**2. The measurement and organization of visual cortex in healthy young adults**

In order to carefully evaluate alterations of visual cortex in healthy aging and age-related neurodegenerative diseases such as AD, we must first have an accurate and detailed under‐ standing of the characteristics of visual cortex in healthy young adults. Visual cortex encom‐ passes nearly 20% of the human cortical sheet, and studies are demonstrating that nearly all of it is organized into VFMs [1, 3]. Because many calculations are necessary to produce our visual world, our brains have many specialized VFMs which perform one or more of those calculations [3, 89]. Most—if not all—of these calculations are performed across the entire visual scene; color perception, for example, occurs throughout the visual field, not just in the lower right quadrant. Ultimately, it will be very interesting to investigate changes in visual cortex among all levels of the cortical visual hierarchy, from low-level visual processing in primary visual cortex (V1) to mid-level specialization like motion processing to higher-level processing like face recognition and visual-spatial attention [2, 90]. For now, studies of plasticity and neurodegeneration primarily investigate the lower level visual areas like V1, V2, V3, and hV4, as these VFMs are particularly well established in healthy young adults, relatively uncontroversial, and typically easily measured across most types of patients [1, 3, 8, 12, 14, 15, 37, 91].

**Figure 1. Schematic of measurements of an individual voxel. (A)** Within a typical voxel measured with a standardly used 3T MRI scanner, there are on the order of ∼1 million neurons, depending on the size of the voxel. For voxels in retinotopic visual cortex, the neurons each have similarly located spatial receptive fields (*teal circles with black outlines*) with preferred centers (*black dots*). Note how the overlapping receptive fields concentrate coverage in one region of visual space (*darker teal*) corresponding to the average receptive field of the group, as shown in **(D)**. **(B)** Each typical voxel is on the order of 1 × 1 × 1 mm for structural measurements and 1 × 1 × 3 mm for visual field mapping functional experiments, though voxels are often slightly larger (e.g., 3 × 3 × 3 mm) for other types of functional MR studies. **(C)** Traveling wave retinotopy (TWR) utilizes the organization of retinotopic cortex, in which neighboring neurons have preferred centers (*black dot*) representing similar portions of visual space, to estimate the average preferred representa‐ tion (center) for the population of neurons in a given voxel (e.g., [1, 3]). **(D)** Population receptive field (pRF) modeling similarly utilizes retinotopic organization to estimate not only the preferred center (*black dot*) in a given voxel, but also the average receptive field—the population receptive field (*teal circle*)—for the voxel's population of neurons (e.g., [1, 3, 92]).

#### **2.1. Human visual field map measurements use simple stimuli and tasks, but provide exquisite detail of cortical organization**

We briefly review here two of the most powerful fMRI techniques for very detailed measure‐ ments of VFMs in individual subjects, traveling wave retinotopy (TWR) [87] and population receptive field (pRF) modeling [92], to demonstrate what types of measurements are possible and how to interpret the existing literature regarding changes in aging or damaged visual cortex (**Figures 1**–**4**).

**2. The measurement and organization of visual cortex in healthy young**

In order to carefully evaluate alterations of visual cortex in healthy aging and age-related neurodegenerative diseases such as AD, we must first have an accurate and detailed under‐ standing of the characteristics of visual cortex in healthy young adults. Visual cortex encom‐ passes nearly 20% of the human cortical sheet, and studies are demonstrating that nearly all of it is organized into VFMs [1, 3]. Because many calculations are necessary to produce our visual world, our brains have many specialized VFMs which perform one or more of those calculations [3, 89]. Most—if not all—of these calculations are performed across the entire visual scene; color perception, for example, occurs throughout the visual field, not just in the lower right quadrant. Ultimately, it will be very interesting to investigate changes in visual cortex among all levels of the cortical visual hierarchy, from low-level visual processing in primary visual cortex (V1) to mid-level specialization like motion processing to higher-level processing like face recognition and visual-spatial attention [2, 90]. For now, studies of plasticity and neurodegeneration primarily investigate the lower level visual areas like V1, V2, V3, and hV4, as these VFMs are particularly well established in healthy young adults, relatively uncontroversial, and typically easily measured across most types of patients [1, 3, 8, 12, 14, 15,

