AN fmri EXAMINATION OF VISUAL INTEGRATION IN SCHIZOPHRENIA

Transcription

1 Journal of Integrative Neuroscience, Vol. 8, No. 2 (2009) c Imperial College Press Research Report AN fmri EXAMINATION OF VISUAL INTEGRATION IN SCHIZOPHRENIA STEVEN M. SILVERSTEIN,,, SARAH BERTEN,, BRIAN ESSEX, ILONA KOVÁCS, TERESA SUSMARAS and DEBORAH M. LITTLE, University of Medicine and Dentistry of New Jersey University Behavioral HealthCare and Department of Psychiatry Robert Wood Johnson Medical School Department of Psychiatry, University of Illinois at Chicago Department of Cognitive Science Budapest University of Technology and Economics Departments of Neurology and Rehabilitation Anatomy and Cell Biology, Ophthalmology and Visual Sciences and Psychology, University of Illinois at Chicago silvers1@umdnj.edu Received 23 February 2009 Accepted 2 April 2009 Behavioral and electrophysiological studies of schizophrenia have consistently demonstrated impairments in the integration of visual features into unified perceptual representations. Specific brain regions involved in this dysfunction, however, remain to be clarified. This study used functional Magnetic Resonance Imaging (fmri) to examine the relative involvement of visual cortex areas (involved in form perception) and parietal and frontal regions (involved in attention), in the visual integration impairment in schizophrenia. Fourteen patients with schizophrenia and 14 healthy controls were compared on behavioral performance and data acquired via fmri while completing a contour integration task that had previously been used to identify a visual integration deficit in schizophrenia. The schizophrenia patients demonstrated poorer visual integration than controls. Analyses of peak signal change indicated that while the groups were equivalent in area V1, the schizophrenia group demonstrated reduced signal in areas V2 V4, which are the earliest regions sensitive to global configurations of stimuli. Moreover, whereas the control group demonstrated greater recruitment of prefrontal and parietal areas during perception of integrated forms compared to random stimuli, the schizophrenia group demonstrated greater recruitment of frontal regions during perception of random stimuli. The two groups differed on brain regions involved in form perception even when they were matched on accuracy levels. The visual integration disturbance in schizophrenia involves both deficient basic visual processes (beginning as early as occipital region V2), as well as reduced feedback from visual attention regions that normally serves to amplify relevant visual representations relative to irrelevant information. Keywords: Schizophrenia; fmri; perception; cognition; vision; occipital lobe. Corresponding author. 175

2 176 Silverstein et al. 1. Introduction A consistent finding in the literature on cognitive-perceptual impairments in schizophrenia is that of a reduced ability to integrate stimulus elements into coherent visual representations. While patients with schizophrenia do appear to be able to process continuous contour, whether real [7] or illusory [19, 61, 62], visual binding has been shown to be deficient when noncontiguous elements need to be integrated into a perceptual whole [e.g., 11, 48, 58, 59]. Over the past 30 years, at least 30 studies have demonstrated such reductions in visual feature integration abilities in schizophrenia [69]. Moreover, in 10 of these studies [14, 22, 43, 46, 48 50, 58, 65, 72], the reduced ability to integrate information, and subsequent reduced influence of visual context, led to superior performance, compared to controls, in making decisions about individual features. Therefore, evidence for impairments in visual integration has been convincingly demonstrated independent of a generalized deficit [6, 28, 29, 55]. These impairments also cannot be accounted for by medication, as they have been demonstrated in unmedicated patients [20]. This extensive experimental literature is consistent with earlier clinical descriptions and first-hand patient accounts of fragmented perception in schizophrenia [4, 64, 68]. The importance of visual integration disturbances in schizophrenia is reflected in the inclusion of this cognitive domain as a core construct in the recent NIMH-sponsored Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia (CNTRICS) initiative [3, 5, 23]. In contrast to the wealth of behavioral studies of visual integration dysfunction in schizophrenia, there have been relatively few studies of its underlying neurophysiology. Electrophysiological studies of nonpatients have generally identified that the lateral occipital cortex (LOC) plays a role in visual binding, as part of a network also involving the prefrontal cortex [52, 53]. Electrophysiological studies of schizophrenia [11, 12, 18] identified both ventral and dorsal stream abnormalities associated with reduced P1 amplitude when patients viewed fragmented pictures. A recent study found reduced N150 amplitude to global fragmented targets in a global local task in schizophrenia [26], and the source of that waveform has been localized to V3/V3a within the LOC [10, 26, 70]. Functional magnetic resonance imaging (fmri) studies of visual integration in nonpatients have localized regions subserving intact integration, including both earlier visual areas (e.g., V2), and higher visual areas involved in shape processing (e.g., V3, V4, the LOC and posterior fusiform regions) [1, 2, 30, 38, 42, 52]. Foxe et al. [19] found intact illusory contour processing in schizophrenia, and identified excessive frontal lobe activity during contour processing, which was interpreted as a compensation for impaired ventral stream processing. To date, however, fmri has not been used to examine integration of spatially separated elements into unified perceptual wholes in schizophrenia. Therefore, in this study, we examined integration of visual features into forms using a psychophysically rigorous contour integration task [31 37, 44, 57, 59] that has previously demonstrated sensitivity to both integration deficits and top down effects

