Comparisons of continuous cortical thickness maps provided a comprehensive view of group differences and support a pattern of progressive atrophy from NC to SMCI to MMCI to early AD. Differences in average thickness are displayed for each group relative to the NC group (); statistical analyses were performed on the ROI measures and are described in sections that follow. The surface maps suggest that the SMCI group averages were at least 250 μm thinner in mesial temporal regions (orange/yellow) and 100–150 μm thinner in the lateral temporal gyri (dark red). The MMCI group averages were similar to SMCI in the mesial temporal regions; however, relative to the NC group, the MMCI group was 250 μm thinner in more extensive mesial and lateral temporal regions (orange/yellow), and 150 μm thinner in posterior cingulate, inferior parietal, and midfrontal regions (red). Cortical thickness of NC and AD groups differed by more than 450 μm in mesial temporal regions (yellow) and more than 200 μm in posterior cingulate and temporoparietal association areas (red/orange/yellow); midfrontal regions were also thinner by 150 μm (red). Each MCI subgroup also was compared with the AD group () to support this apparent progression. As expected, the MMCI group was more similar to AD overall; that is, there were fewer differences in thickness. However, the maps demonstrated unanticipated trends toward thinner cortices in MMCI (by ~100 μm or more) relative to the AD group in anterior cingulate, paracentral, and precentral regions (, bottom row in blue). Statistical analyses were performed on the ROI data as follows.
Region-of-Interest Group Comparisons
ROI analyses were performed across all groups, and these group effects are described here; the sections that follow detail regions affected by SMCI and MMCI specifically. Significant group effects were found for ventricular, temporal, posterior and rostral anterior cingulate, and posterior parietal ROIs (P's < 0.001); the group effect for caudal anterior cingulate did not reach significance (F(3,392) = 3.3, P = 0.02). Pairwise comparisons are presented in , and estimated marginal means for several ROIs are graphed in , supporting the continuous thickness maps ( and ). As expected, no significant group effects were demonstrated for the eTIV, cerebellar gray and white matter, caudate, putamen, cuneus, and pericalcarine regions (all F < 4.0; P > 0.05). Lateral occipital cortex and nucleus accumbens, however, were different across groups (F(3,392) = 11.9, P < 0.001; F(3,391) = 7.9, P < 0.001, respectively).
Pairwise comparisons for regions with a significant overall effect of group, in order of SMCI-NC comparison
Further exploratory investigations revealed significant group effects for all frontal ROIs (P's < 0.001), pre-(F(3,392) = 8.3, P < 0.001) and postcentral (F(3,392) = 9.5, P = 0.001) cortices, and a tendency for group differences in the paracentral (F(3,392) = 4.5, P = 0.004) region. Finally, the AD group tended to have less total volume of cerebral white matter (F(3,391) = 8.6; P < 0.001) and more white matter abnormalities (hypointensities on T1) (F(3,391) = 4.9; P = 0.002). For all regions with a significant overall effect of group, pairwise comparisons were performed; these are shown for each patient group relative to the NC group in .
The investigation of hemispheric differences revealed right dominant asymmetries for hippocampal and entorhinal cortices with no significant group × hemisphere interaction (F(1,391) = 4.0; P < 0.05; F(3,392) = 6.1; P < 0.05, respectively). The body of the lateral ventricle was larger on the left than the right (F(1,391) = 10.2, P < 0.005), which tended to become more pronounced with degeneration (F(1,391) = 2.2, P < 0.10). Additional left dominant asymmetries were reported in volumetric measures of the putamen (F(1,391) = 6.7, P = 0.01) and white matter abnormalities (F(1,391) = 11.6, P = 0.001). Within the inferior temporal gyrus, no main effect of hemisphere was present, although there was a significant diagnosis by hemisphere interaction, suggesting a right dominant asymmetry with increasing degeneration (F(1,391) = 3.2, P < 0.05).
Regions Affected in SMCI
Based on the hypothesis that SMCI and MMCI may represent early and later stages of prodromal AD, for the between-group comparisons gray matter volumes and cortical thickness measures were expected to be largest in the NC group, followed by SMCI, MMCI, and AD groups, in that order. Overall, the SMCI cohort evidenced smaller volumes and thinner cortices relative to NC in mesial and lateral temporal, fusiform, inferior parietal and precuneus, rostral posterior cingulate, and caudal midfrontal regions (, top row; ; ). Pairwise comparisons and percent differences between SMCI and NC groups are presented in , in the descending order of effect size for this comparison. All groups differed from each other for hippocampus (), entorhinal cortex (), amygdala, lateral middle () and inferior temporal gyri, banks of the superior temporal sulcus, and inferior parietal () cortices in the expected directions. NC individuals demonstrated significantly thicker parahippocampal cortex relative to all groups; SMCI, MMCI, and AD groups were not different from each other for this region. MMCI, SMCI, and NC groups differed in the expected directions for the following regions: lateral superior temporal cortex, fusiform, temporal pole, rostral posterior cingulate (), precuneus, and caudal midfrontal cortex; MMCI and AD were not significantly different for these same regions. Temporal horn of the lateral ventricle volume was significantly larger in SMCI relative to NC; SMCI and MMCI were both smaller relative to AD but not different from each other. SMCI did not demonstrate any significantly thicker regions relative to the NC group. In summary, the SMCI cohort distinguished itself from controls in temporal, fusiform, inferior parietal and precuneus, rostral posterior cingulate, caudal midfrontal, and temporal horn of the lateral ventricle measures (, top row; ; ), while they were similar to MMCI in volume of the temporal horn of the lateral ventricle and similar to MMCI and AD in parahippocampal thickness. Remaining regions were not significantly different between NC and SMCI.
