Discussion

In this combined EEG and psychophysical study, we investigated the electrophysiological correlates of contour integration. We systematically varied detection difficulty of contours.

First, we presented ideally aligned or, more or less, misaligned contours. Second, we utilised open (CONCAVE, DOUBLE C) and nearly closed (CONVEX) figure types. Third, only dur-ing experiment 1 the critical stimulus was preceded by a random display. The results for all task conditions demonstrate that detectable contours elicited a more negative ERP, especially over occipital electrode sites. The onset and maximum peak latency of this initial discrimina-tive potential varied between task conditions. It occurred first within the N1 time window ap-proximately 150 ms after stimulus onset. Presentation of open contours as well as the preced-ing stimulus durpreced-ing experiment 1 delayed and partially prolonged this negative component. It was always found within the P2 time window and encompassed the N2 time window for all task conditions except for nearly closed contours in experiment 2. During experiment 1, the N2 amplitude was larger for ALIGNED compared to MISALIGNED contours. Only during experiment 1 an FSP was observed. Starting 350 ms after stimulus onset, correctly identified contours elicited a centro–parieto–occipital P3, which was larger for ALIGNED than for MISALIGNED contours.

Studie 1: ERPs are modulated by task demands

Our results show that the contour-specific negative shift encompasses the N1, P2 and N2 time windows and, therefore, closely resembles slow negative shifts as described for the tsVEP (Fahle et al. 2003). Furthermore, similar to the tsVEP the maximum of the contour-specific response occurs occipital (Lamme et al. 1992). The negative shift increase seems to be a fun-damental feature of contour integration as it occurs, although with variable timing and ampli-tude, for all task conditions including detectable contours. This similarity between the tsVEP and the contour-specific EEG-response strongly indicates that detection of borders is medi-ated by similar mechanisms irrespective of whether borders are defined by spatial irregulari-ties of otherwise homogenous surfaces (texture segmentation) or by collinear orientation in otherwise randomly oriented elements (contour integration). A slow negative potential has also been observed during perceptual closure (Doniger et al. 2000; Sehatpour et al. 2006) and grouping by similarity (Han et al. 2001).

The electrophysiological and behavioural results indicate that the introduction of different task conditions successfully varied task demands: although training improved performance to near ceiling levels, the behavioural performance during experiment 1 confirms that detection of ALIGNED contours is easier compared to MISALIGNED contours. It has already been shown that contour detection gradually decreases with misalignment (Field et al. 1993).

Furthermore, psychophysical studies indicate that detection for the CONVEX figure should be easier than for the other two figure types because the CONVEX figure resembles a closed figure with no changes in curvature (Kovacs & Julesz 1993; Pettet 1999). Although in this study detection accuracy does not differ between figure types, the electrophysiological results discussed below indicate that detection of open contours (CONCAVE) was more difficult and delayed the contour-specific response compared to nearly closed contours. Rapid processing for closed compared to open stimuli has also been reported for a visual search task measuring reaction times (Elder & Zucker 1993).

Experiment 3 confirms that contour detection is easier when the critical period is not preceded by a random display. In this case, the degree of misalignment tolerated to still obtain thresh-old performance of 75% correct responses is larger when compared to the presentation se-quence including the preceding random display. While the interference of combined forward

our results emphasise that diminishing the visibility of critical stimuli by pure forward mask-ing significantly interferes with contour integration.

Macknik and Livingstone demonstrated that forward masking suppresses the transient ON-response which is elicited by the unmasked target in monkey V1 cells (Macknik & Living-stone 1998). They also showed that forward masking in humans is most efficient when – as was the case in experiment 1 – the onset of the target follows immediately after the mask ter-minates. In our case, the suppression of the transient ON-response might diminish visibility of the individual elements.

However, the visibility of the target in experiment 1 could have been diminished by additional aspects. The change in luminance contrast might be larger for the sudden onset of the Gabors (experiment 2) than for a mere rotation of the Gabors (experiment 1). Furthermore, the onset of the CRITICAL period elicits an apparent motion perception of all rotating elements and motion signals might interfere with task-relevant signals. Furthermore, the duration of the NON- CRITICAL period was 600 ms and therefore adaptional adjustments to the orientation of the Gabors might have diminished the initial neural response to the onset of the CRITICAL period as well. For a 1 s exposure, adaptation of the neural response to orientation in extrastri-ate visual areas has been shown (Boynton & Finney 2003). Therefore, although all elements rotated, due to the random assignment of the rotation angle the new orientation of some ele-ments might have remained in the preferred orientation range of the adapted cell, especially as adaptation to a given orientation broadens and shifts the peak of neural tuning curves of orien-tation preferences (Dragoi et al. 2000). Thus, several possible factors may have contributed to diminish relevant information at the onset of the CRITICAL stimulus during experiment 1.