3 Visual Integration in Schizophrenia 177 Fig. 1. Samples of images from the 2 alternative forced choice (2AFC) contour integration task. Top left: 0 jitter, top right: 7 8 jitter, center left: jitter, center right: jitter, bottom left: jitter, bottom right: jitter. These panels show only the region of the display containing the contour. The actual stimuli contain approximately 75% additional space that contains only noise elements. on these in schizophrenia [37, 57, 59]. The task involves the detection of a roughly circular shape whose contour is comprised of unconnected elements embedded within a background of similar, but randomly placed elements (see Fig. 1). Each element is an example of a Gabor signal a Gaussian-modulated, sinusoidal luminance distribution that reflects the center-surrounded properties of orientation-selective spatial frequency (feature) detectors in primary visual cortex (V1). Perception of the circular contour in this task requires the linking of individual Gabor elements into an emergent circular shape [36], which requires context-sensitive interactions among neurons in V1 that code the orientation-correlated contour elements [13, 15]. This is thought to involve feedback from higher visual areas, such as V2, V3, and V4, where visual form information is initially processed [54]. The embedded contour cannot be detected simply by analysis of individual Gabor elements, or by neurons with large receptive fields corresponding to the size of the contour [8, 24]. The purpose of this study was to use fmri to investigate abnormal visual integration in schizophrenia, and in particular, to clarify the contributions of activation changes in primary visual cortex, in higher visual areas, and outside of the occipital lobe. Our hypothesis was that the schizophrenia group would demonstrate both

4 178 Silverstein et al. poorer behavioral performance (accuracy and reaction time) on the task compared to controls, and differences in activation of brain regions associated with visual processing of stimulus configurations. Specifically, if the schizophrenia-related visual integration impairment is not secondary to more basic impairments, this would be reflected in abnormal levels of activation in visual cortex regions sensitive to stimulus grouping (e.g., V2 V4), but not in abnormally high or low levels of activity in area V1, where basic feature processing occurs. In addition, we predicted that levels of prefrontal and parietal activation would be reduced compared to controls, demonstrating reduced top down feedback to, or attentional enhancement of stimulus configurations even in conditions where patients achieved adequate contour integration. 2. Methods 2.1. Subjects The sample consisted of 14 outpatients with schizophrenia (9 men) and 14 healthy control subjects (5 men). Patient diagnosis was based on the Structured Clinical Interview for DSM-IV Diagnosis-Patient Version (SCID) [16] which was conducted by the first author or a research assistant who had previously achieved a level of reliability with the first author across SCID items of κ>0.75. Inclusion criteria for the patient group were a diagnosis of schizophrenia, age 18 55, and an ability to successfully complete a practice version of the task prior to the scanning session. Exclusion criteria included active substance abuse or dependence, a current mood disorder, a history of a neurological disorder or head injury with loss of consciousness lasting more than 10 minutes, or documented intellectual impairment. Control subjects were not screened for personal or familial psychopathology. However, controls were screened for present or past neurological and substance abuse conditions during a pre-scan interview. The groups were matched on age and parental education levels. All subjects completed a practice version of the task, and were familiarized with the scanning environment (using a mock scanner) the day before the fmri session. At that session, all patient subjects were also interviewed using the SCID and the Positive and Negative Syndrome Scale (PANSS) [27, 39]. No potential subjects failed the screening version of the contour integration task. Medication level was assessed for patients using published conversion formulas [73] to generate chlorpromazine equivalent daily doses for second-generation antipsychotic medications. All subjects provided written informed consent. Subjects were asked to refrain from alcohol for 24 hours prior to all fmri sessions and abstain from smoking and caffeine for 12 hours prior to the scan. Each subject was then questioned for the use of these substances prior to the scan. One of the patients and two of the controls reported use of caffeine within 6 hours of the MRI scan. Additionally, three patients reported use of nicotine within 6 hours of the MRI.