Regions Affected in MMCI and AD
Overall, the MMCI cohort demonstrated additional regions of cortical thinning relative to NC and SMCI in lateral occipital, lingual, supramarginal, retrosplenial and anterior cingulate, frontal, and central cortices, alongside larger ventricles (, middle row; ). The MMCI reported effects in the frontal and central cortices were the results of exploratory analyses and were not expected to be detected in MMCI. The NC and SMCI groups differed from the MMCI and AD groups for lateral and third ventricle, lateral occipital, lingual gyrus, supramarginal gyrus, medial () and lateral orbitofrontal, rostral midfrontal (), and superior frontal cortices, as well as the pars orbitalis, pars triangularis, pars operculum, frontal pole, pre-, and postcentral cortices. Retrosplenial cortex () was thinnest in AD and significantly thinner in MMCI relative to NC and SMCI. Finally, both MMCI and AD groups demonstrated thinner anterior cingulate cortices relative to NC and SMCI groups. Unexpectedly, the MMCI group demonstrated thinner rostral and caudal anterior cingulate, pre-, post-, and para-central cortices relative to all groups, including AD (MMCI vs. AD, P < 0.05 for caudal anterior cingulate; P < 0.005 for remaining regions; , bottom row—blue regions); the AD group was not significantly different from the NC or SMCI groups on paracentral, caudal anterior cingulate, or left rostral anterior cingulate. In summary, the MMCI cohort distinguished itself from NC and SMCI in lateral occipital, lingual, supramarginal, retrosplenial, anterior cingulate, frontal, and central cortices (, middle row) and lateral and third ventricular measures. MMCI differed from AD primarily in thicker retrosplenial cortex alongside unexpectedly thinner anterior cingulate, pre-, post-, and para-central cortices (, bottom row—blue regions).
Relative Effect Size of ROIs at Different “Stages” of Disease
Presupposing that SMCI individuals may represent the earliest stages of AD, there is great interest in identifying regions that may be most sensitive to these early changes. The regions that differed between NC and SMCI included numerous temporal cortices, the temporal horn of the lateral ventricle, rostral posterior cingulate, and several parietal and frontal regions. Pairwise effect sizes and percent differences for these regions are shown in , supporting the potential use of the bilateral hippocampi, bilateral entorhinal, left amygdala, and left parahippocampal measures in differentiating NC and SMCI. The regions with the largest effect size in the SMCI-NC group comparison included the hippocampus and entorhinal cortex; these regions were on average 10% smaller relative to the NC group, alongside a greater than 15% increase in temporal horn volume. In MMCI, the difference relative to the NC group was greater in these regions, and the lateral inferior, middle, and superior temporal gyri and fusiform cortices demonstrated large effect sizes as well. In the comparison of NC and AD, the effect size for these same regions increased further, alongside larger differences for inferior parietal, banks of the superior temporal sulcus, retrosplenial, and some midfrontal regions. With a clinical diagnosis of AD, the percent difference from the NC group was approximately double that of the difference in SMCI for mesial temporal regions and nearly four times greater for left hemisphere lateral temporal and inferior parietal regions.
Hypothesized Trajectories of Regional Atrophy With Disease Progression
To further explore the hypothesized regional changes that may occur with progression from NC to SMCI to AD, we examined the contribution of linear and quadratic components to a theoretical trajectory of change across these cross-sectional data. We excluded the MMCI group because of unexpected findings described earlier and focused on a subset of regions along the typical path of AD progression. The mesial temporal regions, including the entorhinal cortex and hippocampus, demonstrated only a significant negative linear effect across groups (P < 0.001); other negative linear trajectories included the rostral posterior cingulate and medial orbitofrontal cortices (P < 0.001). Regions with a significant negative quadratic component included lateral middle temporal gyrus, retrosplenial, inferior parietal (P < 0.05), and rostral midfrontal cortices (P < 0.005); these regions demonstrated a steeper negative slope between SMCI and AD than between NC and SMCI, suggestive of a greater acceleration of degeneration “later” in the disease.
Significant group × region interactions supported a larger difference between groups (or a steeper slope) for the entorhinal cortex relative to the rostral posterior cingulate (F(2,297) = 39.1, P < 0.001) and medial orbitofrontal cortex (F(2,297) = 40.5, P < 0.001). There was no significant interaction for rostral posterior cingulate and medial orbitofrontal cortex (F(2,297) < 1.0, P > 0.05) supporting a similar trajectory.
A significant group × region interaction suggested a larger difference, or steeper slope, between groups for the lateral middle temporal gyrus, particularly between NC and SMCI, relative to the retrosplenial (F(2,297) = 3.2, P < 0.05), inferior parietal cortex (F(2,297) = 9.1, P < 0.001), and rostral middle frontal cortex (F(2,297) = 18.4, P < 0.001). There was no significant interaction for retrosplenial and inferior parietal (F(2,297) < 1.0, P > 0.05). There was a trend for an interaction between the inferior parietal and rostral midfrontal cortices (F(2,297) < 3.0, P = 0.05), suggestive of a steeper trajectory between groups in the inferior parietal cortex.