Results

Behavioural data, detailed in Table 1, show that subjects had little difficulty performing the attention task. Only two subjects accomplished fewer than 13 of the 14 nominal target detections in each location, and only two subjects gave more than a single false alarm. (Data from intervals surrounding missed targets and false alarms did not affect the fMRI averages and statistics, since these intervals were excluded from the fMRI attention analysis.) The average accuracy, calculated as the ratio between the number of correctly detected targets and the total number of detection opportunities, was 94% for left targets and 96% for right targets. Average response latency was 697ms (SD 119ms) for left targets and 701ms (SD 144ms) for right targets.

Table 1. Behavioural results. Accuracy, response latency, and numbers of hits, misses, and false alarms for each subject. As the task was designed so that most of the subjects performed at ceiling, d' scores are not mathematically well defined and therefore are not given.

Numbers and amplitudes of saccades for each subject are given in Table 2. Because of noise in the eye position recordings, detection of saccades less than 1.25° in amplitude was not reliable. However, this 1.25° detection threshold was only one fourth of the distance between the fixation point and the targets, and neither real-time video observations nor averages of point-of-regard as a function of time (Figure 2) indicate any systematic deviation in direction of gaze associated with shifts of attention.

Table 2. Numbers of leftwards and rightwards saccades for each subject, with average saccade amplitude and standard deviation in degrees of visual angle.

Figure 2. Grand average of gaze direction as a function of time during the intervals surrounding responses. Solid line, rightward shifts. Broken line, leftward shifts. There was no systematic deviation in gaze direction towards either of the target locations, which were positioned at ±5 degrees.

The task-versus-fixation comparison yielded regions of interest with average (±SD) coordinates as follows: left ventral occipital, -29±7, -69±6, -12±6; right ventral occipital, 29±7, -69±6, -13±7; left intraparietal, -30±4, -69±6, 26±7; right intraparietal, 28±5, -69±6, 26±7; left superior parietal, -20±11, -69±6, 48±7; right superior parietal, 17±14, -69±6, 48±8. The y coordinates are identical because the same coronal slice was used for each region in each subject.

fMRI time series for each region, averaged across subjects, are shown in Figure 3, and z-scores and percent signal change are detailed in Table 3. The analysis of variance on z-scores was significant for an interaction of hemisphere by region of interest (F(2, 54)=8.81, p=0.0005). Post hoc t-tests revealed significant interhemispheric differences in the ventral occipital region (t(20)=3.08, p<0.006), where the mean z-score was -1.00 in the left hemisphere and +0.88 in the right hemisphere, and in the intraparietal region (t(20)=2.90, p<0.009), where the mean z-score was +0.18 in the left hemisphere and -0.83 in the right hemisphere. Both the z-score magnitudes and the time series indicate that this intraparietal difference arose mainly from the right hemisphere. Since positive z-scores denote correlation with leftwards attention, this pattern of z-scores indicates that ventral occipital cortex was activated contralaterally and (right) intraparietal cortex ipsilaterally to the attended hemifield. In the superior parietal region, although the time series contained transient responses in both hemispheres for shifts in both directions, the t test comparing rightward shifts to leftward shifts was not significant.

Figure 3. Grand averages of fMRI time series from left (left column) and right (right column) hemispheres, centred on the time of occurrence of an attentional shift. The vertical axis is percent above the mean signal acquired during the fixation periods, when no attention task was being performed. Time series were averaged from six seconds following the preceding shift or the beginning of the task period, up to the time of the next shift or the end of the task period. Solid line, rightward shifts. Broken line, leftward shifts. Note the large contralateral effects in ventral occipital cortex, the ipsilateral effect in right intraparietal sulcus, and the transients in superior parietal cortex.

Table 3. Activations for each region of interest, as z-scores and as left-right differences in percent signal change, for individual subjects and as group averages.

No other factors or interactions produced significant effects. We note, however, a possible trend differentiating the sexes: mean z-scores for ventral occipital cortex were greater in magnitude for the males (-1.44 left, +1.22 right) than for the females (-0.63 left, +0.60 right), while intraparietal mean z-scores were greater in magnitude for the females (+0.19 left, -1.14 right) than for the males (+0.18 left, -0.46 right).

The analysis of percent signal change was consistent with the z-score analysis, though the finding in ventral occipital cortex was somewhat weaker. The analysis of variance was again significant for an interaction of hemisphere by region of interest (F(2, 54)=5.03, p=0.0095). Post hoc t-tests showed a trend in the ventral occipital region (t(20)=1.94, p<0.07), where the mean difference in percent signal change was -0.239 in the left hemisphere and +0.209 in the right hemisphere, and in the intraparietal region (t(20)=3.01, p<0.007), where the mean difference was +0.057 in the left hemisphere and -0.159 in the right hemisphere. Data from the superior parietal region were again not significant.

Because the behavioural performance of subjects 4 and 5 was somewhat worse than that of the other subjects, the fMRI data were re-analysed in order to determine whether the exclusion of these two subjects would affect the group results. For the analysis of z-scores, the data were again significant for an interaction of hemisphere by region of interest (F(2, 48)=10.67, p=0.0001), with significant differences in the ventral occipital (t(16)=3.65, p=0.002) and intraparietal (t(16)=2.44, p=0.027) regions. For the analysis of percent signal change also, the interaction remained significant (F(2, 48)=7.85, p=0.0011), with significant ventral occipital (t(16)=2.81, p=0.013) and intraparietal (t(16)=2.58, p=0.020) effects. In both analyses, the exclusion of these two worst-performing subjects increased the ventral occipital effect and decreased the intraparietal effect.

Discussion