5 Visual Integration in Schizophrenia Contour integration task The stimuli consisted of a closed chain of Gabor elements forming a left- or rightpointing egg-like shape within a background of randomly oriented Gabor elements. The spacing between the contour elements was kept constant at eight times the wavelength of a single element. The ratio of average adjacent background element spacing to contour element spacing, or signal-to-noise ratio, in each image was 0.9 a level that requires long-range, horizontal, excitatory interactions between the neurons coding the correlated orientations of the contour elements for intact contour perception [13, 36]. Task difficulty was manipulated across 6 conditions, by randomly jittering the orientation of the contour elements + or 0, 7 8, 11 12, 15 16, 19 20, or (see Fig. 1). As contour element orientation deviates from 0 tangent to its original position, the correlations between orientations of adjacent elements are reduced, and ability to perceive the circular shape is reduced. The ability of the visual system to rapidly form the long-range, excitatory, context-dependent circuits that allow for visual integration can therefore be probed by examining behavioral performance across these conditions, and specifically, examining the threshold at which performance reaches a level mid-way between chance and perfect performance (75%). A reduced ability to activate the neural circuitry involved in visual integration, as is predicted to characterize schizophrenia, would result in a lower jitter value at threshold and reduced activation in cortical areas involved in the formation and processing of configurations of elements [1]. A set of 40 images, evenly divided between left- and right-pointing stimuli, was presented in each block at each of the six difficulty levels. A new shape and background were generated for each card, but all of the contours had the same general size and shape. There were 240 images in total. For all images, the Gabor carrier spatial frequency was 5 cycles per degree, and contrast was approximately 95%. The blocks of images were presented in increasing order of orientation jitter. For each trial, the circular contour was placed horizontally with its narrower part pointing either to the right or to the left. The number of the images pointing to the right or to the left was equal within each block of 40 images, although the order of presentation of left and right stimuli was randomized within the different blocks. The contour was always positioned centrally around a fixation point that appeared in the center of a solid gray display during the one-second interstimulus interval (ISI). The duration of the stimulus presentation was 2 seconds. The subjects had to make a decision within that time frame. We used a two-alternative forced-choice (2AFC) method. The subject s task was to indicate, by a two-button box, which side of the screen the narrower part of the shape was pointing to. Prior to the first jitter condition and then following each condition a period requiring central fixation was presented for 30 seconds. The subjects were tested binocularly inside the scanner. Screen size was 8.42 cm high by cm wide. The screen was approximately 16 cm from the bridge of the nose. Stimuli (contour and noise) subtended 14 visual angle horizontally and 11.3 vertically. The portion of the stimulus that contained the

6 180 Silverstein et al. contour subtended 5.9 horizontally and 4.68 vertically at its center (with a gradually decreasing visual angle approaching the point). The stimuli were displayed in white/gray/black (see Fig. 1) (56 cd/m 2 )onagray background (22 cd/m 2 ). The program to administer the task and record data were implemented in Flash and synchronized with image acquisition. Because stimuli were blocked by condition, and to keep the session as brief as possible, a block design, rather than an event-related design, was used. Although the latter may be more sensitive to the nature of single-trial data, its typical requirement for longer intertrial intervals also raises the risk of confounding effects of reductions in sustained attention, which is an issue in schizophrenia. Also, as this was the initial fmri study of contour integration in schizophrenia, we wanted to use the same version of the task,withthesametimingparametersthathadbeenusedinourearlierbehavioral study demonstrating sensitivity of this task [37] Brain imaging All images were acquired using a 3.0-Tesla whole body scanner (Signa VHi, General Electric Medical Systems, Waukesha, WI) performing serial gradient echo, echoplanar imaging (epirt, plane = axial, TR = 3000 ms, TE = 30.7ms, flip angle = 90, matrix = 64 64, FOV = 20 cm 2, voxel size = mm, slice thickness = 4 mm, gap = 1 mm, NEX = 1, bandwidth = 62 khz) for all image acquisition. The duration of the acquisition was 15.5 minutes during which 28 slices were acquired per volume with a total of 310 brain volumes. The functional paradigm was then followed by acquisition of a high resolution 3D inversion recovery fast spoiled gradient recalled echo sequence (SPGR, plane = axial, TR = 9 ms, TE = 2.0ms, flip angle = 25,NEX=1,bandwidth=15.6 khz, acquisition matrix = , FOV = cm 2, slice thickness/gap = 1.5/0 mm/mm, slices = 124). The contour integration task was presented in the scanner and coordinated with behavioral measurements by a custom designed MRI synchronization control system. All scan sessions were conducted between 9 AM and 12 noon. Prior to each session of data collection, quality assurance protocols were conducted to examine scanner quality in terms of signal stability, level of ghosting, and signal-to-noise ratio. Quality assurance on the projection equipment and visor used for stimulus display were conducted at the beginning of each week to ensure stability of stimulus contrast across the study. No differences in either quality assurance protocol were noted across the study duration Image preprocessing and analysis fmri analysis was conducted using Statistical Parametric Mapping (SPM2) [71]. Data from each individual subject were initially corrected for head motion and none exceeded 3 mm of in-plane motion. The functional data were then coregistered with the corresponding anatomical images which were then spatially normalized to the Montreal Neurological Institute (MNI) template. The normalized functional

