A brief psychometric test reveals robust shape completion deficits in schizophrenia that are less severe in bipolar disorder

Shape completion extracts fundamental object properties such as shape, number, and size from camouflaged and overlapping distal layouts (Keane, 2018). The process relies upon a sparse but densely inter-connected set of cortical regions that are centered in the secondary visual network and that incorporate parts of the dorsal attention, default mode, and frontoparietal networks (Keane et al., 2021). The process is of clinical interest because, relative to well-matched healthy controls, schizophrenia patients poorly discriminate visually completed “illusory” shapes, despite normally discriminating similar configurations that lack illusory contours (Keane et al., 2019). Intriguingly, patients also respond normally to alterations in illusory contour salience. For example, distractor lines near illusory contours similarly impair shape discrimination in patients and controls, suggesting that patients can form but not properly use illusory contours for discerning global shape (Keane et al., 2014).

In this pilot study, we inquired whether similar results would emerge in bipolar disorder. Addressing this question could suggest a new marker distinctive to schizophrenia; it could also could clarify how perceptual disturbances vary across diagnostic categories. Certain perceptual and cognitive impairments attenuate from schizophrenia to bipolar disorder to no psychiatric disorder (Jahshan et al., 2014; Schallmo et al., 2015; Sheffield et al., 2018) and so qualitatively the same results might occur for shape completion. A secondary objective was to better establish the robustness of these effects.

The sample comprised 23 healthy controls, 23 schizophrenia patients (including 3 with schizoaffective disorder), and 26 patients who had type I, type II, or not otherwise specified variants of bipolar disorder (n=20, 4, 2, respectively). The sample excluded one bipolar participant, two controls, and one schizophrenia patient who either did not complete the task or who exhibited poor fragmented performance (thresholds falling beyond 2.5 SD of the control group’s mean). Groups were undifferentiated on age, sex, paternal/ maternal education (years), and handedness (see Supplementary Table). The study was approved by the Rutgers Institutional Review Board and all subjects signed a written informed consent.

Participants judged on each trial whether four sectored circles (pac-men) formed a fat or thin square (illusory condition; 6 deg of visual angle per side) or whether four downward pointing pac-men were rotated left or right (fragmented condition; Fig. 1A) (Ringach and Shapley, 1996). For each condition, there were 40 practice trials followed by 80 test trials, the latter half of which contained spatially-fixed distractor lines. Task difficulty depended on pac-man rotational magnitude, with larger rotations making the alternatives easier to distinguish. For practice trials, there was a fixed set of progressively smaller rotational differences (10, 8, 6, and 4 degrees). For test trials, a Bayesian adaptive “Psi” procedure recommended a rotational magnitude on each trial so as to minimize entropy (uncertainty) of the slope and threshold estimates of the log-Weibull psychometric function (Kontsevich and Tyler, 1999). (For other Method details, see Keane et al. (2014))

Fig. 1.

(A) Stimuli; (B) Thresholds for each group and condition; (C) Distractor line interaction was similar between groups; (D) Shape completion group comparisons using fragmented – illusory threshold differences; (E) Shape completion group comparisons using illusory – fragmented accuracy differences (from practice trials); (F) Shape completion was also computed for each group on a per trial basis (e.g., first fragmented trial threshold minus the first illusory trial threshold).

A 2 (stimulus: illusory, fragmented) × 2 (distractor: no lines, lines) x 3 (group) ANOVA revealed a stimulus by distractor interaction (F(1,69)=27.64, p<.001, η²_p =0.29), replicating previous findings that adding distractor lines hurt performance (i.e., raised thresholds) in the illusory but not fragmented condition (Keane et al., 2014). The foregoing interaction neither differed between groups (three-way interaction: F(2,69)=.85, p>.43; Fig. 2C) nor monotonically changed across groups (t_K,=−.14, p=.24), suggesting that the three groups were similarly responsive to changes in illusory contour appearance.

Consistent with prior studies, “shape completion” was defined as the illusory/fragmented threshold difference in the absence of distractors (Keane et al., 2014). Shape completion differed between groups on an ANOVA (stimulus x group interaction: F(2,69)=7.5, p=.001, η²_p =0.18) and worsened monotonically from schizophrenia to bipolar disorder to no disorder (t_K,=−.43, p<.001). Direct group comparisons revealed a significant Ctrl/SZ shape completion difference (t(44)=4.08, p<.001, g = 1.18), a marginal bipolar/SZ difference (t(47)=2.01, p=0.050, g = 0.57), and a marginal Ctrl/Bipolar difference (t(47)=1.88, p=0.07, g = 0.54).

SZ/Ctrl shape completion differences were robust. First, they could be demonstrated with only a small number of trials. The group difference became continually significant after 16 trials and plateaued in magnitude after ~30 trials per condition (g=1.3; Fig. 1E). Shape completion differences could even be found in the practice trial data (t(44)=−2.28, p=.03, g=.67), despite the inherent ceiling effects (mean fragmented accuracy >87% in each group) and despite not being tailored to each subject’s own ability (Fig. 1F). Note that shape completion impairments could not be ascribed to poor task understanding since subjects were shown comprehensive instructions beforehand and since illusory practice trial performance in the schizophrenia group was well above chance (M = 74%, SD =.15).

Shape completion and medication dose were uncorrelated in the schizophrenia group (p=.54) but correlated in the bipolar group (p=.043), although the latter would not survive statistical correction. Across patients, shape completion impairments did not correlate with 5-Factor PANSS scores (Wallwork et al., 2012), however, the disorganization correlation was within range (r=−.2, p=.20, 95% CI: [−.43, .08]) of previous reports (r=−.32, p=.002, N=94, Keane et al., 2019). Given the modest size of these correlations, other illness factors are likely also playing role and merit further investigation.

To summarize, shape completion deficits are large in schizophrenia, diminish in severity from schizophrenia to bipolar disorder to no disorder, and can be revealed with fewer than 40 test trials. These deficits cannot be attributed to poor sensitivity to illusory contours or poor overall task performance. Shape completion deficits thus constitute a potentially useful biobehavioral marker for schizophrenia.