Mapping the “What” and “Where” Visual Cortices and Their Atrophy in Alzheimer's Disease: Combined Activation Likelihood Estimation with Voxel-Based Morphometry

The human cortical regions for processing high-level visual (HLV) functions of different categories remain ambiguous, especially in terms of their conjunctions and specifications. Moreover, the neurobiology of declined HLV functions in patients with Alzheimer's disease (AD) has not been fully investigated. This study provides a functionally sorted overview of HLV cortices for processing “what” and “where” visual perceptions and it investigates their atrophy in AD and MCI patients. Based upon activation likelihood estimation (ALE), brain regions responsible for processing five categories of visual perceptions included in “what” and “where” visions (i.e., object, face, word, motion, and spatial visions) were analyzed, and subsequent contrast analyses were performed to show regions with conjunctive and specific activations for processing these visual functions. Next, based on the resulting ALE maps, the atrophy of HLV cortices in AD and MCI patients was evaluated using voxel-based morphometry. Our ALE results showed brain regions for processing visual perception across the five categories, as well as areas of conjunction and specification. Our comparisons of gray matter (GM) volume demonstrated atrophy of three “where” visual cortices in late MCI group and extensive atrophy of HLV cortices (25 regions in both “what” and “where” visual cortices) in AD group. In addition, the GM volume of atrophied visual cortices in AD and MCI subjects was found to be correlated to the deterioration of overall cognitive status and to the cognitive performances related to memory, execution, and object recognition functions. In summary, these findings may add to our understanding of HLV network organization and of the evolution of visual perceptual dysfunction in AD as the disease progresses.


