OUP user menu

Visual cortical activity reflects faster accumulation of information from cortically blind fields

Tim Martin, Anasuya Das, Krystel R. Huxlin
DOI: http://dx.doi.org/10.1093/brain/aws272 3440-3452 First published online: 20 November 2012

Summary

Brain responses (from functional magnetic resonance imaging) and components of information processing were investigated in nine cortically blind observers performing a global direction discrimination task. Three of these subjects had responses in perilesional cortex in response to blind field stimulation, whereas the others did not. We used the EZ-diffusion model of decision making to understand how cortically blind subjects make a perceptual decision on stimuli presented within their blind field. We found that these subjects had slower accumulation of information in their blind fields as compared with their good fields and to intact controls. Within cortically blind subjects, activity in perilesional tissue, V3A and hMT+ was associated with a faster accumulation of information for deciding direction of motion of stimuli presented in the blind field. This result suggests that the rate of information accumulation is a critical factor in the degree of impairment in cortical blindness and varies greatly among affected individuals. Retraining paradigms that seek to restore visual functions might benefit from focusing on increasing the rate of information accumulation.

  • functional MRI
  • stroke
  • brain lesions
  • motion processing
  • visual cortex

Introduction

Unilateral damage to the primary visual cortex (V1) or optic radiations causes a profound loss of visual function in the contra-lesional visual field, termed partial cortical blindness. The visual defect is typically homonymous, affecting a similar portion of the visual field in each eye. Because of this, cortical blindness is one of the leading causes of disability in survivors of damage to the occipital pole. Cortical blindness is associated with difficulties in many activities of daily living, including driving, reading, visual search, locomotion and navigation (Pambakian and Kennard, 1997; Cole, 1999; Kerkhoff, 2000; Vargas-Martin and Peli, 2002; Gutteridge and McDonald, 2004; McDonald et al., 2006; Bowers et al., 2009).

According to classical models of connectivity (Hubel and Wiesel, 1962, 1965; Van Essen and Maunsell, 1983), V1 acts as a gateway of visual information transfer to extrastriate visual cortical areas. Virtually all retinal information passes through the dorsal lateral geniculate nucleus to V1. Therefore, if V1 is damaged, all visual areas downstream of it are also deprived of their main feed-forward visual input. Thus, it is no surprise that cortical blindness should be devastating and should affect all visual perceptual modalities.

However, many cortically blind humans appear to possess residual abilities for processing visual stimuli in their blind field (Riddoch, 1917; Holmes, 1918; Pöppel et al., 1973; Perenin and Jeannerod, 1975; Weiskrantz et al., 1995; Azzopardi and Cowey, 1998, 2001; Zeki and Ffytche, 1998). This phenomenon was termed blindsight by Weiskrantz et al., (1974). Specific visual functions spared in blindsight vary between people, but can include localization of stimuli (Pöppel et al., 1973), shape or orientation discrimination (Weiskrantz et al., 1974), spectral sensitivity (Stoerig and Cowey, 1991) and, most commonly, motion and flicker perception (Barbur et al., 1980).

Some of the literature on blindsight attributes spared functions to the existence of a visual circuit involved in eye movement control (Mohler and Wurtz, 1977; Schiller et al., 1980; Dorris et al., 2007; Berman et al., 2009), which bypasses striate cortex via a projection from the superior colliculus to pulvinar, thence to area MT (Cowey and Stoerig, 1991; Stoerig and Cowey, 1997; Sincich et al., 2004). Evidence for this came from monkeys in which selective lesion or inactivation of V1 appeared to cause a form of cortical blindness similar to that experienced in humans. In such monkeys, neurons in ipsi-lesional areas MT and V3/V3A continued to respond to blind field stimulation (Rodman et al., 1989; Girard et al., 1991, 1992; Zeki and Ffytche, 1998), although direction-tuning curves in MT were significantly altered (Girard et al., 1992). When the superior colliculus was damaged, their responsiveness was abolished (Mohler and Wurtz, 1977; Rodman et al., 1990).

However, there are also direct projections from the dorsolateral geniculate nucleus to extrastriate areas V2 (Hendry and Reid, 2000), MT (Sincich et al., 2004) and V4 (Cowey and Stoerig, 1989). These pathways are smaller than the main pathway from dorsolateral geniculate nucleus to V1, but recent evidence shows that inactivating the dorsolateral geniculate nucleus also decreases visually evoked responses in extrastriate visual areas (including V2, V3, V4 and MT) that remain active after V1 lesions in monkeys (Schmid et al., 2010). The relative roles of the superior colliculus/pulvinar/extrastriate cortex versus dorsolateral geniculate nucleus/extrastriate cortex pathways in mediating blindsight after V1 damage remains to be elucidated.