**Figure 1. Schematic of measurements of an individual voxel. (A)** Within a typical voxel measured with a standardly used 3T MRI scanner, there are on the order of ∼1 million neurons, depending on the size of the voxel. For voxels in retinotopic visual cortex, the neurons each have similarly located spatial receptive fields (*teal circles with black outlines*) with preferred centers (*black dots*). Note how the overlapping receptive fields concentrate coverage in one region of visual space (*darker teal*) corresponding to the average receptive field of the group, as shown in **(D)**. **(B)** Each typical voxel is on the order of 1 × 1 × 1 mm for structural measurements and 1 × 1 × 3 mm for visual field mapping functional experiments, though voxels are often slightly larger (e.g., 3 × 3 × 3 mm) for other types of functional MR studies. **(C)** Traveling wave retinotopy (TWR) utilizes the organization of retinotopic cortex, in which neighboring neurons have preferred centers (*black dot*) representing similar portions of visual space, to estimate the average preferred representa‐ tion (center) for the population of neurons in a given voxel (e.g., [1, 3]). **(D)** Population receptive field (pRF) modeling similarly utilizes retinotopic organization to estimate not only the preferred center (*black dot*) in a given voxel, but also the average receptive field—the population receptive field (*teal circle*)—for the voxel's population of neurons (e.g., [1, 3,

**2.1. Human visual field map measurements use simple stimuli and tasks, but provide**

We briefly review here two of the most powerful fMRI techniques for very detailed measure‐ ments of VFMs in individual subjects, traveling wave retinotopy (TWR) [87] and population

**adults**

276 Update on Dementia

37, 91].

92]).

**exquisite detail of cortical organization**

**Figure 2. Phase-encoded fMRI paradigm for visual field map measurements. (A)** An example phase-encoded experi‐ mental design [87]. Top diagram shows the components of a single block of one stimulus presentation (*teal*) for one position (i.e., stimulus phase) of an expanding ring stimulus composed of a black and white moving checkerboard pat‐ tern. Middle diagram shows six blocks (*teal*) grouped together into one stimulus cycle (*purple*). The progressively larg‐ er ring stimulus is shown above each block. Each block in one cycle presents a specific stimulus in the "phaseencoded"—or "traveling wave"—sequence. Each cycle then repeats the same set of blocks. Lower diagram shows a full, single scan comprising six cycles; each cycle is one purple block. **(B)** Schematic diagram of three example phaseencoded time series with different stimulus responses. Each row depicts the time series measurement for a single voxel arising from a single, six-cycle scan using one experimental stimulus (e.g., polar angle). Simulated raw data points showing percent blood-oxygen-level-dependent (BOLD) modulation (i.e., response amplitude) are denoted by the black dots. The teal dotted lines represent sinusoidal fits of the simulated data points; each teal line characterizes the average BOLD activation in a different example voxel. The red lines indicate the peak activations per cycle for this imaginary set of voxels. Top and middle rows represent time series of voxels with the same %BOLD modulation, but different timing of peak responses, which indicates different stimulus selectivity (i.e., responses to different "phases" of the stimulus). Note the offset of the red lines between the two rows. For example, the top row might represent a voxel with a preferred eccentricity tuning of 2° eccentric to fixation, whereas the middle row might have a preferred tuning of 5° eccentric to fixation. Middle and bottom rows represent time series of voxels with the same timing of peak responses, indicating matching stimulus selectivity; i.e., both might have a preferred eccentricity tuning of 5° eccentric to fixation. However, the bottom row has much lower %BOLD modulation than the middle row. Such a difference in response amplitude can be due to several factors, such as differences in local vasculature or receptive field tuning. **(C)** Diagram of three example Fourier power spectra corresponding to the schematic time series in **(B)**. In the phase-encod‐ ed paradigm, only BOLD responses that match the stimulus frequency of six cycles per scan (*red peak*) are considered as data. The responses must also be above a predetermined statistical threshold, typically measured in coherence or percentage variance explained [3, 87, 92]. **(D)** Diagram of three example averaged stimulus cycles corresponding to the schematic Fourier spectra in **(C)** and to averages of the time series in **(B)**. Each teal dotted line represents the sinusoidal fit for the average, while the peak activation is again marked by the red line. The timing of the peak of each averaged cycle is used to calculate the phase of the preferred stimulus independently for each voxel. Typical pseudocolor over‐ lays on 3D or flattened brain renderings as shown in **Figures 6** and **12** use color to denote cortical responses to this peak activation. Note how the top measurement has an earlier peak (*red line*) that corresponds to an earlier phase of the stimulus (i.e., an earlier presentation time in the cycle) while the middle and bottom measurements' peaks are shifted to later in time (e.g., [3]). The bottom example has a lower %BOLD modulation than the other two schematics, but the same peak activation as the middle example. For additional discussion, see [1, 3, 87, 91, 92].