INTRODUCTION
The human visual cortex is primarily located in, but not confined to, the occipital lobe. It extends into the temporal and parietal lobes and shows complicated cortical distribution. Understanding of the human visual cortex has grown with the development of functional magnetic resonance imaging (fMRI). To date, over a dozen cortical areas have been identified to be involved in visual functions (Sereno et al., 1995;Tyler et al., 2005;Wandell et al., 2005), and the pathway model was introduced to divide these functional areas into dorsal and ventral processing streams (Haxby et al., 1991;Pihlajamaki et al., 2005).
The dorsal pathway, also known as the "where" stream, is an occipito-parietal network which lies between the early visual cortex and those specialized cortical structures involved in the processing of spatial and motion information (Kravitz et al., 2013). The ventral pathway, also known as the "what" stream, is an occipito-temporal network that bridges the early visual cortex and is involved in processing visual identity and feature information (e.g., faces, object identities, colors, and words; Kravitz et al., 2013). To date, a growing number of studies have investigated the anatomical distributions of the ventral and dorsal streams as well as their specializations and conjunctions in the processing of visual perceptions for the different categories Downing et al., 2006;Cichy et al., 2011;Mano et al., 2013). For example, the motion-selective area of V5 (hMT+; Morrone et al., 2000;Huk et al., 2002), the spatial specific brain region of V3a (Tootell et al., 1997;Backus et al., 2001), as well as areas for words (Liu et al., 2008) and faces , are anatomically distinct in fMRI.
In recent years, emerging studies have begun to investigate the neural basis of disrupted high-level visual (HLV) functions in Alzheimer's disease (AD). Neuropsychological and neuroimaging findings have demonstrated that both the dorsal (Kavcic et al., 2011;Mandal et al., 2012) and ventral (Kurylo et al., 1996;Adlington et al., 2009) visual streams are impaired during the progression of AD (Delbeuck et al., 2003;Mandal et al., 2012). Because successfully exercising cognitive functions such as memory, executive function and motion perception usually depends upon intact visual perceptual function (Reisberg, 2010;Culham, 2015), it is reasonable to suspect that declines in HLV functions may also contribute to poor cognition in AD. Several studies have demonstrated that deficits in visual perception significantly correlate with poor performance in instrumental activities of daily living (Eslinger et al., 1985), as well as with impaired general cognitive ability (Silveri and Leggio, 1996;Rizzo et al., 2000). This demonstrates the importance of investigating alterations of HLV cortices in AD.
However, due to several sources of study limitations, there are relatively few widely accepted models of alterations of HLV cortices during AD-related pathological progression. First, unlike with primary visual cortices, the "what" and "where" visual streams include the emergence of a series of categories, and processing these types of visions may involve different cortical regions. Because a single fMRI study cannot include a wide variety of paradigms corresponding to different visual categories, it is difficult to obtain a comprehensive overview for the cortical locations of the dorsal and ventral streams. In addition, the issue of crosstalk between the dorsal and ventral streams has been discussed extensively in recent years (Schenk and McIntosh, 2010). Thus, there exist possible regions of conjunction in stream process, and overlaps may even complicate the cortical mapping of "what" and "where" visual functions in intact individuals. Second, due to the variance in methods and relevant parameters used in prior fMRI studies, the anatomical locations and boundaries of the functional areas of the dorsal and ventral streams remain ambiguous. In other words, it remains unclear whether different functional regions across different fMRI studies reflect a true difference in regions of neural activations across different categories of visual targets or are caused by the methodological differences among the studies (e.g., methods of data preprocessing). Overall, due to the lack of comprehensive, systematic, and accurate cortical localizations for HLV cortices, their alterations in AD-related pathological progression have been less than fully investigated to date.
Therefore, this study aims to provide a functionally sorted overview of the HLV cortices of the "what" and "where" visions, and, on this basis, comprehensively investigate the atrophy of these HLV cortices in AD and mild cognitive impairment (MCI). First, the HLV cortices responsible for processing different categories of visual perception were mapped by performing activation likelihood estimation (ALE) on previous relevant fMRI studies. This step was used to provide a functionally sorted overview of the HLV network of the "what" and "where" visions. ALE is a coordinate-based analysis approach for neuroimaging data, which is based upon calculating the overlap between reported foci in different studies, modeling them as probability distributions centered at given coordinates. Therefore, the created ALE map can then be used to indicate inter-study regions of consistent brain activation during a cognitive process which would not be obvious upon initial inspection (Turkeltaub et al., 2002). Second, based upon the subsequent contrast analyses, the resulting specific and conjunctive brain regions responsible for processing different categories were further investigated. It was thus possible to shed light on how interactions among the different categories of visual processing were organized within the brain. For example, "Which brain areas were constantly activated for both 'what' and 'where' visual stimuli?". Third, gray matter (GM) volume of these HLV cortices (mapped in previous steps) was measured using voxel-based morphometry (VBM) method, and compared among AD, MCI, and normal control participants. Then, the GM volume of visual cortices identified with atrophy was further correlated to the cognitive scoring. This was to provide a dynamic observation of alterations to the HLV cortices throughout AD progression, as well as their relation to the cognitive deterioration. Therefore, these findings may help to better elucidate the cortical organization of the HLV cortices, as well as the evolution of visual perceptual dysfunction during the progression of AD.

MATERIALS AND METHODS
Mapping the Cortices of "What" and "Where" Visions Using ALE

Literature Selection and Exclusion Criteria
We searched for studies that investigated HLV functions in the PubMed database (http://www.ncbi.nlm.nih.gov/pubmed) using the search terms: " (((((((((('Visual Perception'[Mesh] (1) Papers should be published in English, and studies must explicitly refer to the visual cortex as involved in the processing of visual stimuli. Subjects should be healthy adults. Those papers that did not meet the first two criteria were excluded. Only 1558 of the 3430 papers met these criteria.
(2) Studies that used fMRI were included. Those only employing other techniques, e.g., positron emission tomography (PET), single-photon emission tomography (SPECT), magnetoencephalography (MEG), transcranial magnetic stimulation (TMS), behavioral measures, and review articles, were excluded. The meta-analysis was limited to fMRI studies because of their comparability in the spatial and temporal resolutions for the ALE analyses. Only 1401 of the remaining 1558 papers met this criterion.
(3) Studies that did not image the whole brain (i.e., from the top of the brain to the cerebellum) or did not report the coordinates of the activation clusters were excluded. Any studies using only a region of interest (ROI) analytic method instead of whole-brain analyses were excluded. Only 689 of the remaining 1401 papers survived. (4) Studies that did not report coordinates in the normalized stereotaxic reference systems of either Talairach and Tournoux (1988) or the MNI or that reported coordinates that were not in the Cartesian (XYZ) format were excluded. Seven of the remaining 689 papers were excluded. (5) Studies in which the authors did not attribute their fMRI results directly to a particular visual function or task, or which involved more than one visual function (e.g., color and depth visions) simultaneously, were excluded. Studies using low-level contrasts (target vision vs. baseline, noise or scrambled meaningless images) were included to maintain the homogeneity of the meta-analysis. Studies that only provided the coordinates for comparison with high-level contrasts were excluded. Ultimately, 93 papers fulfilled all of these criteria.

