|Home | About | Journals | Submit | Contact Us | Français|
Background: Higher levels of facial processing, such as recognition of the individuality and emotional expression of faces, are abnormal in schizophrenia. It is unknown, however, whether the visual detection of a face as face is impaired as well. Methods: We examined the performance of schizophrenia patients (n=29) and normal controls (n=28) in locating a line-drawn face on the left or the right side of a larger line drawing. To prevent the normal formation of general facial impressions, stimulus presentations were brief (13–104 ms). The face stimuli were either displayed upright or inverted in order to study the face inversion effect, ie, the specific effect of stimulus inversion on face processing. Results: Schizophrenia patients showed a significantly reduced face inversion effect, resulting primarily from significantly lower accuracy in detecting upright faces than normal controls. In tree detection, a comparison task that was also administered, the stimulus inversion effect was similarly small in both groups. Conclusion: Given the primitive nature and brief duration of the stimuli, and the simplicity of the task, these results indicate that at the initial visual detection stage, facial processing is inefficient in schizophrenia. By isolating face detection from other aspects of face recognition, this study identifies a face-specific visual deficit in schizophrenia, which may ultimately contribute to impaired face-related cognitive and emotional processing and social interaction.
While recognizing a face appears to be a natural and usually effortless process, deficiency in this behavioral capacity can have profound consequences in everyday life. Because of the great importance of recognizing faces, a distinct and highly efficient brain system has been developed for processing and utilizing facial information.1–4 In schizophrenia, impairments in recognizing different aspects of facial information have been reported,5–8 and such impairments may play a crucial role in poor social interaction in patients. Understanding the nature of facial processing impairment can facilitate the development of effective intervention to improve the quality of patients' social life.
Facial processing comprises multiple functional components, namely, visual processing, analysis of facial identity, and recognition of facial expressions.9 The visual processing component presumably involves detection of physical features of faces, not the features used in other component processes. The visual detection of a face may occur prior to or without the acquisition of other features embedded in faces.10,11 Previous studies have shown that schizophrenia patients were deficient in judging facial identity12 and facial emotion,5,6,13 suggesting that compromised facial processing is not limited to a single component. Yet, the functional integrity of visual processing of facial information, which provides sensory signals for analyzing facial identity and emotional expression, remains an open question. Parsing impaired behavioral performance, such as face recognition, into elementary components, such as visual detection and cognitive and emotional analysis, will help in probing into the pathophysiological processes underlying schizophrenia.14
Visual processing in schizophrenia is altered in several domains including responses in the early visual pathway (contrast detection and backward masking),15,16 trajectory detection,17 and motion discrimination.18–20 Presumably, the abnormalities in the visual system could compromise sensory processing of facial information, which might provide deviant signals to subsequent cognitive and emotional processes involved in face recognition. It is known that the processes involved in face recognition are qualitatively different from those involved in recognition of other visual objects.21–23 Examining the visual component involved in facial processing in schizophrenia is important not only for identifying the mechanisms for detecting a face as a special class of visual objects but also for understanding the sensory, cognitive, and emotional processes associated with the mental disorder. Further, the knowledge about the visual mechanisms of facial processing may provide clues for developing targeted intervention strategies that help patients to improve their social functioning.
The focus of the present study is to examine the process involved in analyzing the visual information fundamental for perceiving a face as such in schizophrenia. To differentiate from other components of face recognition, such as identity or expression information analysis, we use the term face detection to indicate the awareness of the mere presence of a face.
In order to examine the visual process for face detection, it is crucial to limit those clues used in analysis of facial identity and emotion, while probing into face-specific processes. We employed several experimental strategies.
Twenty-nine schizophrenia patients and 28 normal controls participated in the study. General inclusion criteria for both groups were (1) no history of any neurological disorders (such as seizure or stroke) or head injuries, (2) IQ more than 70, and (3) no substance abuse in the last 6 months. The patients, who had been treated in a private psychiatric hospital, met Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria for schizophrenia or schizoaffective disorder, based on a standardized interview and a review of all available medical records. The interviews were conducted by experienced clinicians who were blind to the purposes of this study. Ten patients received a diagnosis of schizophrenia and 19 patients received a diagnosis of schizoaffective disorder according to the DSM-IV.26 All patients were medicated on antipsychotic drugs (mean chlorpromazine =699 mg [627 mg]).27 The Positive and Negative Syndrome Scale (PANSS)28 was administered in the patient group (positive subscale: 16.3 ± 8.2, negative subscale: 12.5 ± 5.6, general subscale: 29.6 ± 10.6). The normal controls were recruited through advertisements in the local community. None of the normal controls had a history of Axis I psychiatric disorders as defined by the DSM-IV.
