Normative cerebral cortical thickness for human visual areas

Studies of changes in cerebral neocortical thickness often rely on small control samples for comparison with specific populations with abnormal visual systems. We present a normative dataset for FreeSurfer-derived cortical thickness across 25 human visual areas derived from 960 participants in the Human Connectome Project. Cortical thickness varies systematically across visual areas, in broad agreement with canonical visual system hierarchies in the dorsal and ventral pathways. In addition, cortical thickness estimates show consistent within-subject variability and reliability. Importantly, cortical thickness estimates in visual areas are well described by a normal distribution, making them amenable to direct statistical comparison. Highlights Normative neocortical thickness values for human visual areas measured with FreeSurfer A gradient of increasing neocortical thickness with visual area hierarchy Consistent within- and between-subject variability in neocortical thickness across visual areas

While direct measurement of cortical thickness is not currently possible in vivo, indirect 59 quantification has become possible through MRI methods, particularly surface-based cortical 60 ribbon reconstruction (Fischl and Dale, 2000). This approach has seen significant refinement 61 with increased spatial resolution and improved signal-to-noise ratio in structural MRI (Fischl,  In the human visual system, cortical thickness varies significantly between retinotopically 68 organized areas. The primary visual cortex (V1) stands out for its unusually thin band of cortex 69 in contrast to surrounding cortical territory. Cortical thickness may then serve as a marker 70 segregating 'lower' sensory from 'higher' associative and integrative regions. In addition, Voss and Zatorre, 2012). This common approach potentially underestimates the variability of 79 cortical thickness in the population at large, either by measuring too small a sample, or by 80 biasing the control group towards a particular age, gender, ethnicity or other genotypic or 81 phenotypic descriptor in order to match the clinical sample. 82 In the interest of providing normative values for cortical thickness that better reflect the inter-83 individual variability in the population, we present a normative dataset of cortical thickness for 84 from the normal distribution were assessed with the Anderson-Darling test, with FDR 159 correction for multiple comparisons. The model goodness of fit was assessed with the 160 coefficient of determination (R 2 ) sampled in 100 bins across subjects. The effects of 161 demographic factors on mean cortical thickness were assess with a mixed-factorial ANOVA (3 162 factors), with age and gender as between-subject factors and visual area as the within-subject 163 independent factor. 164 Within-subject variability in the cortical thickness was assessed by subjecting it to a ANCOVA, 165 with visual area as the within-subject factor, subject identity as the between-subject factor, 166 while controlling for mean cortical thickness. Reliability of the within-subject mean cortical 167 thickness estimate was assessed with a leave-p-out resampling procedure (Celisse and Robin, 168 2008). In a given visual area, 90% of vertices were drawn with no replacement and averaged 169 to create one sample. For each subject 1,000 samples were obtained, and the span of the 95% 170 confidence interval over the resampled means was then taken as the reliability estimator. 171 Hemispheric asymmetric in cortical thickness was assessed with the two-way, single score 172 interclass correlation coefficient (ICC) (McGraw and Wong, 1996). Within-subject agreement 173 was obtained by comparing left and right hemisphere cortical thickness within subjects. 174 Between-subject agreement was obtained by comparing the left (and right) hemisphere 175 cortical thickness of each subject against the matching hemisphere from every other subject. 176 In addition, a hemispheric bias metric was calculated by subtracting the mean cortical 177 thickness, in mm, for the left hemisphere against the right hemisphere, in each subject, in each 178 visual area. Hemispheric biases were assessed with a series of one-sample t-tests to ascertain 179 if the mean bias across participants originated from a distribution with a mean of zero. FDR 180 correction for multiple comparisons was applied.

Between-subject variability in visual area cortical thickness 214
Establishing a baseline for between-subject variability in cortical thickness is critical for 215 statistical assessment in small-sampled studies. The large sample size of the HCP dataset allows 216 us to estimate the inter-individual variability in cortical thickness across an adult population 217 with varied demographic backgrounds, as a benchmark for studies of cortical thickness. 218 Surface-corrected mean cortical thickness estimates for each subject, at each visual area across 219 both hemispheres, are shown in Figure 3A. Here, we observe the group variability in cortical 220 thickness is approximately normally distributed in all regions of interest (see Appendix A for 221 quantile-quantile plots). Therefore, we summarize the group distribution as a one-dimensional 222 Gaussian function, also shown in Figure 3B. Performance of the Gaussian model was assessed with the coefficient of determination (R 2 ), 229 calculated over 100 equally spaced bins for each visual area (see Table 1). The model explained 230 ≥82% of variance in all regions of interest, including regions registered as non-normally 231 distributed VO1 (92%), TO1 (91%), IPS4 (87%) and SPL1 (88%). 232

