Identification of the ventral occipital visual field maps in the human brain

1. Data acquisition

1.1. Anatomical data. Anatomical data are needed in order to render retinotopic data on surface representations. For subjects 1–3 depicted in the video presentation, 2–4 whole brain SPGR T1-weighted MRI images were acquired on a 3T GE scanner at 1 mm isotopic resolution using an 8-channel whole-brain coil. The multiple images for each subject were aligned and averaged. Averaging multiple anatomical images is desirable as it increases the contrast of the boundary between the grey and white matter, and therefore aids segmentation and the creation of an accurate cortical surface.

1.2. Functional data. Functional data were acquired as T2*-weighted gradient echo images with either a spiral²⁴ or rectilinear (EPI) trajectory through k-space, with 2.5 mm isotropic voxels and a TR of either 1.5 s or 2 s. (For details see ref 20.)

2. Stimuli

Retinotopy was measured with stimuli designed either for travelling wave analysis²⁵ or population receptive field modeling^26,27. For traveling wave analysis, alternate scans contained contrast patterns viewed through expanding ring apertures or clockwise rotating wedges with a visual field extent of 14 degrees from fixation. (For details see ref 20.) For population receptive field modeling, stimuli were contrast patterns viewed though moving bar apertures that slowly swept across the visual field, with a maximum eccentricity of 6 degrees from fixation. (For details see refs 15,26.)

3. Functional MRI analysis

3.1. Preprocessing. Several standard fMRI preprocessing steps were used prior to retinotopic analysis, including slice timing correction, motion correction, and high pass temporal filtering with a 1/20 Hz cutoff.

3.2. Travelling wave analysis. Data from traveling wave scans (wedges and rings) were analyzed voxel by voxel by applying the Fourier transform to the time series. The phase at the stimulus frequency measures the polar angle or eccentricity of the stimulus that most effectively drives the BOLD response in that voxel. The coherence of the signal at the stimulus frequency indicates the goodness of fit of the response to the periodic design. Coherence is defined as the power at the stimulus frequency divided by the power across all frequencies. (For details see ref 28.)

3.3. Population receptive fields. A population Receptive Field (pRF) model was solved for each voxel using methods described previously²⁶. In brief, each pRF was defined as a 2-D isotropic Gaussian, parameterized by its center (x, y), size (sd of Gaussian), and amplitude (scale factor). The model prediction is the dot product of the pRF and the stimulus aperture, convolved with a hemodynamic response function. The model was fit through a minimization of least squared error between the predicted and observed time series using a coarse-to-fine approach. The pRF center can be converted to polar coordinates to yield an angle and eccentricity map, similar to that obtained from traveling wave analysis. For the purposes of retinotopic mapping in this paper, other pRF parameters were not used (size and amplitude).

4. Segmentation and visualization

Because the retinotopic maps in visual cortex are organized on the 2D surface, not the 3D volume, map data is best visualized on a surface representation, typically rendered as an inflated (smoothed) or un-inflated 3D surface mesh, or as a flattened (2D) surface^29–31. The data used for the surface rendering is usually the border of the gray and white matter, derived from a segmentation procedure that labels anatomical voxels from the T1 image as gray or white matter, and then finds the boundary that separates the two tissues. For the data shown in this paper, we segment the gray and white matter using the Freesurfer auto-segmentation tools³² followed by manual correction using ITK-GRAY software, modified from ITK-SNAP³³. Functional data were then aligned to the whole brain anatomical data and rendered on a smoothed surface using custom software (vistasoft; http://vistalab.stanford.edu/software/).

5. Anatomical landmarks

We find it helpful to identify several sulci and gyri before drawing the retinotopic maps, and follow the naming conventions used by Duvernoy³⁴. The sulci and gyri labeled in Figure 2 can all be seen in post-mortem brains in Duvernoy‘s text (e.g., his Figure 17). Their locations are described briefly below.

Figure 2. Anatomical landmarks in ventral occipitotemporal cortex.

Several sulci and gyri are labeled on the ventral and medial surface of a highly smoothed right hemisphere. These sulci and gyri provide useful landmarks in the identification of the visual field maps. Calc: Calcarine sulcus; Ling: Lingual sulcus/gyrus; CoS: Collateral sulcus; Fus: Fusiform gyrus; ptCoS: Posterior transverse collateral sulcus; IOG: Inferior occipital gyrus.

5.1. Calcarine sulcus. The most useful landmark is the calcarine sulcus (sometimes called the calcarine fissure), which locates primary visual cortex (V1). It is the large sulcus in medial occipital cortex separating the lingual gyrus from the cuneus. It can be found unambiguously in every subject and is always the location of V1. It runs on an anterior-posterior axis, with the anterior end marked by the parietal occipital fissure, and the posterior end approximately at the occipital pole, though there is variability in the posterior extent, with the sulcus wrapping around to the lateral surface in some subjects, and terminating on the inferior surface in others^35,36.