However, one thing is for certain: there is considerable variability in preserved visual abilities between affected individuals. This has led to the description of several variants of blindsight (Weiskrantz, 1996; Danckert and Rossetti, 2005) and begs the question: why? Considerable interindividual variability is also observed in functional brain imaging studies of cortically blind patients, both in terms of the pattern of cortical activation elicited by visual stimulation and in functional connectivity between visual areas. For instance, it appears that subjects who sustained striate cortex damage early in life (e.g. Patients G.Y. or P.K.) exhibit greater than normal visually evoked activity and greater connectivity between spared extrastriate cortex ispilateral to the lesion and visual areas (including subcortical ones such as the pulvinar and dorsolateral geniculate nucleus) in the contralateral intact brain hemisphere (Ptito et al., 1999; Bridge et al., 2008; Silvanto et al., 2009). Cortically blind subjects who sustained their brain damage in adulthood usually exhibit visually evoked activity in extrastriate areas ipsilateral to the damaged V1, including hMT+ and V4/V8 [Subject F.S. in Goebel et al. (2001), the hemianopic subject in Schoenfeld et al. (2002), Subject S.B.R. in Bridge et al. (2010) and Subjects R.A. and J.P. in Morland et al.(2004)]. Unfortunately, the mere presence of extrastriate cortex activity in these reports does not appear to be reliably correlated with awareness of the visual stimuli presented or with ability to discriminate them. Subject F.S. could not reliably detect stimuli in his blind field, whereas the hemianopic patient in Schoenfeld’s study could. Yet, both exhibited extrastriate activation in what appeared to be the same areas. The largest study yet performed with respect to this question was that of Morland et al. (2004) who examined eight cortically blind patients, including Subjects G.Y., R.A. and J.P. Of the eight, only these last three had some awareness of the motion stimuli they were supposed to discriminate in their blind field. The remainder had damage to extrastriate areas in the lateral occipital lobe, presumably including hMT+, in addition to V1 damage; a fact that was used to explain their lack of motion processing in the blind field. All in all, what may be surmised from this body of work is that if a person has sustained V1 damage in adulthood and they exhibit extrastriate activation in areas ipsilateral to the lesion, there is a better chance (though not a guarantee) that this person might have some residual visual processing, with or without awareness of stimuli presented in their blind field (compare with Sahraie et al., 2003).

Building on such observations, the present study examined the hypothesis that preservation of visually evoked activity in extrastriate visual cortical areas should be positively correlated with better visual discrimination in cortically blind fields. Specifically, we asked whether the presence of blood oxygen level-dependent (BOLD) signal in spared regions of V1 and V2, as well as in motion-processing visual areas V3/V3A and hMT+ ipsilateral to the lesion, was associated with improved global direction discrimination performance in cortically blind fields. Fundamentally, a person confronted with a visual task to be performed in their blind field makes a decision based on what little visual information is getting through and being processed by their intact visual cortical areas. Even if the person is guessing, we posit that this guess should be biased by sensory processing taking place in these spared areas, no matter how minimal or suboptimal.

Given that cortically blind observers are making a decision based on limited or absent visual input, this process may be affected in several different ways. The most obvious prediction is that response times will be slowed and accuracy will be lower, but these phenomena could be caused by several different mechanisms. For example, cortically blind observers may adopt a more conservative bias when forced to respond to blind field stimulation. Alternatively, they might adopt a strategy of guessing under the assumption that their task is hopeless. It might be the case that early (predecision) perceptual processes or post-decision motor processes are slowed down by the lesion. Finally, it is possible that information still accumulates from the blind hemifield, but does so more slowly than from an intact hemifield. To add to the complexity, these possibilities are not mutually exclusive or exhaustive.

To inform these issues, we analysed behavioural aspects of the visual decision-making process in our subjects using a simplified variant of the diffusion model (Ratcliff, 1978), the EZ-diffusion model (Wagenmakers et al., 2007). The full diffusion generalization model includes seven parameters for each experimental condition, and until recently, fitting the model to data has been a difficult process (Ratcliff and Tuerlinckx, 2002). The EZ-diffusion model reduces this complexity to three parameters per experimental condition (drift rate, boundary distance and non-decision time) by making certain assumptions about the data. In particular, it assumes that the distances from the starting point of the diffusion process to the boundaries associated with correct and incorrect decisions are equal, and that there is no variability in these parameters across trials. The decision process is illustrated in Figure 3. According to this model, the onset of a visual stimulus triggers a diffusion process that moves towards a decision boundary corresponding to the correct response. The rate of drift of this process depends on observer factors, such as alertness and attention, and the fidelity of the signal. Noise in the diffusion process can also cause it to drift towards and cross a second decision boundary, resulting in incorrect responses. The diffusion process has a starting location relative to the decision boundaries, and the distance from the starting point to the boundaries is under the observer’s control, akin to bias in signal detection theory. The drift rate and boundaries of the diffusion process together control the decision time. Finally, non-decision time includes both perceptual and motor response factors.

By partitioning behaviour into component cognitive processes, we can gain additional insights into visual signal processing in the absence of an intact V1, beyond the obvious prediction that subjects should be slower and less accurate when responding to stimuli in their blind fields (compare with Kayser et al., 2010). For example, more conservative criteria for making a decision would result in wider boundary estimates from the diffusion model. Differences in early (predecision) perceptual processes that are abnormal or even absent in the damaged hemisphere would result in increased non-decision time. Slower accumulation of information would decrease drift rates in response to blind field stimulation.

The simultaneous use of functional MRI and the EZ-diffusion model also affords us the opportunity to relate changes in component decision processes to specific brain regions. Although areas V1 through MT may not be the actual sites of decision making on our task, we were interested in assessing to what degree activity in these early visual areas might influence behavioural performance.

Materials and methods

Participants

The participants were nine stroke survivors with homonymous visual loss and nine neurologically intact age-matched controls. Details about selection and testing can be found in Huxlin et al. (2009). All stroke survivors underwent Humphrey and Goldmann perimetry, as well as a neuro-ophthalmological examination to rule out conditions such as attentional neglect and ocular disease that might affect visual performance. None of the cortically blind subjects were aphasic, and only Patient CB1 had residual hemiparesis.

  • Patient CB1 was a 50-year-old woman who had a posterior cerebral artery stroke 4 years before enroling in the study. The stroke caused a left partial hemianopia. There was also some lingering hemiparesis and a history of vasculitis.

  • Patient CB2 was a woman, 77 years of age, with a left homonymous hemianopia following a right posterior cerebral artery stroke. CT reports written immediately following the stroke also implicated medial parietal lobe involvement.

