Elsevier

NeuroImage

Volume 27, Issue 1, 1 August 2005, Pages 95-105
NeuroImage

Multifocal fMRI mapping of visual cortical areas

https://doi.org/10.1016/j.neuroimage.2005.01.046Get rights and content

Abstract

The multifocal mapping of electroretinograms and visual evoked potentials has established an important role in both basic research and in diagnostic procedures. We have developed a multifocal mapping method for fMRI, which allows detailed analysis of multiple local visual field representations in the cortex with excellent spatial resolution. Visual field was divided into 60 regions in a dartboard configuration, scaled according to the human magnification factor. Within blocks of 7 s, half of the regions were stimulated with checkerboard patterns contrast reversing at 8 reversals per second, while the other half remained inactive at uniform luminance. The subset of active regions changed with each 7-s block, according to an orthogonal design. Functional MRI was done with a 3-T GE Signa and analyzed with SPM2. A general linear model was fitted producing activation maps for each of the 60 regions, and local signal changes were quantified from V1. These activation maps were next assigned to 3D surface models of the cortical sheet, and then unfolded, using the Brain à la Carte software package. Phase-encoded retinotopic analysis of conventional design served as qualitative comparison data. With multifocal fMRI, all regions were mapped with good signal-to-noise ratio in V1, and subsets of regions showed activation in V2 and V3. This method allows rapid and direct exploration of multiple local visual responses, and is thus able to give complementary information to phase encoded mapping of retinotopic areas.

Introduction

In the primary visual cortex, V1, the visual environment is represented retinotopically, i.e., nearby locations on the retina project to nearby points in the cortex, with each visual hemifield being mapped in the contralateral hemisphere. The successive areas in the anatomical hierarchy, such as V2, V3/VP, V3a, and V4v, are also retinotopically organized, but at lower magnification and with larger receptive fields. The standard mapping procedure for these areas includes stimulation with flickering and slowly moving checkerboard, separately for polar coordinate and eccentricity, and then Fourier transform of the resulting data. After Fourier transform, the phase of the oscillatory response represents the position in the visual field. This method is called the phase-encoded or traveling wave method. Whereas the borders of retinotopic areas can be robustly mapped with phase-encoded visual stimuli (DeYoe et al., 1996, Engel et al., 1994, Engel et al., 1997, Sereno et al., 1995, Warnking et al., 2002), it is unclear how well the phase information represents local activation within each retinotopic area. Engel et al. (1997) showed that a traveling wave of activity can be replicated within individual subjects with 1.1 mm precision along the cortical surface. However, the estimated phase, and thus the retinotopic position on the cortex, depends on the noise, and on the temporal and spatial sampling limits of the BOLD signal. Local retinotopic representation can be evaluated from phase-encoded retinotopic maps, but this is very difficult from raw phase maps (Dougherty et al., 2003). Instead, a global model assuming continuity in the retinotopic representation is important for accessing the local representation.

In contrast, by stimulating discrete regions of the visual field, one can reliably define the center of mass of activated cortex for each region, and thus the corresponding retinotopic representation. In addition, the local quantity of BOLD signal can be estimated for each region. In a number of studies, a small number of visual field regions have been sequentially stimulated in separate blocks (Fox et al., 1987, Schneider et al., 1993, Shipp et al., 1995, Tootell et al., 1995). The practical limitation of local stimulation with block or event-related designs is the excessive measurement time needed to map multiple positions.

Multifocal methods have been developed for the electrophysiological analysis of visual systems to allow a large number of visual field regions to be assessed in a reasonable recording time. Multifocal methods refer to the concurrent presentation of sequences of stimuli in multiple visual field locations (foci), with an analysis procedure that decomposes the resulting compound response signal into components attributable to each stimulus region. They have been widely used to study maps of electroretinogram response across the retina, generally based on the work of Sutter using maximal length shift register (m-sequence) stimuli (Sutter, 2001, Sutter and Tran, 1992). Multifocal m-sequences have also been used to study cortical visual evoked potentials (Baseler and Sutter, 1997, Baseler et al., 1994, Slotnick et al., 1999, Slotnick et al., 2001), including diagnostic procedures for disorders of the visual system (Hood and Zhang, 2000, Hood et al., 2000a, Hood et al., 2000b, Klistorner et al., 1998).