Age and gender correlates 281
One factor that predicts cortical thickness is a person's age. Beyond the initial cortical 282 expansion during childhood development (Sowell et al., 2004), in adults, age is correlated with 283 a decrease in cortical thickness (Salat et al., 2004), and the rate of decrease is non-uniform  Wierenga et al., 2014). In order to assess these influences, we examined mean cortical 288 thickness in each region of interest with a mixed ANOVA model, with age in years and self-289 reported gender as between-subject variables and visual area as the within-subject variable. 290 There was a significant, albeit small, effect of age (F(15, 929) = 2.07, p = 0.009, η² = 0.03), with 291 a mean decrease in cortical thickness of 0.002 mm (± 0.001 SD) per year. This effect is 292 comparable to previous reports of cortical thinning due to normal aging (e.g. (Salat et al., 2004). Madan and Kensinger, 2017), as well as within the reliability estimates for this dataset (see 296 Figure 6), and must therefore be interpreted with caution. 297 We detected no effect of gender (F(1,929) = 0.12, p = 0.729, η² = 10 -3 ), and no interaction 298 between age and gender on cortical thickness (F(14,929) = 1.27, p = 0.219, η² = 0.02). Summary 299 plots of age and gender effects shown in Appendices D and E, respectively. 300

Within-subject variability in visual area cortical thickness 301
The estimation of cortical thickness in a cortical area typically involves assessing the distance 302 between white matter and cerebrospinal fluid at each vertex of the cortical surface. While the 303 individual vertex estimates are unreliable in isolation, over the extent of a region of cortex they 304 form a reliable indicator of the average cortical thickness for that region. However, the within-305 subject variance for the same cortical region is an informative metric, as it allows us to assess 306 the reliability of the average estimate. Here, we assess the within-subject variability of cortical 307 thickness estimates within each visual region. 308 We wish to address two questions; first, is within-subject variability comparable across 309 subjects? Second, are the individual mean cortical thickness estimates reliable? 310 To answer the first question, we looked at the standard deviation of cortical thickness within 311 subjects. For a given visual area, each vertex contains an independent estimate of cortical 312 thickness for that region, and therefore variability in the estimate is observable within areas, 313 on an individual basis. Figure 5 shows the standard deviation of vertex-wise cortical thickness 314 on a subject-by-subject basis. The standard deviation of the within-subject estimate was also 315 assessed with a ANCOVA model, introducing subject identity as the between-subject variable, 316 visual area as a within-subject variable while controlling for mean cortical thickness. No main 317 effect of subject was found (F(1, 934)   The second question addresses the issue of reliability in the subject-level outcome measure for 333 each visual area, the mean cortical thickness. If a small number of outlier values strongly skews 334 the mean cortical thickness, the outcome measure becomes unreliable. In order to assess the 335 effect of outliers, a leave-p-out resampling procedure (Celisse and Robin, 2008) was used where 336 for each visual area, 90% of vertices were drawn without replacement and averaged to create 337 a sample of cortical thickness. For each subject, 1,000 samples were drawn and the span of the 338 95% confidence interval taken as an estimator of within-subject reliability ( Figure 6).

Population estimates of cortical thickness 375
Accurate estimation of the natural distribution of cerebral cortical thickness in the population 376 at large is important for studies wishing to compare cortical thickness measured in special 377 populations against a normative baseline. This dataset serves as a baseline for such studies. complex intra-area variability in neuronal density, and of potential interest for further study. 404

Clinical relevance 405
There are a number of ways in which this type of data can be useful to the interpretation of 406 clinical data. Visual conditions that affect the cortex rather than the retina can be difficult to 407 investigate, particularly when V1 is not affected. Where dysfunction is at the level of V1 or 408 earlier in the visual hierarchy, the resulting visual complaint is usually a loss of visual field. In between the analysis pipeline and that of the normative dataset, or in the case of systematic 444 relationships between pipelines, by applying a scaling factor (Redolfi et al., 2015). Finally, the 445 investigator may wish to examine group differences between a special population, e.g. a clinical 446 group, and the normative data presented here. In such cases, attention must be paid to group 447 differences extraneous to the variable of interest, such as participant motion in the scanner 448  reveal preservation of brain architecture in "visual" cortex. Brain 132, 3467-3480.