  • Patient CB3 was a 69-year-old woman with left homonymous hemianopia following a right posterior cerebral artery stroke.

  • Patient CB4 was a 35-year-old woman with a right homonymous hemianopia following a stroke in the left occipital lobe.

  • Patient CB5 was a 74-year-old woman with a left inferior homonymous quadrantanopia following a posterior cerebral artery stroke.

  • Patient CB6 was a 56-year-old man with a left partial hemianopia following a right posterior cerebral artery stroke.

  • Patient CB7 was a 79-year-old man with a right partial hemianopia following a left posterior cerebral artery stroke.

  • Patient CB8 was a 57-year-old man with a right superior quadrantanopia following a left posterior cerebral artery stroke.

  • Patient CB9 was a 53-year-old man with a right homonymous hemianopia following a left posterior cerebral artery stroke.

Each patient was at least 1 year post-stroke at the time of scanning. Humphrey automated perimetry results and structural magnetic resonance scans illustrating each patient’s lesion are given in Fig. 1. The mean age of cortically blind subjects was 61.78 years, SD = 14.39, ranging from 37 to 79 years. In the control group, ages ranged from 34 to 79 years, mean = 62.22 years, SD = 11.99, and there were four men and five women.

Figure 1

T1-weighted structural MRIs of the nine V1-damaged patients (CB1–9) recruited for this study, illustrating the location of their V1 damage in both horizontal and coronal planes. Next to each patient’s brain scan are representations of his/her Humphrey visual field perimetry (24-2 test) averaged across the two eyes. The grey scale indicates average detection sensitivity in dB. Light grey circles represent the visual field locations and sizes of random dot stimuli the subjects were asked to discriminate. MRIs are oriented so that the left side of the image corresponds to the left side of the brain.

Functional magnetic resonance imaging procedure

All procedures were carried out in accordance with the Declaration of Helsinki and were approved by the Institutional Review Board of the University of Rochester Medical Centre.

Participants were informed of the procedures, and written informed consent was obtained. An event-related design was used (Fig. 2), which consisted of repeated trials presented in one of seven runs. Note that trial onset times were not synchronized to the transistor–transistor logic pulse sent by the scanner allowing us to sample the BOLD signal at sub-repetition time resolution. Trials were initiated by the appearance of a fixation spot after a random intertrial interval of 5–7 s, lasting 1 s. After the fixation spot disappeared, a random-dot stimulus would appear. The observer’s task was to discriminate the global direction of motion (right or left) of this stimulus. No feedback was given as to the correctness of the responses, and subjects were asked to guess if they could not discriminate the direction of motion. The random dots were presented in an aperture 5° in diameter, with a density of 1.6 dots/°2, moving at 10°/s. This 5° aperture appeared randomly at one of four visual field locations, whose exact distribution depended on the location and size of the scotoma in individual patients (Fig. 1). For the controls, the stimulus locations were matched with those of the one patient with whom they were most closely age matched. The duration of the moving dots was 500 ms. For the purpose of the present study, all analyses are collapsed across the two locations within the left or right hemifield. Dots in the stimuli presented moved either with a direction range of 0° (completely coherent motion) or near each participant’s threshold for discrimination. The threshold for discrimination was defined for this purpose as the maximum range of dot directions that resulted in 75% correct performance on this global direction discrimination task. For the patients, this point was based on laboratory measurements taken before the experiment, and ranged from 290–320° of range in the intact hemifield of vision. For control participants, a direction range of 320° was chosen based on our experience of a typical discrimination threshold for visually normal observers. Thus the design was a 4 (location) × 2 (direction range) × 2 (group) mixed factorial. Fifty-six trials of each type were collected, for a total of 448 trials.

Figure 2

Schematic illustration of the global direction discrimination task performed during functional MRI. On each trial, moving dots were presented in a 5° window at one of four locations. Dots could move in the same direction, or randomly within a specified range (usually near 320°) of directions centred on a single direction (right or left), with the range near the threshold for discriminating direction of motion for each participant, as measured psychophysically before scanning.

Figure 3

Schematic illustration of the diffusion model decision process. Information begins to accumulate from a starting point at 0, and drifts towards a decision boundary with some degree of stochastic fluctuation. The decision time is determined by how quickly information accumulates (drift rate) and how much information must accumulate before the observer initiates a response (boundary separation). RT = reaction time.

Magnetic resonance imaging measurements

MRI data were acquired on a 3-T Siemens Trio Scanner. T1-weighted anatomical images (T1-weighted magnetization prepared rapid acquisition gradient echo (MPRAGE) volumes, 1 × 1 × 1 mm) were acquired. Functional scans were T*2-weighted transverse echoplanar images (echo time = 30 ms) with BOLD contrast in an ascending direction, covering the whole head except for the bottom half of the cerebellum. Each volume included 69 slices with 4 × 4 × 3 mm voxels and a repetition time of 2 s. The first two volumes of each scan were discarded to compensate for saturation effects.

Behavioural data analysis

Response latency and accuracy (in terms of per cent correct performance) were assessed for each subject using SPSS 11.0. EZ-diffusion model parameters (Wagenmakers et al., 2007) were estimated for each subject, side of presentation (left and right, collapsing across locations within sides) and direction range level. The EZ model uses accuracy, reaction time and the variance in reaction time for correct trials to estimate parameters of component operations that occur during perceptual decisions.

In the present data set, accuracy, response latency and EZ-diffusion parameter estimates were analysed with 3 (group) × 2 (direction range) × 2 (side of stimulation) mixed factorial ANOVAs. For all ANOVAs, the multivariate approach to repeated measures analysis was used (Maxwell and Delaney, 2003).