The patient and control groups were similar in term of age and sex. The Wechsler Adult Intelligence Scale (verbal)29 was administered for all subjects. Table 1 provides demographic information of the subjects.
The study protocol was approved by the Institutional Research Board (IRB) of McLean Hospital. Written informed consent in accordance of the IRB guidelines was obtained from all subjects prior to participation.
The target for detection was a line-drawn face (figure 1) or a line-drawn tree, embedded in a scrambled line drawing. The target was positioned either on the left or the right side of the drawings. The drawings were displayed on a computer screen for 13, 26, 52, or 104 ms. The target was oriented either upright or inverted.
The task was to judge whether the target, a face or a tree, was located on the left or the right side of each drawing. The advantages of this 1-interval and 2-alternative forced choice method include (1) no memory requirement and (2) minimal subjective criterion bias which is usually associated with other methods, such as the Yes/No. For each stimulus presentation, subjects gave their judgments by pressing one of 2 designated keys on a keyboard. To ensure no ambiguity about what stimulus was expected to see, the stimulus presentations were blocked into a session according to 3 conditions: (1) duration (13, 26, 52, or 104 ms), (2) orientation (upright or inverted), (3) and type (face or tree) of target. The presentation of each stimulus was repeated 42 times in the face detection task and 21 times in the tree detection task. Presentation order of the sessions was randomized across subjects. Stimulus presentation and response recording were programmed within the VisionShell and controlled by a Mac G3 computer system (The Apple Computer Inc).
Accuracy and reaction time were the raw data collected during testing. Ratios of accuracies in detecting upright and inverted targets, the index of stimulus inversion effect, were analyzed to evaluate the performance on face detection vs detection of nonface visual objects. All the analyses above included IQ and education, the only 2 demographic variables that differed significantly between the groups, as covariates.
Figure 2 shows the mean accuracies in detecting upright and inverted faces across stimulus duration (a) and the ratios between the two accuracies (b) A 2-way analysis of variance (ANOVA) examining the effects of subject group (patient vs control) and stimulus duration (13, 26, 52, and 104 ms) revealed that the accuracy ratios were significantly lower in schizophrenia patients than in normal controls (F1,3=9.76, P=.002) and that they differed significantly across stimulus durations (F1,3=4.02, P=.009). The interaction between group and stimulus duration was not significant. This result indicates that the stimulus inversion effect in face detection is reduced in schizophrenia, independent of stimulus duration.
Figure 3 shows the mean accuracies in detecting upright and inverted trees across stimulus duration (a) and the ratios between the two accuracies (b) A 3-way ANOVA with task (face detection vs tree detection), group, and stimulus duration revealed that the ratios were significantly higher (1) in face than in tree detection (F1,1,3=354.4, P < .001) and (2) in normal controls than in patients (F1,1,3 = 8.15, P = .005). The ratios were significantly different across durations (F1,1,3 = 2.69, P = .046). There were significant interactions between group and task (F1,1 = 6.27, P = .013) and between stimulus duration and task (F1,3 = 5.04, P = .002). Other interactions were not significant. The result of the group × task interaction indicates that the reduced stimulus inversion effect in schizophrenia, as shown in “Stimulus Inversion Effect in Face Detection,” depends on the type of task, ie, occurring primarily in face but not in tree detection. Post hoc tests showed that accuracy ratios in tree detection did not differ significantly between two groups (t = 0.786, P = .447).
A 3-way ANOVA with stimulus orientation (upright vs inverted), group, and stimulus duration revealed that (1) accuracy was not significantly different between groups (F1,1,3=0.224, P=.636), 2) but was significantly different between stimulus orientations (F1,1,3=682.57, P < .001), and (3) across durations (F1,1,3 = 5.88, P = .001). The interaction was significant between group and stimulus orientation (F1,1 = 4.57, P = .033). Other interactions were not significant. The group × stimulus orientation interaction indicates that any group difference in face detection depends on stimulus orientation, consistent with the ratio analysis in “Stimulus Inversion Effect in Face Detection” and “Stimulus Inversion Effect in Face and Tree Detection.”
A 3-way ANOVA with stimulus orientation, group, and stimulus duration revealed that (1) accuracy was significantly different between groups (F1,1,3=27.1, P < .001), 2) between stimulus orientations (F1,1,3 = 29.9, P < .001) and (3) across stimulus durations (F1,1,3 = 9.42, P < .001). No interactions were significant. This result indicates that tree detection is more accurate in normal controls than in patients, but this group difference is independent of stimulus orientation.
