پردازش ابراز هیجانی و هویت صورت در اسکیزوفرنی

Abstract Previous studies showed that schizophrenic patients have a deficit in facial information processing. The purpose of the present study was to test the abilities of patients with schizophrenia and normal controls in emotion and identity matching when these two dimensions were varied orthogonally. Subjects (20 schizophrenic patients and 20 controls) had to report if two faces had the same emotion or belonged to the same person. When the task concerned one type of information (i.e. emotion or identity), the other one was either constant (same person or same emotion) or changed (different person or different emotion). Schizophrenic patients performed worse than controls for both kinds of facial information. Their deficit was more important when the secondary factor was changed. In particular, they performed at chance level when they had to match one emotion expressed by two distinct persons. Finally, correlation analysis indicated that performance/deficit in identity and emotion matching co-varied and that in such tasks performance is negatively correlated with the severity of negative symptoms in patients. Schizophrenic patients present a generalised deficit for accessing facial information. A facial emotion and an identity-processing deficit are related to negative symptoms. Implications for face-recognition models are discussed.

1. Introduction Patients with schizophrenia show abnormal performance on facial emotion recognition and identity matching (Cutting, 1981 and Salem et al., 1996). These deficits, which affect the ability of patients to interpret others' intentions or desires, might play a role in the social disorders associated with schizophrenia. Abnormal performance on facial emotion recognition tasks have been reported in schizophrenia (Cutting, 1981, Habel et al., 2000 and Salem et al., 1996). The neural correlates of facial emotional expression are less active during emotion perception among patients with schizophrenia (Gur et al., 2002 and Phillips et al., 1999), but performance improves after training in affect recognition (Frommann et al., 2003). The deficit could be more important for fear and sadness (Edwards et al., 2001). Some reports also indicated that this deficit extended to all facial information processing (Kerr and Neale, 1993, Archer et al., 1994, Salem et al., 1996, Franck et al., 2002 and Hooker and Park, 2002), but some patients with schizophrenia seem to have greater difficulty in processing emotion-related information than other kinds of facial information (Gooding et al., 2001, Gooding and Tallent, 2002 and Hooker and Park, 2002). Finally, the impairment is not restricted to facial affect but might also extend in the recognition of affective prosody (Edwards et al., 2001). Studies of individual cases of schizophrenia have not revealed any consistent pattern of deficit: patients typically exhibit quite specific deficits, but not necessarily for the same facial information, e.g., in face recognition, emotion recognition or identity matching (Archer et al., 1994 and Evangeli and Broks, 2000). Archer et al. (1994) interpreted these observations in accordance with Bruce and Young's (1986) model of face recognition. Bruce and Young (1986) proposed that there are separate functional components for the processing of facial emotion, for the processing of unfamiliar faces, and for familiar face recognition. These components are believed to be independent. This assumption is supported by experimental studies in healthy participants (Bruce, 1986 and Campbell et al., 1996), electrophysiological recordings in monkeys (Hasselmo et al., 1989) and functional imaging studies (Phillips et al., 1998 and Sergent et al., 1994). Neuropsychological studies (Parry et al., 1991, Humphreys et al., 1993 and Schweich and Bruyer, 1993) showed that these processes could be selectively damaged (for a model of facial information processes and their neuroanatomical correlates, see Haxby et al., 2000 and Haxby et al., 2002). Nevertheless, some recent studies suggest that facial emotion processing and face recognition may interact (Baudouin et al., 2000a, Baudouin et al., 2000b, Dolan et al., 1996, Schweinberger and Soukup, 1998 and Tiberghien et al., 2003). For example, healthy participants were not able to pay attention to emotion when identity was varied (Schweinberger and Soukup, 1998). The experiment of Baudouin et al. (2002) extends this observation to patients suffering from schizophrenia. They observed that schizophrenic participants could not selectively attend to facial emotion regardless of the identity displayed. The ability/deficit in classifying faces according to emotion was significantly correlated with the ability/deficit in classifying faces according to identity. Similarly, Young et al. (1996) studied a patient with a partial bilateral amygdalectomy who was poor at recognizing facial emotions. She was not impaired on face-recognition and identity-matching tasks, except in the special case where she had to recognize the same person with two distinct facial emotions. Under those circumstances, she tended to perceive two different persons; the deficit in facial emotion recognition led to a deficit in identity processing. Thus, it appears that the variation of one kind of facial information may interfere with the processing of another. Moreover, other studies reported a positive correlation in performance for emotion and identity tasks, for right brain-damaged patients (Weddell, 1989), for controls, and for lobotomized participants (Braun et al., 1994). These observations do not favor the independence hypothesis. By contrast, they indicate that identity and emotion processing are interrelated. To reconcile these two opposing views, one may suggest that a deficit in the processing of one kind of facial information is concomitant with some decrease of the ability to process the other kind, which nevertheless does not reach an “impaired” level. This hypothesis would explain both observations of correlation and dissociation cases. Finally, these observations suggest that the ability to respond selectively either to facial identity or emotion without interference from the other (e.g., facial expression) does not result from implementing independent processes, but rather that these processes interact and may reciprocally interfere. Thus, attention to one kind of facial information would require the intervention of active dissociation processes that allow selective responses to each kind of facial information. These attentional processes would need to be used to disregard information that is not relevant for the task. In previous studies, each type of information was generally tested individually, i.e., participants did not need to rule out the variations of irrelevant facial information to complete the task—for example, the emotion was not varied when the task involved identity judgments (Archer et al., 1994). The purpose of the present study was to test the abilities of patients with schizophrenia and of controls in emotion and identity matching when these two dimensions were varied orthogonally. Exactly the same photographs were used in two delayed matching tasks (identity matching and emotion matching). Pairs of faces, belonging to the same person or not, and expressing the same emotion or not, were presented to subjects during two tasks. The instruction was the only difference between the two tasks (“is the person the same?” or “is the emotion the same?”). Contrary to previous studies, the emotion was varied when the instruction was about identity, and reciprocally. Thus, participants had to disregard the irrelevant information. The independence hypothesis would predict that when the task concerned one kind of information, variation in the other would not interfere with performance. Conversely, the interactive hypothesis, which we favor, would predict that variation of identity/emotion might interfere in the matching of emotion/identity. This interference would be greater for patients with schizophrenia because of their attentional problems. The relation between a deficit in facial information processing and the severity of schizophrenic symptoms, evaluated with the Scale for the Assessment of Negative Symptoms (SANS; Andreasen, 1983) and the Scale for the Assessment of Positive Symptoms (SAPS; Andreasen, 1984), was also investigated.