Functional magnetic resonance imaging data analysis

Analysis of functional MRI data was performed using SPM5 (Wellcome Department of Cognitive Neurology, London, UK). Functional scans were realigned using a 2-step procedure in which all scans were aligned to the first scan, a mean image was calculated and then all scans were aligned with the mean image. Scans were then normalized to the Montreal Neurological Institute stereotaxic atlas (Ashburner and Friston, 1999), and spatially smoothed with an 8-mm full-width half-maximum Gaussian filter. The T1-weighted structural scans were similarly normalized to the Montreal Neurological Institute atlas (Ashburner and Friston, 1999). Following this spatial preprocessing, functional data were analysed by the standard general linear model approach (Friston et al., 1994). To at least partially account for possible abnormalities in the BOLD response in the cortically blind, a theoretical haemodynamic response function (i.e. the canonical haemodynamic response function), along with its first and second derivatives, was convolved with trial onset times (Friston et al., 1998). The canonical haemodynamic response function, plus first and second derivatives, allows the estimation of temporal and dispersion deviations of the observed BOLD response from the canonical form (Friston et al., 1998). Specifically, the first derivative allows for deviations in the latency of the peak BOLD response, whereas the second derivative allows for estimation of the spread of the BOLD response about the peak.

These individual level models were then interrogated using several contrasts. The relative activation in response to stimulation on each side was assessed with the contrasts LEFT > RIGHT and RIGHT > LEFT, collapsing across the two locations on each side and across motion direction ranges (0° or 320°). In the current analysis, we used a significance threshold of 0.001, uncorrected for multiple comparisons, plus an extent threshold of 30 voxels.

Following computation of the contrasts, locations of significant activity were identified by referencing them with known anatomical landmarks and with previously published Montreal Neurological Institute atlas coordinates for brain regions known to respond to visual motion stimuli (i.e. V1, V2, V3, V3A, V4 and V5/hMT+). The Anatomy Toolbox (Eickhoff et al., 2005, 2006, 2007) for Statistical Parametric Mapping was used as an additional source of guidance to suggest likely interpretation of the location of activity. This toolbox co-registers clusters of activation on a probabilistic cytoarchitectonic map derived from the analysis of post-mortem human brains, and estimates the degree of overlap of these clusters with averaged anatomical regions (Brodmann areas and other structures). We used this information to assess the frequency with which regions of activation overlapped with visual areas of interest (V1, V2, V3, V3A, V4 and V5/hMT+). We considered an activation cluster to reflect an anatomical region of interest if the Anatomy Toolbox indicated that the cluster overlapped at least 10% of the region on the probabilistic map (version 17).

Results

Effect of V1 damage on global direction discrimination performance

In general, global direction discrimination of coherent random-dot stimuli (DR0) was ∼95% correct in both hemifields of controls and in the intact hemifield of V1-damaged patients (Fig. 4A and B). Performance was around chance (50% correct) in the blind field of cortically blind subjects as a group (Fig. 4B), although some patients (Patients CB1 and subjects with perilesional BOLD activity, see later for details) scored significantly better than chance.

Figure 4

Summary of behavioural measures. Error bars represent 95% confidence intervals, using pooled error estimates from a multivariate repeated measures ANOVA of the data. See text for statistical results. NDT = Non-decision time.

For stimuli containing 320° range of dot directions (DR320), performance was slightly better than chance in both hemifields in controls and in the intact hemifield of patients, but patients were at chance in response to their blind hemifields (Fig. 4A and B). Thus, for accuracy, there was a 3-way interaction, F(2,15) = 35.828, P < 0.0005 and partial η2 = 0.827. Side and group interacted, F(2,15) = 22.756, P < 0.0005 and partial η2 = 0.752, and group and direction range interacted, F(2,15) = 4.434, P = 0.031 and partial η2 = 0.372. Finally, there was a main effect of direction range, F(1,15) = 95.15, P < 0.0005 and partial η2 = 0.864. Inspection of the means in Fig. 4 suggests that the group × direction range interaction is different for different sides. Specifically, direction range has less of an effect on patient performance when the stimulus is in their blind field, than when it is in their sighted field, and this interaction depends on which hemifield is blind.

In the subjects’ intact fields of vision (controls and patients combined), reaction times were faster for discriminating the global direction of motion of random-dot stimuli in which the direction range = 0° (mean = 901 ms, standard error = 207 ms), than for stimuli in which the direction range = 320° (mean = − 1041 ms, standard error = 260 ms), Fig. 4C and D. This difference disappeared when the stimuli were presented to the blind field of the cortically blind (Fig. 4D), with the mean reaction time becoming similar to that elicited by the more difficult stimulus (direction range = 320°) in intact hemifields of vision. Thus, for reaction time, there was a significant 3-way interaction, F(2,15) = 4.048, P = 0.039 and partial η2 = 0.351. There was a significant main effect of direction range, F(1,15) = 12.485, P = 0.003 and partial η2 = 0.454. Inspection of the means suggests that the 3-way interaction is because of the group × stimulus side interaction being different at each direction range (Fig. 4C and D). Specifically, when the direction range = 0°, the side of presentation had a different effect on each group, with little or no difference for controls. In blind subjects, there were faster reactions on the right for those with a left-side scotoma and faster reactions on the left for those with a right-side scotoma. When the direction range = 320°, the side of presentation was not as effective in modulating response times. A follow-up analysis confirmed this. Analysing the side of presentation separately for the DR0 and DR320 conditions with Bonferroni-adjusted alphas of 0.05/2 = 0.025 showed a group × side interaction at DR0, F(2,15) = 5.589, P = 0.013 and partial η2 = 0.439, but not at DR320, F(2,15) = 0.793, P = not significant and partial η2 = 0.096.