**Figure 3. Population receptive field modeling.** The parameter estimation procedure for the population receptive field (pRF) model is shown as a flow chart [92]. The pRF parameters are estimated from time series measurements using a linear spatiotemporal model of the fMRI response, which is a reasonable approximation over a wide range of spatio‐ temporal conditions [94–96]. The neuronal model is estimated by finding the model parameters that best predict the fMRI measurements. **(A)** The pRF modeling analysis is applied to all voxels (1:*n*) within the defined region of interest (ROI; e.g., occipital lobe, V1). **(B)** Multiple models of the expected average receptive field of the neuronal population may be used. Most commonly, a two-dimensional Gaussian is estimated, which is defined by three parameters, *x*, *y*, and *σ*, where (*x*, *y*) denotes the pRF center within the visual field, and *σ* is the Gaussian spread (i.e., pRF radius). Note that these parameters are stimulus-referred in degrees of visual angle. **(C)** The example moving bar stimulus com‐ posed of a black and white moving checkerboard pattern moves systematically across visual space. **(D)** The overlap between the pRF and effective stimulus is determined. **(E**, **F)** Next, the predicted pRF response is calculated for a given pRF model and effective stimulus location. The time series model prediction **(F)** is estimated by convolving the pRF activation with a model of the hemodynamic response function (HRF) as in **(E)** [94, 97]. The goodness-of-fit is estimat‐ ed by computing the residual sum of squares (RSS) between this prediction and the data (*black dots*) **(G)**. **(H)** 100,000 different fMRI time series predictions are iteratively tested by varying the pRF model parameters across a wide range of plausible values (e.g., locations across visual space covered by the stimulus; varying pRF sizes). Ultimately, the opti‐ mal pRF parameters (*x*, *y*, *σ*) are found for each voxel independently by minimizing the RSS using a two-stage, coarseto-fine search. Adapted from **Figure 2** in [92].

#### *2.1.1. Traveling wave retinotopy*

Developed in the 1990s, TWR is still the primary fMRI paradigm used to measure early VFMs like V1–3 (**Figures 1** and **2**) [13, 15, 87, 91, 93]. This technique uses two types of periodic stimuli that move smoothly across a contiguous region of visual space to measure the orthogonal dimensions of polar angle (i.e., "around the clock") and eccentricity (i.e., center to periphery; **Figure 5**). These stimuli are typically composed of a set of high-contrast, flickering checker‐ board patterns that are designed to maximally stimulate primary visual cortex and generally elicit an fMRI signal modulation on the order of 1–3% (**Figure 2A**). This modulation is typically 15–20 standard deviations above the background noise. Each voxel's preferred polar angle representation is measured by a rotating wedge stimulus, which extends from the central fovea to more peripheral regions and covers a small section of polar angles (**Figure 6C**, *central* *inset*). This wedge stimulus revolves either counterclockwise or clockwise in discrete steps around the central fixation point, successively stimulating distinct polar angle representations of visual space [3]. Each voxel's preferred eccentricity representation is measured by an expanding ring stimulus, which expands in discrete even steps between the central fovea and the periphery of the visual field (**Figure 6B**, *central inset*). These functional data are represented as color-coded overlays on anatomical data that demark the voxel's preferred polar angle or eccentricity (**Figure 6B** and **C**). The accurate delineation of VFMs relies upon the measurement of these two, orthogonal dimensions—polar angle and eccentricity, which produce a unique mapping between a location in visual space and the preferred responses of the neurons within a single voxel in cortex. If only a single measurement is obtained (e.g., of only polar angle or only eccentricity), the representation in cortex could only be associated with a wide section of visual space and would not lead to the correct definition of VFM boundaries (**Figure 5**) [3].

**Figure 3. Population receptive field modeling.** The parameter estimation procedure for the population receptive field (pRF) model is shown as a flow chart [92]. The pRF parameters are estimated from time series measurements using a linear spatiotemporal model of the fMRI response, which is a reasonable approximation over a wide range of spatio‐ temporal conditions [94–96]. The neuronal model is estimated by finding the model parameters that best predict the fMRI measurements. **(A)** The pRF modeling analysis is applied to all voxels (1:*n*) within the defined region of interest (ROI; e.g., occipital lobe, V1). **(B)** Multiple models of the expected average receptive field of the neuronal population may be used. Most commonly, a two-dimensional Gaussian is estimated, which is defined by three parameters, *x*, *y*, and *σ*, where (*x*, *y*) denotes the pRF center within the visual field, and *σ* is the Gaussian spread (i.e., pRF radius). Note that these parameters are stimulus-referred in degrees of visual angle. **(C)** The example moving bar stimulus com‐ posed of a black and white moving checkerboard pattern moves systematically across visual space. **(D)** The overlap between the pRF and effective stimulus is determined. **(E**, **F)** Next, the predicted pRF response is calculated for a given pRF model and effective stimulus location. The time series model prediction **(F)** is estimated by convolving the pRF activation with a model of the hemodynamic response function (HRF) as in **(E)** [94, 97]. The goodness-of-fit is estimat‐ ed by computing the residual sum of squares (RSS) between this prediction and the data (*black dots*) **(G)**. **(H)** 100,000 different fMRI time series predictions are iteratively tested by varying the pRF model parameters across a wide range of plausible values (e.g., locations across visual space covered by the stimulus; varying pRF sizes). Ultimately, the opti‐ mal pRF parameters (*x*, *y*, *σ*) are found for each voxel independently by minimizing the RSS using a two-stage, coarse-