7 Visual Integration in Schizophrenia 181 data were smoothed with a 6 mm Gaussian smoothing kernel which was approximately two times the original voxel dimensions. The preprocessed functional data for each individual were then analyzed with a general linear model using 6 experimental regressors corresponding to each degree of jitter (0,7,11,15,19,23 ). The onset times of these regressors were convolved with the hemodynamic response function (HDF). Random effects analyses directly comparing the two subject groups were then conducted on the activation maps for the 28 subjects so that a group activation map could be extracted to identify significant effects across regions for each appropriate statistical contrast. To correct for multiple comparisons on the image data, a false discovery rate (FDR), q, of 0.05 was applied to all contrasts [21]. The application of an FDR controls the proportion of false positives only among those voxels that exceed the statistical threshold and therefore reduces the likelihood of Type I errors [17]. A cluster threshold requiring a suprathreshold volume of 15 mm 3 was applied after the FDR to further reduce the likelihood of Type 1 errors [17]. Analysis of functional data was accomplished in two phases. First, voxel-wise comparisons were made between all jitter conditions and between the control and patient groups. These comparisons were used to identify the networks that were activated, and to examine any gross changes between conditions and between groups in clusters of activation. However, as one primary hypothesis involves the modulation of signal in the visual cortical regions, we also carried out a directed region of interest analysis on the individual subject data to extract peak signal changes for each subject. As differentiation between primary, secondary, and tertiary visual cortical regions is not clearly defined by anatomical markers, the Wake Forest University Pick Atlas Tool was used [40]. Within these regions of interest, peak signal change for each subject was extracted for those voxels which were significantly activated on the group maps. In other words, a cluster was defined as the voxels which exceed the statistical threshold on the random effects analysis for each group of subjects. This cluster was then used as a mask and applied to the smoothed, normalized data from each subject. Peak values (beta weights) and average signal change (from the normalized, smoothed, time course) were extracted from each voxel identified by the mask from each subject. Visual inspection of the region of interest relative to the SPGR was carried out on each individual subject to ensure that the borders of primary and secondary visual cortices occurred along the calcarine fissure to the cuneus and lingual gyrus [9] and that tertiary visual cortex was defined from the borders of V1/V2 to the middle occipital gyrus [9]. 3. Results 3.1. Demographic data The patient and control groups did not differ significantly in age [schizophrenia = (8.92), control = (6.47), t(25) = 0.85, p > 0.40], education level [schizophrenia = (3.30), control = (3.46), t(25) = 0.89, p>0.38], maternal

8 182 Silverstein et al. education [schizophrenia = (2.90), control = (2.00), t(25) = 0.26, p>0.79], or paternal education [schizophrenia = (2.60), control = (2.45), t(25) = 1.36, p>0.18]. There was a trend towards a group difference on gender composition [χ 2 (1) = 3.59, p =0.06] Clinical variables Thirteen of the 14 patient subjects were taking antipsychotic medication. For these patients, the mean daily dose (in CPZ equivalents) was mg (SD = 193, minimum = 100, maximum = 700). Mean symptom ratings on the five PANSS factors were as follows: Positive = 2.92 (mild), Negative = 2.63 (minimal to mild), Cognitive = 2.08 (minimal), Excitement = 2.04 (minimal), and Depression = 2.65 (minimal to mild) Performance Overall, both groups of subjects showed reduced accuracy [controls: F (5, 65) = 69.14, p < 0.001; patients: F (5, 60) = 39.56, p < ] and increased latency [controls: F (5, 65) = 43.95, p<0.001; patients: F (5, 60) = 3.59, p =0.001] as the degree of jitter increased. However, as can be seen in Fig. 2A, there was no interaction between degree of jitter and group membership in accuracy of response [F (5, 125) = 1.083, p>0.37] although controls performed significantly better than patients overall [control mean = 73.4%, SEM = 1.9; patient mean = 67.0%, SEM = 2.0; F (1, 25) = 5.378, p<0.05]. Although the patients did not differ from controls on the easiest or most difficult conditions (where ceiling and floor effects, respectively, are expected) [0 : t(26) = 0.92, p>0.37; 7 : t(26) = 1.32, p>0.20; 23 : t(26) = 0.16, p>0.88], they demonstrated a consistent level of relatively reduced accuracy in the more discriminating, intermediate jitter conditions [11 : t(26) = 2.53, p<0.05; 15 : t(26) = 2.22, p<0.05; 19 : t(26) = 2.16, p<0.05]. As is seen in Fig. 2B, there was a group condition interaction for response latencies [F (5, 125) = 5.19, p<0.001]. Both groups showed an overall increase in latency as degree of jitter increased, with a differential increase for controls [F (5, 65) = 43.95, p<0.001] relative to patients [F (5, 60) = 3.59, p =0.01]. As jitter increased and accuracy was reduced, the controls showed a systematic increase in reaction time (RT). Although the patients also showed reduced accuracy with increased jitter, they did not show a corresponding systematic increase in latency. There was a trend for controls to show increased latency relative to patients on the 3 largest degree of jitter conditions (15,19,23 ), however, the only significant difference was observed in the 19 jitter condition [t(26) = 1.71, p =0.05]. Effects of subject gender were examined on task performance (accuracy and response latency) for each group of subjects independently as well as across both groups (2-way ANOVA with gender and subject group as between-subjects factors). None of the results were statistically significant. For example, for the sample as