Results 3.1. Percentage accuracy and latencies for correct responses The factor group was significant for percentage and latencies: patients with schizophrenia performed significantly less accurately (82.2% vs. 91.8%; F1, 38=15.64, P<0.001) and more slowly than controls (1317 vs. 949 ms; F1, 38=14.99, P<0.001). The factor of matching dimension was also significant for both measures: participants performed significantly less accurately and more slowly in the emotion-matching task than in the identity-matching task (81.3% vs. 92.6%: F1, 38=38.23, P<0.0001; 1246 vs. 1020 ms: F1, 38=29.75, P<0.0001). The factor response was significant for both measures: participants performed better and faster when they had to say “different” rather than “same” (89.5% vs. 84.5%: F1, 38=5.55, P<0.05; 1112 vs. 1154 ms: F1, 38=4.78, P<0.05). The factor change was also significant for percentage and latencies: performances were more accurate and faster when the second dimension did not change (92.7% vs. 81.2%: F1, 38=28.64, P<0.0001; 1054 vs. 1212 ms: F1, 38=122.08, P<0.0001) ( Table 3). Table 3. Mean percentages and latencies (ms) for correct emotion and identity matching of schizophrenic patients and controls Task Emotion Identity Response Same Different Same Different Change of the other dimension No Yes No Yes No Yes No Yes Percentage (%): Schizophrenic patients 92.4 48.4 81.4 80.1 90.6 80.5 91.9 92.0 (SD) (13.1) (27.6) (21.1) (18.8) (8.6) (25.7) (10.0) (11.8) Controls 98.1 71.5 93.9 84.8 98.3 96.1 95.3 96.4 (SD) (2.3) (12.3) (8.5) (10.0) (2.3) (3.8) (5.6) (3.8) Latencies (ms): Schizophrenic patients 1138 1643 1426 1426 1121 1242 1312 1225 (SD) (255) (708) (366) (394) (442) (548) (895) (582) Controls 833 1283 1041 1176 747 886 811 816 (SD) (218) (298) (235) (321) (204) (248) (219) (227) Table options The following interactions were significant for both percentage and latencies: matching dimension * change (respectively, F1, 38=43.49, P<0.0001 for percentage; F1, 38=46.36, P<0.0001 for latencies), response * change (respectively, F1, 38=14.83, P<0.0001 for percentage and F1, 38=96.18, P<0.0001 for latencies) and matching dimension * response * change (respectively, F1, 38=9.96, P<0.001 for percentage; F1, 38=17.80, P<0.001 for latencies). This last three-way interaction indicated that the interaction of response * change was significant when the matching dimension was on emotion (F1, 38=31.58, P<0.0001 for percentage; F1, 38=75.54, P<0.0001 for latencies). When the matching dimension was on identity, the interaction response * change was significant for latencies (F1, 38=29.41, P<0.0001) but only tended to be significant for percentage (F1, 38=4.02, P<0.06). When the matching dimension was on emotion, changing identity had a significant detrimental effect when the response was “same” for both accuracy and latencies (59.9% vs. 95.3%: F1, 38=28.89, P<0.0001; 1463 vs. 986 ms: F1, 38=115.76, P<0.0001). When the response was “different”, changing identity had a significant detrimental effect for latencies (1301 vs. 1234 ms: F1, 38=5.10, P<0.05) but not for accuracy (82.4% vs. 87.6%: F1, 38=1.00, NS). Thus, changing identity in emotion matching interfered with accuracy and latencies when the two faces expressed the same emotion (“same” response) and interfered with latencies when the two faces did not have the same emotion (“different” response). When matching dimension was on identity, changing emotion had a significant detrimental effect on latencies when the response was “same” (1064 vs. 934 ms: F1, 38=69.21, P<0.0001) but had no effect when the response was “different” (F1, 38=0.53, NS). Elsewhere, changing identity in emotion matching had a more detrimental effect than changing emotion in identity matching when the response was “same” (difference of 35.4% vs. 6.1% and 477 vs. 130 ms). The interaction group * matching dimension was significant for percentage (F1, 38=6.10, P<0.05). It indicated that both controls and patients with schizophrenia were more accurate in identity matching than in emotion matching (96.5% vs. 87.1% for controls: F1, 38=6.89, P<0.01; 88.8% vs. 75.6% for patients with schizophrenia: F1, 38=37.44, P<0.0001), but the difference between the two tasks was stronger for patients with schizophrenia (13.1% vs. 9.4%). The performances for patients with schizophrenia were nevertheless worse than for controls both for identity (F1, 38=9.60, P<0.01) and emotion (F1, 38=13.80, P<0.001). The interaction between group * change was significant for percentage (F1, 38=6.05, P<0.05), but it was modulated