Developed in the 1990s, TWR is still the primary fMRI paradigm used to measure early VFMs like V1–3 (**Figures 1** and **2**) [13, 15, 87, 91, 93]. This technique uses two types of periodic stimuli that move smoothly across a contiguous region of visual space to measure the orthogonal dimensions of polar angle (i.e., "around the clock") and eccentricity (i.e., center to periphery; **Figure 5**). These stimuli are typically composed of a set of high-contrast, flickering checker‐ board patterns that are designed to maximally stimulate primary visual cortex and generally elicit an fMRI signal modulation on the order of 1–3% (**Figure 2A**). This modulation is typically 15–20 standard deviations above the background noise. Each voxel's preferred polar angle representation is measured by a rotating wedge stimulus, which extends from the central fovea to more peripheral regions and covers a small section of polar angles (**Figure 6C**, *central*

to-fine search. Adapted from **Figure 2** in [92].

*2.1.1. Traveling wave retinotopy*

278 Update on Dementia

**Figure 4. White/gray matter segmentation for young, healthy aging, and mild Alzheimer's disease subjects.** Each panel is a T1-weighted 3D MPRAGE image showing a sagittal slice near the midline of the brain. Green-colored over‐ lay represents white matter identified by an automated algorithm [104] and adjusted by hand-editing to minimize seg‐ mentation errors [1, 3, 92]. White regions below the green overlay represent unsegmented white matter within the cerebellum. Gray matter is shown as the gray regions along the surface of the green overlay and adjacent to the black cerebral spinal fluid (CSF; *black regions within the skull*). Middle inset displays approximate anatomical directions. CaS: calcarine sulcus; POS: parietal-occipital sulcus. **(A)** Left hemisphere of healthy young subject. **(B)** Left hemisphere of healthy aging subject. **(C)** Left hemisphere of a subject with mild Alzheimer's disease (AD-S2). **(D)** Right hemisphere of a second subject with mild Alzheimer's disease (AD-S1). Note how AD-S1's anatomy is relatively intact, but the vis‐ ual field map measurements from this subject shown in **Figure 12** are perceptibly abnormal. Also compare the striking‐ ly increased CSF-filled space in this T1 image of AD-S2 to the relatively normal visual field map organization for this subject in **Figure 12**. Data were adapted from [36, 37].

**Figure 5. Two orthogonal gradients are required to define a cortical field map. (A)***Left:* The graph demonstrates measurements of three stimulus values—1: low (L, *red*); 2: medium (M, *green*); 3: high (H, *blue*)—for one sensory di‐ mension (e.g., eccentricity). *Right:* The graph demonstrates measurements of three stimulus values—1: low (L, *orange*); 2: medium (M, *aqua*); 3: high (H, *purple*) for a second sensory dimension (e.g., polar angle). *Middle:* Schematic of the orthogonal gradients defining a cortical field map shows how measurements of the cortical representation of dimen‐ sion 2 change along measurements of the representation of a single value of dimension 1, and vice versa. **(B)** Diagrams demonstrate how a single set of orthogonal gradients (one for each dimension) defines a single cortical field map. Black dotted lines demark the edge of the gradient representations. **(C)** Diagrams here demonstrate how a reversal in the dimension 1 gradient representations (*left*) divides up the single representation of the dimension 2 gradient (*right*) into two cortical field maps (*middle*). Yellow dotted lines demark a boundary defined by a gradient reversal, and black dotted lines again denote the edge of the gradient representations. A reversal in the dimension 2 gradient representa‐ tions would similarly divide up a single representation of the dimension 1 gradient (*not shown*). **(D)** Cloverleaf cluster organization of multiple cortical field maps. *Left:* Schematic of an example macrostructural organization of dimension 1 (e.g., eccentricity) across a region of the cortical surface, with low (*red*) to medium (*green*) to high (*blue*) stimulus val‐ ues represented in concentric circles. *Right:* Schematic of an example macrostructural organization of orthogonal di‐ mension 2 (e.g., polar angle) across the same region of the cortical surface, with low (*orange*) to medium (*aqua*) to high (*purple*) stimulus values represented in wedges running "around the clock". *Middle:* Schematic shows the four cortical field maps defined by these orthogonal gradients and arranged in a cloverleaf cluster [3, 111–113]. Black dotted lines demark the edge of the gradient representations. Yellow dotted lines demark boundaries defined by a gradient rever‐ sal. For additional discussions, see [1, 3, 88].