Data Analysis
The 93 studies fitting within the established criteria were first divided into two major categories of "what" and "where" visions. These included studies of face, object, alphabetic word/letter, logographic word/symbol, scene, body, motion, and spatial visions, respectively. These studies were further classified into eight basic visual categories of motion, space, object, face, alphabetic word/letter (word for short), logographic word/symbol, scene, and body visions (Supplementary Table 1). To ensure the power of the results, visual functions with less than 10 experiments (i.e., body, scene and logographic word visions) were not used to establish an independent category. Consequently, five categories remained.
To identify the specific brain regions responsible for particular visual functions, we subsequently conducted a contrast analysis of each pair from the task categories of face-word, face-object, object-word and "what"-"where" vision. Contrast analyses between motion and spatial vision were not performed due to the large difference in the number of included studies between these two categories, which might have impacted the power of the statistical results.
The Ginger ALE software program, version 2.0 (available at http://www.brainmap.org/ale) with revised ALE algorithm implemented in was used. ALE is a coordinate-based meta-analysis tool that treats significant foci reported in neuroimaging studies not as single points but as spatial probability distributions centered at the given coordinates. ALE maps are then produced by computing the union of activation probabilities for each voxel Turkeltaub et al., 2012). First, coordinates of the original studies reported in the Talairach brain map were transformed into the MNI space using the Lancaster transform (Lancaster et al., 2007). The foci from each individual study were smoothed by a full width half maximum (FWHM) value scaled by the study's sample size to model the uncertainty in the spatial location of the activations (Eickhoff et al., 2009). An analytical method was used to determine the null distribution of the ALE statistic, and correction for multiple comparisons was then applied using the false discovery rate (FDR pN; Laird et al., 2005), with p < 0.01. A minimum cluster volume of 200 mm 3 was applied. The statistical level for contrast analysis was also set at p < 0.01 (FDR pN corrected) with a minimum volume ≥ 200 mm 3 . The generated ALE maps were overlaid on a Colin_tlrc_2 × 2 × 2 template using the MANGO software program for visualization (http://rii. uthscsa.edu/mango/).

Subjects
All of the imaging and neuropsychological test data were taken from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). For more information about the ADNI dataset, as well as its subject inclusion and exclusion criteria, please see the ADNI dataset for more details in the Supplementary Material and the ADNI website at http:// adni.loni.usc.edu/methods/documents/. For the current study, subjects with glaucoma, congenital blindness, or excessive head motion during MRI scanning were additionally excluded due to the nature of this study. Moreover, to ensure comparability between structural and functional results in future studies, only subjects with both three-dimensional T1-weighted imaging (3D T1WI) and resting-state functional MRI scans were included. Finally, 44 CN (normal control), 52 EMCI (early MCI), 35 LMCI (late MCI), and 30 AD subjects were included.