9 Visual Integration in Schizophrenia 183 Fig. 2. Accuracy (top, A) and reaction time (bottom panel, B) are presented for patients (white circles) and controls (black circles) for each degree of jitter (x-axis). Threshold refers to the point midway between perfect and chance-level performance, and is the point of most reliable discrimination between individuals and groups. a whole, the between-subjects analyses involving gender produced the following results: 0 condition: F (1, 26) = 0.009, p =0.93; 7 condition: F (1, 26) = 0.122, p = 0.73; 11 condition: F (1, 26) = 0.18, p =0.68; 15 condition: F (1, 26) = 0.66, p =0.42; 19 condition: F (1, 26) = 0.001, p =0.99; 23 condition: F (1, 26) = 0.24, p = Although we are underpowered to detect a difference on behavioral performance if one does exist, the high p-values suggest that the lack of significant findings may represent a true nondifference between males and females on this specific task Baseline networks Activation maps for data collected during the 0 condition (i.e., 0 compared to central fixation) are presented in Fig. 3 (detailed coordinates of peak activation from each region are presented in Table 1). Relative to periods of central fixation, both controls (Fig. 3A) and patients (Fig. 3B) showed a widespread network of activation during the forced choice task that included the left and right frontal eye fields (middle frontal gyri), left and right superior and inferior parietal lobules, superior, middle,

10 184 Silverstein et al. Fig. 3. Activation maps for the 0 jitter condition relative to central fixation for controls (A), patients (B). Regions that were differentially greater in controls relative to patients (C) and patients greater than controls (D) are also presented. Significant voxels are presented corrected for multiple comparisons with a p-valuefdr < and inferior temporal gyri, visual cortices (primary, secondary, tertiary) and regions within the cerebellum. The random effects analyses comparing patients to controls are presented in Figs. 3C and 3D. Controls showed differentially greater activation than patients bilaterially in regions identified as dorsolateral prefrontal cortex, right ventrolateral prefrontal cortex, left and right frontal eye fields, left and right superior parietal lobules, V3, and in right V1 (Fig. 3C). Additionally, in this contrast a cluster of activation was also identified in the anterior cingulate. Examination of the time course in the anterior cingulate confirmed that this activation was not the result of increased activity in the controls but instead of decreased signal in the patients. In comparison, patients showed increased activity in right superior temporal gyrus, head of the left caudate nucleus, and left middle temporal gyrus (Fig. 3D). As a further test of potential gender effects, we examined the effects of gender on activation maps within each jitter condition relative to fixation. Even when we removed the correction for multiple comparisons and set the p-value to 0.05, there were no significant clusters of activation between the gender groups. We also re-ran this analysis without a cluster threshold and without a correction for multiple comparisons. At this threshold, significant voxels were identified within V2 on the right,

11 Visual Integration in Schizophrenia 185 Table 1. Magnitude of peak activity (maximum z-statistic) and coordinates in MNI space for each significant cluster of activation. Corresponding Brodmann areas (BA) are also included. Peak values are presented for baseline networks for controls (A), patients (B) and for contrasts where controls showed greater activation than patients (C) and conversely, regions with greater activity in patients than in controls (D). BA areas relevant to visual processing are 17 (V1), 18 (V2/V3), 19 (V4) and 20 (IT). Denotes an inability to differentiate left from right. These numbers are identical because of this difficulty. Denotes no significant clusters in the identified region. (A) Controls 0 (B) Patients 0 BA MNI Coordinates k Max. MNI Coordinates k Max. x y z Z x y z Z Dorsolateral Prefrontal Cortex Right Left Orbitofrontal Cortex Right 11/ Left 11/ Supplementary Motor Area (SMA) Frontal Eye Fields (FEF) Right Left Superior Parietal Lobule Right Left Inferior Parietal Lobule Right Left Thalamus Right Left Superior Temporal Gyrus Right Left Middle Temporal Gyrus Right Left Inferior Temporal Gyrus Right Left

12 186 Silverstein et al. Table 1. (Continued) (A) Controls 0 (B) Patients 0 BA MNI Coordinates k Max. MNI Coordinates k Max. x y z Z x y z Z Visual Cortex Right 17/18/ Left 17/18/ Cerebellum Right Left (C) Controls > Patients 0 (D) Patients > Controls 0 Dorsolateral Prefrontal Cortex Right Left Frontal Eye Fields (FEF) Right Left Superior Parietal Lobule Right Left Caudate (head) Right Superior Temporal Gyrus Right Middle Temporal Gyrus Left Visual Cortex Right 17/18/

13 Visual Integration in Schizophrenia 187 and within the anterior cingulate for controls, with women showing greater activation than men. When examined in the other direction (men > women) significant voxels were identified within the horn of the lateral ventricles. Overall, however, it appears that there are no meaningful gender differences on this task, and none of these reach statistical significance when the data are corrected for multiple comparisons Networks subserving object recognition To examine the networks implicated in recognition of coherent objects and the decision making which follows, we contrasted activation during the 0 jitter condition relative to the (least organized) 23 jitter condition in controls and in patients. Because performance was near 90% in the 0 condition and at chance in the 23 condition for both controls and patients, contrasts between these conditions represent recognition of a coherent object relative to an unorganized stimulus field. Resulting activation maps are presented in Fig. 4, and detailed coordinates of peak activation from each region are presented in Table 2. In controls, there was greater activity in the 0 condition relative to the 23 condition involving regions in prefrontal cortex Fig. 4. Activation maps demonstrating activity during 0 jitter compared to a condition in which contours were not visible (i.e., 23 jitter), in controls (A) and in patients (C) and activation maps depicting greater activation in the 23 jitter condition relative to 0 in controls (B) and in patients (D). Replication of these findings with contrasts between the 0 and 19 conditions is presented in the bottom panel. Significant voxels are presented corrected for multiple comparisons with a p-value FDR < 0.05.