**Figure 6. Visual field map measurements in healthy young and aging subjects.** The images show example measure‐ ments of VFMs V1, V2, V3, and hV4 from a healthy young adult's left hemisphere (*top row*) and from a healthy aging adult's left hemisphere (*bottom row*). **(A)** Cortical locations of VFMs (black lines) are show on 3D renderings of each subject's left hemisphere. Central inset shows the approximate location of the cropped sections of the hemispheres. For cortical surfaces, dark gray represents sulci, and light gray represents gyri. "**\***" denotes the approximate location of the occipital pole; CaS: calcarine sulcus. **(B**, **C)** Flattened views of the cortical surfaces surrounding CaS are shown for the measurements of eccentricity **(B)** and polar angle **(C)** in each subject. The pseudocolor overlay on each flattened section of cortex signifies the location in visual space that creates the highest activity at each cortical position (see colored leg‐ end insets, *middle*). The wedge and ring stimuli each maximally spanned the central 11° radius of visual space. For sim‐ plicity, cortical representations are only shown for V1, V2, V3, and hV4—the VFMs of interest. Also note that functional data are only shown for voxels with a strong response at a coherence ≥0.20. Flattened renderings are scaled to the same size (see 1 cm scale bar, *middle inset*). Approximate anatomical directions for the flattened representations are shown in the central inset. Data were adapted from [36, 37].

**Figure 5. Two orthogonal gradients are required to define a cortical field map. (A)***Left:* The graph demonstrates measurements of three stimulus values—1: low (L, *red*); 2: medium (M, *green*); 3: high (H, *blue*)—for one sensory di‐ mension (e.g., eccentricity). *Right:* The graph demonstrates measurements of three stimulus values—1: low (L, *orange*); 2: medium (M, *aqua*); 3: high (H, *purple*) for a second sensory dimension (e.g., polar angle). *Middle:* Schematic of the orthogonal gradients defining a cortical field map shows how measurements of the cortical representation of dimen‐ sion 2 change along measurements of the representation of a single value of dimension 1, and vice versa. **(B)** Diagrams demonstrate how a single set of orthogonal gradients (one for each dimension) defines a single cortical field map. Black dotted lines demark the edge of the gradient representations. **(C)** Diagrams here demonstrate how a reversal in the dimension 1 gradient representations (*left*) divides up the single representation of the dimension 2 gradient (*right*) into two cortical field maps (*middle*). Yellow dotted lines demark a boundary defined by a gradient reversal, and black dotted lines again denote the edge of the gradient representations. A reversal in the dimension 2 gradient representa‐ tions would similarly divide up a single representation of the dimension 1 gradient (*not shown*). **(D)** Cloverleaf cluster organization of multiple cortical field maps. *Left:* Schematic of an example macrostructural organization of dimension 1 (e.g., eccentricity) across a region of the cortical surface, with low (*red*) to medium (*green*) to high (*blue*) stimulus val‐ ues represented in concentric circles. *Right:* Schematic of an example macrostructural organization of orthogonal di‐ mension 2 (e.g., polar angle) across the same region of the cortical surface, with low (*orange*) to medium (*aqua*) to high (*purple*) stimulus values represented in wedges running "around the clock". *Middle:* Schematic shows the four cortical field maps defined by these orthogonal gradients and arranged in a cloverleaf cluster [3, 111–113]. Black dotted lines demark the edge of the gradient representations. Yellow dotted lines demark boundaries defined by a gradient rever‐

sal. For additional discussions, see [1, 3, 88].

280 Update on Dementia

The term "traveling wave" arises from the repeating pattern of cortical activity that is activated from one end of the VFM to the other along iso-angle or iso-eccentricity lines in response to these periodic stimuli (**Figure 2B**) [87, 91]. Thus, the time, or phase, of the peak modulation induced by the stimulus varies smoothly across the cortical surface. This phase defines the most effective stimulus eccentricity (i.e., ring) and polar angle (i.e., wedge) to activate that region of cortex, giving TWR its description as "phase-encoded retinotopy" (**Figure 2D**). This paradigm only produces activity in regions that are retinotopically organized and is excellent for measuring early VFMs such as V1, V2, and V3.

In neuroimaging experiments, cortical activity driven by a particular stimulus is typically differentiated from unrelated activity and noise by setting well-defined statistical thresholds [94]. The statistical threshold for cortical activity arising from the TWR paradigm is commonly determined by coherence, which is equal to the amplitude of the blood-oxygen-level-depend‐ ent (BOLD) signal modulation at the frequency of stimulus presentation (e.g., six stimulus cycles per scan), divided by the square root of the power over all other frequencies except the first and second harmonics. Note that the activity must be correlated with the stimulus modulation frequency; unrelated activity at other frequencies is not included in the coherence measure. For each stimulus condition (e.g., wedge or ring), each voxel is independently assigned a coherence value, which denotes the strength of the BOLD response of that voxel for that particular stimulus (**Figure 2C**). Only voxels with a coherence above a chosen threshold —typically a coherence of 0.15–0.30—are further evaluated to determine the organization of cortical visual-spatial representations into specific VFMs [1, 3].