5.2. Lingual gyrus and lingual sulcus. The lingual gyrus is inferior to the calcarine sulcus and is the location of the ventral portions of V2 and V3. Its inferior boundary is the collateral sulcus. It stretches approximately from the posterior pole to the parahippocampal gyrus. The lingual gyrus contains one or more sulci together, considered the lingual sulcus.

5.3. Collateral sulcus. The collateral sulcus separates the lingual and parahippocampal gyri on the one side from the fusiform gyrus on the other. Its posterior end is usually denoted by a transverse portion of the sulcus, called the posterior transverse collateral sulcus (ptCoS). The ptCoS is an important landmark because it usually marks the division between the hV4 map and the VO-1 map²⁰.

5.4. Fusiform gyrus. The fusiform gyrus is inferior to the lingual and parahippocampal gyrus, and is bounded medially by the collateral sulcus. The foveal representation of the ventral-occipital maps (VO-1/2) lies in the posterior portion of the fusiform gyrus.

5.5. Inferior occipital gyrus. The inferior occipital gyrus lies posterior to the fusiform gyrus and collateral sulcus and inferior to the lingual gyrus. It is generally separated from the posterior portion of the posterior fusiform gyrus by the posterior transverse collateral sulcus. The lateral portion of the hV4 map often lies here. This gyrus is typically in close proximity to the transverse venous sinus, which can lead to fMRI artifacts that obscure this portion of the hV4 map¹⁵.

6. Drawing the maps

6.1. V1–V3 maps. Identification of the first three visual field maps in humans was one of the first accomplishments in visual neuroscience following the advent of fMRI^25,28,37. Delineation of these map boundaries is now routine, although tracing the maps through the foveal confluence can be difficult in the absence of methods developed specifically for this purpose¹⁰. In brief, the V1 map is centered on the calcarine sulcus, and its borders with V2 lie on retinotopic representations of the vertical meridian. The peripheral boundary is usually not identified with fMRI because the field of view that can be achieved during scanning is less than the field of view represented in the maps. Hence the peripheral boundary of a region of interest is usually chosen as the most anterior region of the maps where the signal is good, according to a threshold in coherence (traveling wave analysis) or variance explained (pRF analysis). If there are clear reversals in the angle map all the way back to the occipital pole, then it may be possible to trace V2 and V3 in concentric ‘V’ shapes around V1; otherwise the most foveal portion of the maps is typically labeled as a non-specific ‘foveal confluence’, and each map is traced as far into the fovea as the resolution in the angle maps allow, usually one or two degrees from fixation, such that the two arms of the ‘V’ are not connected, and labeled as the dorsal and ventral portions of the map The dorsal and ventral V2/V3 borders are marked by a representation of the horizontal meridian in the angle map.

6.2. The hV4 map. The hV4 map lies on the ventral surface of the occipital lobe and borders at least two other visual field maps, the ventral part of V3 (V3v) and VO-1.

6.2.1. The anterior boundary: hV4/VO-1 The boundary between hV4 and VO-1 can be identified by both anatomical and retinotopic data. The retinotopic feature that defines this border is a reversal in the eccentricity map (Figure 3). This peripheral eccentricity band dividing hV4 from VO-1 usually coincides with a particular anatomical feature, the ptCoS²⁰.

Figure 3. Eccentricity map.

The color overlay on the smoothed cortical mesh in the main panel shows the stimulus eccentricity that most effectively drives each cortical location. The eccentricity was derived from solving a pRF model, and the stimulus extent was limited to 6 degrees. The asterisk indicates the confluent foveal representation. Black and white lines mark the boundaries between visual field maps. The white line divides hV4 from VO-1 and is the only boundary line in this figure that is derivable from the eccentricity map. This line coincides with the ptCoS (inset) and with the eccentricity reversal (blue in color overlay). Hence the white line also divides the ventral occipital maps into two clusters, one that includes V1-hV4, and one that includes VO-1/2¹⁷. Data are limited to voxels in which the pRF prediction accounted for at least 10% of the variance explained in the time series and to a posterior mask that includes the 6 labeled visual field maps: V1, V2 and V3 ventral, hV4, VO-1, and VO-2. The inset is the same cortical mesh with labels indicating the major sulcal and gyral patterns, as in Figure 2.

6.2.2. The medial boundary: hV4/V3v One of the borders of hV4 is shared with the ventral portion of V3 (V3v). This border is defined by an upper vertical meridian polar angle reversal (Figure 4). Because the V1–V3 maps are typically well defined by the retinotopic data, it is usually also clear where this boundary is found.