14 188 Silverstein et al. Table 2. Magnitude of peak activity (maximum z-statistic), coordinates in MNI space, and corresponding Brodmann areas (BA) for each significant cluster of activation for contrasts between extreme jitter conditions for controls (A: 0 > 23 and B: 23 > 0 ) and patients (C: 0 > 23 and D: 23 > 0 ). Denotes no significant clusters in the identified region. (A) Controls 0 > 23 (B) Patients 0 > 23 BA MNI Coordinates k Max. MNI Coordinates k Max. x y z Z x y z Z Prefrontal Cortex (PFC) Right Left Anterior Cingulate Frontal Eye Fields (FEF) Right Left Caudate Right Left Thalamus Right Left Superior Temporal Gyrus Right Left Middle Temporal Gyrus Right Left Inferior Temporal Gyrus Right Left Visual Cortex Right 17/18/ Left 17/18/ Fusiform Gyrus Right Left Parahippocampal Gyrus Right Left

15 Visual Integration in Schizophrenia 189 Table 2. (Continued) (C) Controls 23 > 0 (D) Patients 23 > 0 BA MNI Coordinates k Max. MNI Coordinates k Max. x y z Z x y z Z Orbitofrontal Cortex Right 11/ Left 11/ Prefrontal Cortex Right Left Middle Temporal Gyrus Right Left Visual Cortex Right 17/18/ Left 17/18/

16 190 Silverstein et al. (bilaterally), anterior cingulate, frontal eye fields, superior, middle, and inferior temporal gyri, posterior cingulate, V2/V3 and fusiform gyri. In contrast, only regions in visual cortex were more active in the 23 condition relative to the 0 condition in controls. In patients, two regions identified as left and right parahippocampal gyri and another region identified as right fusiform gyrus were differentially greater in the 0 condition relative to 23. When the 0 condition was subtracted from the 23 condition in patients, a network which included left orbitofrontal cortex, left and right caudate (head), left and right medial prefrontal cortex, left middle temporal gyrus, and bilateral primary visual cortex was identified. As a check on the reliability of the results from the 0 vs. 23 condition contrast, we also compared the groups on the 0 vs. 19 contrast, and this revealed a similar pattern of results (see Fig. 4). The controls showed greater activity in the 0 condition in prefrontal regions, the anterior cingulate, superior and inferior parietal lobules, and in the superior, middle, and inferior temporal regions. In contrast, patients showed only small clusters of significant activity in the thalamus, superior temporal gyri, and parahippocamal gyri in the 0 condition when contrasted with the 19 condition. For controls, there was increased activity during the 19 condition (compared to the 0 condition) primarily in the head of the caudate and visual regions. Similar to contrasts between the 23 and 0 conditions, patients showed increased activity in the 19 (relative to 0 ) condition across a widespread of network including prefrontal, inferior frontal, superior, middle, and inferior temporal regions, visual cortices, and regions in the fusiform gyri Signal modulation in visual cortices One primary hypothesis in the current investigation is that the signal in visual cortex regions that are maximally sensitive to basic shape features may differ in patients relative to controls. To examine this directly we extracted peak intensities from regions associated with visual processing and object integration [V1, V2/V3, V4, inferotemporal cortex (IT)] in each condition (see Fig. 5). Comparisons between peak activation within each hemisphere and degree of activation Fig. 5. Regions of interest for extraction of the peak signal change within V1 (green), V2/V3 (red), V4 (blue), and inferotemporal cortex (IT) (yellow). Peak signal intensity for controls (black circles) and patients (white circles) for areas involved in visual processing.

17 Visual Integration in Schizophrenia 191 (cluster size) were conducted. None of these comparisons reached statistical significance (p>0.05) although the comparison between hemispheres for peak activation in V2/V3 trended to show an increase in peak activity with right greater than left for controls [F (1, 26) = 2.96, p =0.071]. As such, all further comparisons collapsed across cerebral hemispheres. In primary visual cortex (V1) there was a main effect of condition [F (5, 130) = 5.80, p<0.001]. There was no main effect of subject group [control, patient; F (1, 26) = 0.33, p > 0.57] or interaction between group and degree of jitter [F (5, 130) = 0.38, p > 0.86]. In V2/V3, signal change also increased as degree of jitter increased [F (5, 130) = 6.98, p<0.001]. However, there was a main effect of subject group such that controls showed higher peak signal changes than patients did [F (1, 26) = 5.68, p =0.025]. The degree of increased signal change in controls relative to patients was consistent across all conditions of jitter, as revealed by a lack of group condition interaction [F (5, 130) = 0.38, p > 0.86]. Similar effects were found for V4 with main effects of subject group [F (1, 26) = 4.89, p<0.05] and condition [F (5, 130) = 5.09, p<0.001] but no group condition interaction [F (5, 130) = 1.08, p>0.72]. There were no significant effects of peak signal change for subject group or condition for IT within the inferior temporal lobe Networks matched on accuracy To ensure that the above results were not confounded by correlates of poor performance in the patient group, a final set of analysis compared the groups across three contrasts where, on each one, they were matched on performance (see Fig. 2): (1) patients at 7, controls at 11 ; (2) patients at 11, controls at 15 ;and(3)patients at 15, controls at 19 (see Fig. 6). As can be seen, 6A C controls showed greater activation (even when matched on behavior) in the head of the caudate (Figs. 6B, 6C), both inferior and superior parietal lobules (Figs. 6A C), left and right putamen (Figs. 6A, B), right dorsolateral prefrontal cortex (Figs. 6A, C), and regions in V2/V3. In contrast, patients showed Fig. 6. Contrasts between controls and patients when matched for accuracy (A: 11 in controls and 7 in patients; B: 15 in controls and 11 in patients; C: 19 in controls and 15 in patients). Greater activity for controls (top) and greater activity for patients (bottom) is presented. Significant voxels are presented corrected for multiple comparisons with a p-value FDR < 0.05.