#### *2.1.2. Population receptive field modeling*

A newer method of measuring VFMs called population receptive field (pRF) modeling has been introduced not only for measurements of early visual cortex (e.g., V1–3), but also to improve measurements of the visual-spatial organization of higher-order regions that contain larger RFs (**Figures 1** and **3**) [92]. This model relies on the fact that the population of RFs in each voxel of retinotopically organized regions of cortex is expected to have similar preferred centers (i.e., location in visual space driving the peak neural responses) and sizes (i.e., the degrees of visual angle driving significant neural responses), allowing their combined pRF to be estimated as a single, two-dimensional Gaussian RF (**Figure 3B**). The pRF method does not require two distinct stimuli to measure orthogonal dimensions of visual space as in TWR, but can use any stimulus that systematically traverses the entire field of view. Most commonly in current measurements, this is a moving bar composed of the same checkerboard pattern with neutral gray blocks inserted at a nonstimulus frequency (**Figure 3C**). The neutral gray blocks allow for an estimation of a voxel's response to any visual stimulus versus just the preferred visual stimulus, which is necessary for the accurate measurement of pRF sizes.

For the first part of the analysis, the pRF model generates a database of many possible pRF centers and sizes within the field of view covered by the stimulus. The analysis then convolves the standard hemodynamic response function (HRF) with each of the possible pRFs (**Fig‐ ure 3E**). Lastly, a least-squares fitting method is used to iteratively test the actual data collected against each of the possible pRFs for each voxel independently. The best-fit pRF position and size is then allocated as the pRF parameters for that voxel. As in TWR, further analysis is only performed on voxels with responses above a specific threshold of variance explained—which can be converted to coherence—as determined by the model are included for further analysis (for additional discussions, see [1, 3, 92]).

#### *2.1.3. VFMs are defined using data from individual subjects*

paradigm only produces activity in regions that are retinotopically organized and is excellent

In neuroimaging experiments, cortical activity driven by a particular stimulus is typically differentiated from unrelated activity and noise by setting well-defined statistical thresholds [94]. The statistical threshold for cortical activity arising from the TWR paradigm is commonly determined by coherence, which is equal to the amplitude of the blood-oxygen-level-depend‐ ent (BOLD) signal modulation at the frequency of stimulus presentation (e.g., six stimulus cycles per scan), divided by the square root of the power over all other frequencies except the first and second harmonics. Note that the activity must be correlated with the stimulus modulation frequency; unrelated activity at other frequencies is not included in the coherence measure. For each stimulus condition (e.g., wedge or ring), each voxel is independently assigned a coherence value, which denotes the strength of the BOLD response of that voxel for that particular stimulus (**Figure 2C**). Only voxels with a coherence above a chosen threshold —typically a coherence of 0.15–0.30—are further evaluated to determine the organization of

A newer method of measuring VFMs called population receptive field (pRF) modeling has been introduced not only for measurements of early visual cortex (e.g., V1–3), but also to improve measurements of the visual-spatial organization of higher-order regions that contain larger RFs (**Figures 1** and **3**) [92]. This model relies on the fact that the population of RFs in each voxel of retinotopically organized regions of cortex is expected to have similar preferred centers (i.e., location in visual space driving the peak neural responses) and sizes (i.e., the degrees of visual angle driving significant neural responses), allowing their combined pRF to be estimated as a single, two-dimensional Gaussian RF (**Figure 3B**). The pRF method does not require two distinct stimuli to measure orthogonal dimensions of visual space as in TWR, but can use any stimulus that systematically traverses the entire field of view. Most commonly in current measurements, this is a moving bar composed of the same checkerboard pattern with neutral gray blocks inserted at a nonstimulus frequency (**Figure 3C**). The neutral gray blocks allow for an estimation of a voxel's response to any visual stimulus versus just the preferred

visual stimulus, which is necessary for the accurate measurement of pRF sizes.

For the first part of the analysis, the pRF model generates a database of many possible pRF centers and sizes within the field of view covered by the stimulus. The analysis then convolves the standard hemodynamic response function (HRF) with each of the possible pRFs (**Fig‐ ure 3E**). Lastly, a least-squares fitting method is used to iteratively test the actual data collected against each of the possible pRFs for each voxel independently. The best-fit pRF position and size is then allocated as the pRF parameters for that voxel. As in TWR, further analysis is only performed on voxels with responses above a specific threshold of variance explained—which can be converted to coherence—as determined by the model are included for further analysis

for measuring early VFMs such as V1, V2, and V3.

282 Update on Dementia

cortical visual-spatial representations into specific VFMs [1, 3].

*2.1.2. Population receptive field modeling*

(for additional discussions, see [1, 3, 92]).

