed by Wolfe (1994) who suggests that, in the presence of similar appearing distractors, search is generally initiated in “parallel”, and then proceeds in “serial” fashion. The remainder of the eye movements made by all subjects in this study strongly suggests the use of a serial search strategy and this is consistent with eye movement data reported by other investigators (e.g. Williams, Reingold, Moscovitch, & Behrmann, 1997). The most notable finding in this analysis is the pattern of fixations produced by the Right Lesion group in the absence of clinical neglect. This pattern is best characterized by an ipsilesional remapping of the “significant work space” (Mesulam, 2000) and is consistent with the results of several studies on patients with neglect by Karnath and colleagues who conclude that the salient behavioral manifestation is an ipsilesional shift in the egocentric frame of reference (Karnath & Fetter, 1995; Karnath, Niemeier, & Dichgans, 1998). On visual inspection, the Right Lesion group’s pattern of fixation distribution was not a linear gradient, but appeared to be a skewed variant of a normal distribution that had been translated across the horizontal plane. We chose to implement only the fundamental principle of the gradient model, namely that the distribution of attention in the global workspace follows a linear gradient from left to right. We did not attempt to restrain the model further by setting a y-intercept in the linear fit to the data. While we acknowledge that there are complexities in both the initial conceptualization of the gradient model (Kinsbourne, 1970a,b) and in the subsequent work on unilateral neglect that are not addressed in this rather limited conceptualization, nonetheless, we believe that the basic premise of the linear gradient notion was not supported in this population of patients with unilateral lesions and no clinically observable neglect. Variations on the experimental paradigm used in the present study could be used to explore the more subtle predictions of the gradient model by varying hemispace of stimulus presentation and the nature of the stimulus (verbal versus non-verbal) (Reuter-Lorenz, Kinsbourne, & Moscovitch, 1990). The distribution appeared to be most similar to the prediction made by the salience model (Anderson, 1996). When directly compared to the distribution of fixations produced by the Right Lesion group, the salience model resulted in a smaller root-mean-squared value than the gradient model

1408

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409

and was therefore determined to better describe the actual data obtained in this experiment. The fixation distribution of the Right Lesion group in the current study is in partial agreement with those of Behrmann et al. (1997), who studied eye movements while nine patients with neglect performed a visual search task. The subjects in the Behrmann study demonstrated a distinctive pattern of fixations across the global workspace consisting of an increasing linear gradient from left to right (Behrmann et al., 1997). However, the neglect subjects demonstrated one important deviation from a perfect linear gradient: a distinct drop-off in the rightmost edge of space. With this drop-off, the data do not fully support one of the most important predictions of the gradient theory, namely, that the maximum weight of attention occurs in the rightmost space. Finally, the parameter values obtained from the best fit of the salience model to the actual data produced by the Right Lesion group in this study should be useful to drive future research on the characterization of the spatial distribution of attention following unilateral lesions.

Acknowledgements This research was supported by grant NS20285 (M.-M. Mesulam, PI) from the National Institute of Neurologic Disorders and Stroke, and, in part, by grants AG14068 from the National Institute on Aging (S. Weintraub, PI) and H133B980 (E. Roth, PI) from the Department of Education to the Rehabilitation Institute of Chicago. The authors thank Alissa Hays and Kathleen Schelble for technical assistance, and Marc Dublin for statistical assistance. References Anderson, B. (1996). A mathematical model of line bisection behaviour in neglect. Brain, 119, 841–850. Barton, J. J., Behrmann, M., & Black, S. (1998). Ocular search during line bisection: the effects of hemi-neglect and hemianopia. Brain, 121, 1117–1131. Behrmann, M., Watt, S., Black, S., & Barton, J. J. S. (1997). Impaired visual search in patients with unilateral neglect: an oculographic analysis. Neuropsychologia, 35, 1445–1458. Bisiach, E., Cornacchia, L., Sterzi, R., & Vallar, G. (1984). Disorders of perceived auditory lateralization after lesions of the right hemisphere. Brain, 107, 37–52. Bowen, A., McKenna, K., & Tallis, R. C. (1999). Reasons for variability in the reported rate of occurrence of unilateral spatial neglect after stroke. Stroke, 30, 1196–1202. Chedru, F., Leblanc, M., & Lhermitte, F. (1973). Visual searching in normal and brain-damaged subjects (contribution to the study of unilateral inattention). Cortex, 9, 94–111. Coull, J. T., & Nobre, A. C. (1998). Where and when to pay attention: the neural systems for directing attention to spatial locations and to time intervals as revealed by both PET and fMRI. Journal of Neuroscience, 18, 7426–7435. Deubel, H., & Scheider, W. X. (1996). Saccade target selection and object recognition: evidence for a common attentional mechanism. Vision Research, 36, 1827–1837.