18 192 Silverstein et al. greater activation in the anterior cingulate, left dorsolateral prefrontal cortex (but more medially than the regions of greater activity in controls), fusiform gyri, superior temporal gyri, middle temporal gyri, and V5. Peak activation within each region and corresponding coordinates are presented in Table Relationships between behavioral data, medication, and symptoms For correlational analyses, indices of visual integration were constructed from the behavioral data by examining the differences between performance at the 0 jitter condition and performance in other conditions. Relationships between these performance-based contrast values and medication dose (in CPZ equivalents) were examined by correlating these values with CPZ equivalent for each subject with schizophrenia (reliable medication data could not be obtained for two patients). The r values for these correlations were: 0.03 (7 ), 0.08 (11 ), 0.15 (15 ), 0.11 (19 ), and 0.20 (23 ), consistently indicating an independence of visual integration and medication level. Correlations between symptom levels and these performance indices were not significant. 4. Discussion There were two major findings from the fmri data. First, while in all cases except one controls and patients were equivalent in their signal change in area V1, signal strength was consistently relatively reduced in patients in higher visual cortex areas (e.g., V2, V3, V4). Because V1 is most sensitive to individual features (as opposed to emergent holistic features), whereas areas V2 V4 are maximally sensitive to global stimulus organization, these data suggest that the performance deficits reflect a true weakening of visual integration processes, and are not secondary to impaired sensory registration or feature integration. This hypothesis is consistent with fmri data from healthy subjects, and from monkey studies, demonstrating that activation in V1 is generally not increased during perception of contours, and that even when V1 activity is found [1], signals are typically strongest in higher tier areas such as V3 and V4, and in more anterior visual areas [42, 45]. These data are also consistent with ERP findings of reduced N150 during a global local task in schizophrenia [26], data localizing the N150 source to visual areas V3/V3a [10, 26, 70], and findings of reduced occipital lobe gamma oscillations during a visual processing task, even when behavioral performance is equivalent to that of controls [63]. It is noteworthy that controls demonstrated more activation in areas V2 and V3 even when matched to patients on accuracy level, and that patients demonstrated reduced activation even in the 0 and 23 jitter conditions. The 0 condition result suggests that even when the contour stimulus was maximally structured, allowing for highly accurate contour perception, patients nevertheless were characterized by a weaker neural signal when coding the stimulus. As stimulus structure was weakened by introducing orientation

19 Visual Integration in Schizophrenia 193 Table 3. Magnitude of peak activity (maximum z-statistic), coordinates in MNI space, and corresponding Brodmann areas (BA) for each significant cluster of activation for contrasts between controls and patients matched on behavioral performance. Denotes an inability to differentiate left from right. These numbers are identical because of this difficulty. Denotes no significant clusters in the identified region. Controls 11 > Patients 7 Patients 7 > Controls 11 BA MNI Coordinates k Max. MNI Coordinates k Max. x y z Z x y z Z Dorsolateral Prefrontal Cortex Right Left Prefrontal Cortex Right Left Anterior Cingulate Right Left Superior Parietal Lobule Right Left Inferior Parietal Lobule Right Left Putamen Left Fusiform Gyrus Right Left Middle Temporal Gyrus Right Left Visual Cortex (V1, V2, V3) Right 17/18/ Left 17/18/

20 194 Silverstein et al. Table 3. (Continued) Controls 15 > Patients 11 Patients 11 > Controls 15 BA MNI Coordinates k Max. MNI Coordinates k Max. x y z Z x y z Z Dorsolateral Prefrontal Cortex Right Left Prefrontal Cortex Right Left Caudate Right Left Superior Parietal Lobule Right Left Inferior Parietal Lobule Right Left Putamen Left Fusiform Gyrus Right Left Middle Temporal Gyrus Right Left

21 Visual Integration in Schizophrenia 195 Table 3. (Continued) Controls 19 > Patients 15 Patients 15 > Controls 19 BA MNI Coordinates k Max. MNI Coordinates k Max. x y z Z x y z Z Dorsolateral Prefrontal Cortex Right Ventrolateral Prefrontal Cortex Left Caudate Right Left Superior Parietal Lobule Right Left Inferior Parietal Lobule Right Left Posterior Cingulate Right Left Middle Temporal Gyrus Right Left Visual Cortex (V1, V2, V3) Right 17/18/