As expected, drift rate was greater in the DR0 than the DR320 condition, but this effect was greatly attenuated or absent in the blind fields of the cortically blind subjects, where drift rates were essentially similar to those in the DR320 condition (Fig. 5A). The ANOVA confirmed that there was a significant 3-way interaction, F(2,15) = 12.34, P = 0.001 and partial η2 = 0.622, and a significant interaction between group and stimulus side, F(2,15) = 18.52, P < 0.0005 and partial η2 = 0.712. There was also a significant interaction between group and direction range, F(2,15) = 5.974, P = 0.012 and partial η2 = 0.443, and a main effect of direction range, F(1,15) = 36.683, P < 0.0005 and partial η2 = 0.708, and group, F(2,15) = 7.149, P = 0.007 and partial η2 = 0.488. The 3-way interaction observed appears to mirror that for accuracy, in that the 2-way interaction between group and direction range appeared to depend on side of stimulation. Controls had lower drift rates to DR320 stimuli, suggesting a longer time to decision for stimuli that were more difficult to discriminate. V1-damaged patients also showed this effect in their sighted hemifields, but had equally low drift rates for DR0 and DR320 stimuli in their blind fields, suggesting that they possessed equivalent visibility in the blind field.

Figure 5

Results of the Right–Left (R–L) and Left–Right (L–R) t-contrasts in a representative set of participants (two controls and two subjects with cortical blindness). The R–L contrast is colour-coded in blue-white, whereas the L–R contrast is coded orange-yellow. Numbered arrows indicate common anatomical locations of detectable BOLD clusters around the calcarine sulcus (putative V1/V2), the inferior occipital lobe (putative V3v/V4), V3A and the occipital/temporal/parietal junction (putative hMT+). (A) Subject C3 is a control participant stimulated in the upper visual field, showing activation of the lower bank of the calcarine fissure (putative V1/V2) and a relatively symmetric response to left and right stimulation. (B) Subject C9 is also a control participant stimulated in the lower visual field to match Patient CB9, showing activation of the upper bank of the calcarine fissure (putative V1/V2), as well as putative V3A and MT+. The response of the left hemisphere to right visual field stimulation was present but much smaller. (C) Patient CB4: example of a hemianope with activation of perilesional cortex. (D) Patient CB8: example of a hemianope without a detectable response in perilesional cortex. The intact brain hemisphere responds normally, with significant clusters in the lower bank of the calcarine (expected given that stimulus presentation was in the upper hemifield), the ventral occipital cortex and putative hMT+.

There was a significant 3-way interaction for the boundary, F(2,15) = 6.134, P = 0.011 and partial η2 = 0.45 (Fig. 5B). Direction range and stimulus side interacted, F(2,15) = 9.155, P = 0.009, partial η2 = 0.379. The 3-way interaction appears to follow from the difference in the 2-way interaction between coherence and side among groups. Specifically, direction range had an effect regardless of side of presentation for controls, with observers adopting higher boundaries in the DR0 condition, but this effect was greatly attenuated in the blind field of both patient groups.

The analysis of non-decision time revealed a significant 3-way interaction, F(2,15) = 4.447, P = 0.03 and partial η2 = 0.372 (Fig. 5C). There were no other main effects or interactions. Inspection of the means (Fig. 5C) suggests that the interaction was likely because of the effect of coherence on non-decision time, which affected both sides equally in intact controls but affected only the sighted hemifield in cortically blind subjects. Specifically, non-decision times were shorter following DR0 stimuli in sighted fields, but in blind hemifields, direction range had little effect.

Functional magnetic resonance imaging results

In controls, the LEFT > RIGHT (red-yellow shading in Fig. 5) and RIGHT > LEFT (blue-green shading in Fig. 5) contrasts were consistent, with four prominent centres of activity in the hemisphere contralateral to stimulation (Fig. 5 and Table 1). In some participants, these were isolated clusters of statistically significant voxels, whereas in others, they were local maxima in a single contiguous cluster spanning several brain regions in occipital lobe and occasionally extending into temporal lobe. One of these areas lay along the calcarine sulcus contralateral to stimulation and presumably reflects activity in visual areas V1 and V2. A second extended inferiorly from this putative V1/V2 to putative V3/V4. A third focus centred around the superior occipital cortex in the cuneus gyrus and likely represents V3/V3A (Braddick et al., 2001; Zeki, 2003). The fourth source was located in lateral occipital/temporal cortex at coordinates consistent with human hMT+/V5 (Watson et al., 1993; Dupont et al., 1994; Orban et al., 1995; Rees et al., 2000). The Anatomy Toolbox (Eickhoff et al., 2005) confirmed that these were reasonable interpretations of the activated regions, identifying some degree of overlap of significant regions with Brodmann area 17/V1 (8/9 controls), V2 (8/9 controls), V3v (8/9 controls), V3A (7/9 controls), V4 (8/9 controls) and hMT+ (8/9 controls). Table 1 gives the breakdown of which areas were activated in each hemisphere of each subject.