Figure 4. Angle map.

The color overlay in the main panel indicates the angle in the visual field that most effectively drives responses in each cortical location. The visual field map boundaries in white can be identified from the angle map. Otherwise as Figure 3.

6.2.3. The ventral/lateral boundary: The ventral/lateral boundary of hV4 is more difficult to identify than the other boundaries because there is no unambiguous feature of the retinotopic data to define it. As described in 6.2.1 and 6.2.2, we know what is on the other side of two of the hV4 boundaries, making those boundaries well defined: The anterior boundary is defined by an eccentricity reversal and is shared with VO-1, and the medial boundary is defined by a polar angle reversal and is shared with V3v. In contrast, there is not a well-established map or retinotopic feature abutting the ventral side of hV4. Furthermore, there is often signal dropout in the fMRI measure on the ventral / lateral aspect of the hV4 map due to the transverse venous sinus, which typically runs near the inferior occipital gyrus¹⁵. In some cases, the signal dropout defines the most ventral/lateral extent of the map that can be identified (Figure 5 and Figure 6).

Figure 5. Variance explained map.

The color overlay in the main panel indicates variance in the voxel time series explained by the pRF model predictions. Otherwise as Figure 3. Some locations with low variance explained are likely due to fMRI artifacts, such as the region indicated by the white line, where the low variance explained is caused by dropout from the transverse sinus. Other regions with low variance explained may have visual field representations outside the stimulus extent, such as the anterior edge of V1, V2, and V3.

Figure 6. Mean signal map.

The color overlay in the main panel indicates the mean fMRI signal. The signal is not uniform across cortex. Certain regions have low mean signal due scanning artifacts. The region indicated by the white line lies near the transverse venous sinus which causes signal dropout. Retinotopic data from these locations must be interpreted with caution. Otherwise as Figure 3.

6.2.4. Iso-eccentricity lines shared by hV4 and V3v: The general organization of the hV4 map and its neighbors can be understood by tracing iso-eccentricity and iso-angle lines. The hV4 eccentricity map is in register with the V1–V3 maps, so that iso-eccentricity lines span the hV4/V3v border (Figure 7). Unlike the V1–V3 maps, the hV4 map appears to contain little representation of the far periphery; this may be because the neurons in hV4 do not respond to stimuli in the periphery, or it may be because peripheral representation is highly compressed into a small amount of cortex, such that responses to more foveal stimuli have a much larger effect on the BOLD time course, masking the peripheral representation. Along the posterior-to-anterior axis, hV4 is therefore much shorter than V1–V3, and the most anterior eccentricity band in hV4 crossed V3v near the middle of the length of V3v.

Figure 7. Iso-eccentricity lines.

Iso-eccentricity lines in visual cortex are continuous within clusters, such as the posterior cluster containing V1, V2, V3, and hV4, and the ventral occipital cluster containing the VO-1/2 maps. The eccentricity overlay and the map boundaries are identical to those in Figure 3.

6.2.5. Iso-angle lines shared by hV4 and VO-1: The iso-angle lines in hV4 are continuous with the VO-1 maps and not the V3v map. The lines usually bend along the hV4/V0-1 border, with the lower meridian representation being the most ventral and the shortest iso-angle line (Figure 8).

Figure 8. Iso-angle lines.

The iso-angle lines in hV4 and VO-1, indicated by white lines, are continuous across the two map clusters. Otherwise as Figure 4.

Regularity of maps and anatomy

Identifying visual field maps is an important component of characterizing the organization and function of visual cortex. It is important to have well-justified and reproducible methods to define the maps. In the case of the V1–V3 maps, the functional and anatomical organization is sufficiently regular and well understood that these maps can be identified using automated procedures (no human intervention)^11,12. The hV4 and VO maps are not yet included in automated fitting procedures; however, recent progress suggests that these maps too have a high degree of regularity. Two boundaries are well defined by retinotopic features (sections 6.2.2, 6.2.3), and one of these also coincides with an anatomical landmark, the ptCoS. Moreover, the internal structure of the hV4 map and its neighbors are well understood, such that the iso-eccentricity and iso-angle lines derived from retinotopic mapping follow regular patterns.

Variability across subjects

While most of the large-scale structures in the ventral occipital retinotopic maps are similar across subjects, there are also individual differences. Some of these differences likely reflect quantitative differences between subjects in the size and layout of the maps. Other differences reflect various sources of measurement noise. In no case will a measured map exactly match a template, and in some cases map boundaries will be ambiguous. We believe that the best approach offered is to simultaneously satisfy as many of the constraints from the anatomy, eccentricity, and angle maps as possible.