Image Acquisition and Data Preprocessing
All of the subjects were scanned using 3T Philips MRI scanners. The high resolution 3D T1WI were acquired using magnetization-prepared rapid acquisition gradient-echo (MPRAGE) with TR = 6.8 ms, TE = 3.1 ms, 170 slices, FOV = 256 mm, matrix size = 256 × 256, and slice thickness = 1.2 mm.
All 3D T1WI data were processed using the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm) implemented in the SPM8 software (http://www.fil.ion.ucl.ac.uk/spm/). Brain images were linearly registered (12-parameter affine) to approximate the MNI space and segmented into GM, white matter, and cerebrospinal fluid images. The GM images were further normalized using DARTEL (Diffeomorphic Anatomical Registration using Exponentiated Lie algebra; Ashburner, 2007), Jacobian modulated and smoothed using an 8 mm FWHM Gaussian kernel. Finally, the mean GM volume was extracted for each visual ROI. Here, all visual ROIs were made by using the clusters obtained from the ALE results in part 3.1, corresponding to the "face, " "object, " "word, " "motion, " and "spatial" visual regions. In total, the GM volume of 36 ROIs (11 for face vision, 7 for object vision, 5 for word vision, 7 for motion vision, and 6 for spatial vision, see Table 3) was calculated for each subject in the CN, EMCI, LMCI, and AD groups.

Statistical Analyses
The demographics and mini-mental state examination (MMSE) scores of the subjects were compared among the CN, EMCI, LMCI, and AD groups by one-way ANOVA (or Kruskal-Wallis H test if the data were not normally distributed), with a statistical significance level at p < 0.05. For the evaluation of GM volume difference among four groups, a multivariate analysis of variance (MANCOVA) was firstly conducted to evaluate the main effect of group. Then, analysis of covariance (ANCOVA) was performed respectively for each ROI to compare the GM volume among four groups. The whole-brain volume was added as a covariate in both the MANCOVA and ANCOVA analyses to ensure that observed GM volume differences in HLV cortices among groups were independent of the variance in whole-brain volume across subjects. A statistical level of 0.0014 was used to correct for 36 times of comparisons for all the ROIs (i.e., Bonferroni method, 0.05/36 ≈ 0.0014), and further Post-hoc analyses between each pair of groups were also corrected with the Bonferroni correction. In addition, the GM volume of each ROI identified with significant atrophy in ADs or MCIs (finally turned out to be 25 ROIs totally in our results) was correlated to the scores of clock drawing test, category fluency-animal, trail making test A & B, logical memory test I & II, Boston naming test, MMSE, and whole-brain mean GM volume using Pearson correlation (or Spearmen correlation if the data were not normally distributed). A statistical level of 0.002 was used to correct for multiple times of correlations (i.e., Bonferroni method, 0.05/25 = 0.002).

RESULTS
Cortices for "What" and "Where" Visions Acquired from ALE Overall, convergent brain regions were identified from five ALE analyses of face, word, object, motion and spatial visual functions (Table 1, Figure 1) and were mainly distributed in the bilateral inferior and middle temporal gyrus (Brodmann area 19, BA 19) and the middle and inferior occipital gyrus (BA 18,19). Moreover, the bilateral frontal lobes were found to be significantly activated, with regions for face, word, object and motion processing. Specific and conjunctive brain regions for different categories of visual functions acquired from contrast analyses are summarized in Table 2