View this table:
Table 1

Average locations of local maxima within clusters corresponding to likely anatomical locations for control participants

Anatomical locationL CalcL Inf OL CGLeft O/T/PR CalcR Inf OR CGR O/T/P
Likely identityV1/V2V3v/V4V3AhMT+V1/V2V3v/V4V3AhMT+
MNI Coordinates
    x−7.55−20.67−23.38−45.4310.3922.252449.5
    y−78.59−71.67−85.08−73.57−79.8−69−85.14−66.75
    z−1.35−10.7524.317.86−1.11−13.7522.864.75
Control participants
    C1XXXXX
    C2XXXXXXX
    C3XXXXXXXX
    C4XXXXX
    C5XXX
    C6XXXXXXXX
    C7XXX
    C8XXXXXXXX
    C9XXXXXXX
  • Coordinates refer to the MNI atlas. Coordinates are averages for the control participants. C1–C9 = control participants; Inf = inferior; L = left, R = right, Calc = Calcarine sulcus, CG = Cuneus, O = occipital lobe, T = temporal lobe, P = parietal lobe. An X denotes that the Anatomy Toolbox for SPM (Eickhoff et al., 2005) indicated some degree of overlap of a cluster with the corresponding anatomical region for that participant in the SPM corresponding to the appropriate side of stimulation (i.e. the SPM of left field stimulation for locations in the right hemisphere).

In cortically blind patients, a similar pattern to that observed in controls was obtained when the sighted hemifield was stimulated (Table 2). Stimulation of the blind field produced no detectable activity, exceeding that from stimulation of the sighted field in Patients CB1–3, CB8 and CB9. Patients CB4, 6 and 7 showed significant activation of perilesional tissue, as well as of superior occipital cortex (putative V3A) and putative hMT+. Patient CB5 had a significantly active region around hMT+ and V3A, but not in perilesional tissue. Finally, Patient CB8 had an inferior occipital active region and medial frontal activity.

View this table:
Table 2

Presence of local maxima within clusters corresponding to likely anatomical locations in participants with cortical blindness

Anatomical locationL CalcL Inf OL CGLeft O/T/PR CalcR Inf OR CGR O/T/P
Likely identityLesionV1/V2V3v/V4V3AhMT+V1/V2V3v/V4V3AhMT+
Patient ID
    CB1R
    CB2RXXXX
    CB3RXXXX
    CB5RXXXX
    CB6RXXXXXXX
    CB4LXXXXXXXX
    CB7LXXXXX
    CB8LXXXX
    CB9LXXXX
No. of lesion hemispheres2/91/92/92/91/91/9
  • CB1–CB9 = hemianopic participants. An X denotes that the Anatomy Toolbox for SPM (Eickhoff et al., 2005) indicated some degree of overlap of a cluster with the corresponding anatomical region for that participant in the SPM corresponding to the appropriate side of stimulation (i.e. the SPM of left field stimulation for locations in the right hemisphere). L = left; R = right; Calc = Calcarine sulcus; CG = Cuneus; O = occipital lobe; T = temporal lobe; P = parietal lobe; inf = inferior.

To assess the effect of criterion threshold on the pattern of results, we assessed each cortically blind participant with a more stringent and a more relaxed threshold. When family-wise error rate was held at 0.05, no perilesional or extrastriate activity was detectable in any cortically blind participant. It is worth noting that at this more conservative criterion, four visually intact controls had no detectable responses in putative area hMT+, and three had no detectable responses in striate cortex. At a more relaxed criterion of α = 0.01/voxel, and with no criterion of contiguous voxels, there was still no detectable perilesional activity in those cortically blind participants who lacked such activity when analysed at our original criterion, and again Patient CB5 was the only participant with detectable extrastriate activity.

Correlation between functional magnetic resonance imaging activity and behaviour

To correlate brain activity with behaviour in cortically blind subjects, three grouping variables were generated, based on whether regions of interest showed significant activation at the chosen threshold or not: perilesional tissue (putative V1/V2), V3A and hMT+. Grouping by hMT+ and V3A activity was completely redundant, and each of these areas in turn was largely redundant with perilesional tissue activation. There was only one exception to this rule: Patient CB5 had significant activity around hMT+ and V3A, but not in perilesional tissue. The tendency for perilesional and extrastriate activity to co-occur was statistically significant (Fisher’s exact test P = 0.048).

Because hMT+ and V3A were entirely redundant (i.e. they were always co-activated), we report the results of two analyses using perilesional activity and extrastriate (hMT+/V3A) activity to define groups. Independent samples t-tests were calculated for the following behavioural measures, separately for DR0 and DR320 stimuli: reaction time, accuracy, drift rate, boundary and non-decision time (Fig. 6). Activation of perilesional tissue was associated with greater accuracy in response to DR0 stimuli, t(7) = 2.825, P = 0.026 and Cohen’s d = 1.875, and DR320 stimuli, t(7) = 2.642, P = 0.033 and d = 2.076. It was also associated with a higher drift rate in response to blind field stimulation at DR0, t(7) = 2.689, P = 0.031 and d = 1.696 and DR320, t(7) = 2.569, P = 0.037 and d = 2.069. In contrast, the diffusion boundary and non-decision time were not detectably affected by the presence of perilesional activity.

Figure 6

Summary of the relationship between activation of perilesional cortex, putative hMT+ and putative V3A with accuracy and drift rate. (A) Comparison of accuracy in response to blind field stimulation between subjects with cortical blindness with and without detectable BOLD responses in perilesional cortex. (B) Comparison of accuracy in response to blind field stimulation between subjects with cortical blindness with and without detectable BOLD responses in putative hMT+. (C) Comparison of drift rate estimated with the EZ-diffusion model in response to blind field stimulation between subjects with cortical blindness with and without detectable BOLD responses in perilesional cortex. (D) Comparison of drift rate between subjects with cortical blindness with and without detectable BOLD responses in putative hMT+. Corresponding t-test results are given in detail in the text. Error bars represent 95% confidence intervals.