It is vital to correctly localize common functional areas across subjects in order to then study which specific computations are carried out by each area and how these functions change with damage and disease; however, the size of functional regions across the cortical surface varies significantly across individuals, complicating measurements across subject groups [14, 92, 98– 102]. The surface area of primary visual cortex (V1) can differ across subjects by a factor of 3 or more; this variation in VFM size is independent of the total brain size [14]. Consequently, the position of each VFM with respect to the underlying structural anatomy varies across individuals. In addition, the amount of variation tends to increase from V1 to regions of visual cortex involved in higher-order computations (e.g., object recognition). Therefore, the common fMRI approach of averaging measurements across subjects does not work in the case of VFM data, as this approach will problematically blur the VFM boundaries to an unusable degree and may even destroy all traces of VFM organization in some regions [3]. Likewise, standar‐ dized brain-template coordinates (e.g., Talairach or Montreal Neurological Institute—MNI coordinates) cannot be used to accurately estimate the position of any VFMs beyond area V1 in group-averaged or individual-subject data. Furthermore, issues like cortical degeneration introduce even greater variability into the match between cortical function and structure. The only accurate approach is to measure VFMs in individual subjects. Functional or structural measurements from each VFM—each region of interest (ROI) —can be obtained from indi‐ vidual subjects and then averaged across the subject group by VFM.

For analysis of such functional imaging data for individual subjects, several neuroimaging software packages are available. We use a signal-processing, Matlab-based software package called *mrVista*, which was originally developed by the Wandell lab at Stanford University and is now commonly available for neuroimaging analysis (http://white.stanford.edu/software/) [103, 104]. Using this software, each subject's cortical white matter is determined—"segment‐ ed"—in that subject's high-resolution anatomical scan, first with an automated algorithm and then by hand-editing to minimize errors for individual-subject analyses (**Figure 4**) [104]. Then a 3–4 mm layer of gray matter is automatically formed arising from the segmented white matter surface. Only data drawn from this identified gray matter are analyzed, in order to reduce extraneous measurements arising in the white matter or noise outside the head. The gray matter of the cortical sheet can then be visualized in three dimensions or as a flat sheet to allow analysis of functional activity within the sulci. For analysis of the functional data, linear trends are removed from the time series during preprocessing. Individual scans are inspected for motion artifacts; issues with motion between individual scans or across scan sessions can be corrected using mutual-information algorithms [105]; however, motion-correction algorithms should not be routinely applied if not needed, as they themselves may create artifacts. Once the preprocessed functional dataset is registered to the high-resolution anatomical scan, the VFM activity can be viewed along the cortical sheet to allow for optimal definition of VFM boundaries (e.g., **Figures 5** and **6**; for detailed discussions, see [1, 3, 88]). Importantly, we never apply spatial smoothing to the data, as smoothing will destroy key details of VFM organiza‐ tion, much like averaging data across subjects.

The description of a cortical "map" is frequently used nonspecifically for topographical gradients or other related cortical representations; however, it is advantageous for research into visual processing to explicitly define a "visual field map" in accordance with very exact criteria: (i) a VFM comprises two (or more) orthogonal, nonrepeating topographical represen‐ tations of fundamental sensory dimensions (e.g., eccentricity and polar angle); (ii) each of these topographical representations must be organized as an generally contiguous, orderly gradient; (iii) each VFM should represent a substantial portion of sensory space (e.g., a hemifield of the visual field); and (iv) the general features of each VFM should be consistent across individuals (**Figure 5**) [1, 3, 13, 15, 88, 106–111]. For additional discussion, see *A Brief Primer on Cortical Field Mapping* in the supplemental material of [88].

#### **2.2. Characteristics of V1, V2, V3, and hV4 in healthy young adults**

Three VFMs known as V1, V2, and V3 occupy the medial wall of occipital cortex in humans and participate in the first stages of visual processing (for detailed reviews, see [1, 3]). Each represents a full hemifield of the opposite—i.e., contralateral—visual field in each hemisphere; the left hemisphere VFMs represent the right visual field, and vice versa (**Figure 6**).

V1 is very reliably located along the calcarine sulcus, bounded on either side by the unique split-hemifield representations of V2 and V3 on the cuneus and lingual gyri. V1, V2, and V3 each contain a representation of the center of visual space located at the occipital pole. Increasingly more peripheral representations form complete eccentricity gradients extending into more anteromedial cortex forming complete eccentricity gradients (**Figure 6A** and **B**; e.g., [1, 3, 13, 15, 91]). The position at which the central—i.e., foveal—representations of these three VFMs come together at the occipital pole is called the foveal confluence [114]. The boundaries between each map are delineated by reversals in polar angle gradients (**Figure 6A** and **C**). The polar angle gradient of V1 represents a contiguous hemifield of visual space. In contrast, the polar angle representations of V2 and V3 each consist of a split-hemifield representation (i.e., a quarterfield of visual space). The separate quarterfield sections are denoted by their positions ventral or dorsal to V1 (e.g., V2d, V2v, V3d, and V3v). For each of these three VFMs, the upper visual quarterfield is represented on the ventral surface, and the lower visual quarterfield is represented on the dorsal surface. The V2 and V3 quarterfields each meet at the fovea, but are otherwise separate [3]. Due to their relatively consistent anatomical locations and unique concentric organization, these three VFMs form the easiest landmarks to be identified in visual field mapping analyses [13–15, 87]. An additional VFM is adjacent to V3v along the ventral occipital surface: human V4, designated hV4 because of the unclear homology to macaque V4 [106]. The eccentricity representation of hV4 merges with the foveal confluence of V1, V2, and V3, and the polar angle representation moves smoothly from the boundary at the upper visual field representation of the V3v quarterfield into a full hemifield representation.