Cortices Corresponding to the Five Categories of Visual Functions
Our ALE analysis of face processing included 318 foci from 27 experiments. Eleven significant activation clusters extending across nine different BA in the bilateral temporal and occipital lobes, inferior frontal lobe and amygdala were found (Table 1, Figure 1A). Comparisons between face and word as well as between face and object processing showed higher activation for face processing in the bilateral occipital gyrus (BA 18), left inferior temporal gyrus (BA 19), and the right middle temporal/occipital gyrus (BA 37; Table 2, Figure 2A).
Sixteen experiments on word vision with 119 foci were included in our ALE meta-analysis, which revealed five significant clusters, including two in the bilateral inferior temporal gyrus (BA 19/37, both extending into the bilateral fusiform gyrus, Table 1, Figure 1B). The rest significant regions were found in the left middle temporal gyrus and the left opercular part of the inferior frontal gyrus, respectively. Contrast analyses of word vs. face and of word vs. object perception revealed that word-specific regions were mainly located in the left inferior temporal gyrus (BA 37) which extended to the middle temporal gyrus (BA 37; Table 2, Figure 2B).
In total, 112 foci from 18 experiments contributed to our ALE analysis of object perception. Seven significant clusters were identified (Table 1, Figure 1C), with five clusters located in the bilateral fusiform gyrus (BA 37), right middle temporal (BA 19/37) gyrus, and left middle occipital gyrus (BA 19). Two clusters were located in the bilateral inferior frontal gyrus (BA 44). Contrast analysis of object vs. face perception showed higher activation for object perception in the right middle temporal gyrus (BA 37), while there was no significant difference in activation for object perception when compared with word perception ( Table 2).
Two hundred and forty-seven foci from 30 experiments contributed to our ALE analysis of spatial perception, and six clusters were found (Table 1, Figure 1D), which were mainly located in the temporal and occipital lobes (BA 18,19,and 37). In addition, the bilateral superior parietal lobules were also found with significant clusters.
Our ALE analysis of motion perception included 205 foci from 14 experiments and found seven significant clusters (Table 1, Figure 1E). In addition to regions in the temporal and occipital Frontiers in Human Neuroscience | www.frontiersin.org  lobes, significant regions were also found in the bilateral postcentral gyrus, left supramarginal gyrus, left superior parietal lobule (BA 7), and the right middle frontal gyrus (BA 6).
"What" vs. "Where" We performed contrast analyses broadly between the "what" and "where" visual streams to obtain specific brain regions. By comparison, the "what" stream showed overall higher activation in the bilateral inferior temporal gyrus and the left fusiform gyrus, while the "where" stream showed higher activation in the left precentral gyrus, right superior parietal lobule, middle occipital gyrus, middle temporal gyrus, and postcentral gyrus (Table 2, Figure 2C).

Conjunctively Activated Brain Regions
The spatial distributions of brain regions with conjunctive activation acquired from the above contrast analyses were largely consistent and summarized here. The bilateral middle and inferior occipital gyrus, middle and inferior temporal gyrus, and fusiform gyrus were found to be the common conjunctive regions among most of the observed visual categories (Figure 3).

Comparisons of GM Volume
Demographics There were no significant differences in gender, age, or education levels among the AD, EMCI, LMCI, and CN groups. Significant differences in MMSE scores were found among CN, EMCI, LMCI, and AD participants, and the AD patients performed the worst (Supplementary Table 2).

Volume Differences in HLV Cortices among Groups
MANCOVA analysis revealed that there was significant difference in GM volume of the 36 ROIs among four groups (p = 9.07 × 10 −5 ). By performing subsequent ANCOVA analyses, 25 ROIs were found with significantly different volume among groups (Table 3). Post-hoc analyses further revealed that the GM volume of 14 ROIs in the AD group was significantly smaller than in any other groups, and the volume of the rest 11 ROIs was significantly smaller than EMCI and CN groups (Figure 4). In LMCI subjects, only three clusters located in the left supramarginal gyrus, left middle, and inferior occipital gyrus were significantly atrophied compared with CN or EMCI (Figure 4). It's worth noting that all of the three clusters were involved in the motion and spatial visual processing. No clusters showed significant difference in GM volume between CN and EMCI subjects.

Correlation Analyses Results
The correlation results between GM volume of the 25 ROIs with significant atrophy and the cognitive test scores and whole-brain mean GM volume are summarized in Table 4. Overall, both the MMSE score and whole-brain mean GM volume were found with a significantly positive correlation to the regional GM volume in most of the 25 ROIs. In addition, the performances of category fluency-animal and trail making test A & B were found to be positively correlated with the GM volume in most ROIs of both "what" and "where" visual cortices. For the logical memory test I & II and Boston naming test, significant correlation between GM volume and test performances was found to be more widespread in "what" visual cortices than in "where" visual cortices. Whereas, in clock drawing test, significant correlation between GM volume and test score was only found in a small portion of ROIs.