4.4.1 Variations in latency of the contour-specific negativity

The earliest occurrence of the contour-specific negativity was during the N1 time window between 100 and 190 ms after stimulus onset. This early effect might indicate that easily de-tectable contours pop out and contour integration can be resolved without target-specific at-tention. This interpretation derives from the involvement of the N1 in visual binding and vis-ual search tasks. More specifically, attentional modulation of the ERP in the N1 time range may be linked to selective perceptual amplification at a pre-specified location (Hillyard &

Studie 1: ERPs are modulated by task demands

al. 2004). Furthermore, an enhanced negativity within the time range investigated in our study was found for Kanizsa figures when compared to stimuli that induce perception of fragmented illusory contours (Herrmann & Bosch 2001), that is, an N1 enhancement might indicate fast integration processes. In accordance, Kanizsa triangles seem to pop out when presented be-tween distracters and this pop out is related to N1 modulations (Senkowski et al. 2005). This interpretation is also in accordance with the finding that in a backward masking experiment with overtrained monkeys a stimulus duration of 30–60 ms is sufficient for above-chance per-formance during contour integration (Mandon & Kreiter 2005).

The contour-specific response occurs later when contour integration becomes more difficult.

The maximum negative response during experiment 2 peaked approximately 180 ms after stimulus onset for nearly closed and 220 ms after stimulus onset for open figures. This indi-cates that for open figures the maximum of the contour-specific response occurs already after the N1 and during the P2 time window. During experiment 1, the maximum negative response peaked 220 ms after stimulus onset for nearly closed and 270 ms after stimulus onset for open contours. This indicates that the preceding stimulus presented in experiment 1 leads to a fur-ther delay of the contour-specific response.

Selective attention and target detection studies both suggest that delay or prolongation of the negative shift indicate that attention has been allocated to achieve contour integration. For example, the SN occurs when stimuli share attended visual qualities with the target, but only the later stages of the SN elicit a larger amplitude for target-relevant feature conjunctions (e.g.

of orientation and spatial frequency; Harter & Aine 1984)). Similarly, the persistence of the increased negativity during the N2 time window for all task conditions except for the CON-VEX figure in experiment 2 also indicates that for higher task demands processing of the con-tour relies on additional resources. An increased N2 has been reported to indicate prolonged visual search (Schubö et al. 2001) and suppression of competing information from surround-ing distracter items in order to improve target detection (Luck & Hillyard 1994). Functional imaging studies also indicate that information about both the contour and its surround is proc-essed to achieve contour integration in a network involving various visual areas (Altmann et al. 2004; Altmann et al. 2003).

4.4.2 Frontal selection positivity

The additionally FSP for ALIGNED contours in experiment 1 is another indication that the contour-specific response becomes increasingly attention demanding with decreasing sali-ency. The FSP is elicited concurrently with the posterior negative enhancement, and this pat-tern of posterior and anterior activation for targets has been found in several target detection tasks (Kenemans et al. 1993; Michie et al. 1999; Potts & Tucker 2001; Smid et al. 1999).

Moreover, enhancement of frontal gamma band activity is related to the intentional allocation of attention to maintain the currently perceived perceptual alternative of a reversible figure (Basar-Eroglu et al. 1996a; Mathes et al. 2006a). Functional imaging studies also suggest a selective role of prefrontal areas during attention demanding visual search (Leonards et al.

2000). The occurrence of a frontal activation therefore emphasises that contour integration during experiment 1 is successfully achieved by allocation of attentional resources.

Importantly, performance was nearly perfect for all ALIGNED and MISALIGNED contours.

A small and significant decrease in performance, however, was observed for slightly mis-aligned contours during experiment 1. It can therefore be assumed that task demands were greatest for MISALIGNED contours during experiment 1. The frontal enhancement, however, is significant only for the ALIGNED stimulus configuration. MISALIGNED contours show no FSP over frontal and a diminished negative enhancement over posterior sites. This reduc-tion of the effects might be related to the non-optimal performance for the MISALIGNED task condition during experiment 1. We assume that, when contour detection performance decreases due to high task demands, the contour-specific modulation of the ERP declines until it is indistinguishable from the CONTROL condition where no information about the localisa-tion of the contour can be derived from the neural response. Similarly, the negative response elicited by perceptual closure is diminished as long as fragmented objects cannot be clearly identified (Doniger et al. 2000).

4.4.3 P3

In all experiments, contours elicited a larger P3 compared to the CONTROL condition. Al-though training improved performance to near ceiling levels for both ALIGNED and MIS-ALIGNED figures, the maximum P3 amplitude was always larger for MIS-ALIGNED compared

Studie 1: ERPs are modulated by task demands

uli and may be related to decision making (Basar-Eroglu et al. 1992; Picton 1992). However, ALIGNED and MISALIGNED contours did not differ in this respect. The amplitude of the P3 is often smaller for difficult compared to easy target discrimination, which is explained by a greater degree of uncertainty for difficult task-relevant decisions (Picton 1992). This indi-cates that despite near-perfect detection the discrimination and localisation of MISALIGNED figures were more difficult.

In conclusion, contour integration elicits a negative shift over posterior electrode sites and contours, therefore, seem to be processed similar to textures. Contour integration might be rapid for salient contours, however, with decreasing saliency, contour integration slows down and possibly relies increasingly on the allocation of additional selective attention.