22 196 Silverstein et al. uncertainty, however, continued weakening of the neural coding occurred, and was associated with impaired behavioral performance. Second, the overall network of brain activity during task performance was more widely distributed in controls, and involved greater recruitment of frontal and parietal areas in the control of visual attention. It is likely that activation in these more anterior areas reflected both feedforward processing of the contours from posterior areas, and the generation of feedback from frontal and parietal to visual areas to strengthen the salience of the overall contour representation relative to background noise [41]. In the control group, reaction time increased as a function of contour element orientation jitter (i.e., as the stimuli became more difficult to integrate). This was associated with an increase in visual cortex activation, presumably reflecting prolonged visual analysis, but not activation in other cortical regions. In contrast, the schizophrenia group demonstrated an attenuated pattern of increasing RTs as a function of orientation jitter. Moreover, they demonstrated increased activation primarily in anterior regions, with less visual cortex activation increase in the condition where contours were least visible. This pattern of results suggests that the normal linkage between the requirement for prolonged visual analysis, increased visual cortex activity, and delayed response preparation is disrupted in schizophrenia. It appears as if, in schizophrenia, enhanced demands for visual analysis led to a more diffuse pattern of brain activity, one that does not support the prolonged maintenance of visual processing required for successful performance, or allow for a delay in behavioral responding until adequate visual analysis is completed. Specific evidence in support of this hypothesis is that the largest differences between patients and controls in RT were in the three most difficult conditions. Visual integration performance was not related to either medication dose or to symptoms. Although the issue of relating performance to medication was not a focus of this study (given consistent evidence of a nonrelationship in past studies [69]), it should be noted that this study was underpowered to detect these relationships with correlational analyses. The same issue of power holds for analysis of symptomperformance relationships. However, as patients in this study were all clinically stable and were experiencing relatively low levels of symptoms (e.g., mean PANSS symptom scores were within the minimal-mild range for all 5 factors), the study did not include a symptomatically heterogeneous enough group of patients to detect these relationships. In general, cognition-symptom relationships are observed more reliably in more symptomatically heterogeneous samples [60], and specifically, visual integration deficits have been linked to clinical disorganization when symptomatic samples are studied [25, 28, 47, 51, 59, 65 67, 69]. The finding of behavioral and neurophysiological evidence for visual integration in a relatively asymptomatic sample, however, suggests that the contour integration task is a sensitive and psychophysically rigorous measure of visual integration, and also that integration dysfunction may be present even in relatively remitted patients although the impairment can

23 Visual Integration in Schizophrenia 197 be expected to be reduced compared to more acutely or chronically symptomatic patients. A potential limitation of this study is that the stimulus presentation used a block design in which the difficulty level increased with each successive block. This potentially confounds the observed results with a nonspecific order effect. The design was chosen based on prior data that schizophrenia patients and controls demonstrated performance differences using this version of the task, and that this version is both a reliable and valid measure of visual integration [37]. Still, a direct comparison between random versus blocked presentation of stimuli could be informative. In a prior behavioral study by our group using a contour detection task (as opposed to the 2AFC version used in this study), both healthy controls and schizophrenia patients demonstrated superior performance when stimuli were presented in increasing order of difficulty, compared to when they were presented in a random order [57]. This suggests that while activation differences might exist between the two conditions, group differences might be similar across the two conditions. However, it is also possible that while both groups might perform better and have increased activation in the blocked condition, the group difference might be of a different magnitude in the random condition, which would be informative regarding top down effects on contour processing. Another limitation of the study involves a potential confound between contour integration and attentional processes involved in visual search in the more difficult conditions. This is especially relevant to the contrasts comparing the 23 and 19 jitter conditions to the 0 condition, where inability to perceive a contour may have led to attentional, or other unknown processes being used by subjects in what we assumed to be a control condition. This can be addressed to some extent in future studies by using passive viewing conditions where no response is required. On the other hand, we consider it unlikely that the greater activation in post-v1 areas in the controls, in the 0 vs. 19 and 0 vs. 23 contrasts, reflected such a confound because: (1) controls demonstrated these effects even when matched to patients on high levels of accuracy, in contrasts in which data from more difficult conditions were not included; (2) in between-group contrasts comparing the groups on peak activation in each jitter condition alone, controls demonstrated greater activation in V2/V3 and V4; and (3) our results from controls are strikingly similar to the nonclinical human and monkey data observed by Altmann et al. [1], using tasks which did not require a subject response, and from analyses that did not contrast either easier or more difficult conditions. Overall, this suggests that the group differences in the between-condition contrasts comparing contour-visible and contour-not-visible conditions reflect true schizophrenia-related visual cortex processing abnormalities, as opposed to confounds related to either their poor motivation or increased effort exerted in the two most difficult conditions. Although there was a trend towards a group difference in male:female ratio, it is unlikely that our findings are an artifact of this sex difference. First, males