V1 is labeled as primary visual cortex, because it receives direct input from the retinal-thalamic pathway and is the first place along this pathway where information from the two eyes is combined [115]. In addition, V1 is an important site of initial basic calculations for such visual processing as orientation, color, and motion [116–120]. Beyond V1, perceptual calculations become more specialized, with V2 subserving relatively simple color and form perception and V3 more selectively supporting motion processing [118, 119, 121–124]. hV4 serves as the next stage of the V1–V2 processing pathway with specialized neuronal populations for somewhat more complicated color and form visual perception [106, 125, 126]. Damage to part of the retina or V1 produces a complete loss of visual perception within the region normally processed by the damaged tissue in most cases [9], while damage to VFMs beyond V1 can produce more selective visual deficits like loss of color vision (i.e., achromatopsia; damage to hV4) or loss of motion perception (i.e., akinetopsia; damage to hMT+) [121, 127–131].

The description of a cortical "map" is frequently used nonspecifically for topographical gradients or other related cortical representations; however, it is advantageous for research into visual processing to explicitly define a "visual field map" in accordance with very exact criteria: (i) a VFM comprises two (or more) orthogonal, nonrepeating topographical represen‐ tations of fundamental sensory dimensions (e.g., eccentricity and polar angle); (ii) each of these topographical representations must be organized as an generally contiguous, orderly gradient; (iii) each VFM should represent a substantial portion of sensory space (e.g., a hemifield of the visual field); and (iv) the general features of each VFM should be consistent across individuals (**Figure 5**) [1, 3, 13, 15, 88, 106–111]. For additional discussion, see *A Brief Primer on Cortical*

Three VFMs known as V1, V2, and V3 occupy the medial wall of occipital cortex in humans and participate in the first stages of visual processing (for detailed reviews, see [1, 3]). Each represents a full hemifield of the opposite—i.e., contralateral—visual field in each hemisphere;

V1 is very reliably located along the calcarine sulcus, bounded on either side by the unique split-hemifield representations of V2 and V3 on the cuneus and lingual gyri. V1, V2, and V3 each contain a representation of the center of visual space located at the occipital pole. Increasingly more peripheral representations form complete eccentricity gradients extending into more anteromedial cortex forming complete eccentricity gradients (**Figure 6A** and **B**; e.g., [1, 3, 13, 15, 91]). The position at which the central—i.e., foveal—representations of these three VFMs come together at the occipital pole is called the foveal confluence [114]. The boundaries between each map are delineated by reversals in polar angle gradients (**Figure 6A** and **C**). The polar angle gradient of V1 represents a contiguous hemifield of visual space. In contrast, the polar angle representations of V2 and V3 each consist of a split-hemifield representation (i.e., a quarterfield of visual space). The separate quarterfield sections are denoted by their positions ventral or dorsal to V1 (e.g., V2d, V2v, V3d, and V3v). For each of these three VFMs, the upper visual quarterfield is represented on the ventral surface, and the lower visual quarterfield is represented on the dorsal surface. The V2 and V3 quarterfields each meet at the fovea, but are otherwise separate [3]. Due to their relatively consistent anatomical locations and unique concentric organization, these three VFMs form the easiest landmarks to be identified in visual field mapping analyses [13–15, 87]. An additional VFM is adjacent to V3v along the ventral occipital surface: human V4, designated hV4 because of the unclear homology to macaque V4 [106]. The eccentricity representation of hV4 merges with the foveal confluence of V1, V2, and V3, and the polar angle representation moves smoothly from the boundary at the upper visual

the left hemisphere VFMs represent the right visual field, and vice versa (**Figure 6**).

field representation of the V3v quarterfield into a full hemifield representation.

V1 is labeled as primary visual cortex, because it receives direct input from the retinal-thalamic pathway and is the first place along this pathway where information from the two eyes is combined [115]. In addition, V1 is an important site of initial basic calculations for such visual processing as orientation, color, and motion [116–120]. Beyond V1, perceptual calculations become more specialized, with V2 subserving relatively simple color and form perception and

*Field Mapping* in the supplemental material of [88].

284 Update on Dementia

**2.2. Characteristics of V1, V2, V3, and hV4 in healthy young adults**