Results using presence/absence of hMT+/V3A activation as the predictor variable were similar, but the effects were smaller and non-significant for the DR0 condition. Extrastriate activity was not significantly associated with greater accuracy in response to DR0 stimuli, t(7) = 2.006 and P = 0.085, but was with DR320 stimuli, t(7) = 2.72, P = 0.03 and d = 1.885. Drift rate in the DR0 condition was also not significantly greater in patients with extrastriate activation, t(7) = 1.946 and P = 0.093, but this effect was significant in the DR320 condition, t(7) = 2.67, P = 0.032 and d = 1.863.

At an individual participant level, we used binomial tests to determine whether accuracy in a given condition deviated significantly from chance. Results are presented in Table 3. In general, participants with perilesional BOLD activity were significantly better than chance in response to stimulation of their blind fields in the DR0 condition, whereas those without perilesional activity were not. The lone exception to this rule was Patient CB1, who had no detectable BOLD response but performed well above chance in both hemifields. Nevertheless, the tendency for perilesional BOLD signal to correspond to better than chance accuracy in response to blind field stimulation was significant, Fisher’s exact test P = 0.048. Extrastriate cortex was similarly associated with greater than chance accuracy, but the association was not as strong. There were two exceptions: Patient CB1 again had no detectable BOLD response but was above chance in the DR0 condition when the blind (left) hemifield was stimulated, and Patient CB5 did have detectable extrastriate BOLD activation but was no better than chance in response to blind field stimulation. As mentioned previously, CB5 was also the lone exception to the co-occurrence of perilesional and extrastriate activation. The association between extrastriate activation and greater than chance accuracy was not significant, Fisher’s exact test P = 0.167.

View this table:
Table 3

P-values of binomial tests of accuracy for individual participants with cortical blindness

Patient IDLesionPeri BOLDExtra BOLDLVF DR0RVF DR0LVF DR320RVF DR320
CB1R0.0000.0000.3530.002
CB2R0.2860.0000.8710.03
CB3R0.2860.0000.5750.000
CB5RX0.1290.0000.4250.000
CB6RXX0.0000.0000.2260.019
CB4LXX0.0000.030.9340.066
CB7LXX0.0000.0000.5750.286
CB8L0.0000.5750.6470.981
CB9L0.0000.9550.0450.647
  • CB1–CB9 = hemianopic participants. LVF = left visual field; RVF = right visual field. Significant deviations from chance (α = 0.05) are bold. All deviations from chance were in the direction of better than chance accuracy.

Although stimulus locations within the blind field were chosen based on Humphrey perimetry to be regions of blindness (mean pattern deviation worse than −5 dB) for each subject with cortical blindness, we further tested the hypothesis that the relationship between accuracy/drift rate and lesioned hemisphere activity could be explained by spared sensitivity. In other words, did we inadvertently choose locations in some patients that were more sensitive than in others? To test this hypothesis, we computed the average sensitivity as measured by Humphrey perimetry for each stimulated location by taking the values from the perimetry record nearest to each stimulated location (measured in left and right eyes) and averaging them. We then repeated the t-tests above using these averages as the dependent variable. Neither perilesional activity [t(7) = 0.704 and P = 0.504] nor hMT + activity [t(7) = 1.942, P = 0.093] was significantly associated with sensitivity as measured by Humphrey perimetry.

Discussion

To our knowledge, this is the first report of an attempt to characterize component processing in response to blind field stimulation in cortical blindness using a diffusion process model, and the first to tie this analysis to brain activity. Brain activity in response to stimulation of the sighted hemifield caused an essentially normal pattern of contralateral activity in the intact hemisphere, with prominent foci in the calcarine sulcus (putative V1 and V2), inferior occipital lobe and cuneus (putative V3A) and lateral occipito-temporal cortex (putative hMT+). Some cortically blind subjects had contralateral (ipsilesional) activity (detectable at the particular threshold used here) in response to blind field stimulation, whereas others did not. Activity in one of the three regions (perilesional, hMT+ and V3A) tended to co-occur with activity in the other two, with only one discrepant cortically blind subject who had extrastriate but not perilesional activity. Cortically blind subjects with perilesional activity were more accurate than those without such activity in discriminating the direction of motion in their blind hemifields, and the diffusion model analysis indicated that this was because they accumulated information to make a decision at a faster rate. At the individual subject level, perilesional activity was associated with above-chance performance in response to blind field stimulation, although accuracy was still below that achieved in response to intact hemifield stimulation.

Activity in perilesional tissue, presumably corresponding to V2 and also possibly to spared regions of V1, predicted greater accuracy in response to blind field stimulation. The diffusion model analysis indicated that this was primarily because of a higher drift rate for the accumulation of information to make a decision, and did not reflect a change in boundary (i.e. a more conservative criterion) associated with perilesional activity. This is consistent with a recent report that found that hemianopic monkeys performing a detection task were slower to reject trials in which a target was presented to their blind field than trials without targets (Cowey and Alexander, 2012). In that case, the accumulation of information never (or rarely in one case) reached the decision boundary for detection, but any accumulation towards the detection decision boundary would delay the eventual crossing of the rejection boundary. This finding has a direct implication for the treatment of hemianopia. If drift rate is a critical factor in the degree of spared vision, then retraining cortically blind subjects to accumulate this information more quickly or giving them more time to accumulate information could potentiate or speed up the rate of training-induced improvements in visual direction discrimination abilities (Sahraie et al., 2006; Huxlin et al., 2009).

