This study utilized methods of computational neuroanatomy and high-dimensional pattern classification to investigate spatial patterns of brain atrophy, as well as their longitudinal change, in a group of 103 MCI individuals from the ADNI study, aiming to determine imaging markers that predict short-term conversion within a period of approximately 15 months, on average. Despite the relatively short clinical follow-up period of this study, significant differences at baseline were measured between MCI-C and MCI-NC. In particular, MCI-C had reduced GM volumes in a number of brain regions, including superior, middle and inferior temporal gyri, anterior hippocampus and amygdala, orbitofrontal cortex, posterior cingulate and the adjacent precuneous, insula, fusiform gyrus, and parahippocampal WM. Moreover, pronounced was the larger size of the temporal horns of the ventricles. It is therefore evident that MCI-C had already reached levels of widespread and significant brain atrophy at baseline. The short follow-up period further emphasizes the importance of this finding, since it is reasonable to assume that many of the MCI-NC are statistically expected to convert to AD in the near future, and are likely to also have significant and widespread atrophy at baseline.
The complexity of the pattern of differential atrophy between MCI-C and MCI-NC suggests that perhaps more sophisticated methods for measuring structural brain changes in MCI, AD, but also in normal aging, can be helpful for diagnosis and prognosis of the disease, compared to the most common approach that has been taken up to date in the neuroimaging literature [4
], namely to examine volumes of a small number of structures typically of the hippocampus and the entorhinal cortex. This is further bolstered by histopathological studies [47
] that have investigated the pattern of deposition of β-amyloid plaques and tau-pathology during the progression of AD, as well as with studies of magnetization transfer that indicated a more than expected widespread distribution of brain pathology [48
An important characteristic of this study is that, in addition to the group analysis, it also used individual patient classification using pattern recognition methods that have been extensively described and tested in the literature during the past few years [16
]. Thus, scores of AD-like patterns of brain atrophy were determined for each individual. The histogram of these scores () clearly indicates that almost all MCI-C effectively had AD-like brains at baseline, since all but 2 scores were well into the positive range. This result agrees with growing evidence of the presence of significant AD pathology in many MCI patients [50
], and further amplifies the need to investigate the development of AD pathology at much earlier stages, ideally when individuals are cognitively normal. Recent studies have reported that very similar spatial patterns of brain atrophy develop in cognitive normal elderly and in very mild MCI individuals years before significant cognitive decline is evident clinically [45
Further elaboration of the high-dimensional pattern classification technique, in which a classifier (Classifier 2) was constructed specifically to identify MCI-C on an individual basis, indicated that a relatively reasonable classification accuracy can be obtained (AUC = 0.77, maximum accuracy 81.5%), despite the fact that MCI-C and MCI-NC had about the same MMSE scores at baseline. In view of the heterogeneity of the MCI-NC group, which is reflected by the bi-modal distribution of the scores measuring AD-like atrophy (), this result is quite encouraging. Prediction of short-term conversion is important, as individuals classified as positive are likely to receive the most aggressive treatment. Moreover, it is clear from that a number of MCI-NC have AD-like structural profiles, which indicates that these individuals might become converters soon. We are following the clinical data posted under the ADNI web site, and will report clinical progression in future studies.
In addition to the higher GM and WM atrophy, and ventricular enlargement found in MCI-C, we also found higher level of periventricular abnormal tissue, which appears gray in T1 images and is known to indicate small-vessel disease. The role of vascular disease in AD is receiving increasing attention in the literature [52
]. Our results suggest that one of the significant differences between the MCI-C and MCI-NC subgroups is likely to be periventricular leukoareosis, and further support the need to examine vascular pathology in tandem with brain atrophy. Regardless of whether or not AD is pathophysiologically related to vascular disease, its clinical manifestation almost certainly depends on the concurrent presence of vascular disease [52
], therefore measuring vascular pathology is bound to be very important in predicting which individuals will convert to AD.
Against our expectation was the fact that the only significant group differences in the longitudinal rate of change were found in measures of periventricular abnormal WM and the temporal horn’s CSF volumes. It is generally believed that rate of change is likely to be a better predictor of clinical progression. Although this might be the case in theory, practically obtaining robust measurements of rate of change from a limited number of imaging measurements is problematic, due to variations in scanner performance and tissue contrast, but also due to noise in the image analysis measurements, including segmentation and spatial normalization. Even though we used the ADNI data that has been corrected for scanner distortions, in this study we found that baseline measurements are far better markers of patterns of atrophy distinguishing converters from non-converters. To some extent this is expected, since baseline atrophy is the cumulative effect of possibly many years of progression of AD pathology. The use of high-dimensional pattern classification methods also helped amplify the predictive value of baseline measurements, since they allowed us to utilize nonlinear contrasts among regional volumetric measurements from many brain regions simultaneously to derive optimally differentiating directions in high-dimensional spaces, in which the two groups become highly separable. Rate of change is nonetheless a significant biomarker in clinical studies evaluating treatment effects.
The finding of reduced WM volumes between the two subgroups is interesting and merits further research, as it has been previously reported in studies comparing CN, AD and MCI in ADNI [25
]. Although most WM reductions were periventricular, which agrees with the observed increase of abnormal periventricular tissue (), some inferior-medial temporal lobe regions also displayed reduced WM. Dense connections existing between the hippocampus and the posterior cingulate, which coupled with the early changes that have been reported in the posterior cingulate [64
], might imply that changes in WM might provide additional markers of disease progression, something that has traditionally not attracted much attention in the AD literature. A growing recent literature using diffusion tensor imaging further supports the importance of examining white matter changes in AD [66
], albeit the majority of these studies have been restricted to measuring quantities such as fractional anisotropy and diffusivity, and therefore have not differentiated between brain atrophy and other tissue changes that can potentially have vascular underpinnings (for example, both fractional anisotropy and diffusivity are known to be lower in leukoareosis). More sophisticated types of analysis of diffusion tensor images [76
] can potentially elucidate alterations of WM connectivity in AD.
In summary, we investigated the use of a high-dimensional pattern classification approach as a means to obtain a sensitive and specific biomarker of AD-like spatial patterns of brain atrophy, and of conversion from MCI to AD. The results of this short-follow-up study are encouraging and indicate that subtle and spatially complex patterns of brain atrophy can be detected and quantified with relatively high accuracy. Further follow-up will determine whether the baseline measurements can predict the time-to-conversion. Additional studies are required to test the generalization ability of this biomarker. However the fact that ADNI is a multi-center study, as well as the cross-validation used in our study, offer promise that this technology will prove to be robust across clinical sites.