In patients, accuracy in detection of upright face, averaged across stimulus duration, was moderately correlated with the general subscale of the PANSS (r=−0.38, P < .05) but not with positive (r = −0.24) or negative subscales (r = −0.17). The accuracy was not significantly correlated with the level of antipsychotic medication received by patients (measured by chlorpromazine equivalent) (r=−0.30). The face detection accuracy was not significantly correlated with verbal IQ (r=0.21) or years of education (r=0.20). The accuracies of patients with schizophrenia (77% [SD: 13%]) and schizoaffective disorder (76% [SD: 15%]) were not significantly different [t (27)=0.12, P=.91].
In the control group, the averaged accuracy in detecting upright faces was moderately correlated with years of education (r=0.40, P < .05), but not with verbal IQ (r = 0.30).
For face detection, a two-way ANOVA with group and stimulus duration revealed that the reaction time ratios (upright over inverted) were not significantly different between patients and the normal controls (F1,3=0.482, P=.49) or across stimulus durations (F1,3=0.322, P=.809). The interaction between group and stimulus duration was not significant (F1,3=0.884, P=.45).
For tree detection, the ratios were significantly higher in patients than in normal controls (F1,3=4.75, P=.031) but were not significantly different across stimulus durations (F1,3=2.08, P=.11). The group × stimulus duration interaction was not significant (F1,3=1.54, P=.21).
Reaction time was not correlated with upright face detection accuracy or with accuracy ratios in either group.
The trial number for face detection was twice as many as for tree detection. The reason for not using as many trials for tree detection as for face detection was to limit the total trial numbers in order to prevent subjects from fatigue. To evaluate the effect of the uneven trials on the results, a subgroup of subjects (5 schizophrenia patients and 5 normal controls) were tested in tree detection with an equal number of trials as in face detection. The additional data yielded a similar result—the significant group difference in detecting upright trees (P < .05) and the insignificant group difference in detecting inverted tree (P > .05) remain after combining the additional set of the data. Because group means and SDs were similar between the conditions with even number of trials and with uneven number of trials, it is unlikely that effect size, and therefore the power to detect a group difference, will change significantly.
This study found that face detection was inefficient in schizophrenia patients, as shown by low accuracy, a reduced stimulus inversion effect, and long reaction times. While also deficient, the performance of schizophrenia patients in tree detection was qualitatively different from that in face detection in that their stimulus inversion effect was minimal, similar to that of normal controls.
Normally, detection of a face is fast and accurate, even when only limited visual information is available. In this study, the normal controls achieved significantly higher accuracy and faster response in detecting upright than in detecting inverted face images for all stimulus durations. The advantage in detecting upright versus inverted faces, consistent with the known inversion effect for face recognition, suggests that a highly efficient system exists for processing faces, even as early as at the visual detection level. In schizophrenia patients, accuracy in detecting upright faces was reduced, and reaction time was prolonged overall. More importantly, the advantage in detecting upright faces over detecting inverted faces, or stimulus inversion effect, was significantly reduced in patients.
The use of the stimulus inversion effect, which is measured by within-subject ratio comparisons, minimized cross-subject performance variance in general visual perception and execution of such behavioral tasks, an important factor when considering “the generalized deficit” associated with schizophrenia patients.30 As a result, the group difference in the ratio variable reflects the face-specific processes evaluated here. This pattern of the results suggests that the efficiency of visual processing of facial information in schizophrenia is compromised.
As an initial stage of facial processing, face detection concerns visual information used to construct faces. While face recognition is categorically different from recognition of other visual objects, it has not been directly demonstrated whether classification of facial images as faces is distinct from classification of other nonface visual objects.9 The results of this study suggest that this is the case. Stimulus inversion effect, as measured by the accuracy ratios, is significantly greater in face detection than in tree detection (“Stimulus Inversion Effect in Face and Tree Detection”). As also can be seen in figures 2b and and3b,3b, the ratios of detecting upright and inverted faces were greater than 1.3, whereas in tree detection, the ratios were hovering between 1.0 and 1.1.
The differences between face and tree inversion effects suggest that at the initial visual detection stage, the mechanisms for processing facial information are already distinct from those for processing information of nonface visual objects.
The stimulus inversion effect is primarily a phenomenon of face perception and is independent of task difficulty level.25 In recognition of many nonface visual objects, lack of stimulus inversion effect is not unusual. Consistent with previous findings, both normal controls and schizophrenia patients in this study showed little inversion effects in detecting trees. Small magnitudes of stimulus inversion effect in tree detection, in contrast with those in face detection, make it difficult to compare stimulus inversion effects across tasks. On the other hand, the differential stimulus inversion effects suggest that the detection of trees is qualitatively different from detection of faces, and thus that any group difference in face accuracy be face specific.