Our results are in good agreement with previous observations of brain activity in response to blind field stimulation. Several functional MRI studies have observed activation of perilesional cortex, usually in addition to extrastriate cortical activity, on presentation of stimuli to the blind field of V1-damaged humans (Sorenson et al., 1995; Rausch et al., 2000; Schoenfeld et al., 2002). In contrast, other studies observed activation of extrastriate cortex, usually in hMT+ in the absence of activation in perilesional cortex (Barbur et al., 1993; Stoerig et al., 1998; Goebel et al., 2001; Nelles et al., 2002). Still others obtained mixed results within the same study (Bittar et al., 1999; Morland et al., 2004). An interesting example was that of Bittar et al. (1999), who examined BOLD responses in three hemispherectomized participants with homonymous hemianopia, only one of whom had blindsight. Presentation of moving gratings against a background of moving dots in the blind field caused no changes in activity in the two participants without blindsight. However, it caused significant activation of the motion-processing areas V3/V3A and hMT+ in the intact hemisphere of the participant with blindsight.

Activity in hMT+ and V3A was associated with better performance when greater motion integration was required as in the case of the DR320 stimuli. We did not see an association between activity in hMT+ and V3A for coherent stimuli possibly because of our small sample size. Moreover, because these regions tended to occur together, and in fact were discrepant in only one V1-damaged patient out of nine, we cannot make any strong assertions regarding the relative contributions of these areas to performance. Previous reports showing good performance (blindsight) without perilesional activity have tended to do so in only one (Stoerig et al., 1998; Schoenfield et al., 2002) or two (Goebel et al., 2001) patients. It is possible that such cortically blind subjects are exceptions to the rule that perilesional responsiveness predicts good performance in visual tasks. Perilesional activity may simply be a more sensitive measure of a ‘wider’ (more diffuse) motion processing network after post-chiasmal lesions. Alternatively, perilesional responsiveness may be in some sense more important for behavioural performance following post-chiasmal lesions.

In the design of the present experiment, the duration of the motion stimulus was kept constant at 500 ms. This was done to keep the design simple, match the duration used in prior studies (i.e. Huxlin et al., 2009) and provide sufficient time for accumulation of information, while minimizing the probability of both stimulus adaptation and ‘cheating’ via saccadic eye movements. Given that the rate of information accumulation was identified as an important factor in performance for individuals with some degree of spared function, it would be important for future studies to test the effects of longer (and shorter) stimulus durations on brain activity and psychophysical performance. If longer durations do indeed afford sufficient accumulation of information for above-chance direction discrimination, then this approach may represent a promising avenue for future rehabilitation strategies.

The criterion we chose for detecting BOLD responses that deviated from those expected by chance was a compromise between the risks of Type I and Type II errors. The most conservative criterion of holding the total family-wise error rate to 0.05 would have indicated no perilesional or extrastriate BOLD responses to blind field stimulation in cortically blind participants, but also would have abolished otherwise reasonable detections in visually intact controls, indicating low statistical power and a concomitantly high risk of Type II errors. On the other hand, even a much more liberal criterion (α = 0.01) resulted in no additional detectable activity in cortically blind participants in the critical regions of interest. Thus, the associations that we report between BOLD responses and behaviour do not appear to depend significantly on the specific criterion we chose, which represented a useful compromise between Type I and II error risks.

The individual binomial tests reported in Table 3 might be argued to be subject to alpha inflation because of multiple tests. We did not correct for alpha inflation in this case for several reasons. First, the purpose of the analysis was to inform the group-level result with the individual-level pattern, not the strong assertion of any one individual result. The application of Bonferroni corrected α = 0.05/36 = 0.0014 would eliminate only one significant result that is important to our argument, that of Patient CB4’s response to left (blind) visual field stimulation in the DR0 condition. The other marginal deviations from chance in Table 3 are in response to stimulation of intact hemispheres. This correction would render the Fisher’s exact test of covariation between detectable lesion activity and above-chance accuracy insignificant, but the qualitative pattern at the individual level and the significant group-level result (higher accuracy with than without detectable perilesional BOLD) remain. Second, the table is analogous to a correlation matrix, which is not typically corrected for the number of correlations. Cohen and Cohen (1983, pp. 57–8) recommend performing an omnibus test of the correlation matrix before assessing individual correlations, to determine whether there is any deviation from chance in the matrix. The family of tests in Table 3 clearly exceeds the number of significant results expected by chance in 36 tests (1.8). Therefore, although it might be argued that Table 3 provides weak evidence in favour of the relationship between BOLD responses to blind field stimulation and accuracy, it is merely an adjunct to the strong evidence of the group-level result.

Conclusion

We observed a relationship between perilesional and extrastriate activity and the rate of information accumulation as estimated by the EZ-diffusion model (Wagenmakers et al., 2007) on stimulation of cortically blind fields with global motion stimuli. These results identify a specific problem with visual information processing in cortically blind fields. They also suggest that preservation of activity in spared striate and extrastriate visual areas, especially those normally involved in motion processing, is significantly correlated with better visual performance within cortically blind fields. Thus, visual rehabilitation that focuses on optimizing the rate of information accumulation might be effective in bringing about improved vision in at least those cortically blind subjects who retain adequate, supporting brain structures. This might be done, for example, by extending the duration of motion stimuli at the beginning of treatment and then incrementally shortening stimulus duration as performance improves at a given visual field location.

Funding

This work was supported by an unrestricted grant from Research to Prevent Blindness to the University of Rochester Eye Institute, a grant from the Pfeiffer foundation (to K.R.H.), National Institutes of Health Training Grant #08T2EY07125C-13 (to T.M.), National Institutes of Health Loan Repayment Program Award L30 EY01773 (to T.M.) and National Institutes of Health Core Grant P30EY0131 (to K.R.H and T.M.).

Acknowledgements

The authors thank Pat Weber for outstanding MRI measurement, and Terry Schaeffer and Dorothea Castillo for the Humphrey visual field perimetry. They also thank Dr. Keith Schneider and Dr. Daphne Bavalier for advice on functional MRI research design.

Abbreviations
BOLD
blood oxygen level-dependent

References

View Abstract