by the interaction of group * response * change (F1, 38=6.06, P<0.05). It indicated that the interaction of response * change was significant for controls (F1, 19=36.57, P<0.0001) and for patients with schizophrenia (F1, 19=10.10, P<0.01). For controls, changing the second dimension had a significant detrimental effect either when the response was “same” (98.2% vs. 83.8%: F1, 19=73.19, P<0.0001) or “different” (94.6% vs. 90.6%: F1, 19=17.43, P<0.001), but this detrimental effect was stronger when the response was “same” (14.4% vs. 4.0%). For patients with schizophrenia, changing the second dimension had a significant detrimental effect when the response was “same” (91.5% vs. 64.4%: F1, 19=13.47, P<0.01) but had no effect when the response was “different” (86.6% vs. 86.1%: F1, 19=0.01). Moreover, when the response was “same”, the detrimental effect of changing the second dimension was stronger for patients with schizophrenia than for controls (27.1% vs. 14.4%). Thus, to summarize these results, performance of all participants was impaired by the change of the second dimension. Nevertheless, performance was more dramatically impaired for patients with schizophrenia than for controls. Table 2 further indicates that for patients with schizophrenia, performance was at chance level when they had to match the same emotions in different faces (correct responses: 48.4% vs. 71.5% for controls). The global interaction (group * matching dimension * response * change) was not significant either for percentage (F1, 38=2.85, NS) or latencies (F1, 38=0.45, NS). 3.2. Correlation analysis We calculated Spearman correlation coefficients between performance for identity matching and emotion matching (see Fig. 2). For accuracy, there was a significant positive correlation between the two tasks both for accuracy (ρ=0.53, P<0.001) and latencies (ρ=0.72, P<0.0001). Thus, the more/less accurate and fast participants on identity matching also tended to be the more/less accurate and fast participants on emotion matching. When we considered patients with schizophrenia alone, the correlation was also significant for both accuracy (ρ=0.53, P<0.05) and latencies (ρ=0.61, P<0.01). Then, when a patient had a deficit in one task, he tended also to have a proportional deficit in the other task. When only controls were considered, the correlation was always significant for latencies (ρ=0.62, P<0.01) but not for accuracy (ρ=0.14, P>0.50). Correlation between percentage of correct responses and latencies in identity ... Fig. 2. Correlation between percentage of correct responses and latencies in identity and emotion matching for schizophrenic patients and controls. Figure options For patients with schizophrenia, we also calculated Spearman correlation coefficients between SANS or SAPS total scores and subscores and performances for identity or emotion matching. Performances were not correlated with total SAPS scores, but the delusions subscore was correlated with both emotion matching (ρ=0.46, P<0.04) and emotion matching when identity changed (ρ=0.68, P<0.01). The SANS score was negatively correlated with identity matching (ρ=−0.48, P<0.05) and tended to be negatively correlated with emotion matching (ρ=−0.41, P<0.08). Moreover for emotion matching, the SANS score was negatively correlated with emotion matching when identity changed (ρ=−0.54, P<0.02). Three SANS subscores were correlated with emotion matching when identity changed (affective flattening: ρ=−0.48, P<0.03; avolition–apathy: ρ=−0.49, P<0.03; inattentiveness: ρ=−0.53, P<0.02). Thus, the deficit in matching one emotion expressed by two distinct persons observed for patients with schizophrenia co-varied with the severity of negative symptoms (see Fig. 3). No other correlation was significant. Correlation between SANS score and percentage of correct responses in emotion ... Fig. 3. Correlation between SANS score and percentage of correct responses in emotion matching (“same” response) when identity was varied for schizophrenic patients. Figure options Fig. 3 also indicates that four patients performed far below chance levels (from 0% to 12.5% correct) in emotion matching from different identities. Two of them also performed below chance when matching the same identity from a different emotion (both 12.5%). All these patients performed above chance levels in other conditions. They evidently mistook a change of identity (or emotion) for a change of emotion (or identity). This finding parallels the observation reported by Young et al. (1996) with an amygdalectomy patient.