Although the performance on tree detection was inferior in patients, the stimulus inversion effect did not differ between the 2 groups (figure 3b). On the other hand, the stimulus inversion effect for faces was significantly reduced in patients. The reduced stimulus inversion effect in the patient, as shown by the significant difference in accuracy ratios and by the significant group and task interaction in accuracy ratios, suggests an impairment specific to facial information processing in schizophrenia. Lacking or altering face-specific mechanisms would make processing facial information no different from processing visual information of other objects like a tree.
The reduced face stimulus inversion effect in patients does not appear to depend on stimulus duration because there was no significant interaction between group and stimulus duration in the accuracy ratio analysis, suggesting that task difficulty level,25 as manipulated by stimulus duration, is not a major factor for the reduced stimulus inversion effect, at least for the range tested in this study. Also note that between 2 adjacent stimulus durations, the accuracy data in face detection, as well as in tree detection, are highly correlated (ranging from 0.65 to 0.85), indicating a good internal consistency of the visual detection tasks used here.
Like other visual objects (houses, stick figures of moving men) that have been used in comparison with faces, trees have limitations in that the outline of their images are not identical to that of face images. Such a difference in visual image property could potentially complicate the comparison of face and tree detection. Future studies need to take into account this factor by adding visual objects with outlines more similar to that of faces.
Facial information is processed in a special brain system that includes the fusiform face area in the temporal cortex (FFA).2,3 In schizophrenia, abnormal structure of FFA was reported31,32 and was linked to the performance on a face memory task.33 The inefficient face detection in schizophrenia patients, found in this study, is consistent with a dysfunction of the FFA. In particular, while many cortical areas are responsive to faces, the FFA is the primary site that is responsible for the face inversion effect.4 In light of this knowledge, the finding that the stimulus inversion effect was reduced in face (but not in tree) detection in patients points to the involvement of the FFA. Note, however, that a recent study showed that FFA responses to general face recognition are preserved in schizophrenia.34 Taken together, these results suggest that in schizophrenia, either a different, and less efficient, cortical mechanism is used for facial processing or the same FFA mechanism becomes compromised when the high efficiency for facial processing is necessary such as under the conditions of very brief stimulus exposures.
Other visual areas, such as the occipitotemporal sulcus and anterior inferotemporal cortex, are also involved in processing of facial information.35,36 Interestingly, the occipitotemporal sulcus overlaps with the visual areas responsible for motion processing, which has been implicated in schizophrenia.37,38 Consistent with the association of cortical processing of face and visual motion, it has been suggested that facial processing relies on information conveyed in low spatial frequency,39 the sort of visual feature that also serves as primary input for motion processing. Thus, it remains a possibility that the involvement of cortical areas other than the FFA contribute to the inefficient face detection in schizophrenia found in this study.
It has been suggested that poor face recognition in schizophrenia may be related to certain psychotic symptoms, such as delusions7. The result of this study, however, does not lend direct support to this notion because face detection performance in the patients was not correlated with the positive subscales of PANSS that include delusion. This suggests that the inefficient face detection is probably more associated with specific processes involved in facial information processing than with particular clinical abnormalities. On the other hand, a moderate correlation between face detection and general subscales of PANSS suggests that a poor performance in recognizing face does have a negative impact on the general mental status of patients.
Face detection provides an assessment for visual processing of face information, without regard to identity (age, gender, or individuality) or expression (happy or sad). This visual input is considered to precede other components of facial processing.9 The inefficient face detection in schizophrenia, as shown in this study, may thus have consequences for other components of facial processing (recognition of identity and expression). How inefficient face detection in schizophrenia contributes to the recognition of identity and expression of faces and to general social functioning should be used as topics for further studies.
By isolating face detection from cognitive, emotional aspects of face recognition, the present study found that the first stage of facial information processing is inefficient in schizophrenia. The inefficient face detection suggests that unlike normal controls, patients with schizophrenia have difficulty in accessing the normally highly efficient, face-specific brain system. Like many other cognitive and emotional deficits, facial recognition impairment in schizophrenia actually has its specific sensory counterparts.
We thank the late Philip Holzman for early inspiration for this line of research. We thank Ken Nakayama and Anne Grossetete for the help during the design of the face detection task, Cinnamon Bidwell for modifying the images used in the tree detection task, and Sarah Taylor for comments on the earlier version of the article. This study was supported in part by grants from National Institutes of Health and Harvard University.