Fullerton, K. J., McSherry, D., & Stout, R. W. (1986). Albert’s test: a neglected test of perceptual neglect. Lancet, 1, 430–432. Gitelman, D. R. (2002). ILAB: a program for post experimental eye movement analysis. Behavior Research Methods, Instruments and Computers, 34, 605–612. Gitelman, D. R., Alpert, N. M., Kosslyn, S., Daffner, K., Scinto, L., & Thompson, W. et al., (1996). Functional imaging of human right hemispheric activation for exploratory movements. Annals of Neurology, 39, 174–179. Goodglass, H., Kaplan, E. F. (1983). The Boston diagnostic aphasia examination. Philadelphia: Lea and Febiger. Greenwood, P. M., Parasuraman, R., & Alexander, G. E. (1997). Controlling the focus of spatial attention during visual search: effects of advanced aging and Alzheimer disease. Neuropsychology, 11, 3–12. Ishiai, S., Koyama, Y., Seki, K., & Nakayama, T. (1998). What is line bisection in unilateral spatial neglect? Analysis of perceptual and motor aspects in line bisection tasks. Brain and Cognition, 36, 239–252. Karnath, H. O., & Fetter, M. (1995). Ocular space exploration in the dark and its relation to subjective and objective body orientation in neglect patients with parietal lesions. Neuropsychologia, 33, 371–377. Karnath, H. O., Niemeier, M., & Dichgans, J. (1998). Space exploration in neglect. Brain, 121, 2357–2367. Kinsbourne, M. (1970a). A model for the mechanism of unilateral neglect of space. Transactions of the American Neurological Association, 95, 143–146. Kinsbourne, M. (1970b). The cerebral basis of lateral asymmetries in attention. Acta Psychologica, 33, 193–201. Kinsbourne, M. (1987). Mechanisms of unilateral neglect. In M. Jennerod (Ed.), Neurophysiological and neuropsychological aspects of spatial neglect (pp. 69–86). Amsterdam: Elsevier. Kugelmass, S., Lieblich, A., & Ehrlich, C. (1972). Perceptual exploration in Israeli, Jewish and Bedouin children. Journal of Cross-Cultural Psychology, 3, 345–352. Mesulam, M.-M. (1981). A cortical network for directed attention and unilateral neglect. Annals of Neurology, 10, 309–325. Mesulam, M.-M. (1985). Attention, confusional states and neglect. In M.-M. Mesulam (Ed.). Principles of behavioral neurology (pp. 125– 168). Philadelphia (PA): FA Davis. Mesulam, M. M. (1990). Large-scale neurocognitive networks and distributed processing for attention, language, and memory. Annals of Neurology, 28, 597–613. Mesulam, M. M. (1998). From sensation to cognition. Brain, 121, 1013– 1052. Mesulam, M. M. (1999). Spatial attention and neglect: parietal, series B: biological sciences. Philosophical Transactions of the Royal Society of London, 354, 1325–1346. Mesulam, M. (2000). Brain, mind, and the evolution of connectivity. Brain and Cognition, 42, 4–6. Moore, T., & Fallah, M. (2001). Control of eye movements and spatial attention. Proceedings of the National Academy of Sciences of the United States of America, 98, 1273–1276. Mosidze, V. M., Mkheidze, R. A., & Makashvili, M. A. (1994). Disorders of visuo-spatial attention in patients with unilateral brain damage. Behavioral Brain Research, 65, 121–122. Nobre, A. C., Sebestyen, G. N., Gitelman, D. R., Mesulam, M. M., Frackowiak, R. S., & Frith, C. D. (1997). Functional localization of the system for visuospatial attention using positron emission tomography. Brain, 120, 515–533. Nobre, A. C., Coull, J. T., Frith, C. D., & Mesulam, M. M. (1999). Orbitofrontal cortex is activated during breaches of expectation in tasks of visual attention. Nature Neuroscience, 2, 11–12. Reuter-Lorenz, P. A., Kinsbourne, M., & Moscovitch, M. (1990). Hemispheric control of spatial attention. Brain and Cognition, 12, 240–266. Rosler, A., Mapstone, M. E., Hays, A. K., Mesulam, M. M., Rademaker, A., & Gitelman, D. R. et al., (2000). Alterations of visual search strategy in Alzheimer’s disease and aging. Neuropsychology, 14, 398– 408.

M. Mapstone et al. / Neuropsychologia 41 (2003) 1396–1409 Schwartz, M., Saffran, E. (1987). The American-NART: Replication and extension of the British findings on the persistence of word pronunciation skills in patients with dementia. Philadelphia, in press. Sheliga, B. M., Riggio, L., Craighero, L., & Rizzolatti, G. (1995). Spatial attention-determined modifications in saccade trajectories. Neuroreport, 6, 585–588. Shepard, M., Findlay, J., & Hockey, R. (1986). The relationship between eye movements and spatial attention. Quarterly Journal of Experimental Psychology, 38, 475–491. Spiers, P. A., Schomer, D. L., Blume, H. W., Kleefield, J., O’Reilly, G., & Weintraub, S. et al., (1990). Visual neglect during intracarotid amobarbital testing. Archives of Neurology, 40, 1600–1606. Treisman, A., & Gelade, G. (1980). A feature integration theory of attention. Cognitive Psychology, 112, 97–136. Weintraub, S. (2000). Neuropsychological assessment of mental state. In M. -M. Mesulam (Ed.). Principles of behavioral and cognitive neurology (pp. 121–173). New York: Oxford University Press..

1409

Weintraub, S., Mesulam, M. -M. (1985). Mental state assessment of young and elderly adults in behavioral neurology. In M. -M. Mesulam (Ed.), Principles of behavioral neurology (pp. 71–123). Philadelphia (PA): FA Davis. Weintraub, S., & Mesulam, M. M. (1987). Right cerebral dominance in spatial attention: further evidence based on ipsilateral neglect. Archives of Neurology, 44, 621–625. Williams, D. E., Reingold, E. M., Moscovitch, M., & Behrmann, M. (1997). Patterns of eye movements during parallel and serial visual search tasks. Canadian Journal of Experimental Psychology, 51, 151– 164. Wolfe, J. M. (1994). Guided search 2.0: a revised model of visual search. Psychonomic Bulletin and Review, 1, 202–238. Zelinsky, G. J., Rao, R., Hayhoe, M. M., & Ballard, D. H. (1997). Eye movements reveal the spatiotemporal dynamics of visual search. Psychological Science, 8, 448–453. Zihl, J., & Hebel, N. (1997). Patterns of oculomotor scanning in patients with unilateral posterior parietal or frontal lobe damage. Neuropsychologia, 35, 893–906.