The m-sequence method relies on particular mathematical properties of these sequences of having near-zero autocorrelation at all lags so that the cross-correlation between stimulus sequence and response signal gives good estimates of the temporal weighting functions, or kernels, of the system under study. A more general scheme was presented by James (2003), allowing a flexible choice of stimulus design tailored to the particular system and questions under consideration, and using multiple linear regression to determine the components due to each stimulus region and condition.

The aim of the present study was to develop a multifocal analysis technique for fMRI. The primary goal is to create a tool, which would be easy to use and would give reliable information about the local representation of the visual field in the retinotopic areas in reasonable measurement time, and thus serve as a complementary tool for phase-encoded mapping. A relatively literal translation of the m-sequence method to the fMRI domain can be done, but will tend to have short presentation times that do not generate a strong detection power (Liu, 2004). Furthermore, the use of cross-correlation does not take into account the presence of autocorrelation in the noise of fMRI data, leading to less efficient estimates. Kellman et al. (2003) have applied this approach, and they were able to distinguish foveal from peripheral BOLD responses. Recently, Hansen et al. (2004) divided the visual field to 24 discrete regions and found linear spatial summation in V1. In our study, we have developed a method tailored to generate a powerful BOLD response, taking advantage of the general linear model (GLM) estimation capabilities present in the SPM package (Frackowiak et al., 1997) to derive efficient estimates of the region-specific activation maps. This has allowed the parallel mapping of 60 visual field regions in reasonable time. Phase-encoded retinotopic mapping served as reference data for defining the borders of functional areas, and in a control experiment these methods were compared directly. Part of this data has been presented earlier in abstract form in HBM 2004 meeting.

Section snippets

Subjects and stimuli

Six subjects (3 females, ages 23–39 years) with normal or corrected-to-normal vision gave informed consent to participate in the study. The study was approved by an ethical committee of the Hospital District of Helsinki and Uusimaa.

The stimulus images were generated with Matlab™ (Mathworks Inc.), and timing was controlled with Presentation™ (Neurobehavioral Systems Inc.) software. The stimuli were presented via a back projection system, with a data projector with three micromirrors (Christie

Results

Fig. 2 shows examples of single region activation SPM(t) maps for Subjects 2 and 5. Fig. 2a illustrates activation due to region 29, in the upper left quadrant, for Subject 2. Localized activation is clearly visible in the MRI images. In Fig. 2b, the 3D activation is assigned to the anatomical reconstruction of the surface, appearing on the ventral lip of the calcarine sulcus of the right hemisphere. In Fig. 2c, the cortex is unfolded and shows the location of activation in relation to the

Discussion

This study extends the potential of multifocal visual stimulation method to functional MRI. We were able to acquire significant local signals in V1 from each of the 60 stimulated regions from all subjects. Because the mffMRI approach measures directly local activation in V1, it can complement the standard phase-encoded mapping, where local retinotopic representation has to be mapped more indirectly, e.g., via modeling the phase distribution (Dougherty et al., 2003). The multifocal method can

Acknowledgments

We thank Raimo Joensuu and Antti Tarkiainen for technical support and Marita Kattelus for help in the measurements. Linda Stenbacka segmented the anatomical MR images and did the phase-encoded analysis of the retinotopic areas on most subjects. We are most grateful for unite mixte INSERM/Université Joseph Fourier 594, who provided the Brain á la Carte toolbox for surface oriented analysis. This work has been supported by Academy of Finland grant 105628, Finnish Medical Foundation, Sigrid

References (36)

Cited by (65)

View all citing articles on Scopus
View full text