DISCUSSION
In this study, we performed a series of ALE analyses to map the "what" and "where" visual cortices, and we systematically differentiated them with regard to five categories visual perceptions: object, face, word, motion, and spatial. Subsequently, brain areas with specific and conjunctive activation for processing these visual functions were identified by contrast  analyses. Finally, based upon the ALE-defined spatial locations, the atrophy patterns of these HLV cortices in MCI and AD patients were evaluated. The results showed that the distribution of atrophied brain regions was widespread in AD patients and localized in MCI subjects, and these observed findings were related to the deterioration of overall cognitive status. To our knowledge, this is the first comprehensive coordinate-based meta-analysis of "what" and "where" visual perceptions, and also the first morphometric analysis focusing on the HLV cortices in AD and MCI patients.
The HLV Networks of the Brain for "What" and "Where" Visions Our whole-brain ALE analyses suggested a series of brain regions that were involved in the HLV functions of "what" and "where" visions. On one hand, convergent distributions of these clusters were found in the occipital and temporal gyrus, which were consistent with the locations of the traditionally considered brain components for HLV processing in previous findings (Grill-Spector and Malach, 2004;Wandell et al., 2007). On the other hand, the bilateral inferior frontal gyrus and superior parietal lobule were found to be significant for most of the "what" and "where" visual processes, respectively, perhaps underlying their involvement in HLV processing. In particular, the inferior frontal gyrus was found to be involved in emotional responses, riskier choices and language comprehension (Sakai, 2005;Vigneau et al., 2006;Sarubbo et al., 2013). The superior parietal lobule plays a pivotal role in many sensory and cognitive processes, including spatial perception (Faillenot et al., 1997;Weiss et al., 2003), integration for motor acts (Culham et al., 2006), motor learning (Weiss et al., 2003) and visuospatial attention (Corbetta et al., 1995), etc. In addition, in Kravitz's review (Kravitz et al., 2013), they also described the superior parietal lobule contains 3D representation of objects for visuospatial processing and closely interconnected with occipital and prefrontal regions to form a framework of visuospatial   processing. Hence, we assume that the inferior frontal gyrus and the superior parietal lobule may collaborate with regions in the occipital and temporal lobes to exercise particularly HLV processing; thus, these two regions may also be included as components constituting a HLV network. These findings are also supported by those of previous studies (de Haan and Cowey, 2011;Kravitz et al., 2013). However, it is worth noting that the current finding in the inferior frontal gyrus and superior parietal lobule may also possibly due to a general response to the tasks (e.g., memory and attention, etc.). Thus, further studies on whether the two regions are activated at the visual perceptual stage or at the subsequent information integration or action guidance stage are encouraged to shed light on this question.

Specific Brain Areas for High-Level Vision
Contrast analyses were used to investigate specific brain regions for particular visual functions. When compared with face and object visions, the brain regions specific for word processing were found in the left inferior temporal gyrus. This result is in accordance with previous evidence that specific activation by word or letter stimuli locates in the left inferior occipito-temporal cortex of the human brain (Cohen et al., 2002;Glezer et al., 2009). It's worth mentioning that to maintain the homogeneity of the meta-analysis, we only included studies of alphabetic word perception. Based on the fact that only a few studies of logographic words are available, it would be interesting for future comparative meta-analyses to investigate specific brain areas for perceiving words in different languages and writing systems. Three of the "face specific" clusters revealed in our study were located in the inferior temporal (BA 19/37) and occipital gyrus (BA 18/19). These anatomical locations correspond well to the previously defined occipital face area (OFA; Yovel and Kanwisher, 2004), which are thought to be specialized in facial information processing. In addition, the posterior portion of right middle temporal gyrus was found as another "face specific" region when compared with word vision. There is also previous evidence that supported this area as a face processing activated region (Hein and Knight, 2008). However, the "fusiform face area, " which is recognized as one of the classical face processing regions, was not found in our contrast result. Alternatively, this region was shown as a conjunctive region for both face and object processing. In Haxby's and Hanson's studies (Haxby et al., 2001;Hanson et al., 2004), they found that the representations of faces and objects in the ventral temporal cortex are highly overlapping, thus, the specificity of cortical response to certain visual category may be difficult to be clarified in this area. Therefore, taken together with our results, we may speculate that the fusiform gyrus may contribute to both face and non-living object recognition. Nevertheless, future fMRI studies with higher spatial and temporal resolution are still required to provide more findings on the specificity of fusiform gyrus in recognition perception.
The contrasts between the two major aspects of HLV functions, i.e., the "what" and "where" streams, revealed that "where" visual perception was related to greater activation in a series of brain regions, including the superior occipital gyrus, superior parietal lobule, posterior portion of middle temporal gyrus, precentral and postcentral gyrus. The results were consistent with the distribution of brain regions that formed the dorsal visual pathway, as described in Kravitz's review (Kravitz et al., 2011). Besides, these brain regions were considered closely interconnected and formed a framework to support the spatial and motion perception. In contrast, "what" visual perception showed a greater involvement of the bilateral inferior temporal gyrus. In accordance with previous findings, this result highlights the role of inferior temporal gyrus in "what" vision processing (Kravitz et al., 2013).

Conjunctive Regions for High-Level Visual Functions
Conjunctive brain regions acquired from contrast analyses among different visual categories were convergently distributed in the right inferior and middle temporal gyrus (BA 19/37) and bilateral middle and inferior occipital gyrus (BA 18, 19, 37), which essentially correspond to the location of the lateral occipital complex (LOC). The LOC has been considered to be the prime locus of object identity representation (Ishai et al., 2000;Cichy et al., 2011), and it is traditionally considered as a part of the "what" visual processing stream. One possible reason for its conjunctive activation in terms of "where" visual processing may be that most spatial and motion visual tasks use certain objects or forms as visual targets, and object identification may be constantly included as a precursor for "where" processing. Therefore, our current findings in the LOC may indicate the Frontiers in Human Neuroscience | www.frontiersin.org existence of interrelationships between "what" and "where" information processing (i.e., the functional crosstalk between the ventral and dorsal stream) and contribute to our understanding of this type of crosstalk.

Atrophied HLV Cortices in AD and MCI
Our morphometric analyses between AD and CN revealed that the atrophy of HLV cortices in AD was widespread. Most of the cortices responsible for both the "what" (i.e., face, object and word visions) and "where" (i.e., motion and spatial visions) visual functions were involved. Previous neuropsychological and fMRI findings have also provided evidence that visual impairments in AD patients arise from both the ventral and dorsal streams (Kim and Park, 2010;Kirby et al., 2010). In addition, compared with controls, more amyloid deposits (one of the major reason for neuronal cell death) were found distributed in parietal, temporal, occipital and frontal regions, and these were related to the impairment of HLV functions (Edison et al., 2007;Rodrigue et al., 2009) in AD patients. These pathological findings further support our results and taken together provide a neurobiological basis for disrupted HLV functions in AD.
In addition, our results may support that the atrophy of HLV cortices develops at a late stage of the disease (i.e., AD/LMCI), because no significant differences were found between EMCI and CN. This might indicate that the atrophy of HLV cortices could be a late-stage outcome of Alzheimer's disease progression. In addition, the distribution of atrophied visual cortices in LMCI patients was comparatively localized (with only three significant regions), while much more widespread in AD patients (with 25 significant regions). Besides, the correlation results also suggested that the observed atrophy in HLV cortices was correlated to the whole GM atrophy and the deterioration of overall cognitive status in MCIs and ADs. Therefore, these finding may indicate that the atrophy of HLV cortices could be in accordance with the AD-related pathological progression. Thus, the extent of the visual cortical atrophy may be of some indicative value for the severity and disease progression of AD.
Meanwhile, it's worth noting that the correlation between MMSE and GM atrophy is found to be weak in most ROIs (with the r-values around 0.2-0.4). This may indicate that the HLV cortices changes are not entirely dependent on the whole cortical atrophy or disease severity of AD and MCI. Alternatively, it may also reflect and be related to some intrinsic or independent alteration pattern of the HLV network per se. In line with this, our results also revealed differential correlations between GM volume of HLV cortices and the cognitive tests of different domains. For example, the observed brain atrophy was found to be related to clock drawing test score in a few HLV regions, while was widely correlated to the performances of category fluency-animal and trail making test A & B (with the r-values around 0.3-0.4). This perhaps underlies a comparatively close relationship between HLV cortical atrophy and the impaired sematic memory and executive function in ADs and MCIs. In addition, for the logical memory test I & II and Boston naming test, it's worth noting that association between GM atrophy and the test performances was found to be more widespread in "what" visual cortices than in "where visual" cortices. On one hand, it is anticipated that Boston naming test score is closely related to the impairment of "what" visual cortices since it involves object recognition function which is primarily processed by ventral stream. On the other hand, findings in logical memory test may also provide evidence to support that atrophied "what" visual cortices may have an close interplay with the memory dysfunction under AD-related neuropathology. However, the specific causal relationship is to be investigated.
Furthermore, another interesting finding was that all of the three significant atrophied regions in LMCI group were included in the "where" visual cortices (i.e., regions related to motion and spatial visual functions), while no significant atrophy was found in the "what" visual regions. It's worth noting that a similar selective impairment pattern of "where" visual cortices in AD has also been reported by previous studies (Mandal et al., 2012). Taken together, we might speculate that "where" visual cortices are perhaps more vulnerable to the AD-related neuropathological changes than "what" visual cortices.

Limitations
Several limitations in the study should be noted. First, the number of included studies for ALE analysis is relatively low. This was mainly due to our criterion that only studies using low-level contrasts were included. It should be noted that most of the previous studies on HLV functions were performed with various types of high-level contrasts, which may be related to heterogeneous brain stimuli. Therefore, although the current inclusion strategy might largely limit the sample size, it will help maintain the homogeneity of meta-analysis. Meanwhile, it should be noted that the number of studies among different visual functions for some contrast analyses was not equivalent. Thus, the statistical power of these contrasts may have been affected. Second, the literature coverage was restricted to studies investigating HLV functions of the "what" and "where" streams, which led to only five categories of visual perception with comparatively large sample sizes. Further meta-analyses focusing on other aspects of the HLV functions will be encouraged to accomplish a more comprehensive probabilistic HLV atlas. Third, to enable a comparison between current VBM results and functional findings in the future study, we only chose the subjects both with 3D T1WI and resting-state fMRI data in the ADNI dataset to increase the comparability. Therefore, the subject number of 161, though acceptable, may be relatively small compared with the whole ADNI dataset. However, it should be noted that the MRI data in ADNI dataset were obtained from different scanners (Philips, GE, or Seimens) with different field strengths (3.0T or 1.5T), while our selected data group would only include those obtained with 3T Philips scanners. Therefore, our current data selection strategy also helped to control the influence of scanning factors.

CONCLUSIONS
In summary, this study combined ALE and VBM analyses to map the HLV cortices and their atrophy in AD progression. The ALE results showed brain regions for processing the visual perceptions of different categories included in the "what" and "where" visions, as well as their conjunctions and specifications. Based on the ALE maps, comparison results of GM volume demonstrated atrophy of HLV cortices in AD and LMCI individuals. In addition, the observed atrophy was correlated to the deterioration of overall cognitive status and to the cognitive domains of memory, execution, and object recognition functions. These results possibly indicate that the visual cortical atrophy may be of some indicative value for the disease severity and progression of AD. Therefore, these findings may enrich our understanding of the organizational patterns of HLV networks in the human brain, and it may also extend our knowledge on the evolvement of visual perceptual dysfunction in AD progression. Moreover, we hope our results could be applied in further studies (e.g., studies investigating functional connectivity based on our coordinates of HLV cortices) of HLV cortices and related diseases.

AUTHOR CONTRIBUTIONS
YD: design of the work, data analysis, manuscript writing, account for all aspects of the work; LS: final article approval; YL: accountable for all aspect of the work; PL: accountable for all aspect of the work; KL: accountable for all aspect of the work; WC: accountable for all aspect of the work; DW: final article approval.