High-dimensional morphometryDisentangling normal aging from Alzheimer's disease in structural magnetic resonance images
Introduction
The objective of computational anatomy when applied to neurodegenerative diseases, such as Alzheimer's disease (AD), is to understand the pathological changes affecting brain morphology (Frisoni et al., 2010, Scahill et al., 2002). However, the morphology of the brain affected by AD is not completely related to the disease, especially in asymptomatic and prodromal stages, because the brain structure is also the result of patient phenotype and clinical history. In a brain affected by AD, we can identify 2 major processes contributing to morphological changes: normal aging and AD pathology itself.
- •
Age-related anatomical changes. It is known that aging is related to progressive impairment of neural mechanisms (Burke and Barnes, 2006), to chemical alterations in the brain, and to changes in cognition and behaviour (Hof and Mobbs, 1984). It has been observed that morphological changes in the aging brain are heterogeneous and primarily lead to gray matter loss in frontal, temporal, and parietal areas (Long et al., 2012, Sowell et al., 2003).
- •
Disease-related anatomical changes. AD is a neurodegenerative disease characterized by the cooccurrence of different phenomena. It starts with the deposition of amyloid plaques and tau proteins in neurofibrillary tangles, which is followed by the development of function brain loss, and finally by widespread structural atrophy (Jack et al., 2010). The typical pattern of brain tissue loss seen in AD mirrors tau deposition (Thompson et al., 2003) and involves primarily hippocampi, the entorhinal cortex, the posterior cingulate, and secondarily the temporal, parietal, and frontal cortices (Frisoni et al., 2010). Aging is the primary risk-factor in AD and leads to patterns of structural loss overlapping the pathological ones. However, the magnitude of brain atrophy caused by AD is generally striking compared with normal aging. As claimed in previous studies, AD is more likely to be a pathological state concurrent to aging, identified by specific biochemical and structural hallmarks (Barnes, 2011, Nelson et al., 2011).
Being able to separately model healthy aging and AD would allow us to describe a given anatomy as being composed of distinct and concurrent factors. Such a decomposition would be extremely interesting not only to improve the understanding of the disease but also for clinical purposes, such as for early diagnosis and for the development of drugs targeting the atrophy specific to the pathology. It is important to notice that, although brought on completely different biological mechanisms, aging and AD often map to common areas, and the correct identification of the respective contributions can be difficult, especially in morphometric studies. Moreover, it is plausible that these phenomena are not completely independent and may overlap to create a positive “feedback” process. Thus, the onset of pathological changes may lead to accelerated global aging in the long term (Fjell et al., 2012), and vice versa.
A reliable estimate of the aging component is also important for modeling the evolution of the disease and for subsequent statistical analysis. When comparing the longitudinal observations from different clinical groups, at different aging stages, it is crucial to correctly position the observations on the time axis. This is not straightforward because the disease appears at different ages and chronologically older brains may have greater structural integrity than younger ones affected by the pathology. Therefore, it might be of practical interest to compute an index of age shift “relative” to a reference anatomical model.
The idea of modeling the time course of AD with respect to clinical and demographic factors was proposed in previous statistical studies (Ito et al., 2012, Samtani et al., 2012, Yang et al., 2011). However, these works were limited to scalar observations such as clinical scores and demographics and thus do not provide an explicit model which relates structural changes in the entire brain to the disease and aging. Moreover, the disease progression was identified by clinical measures and was not therefore explicitly associated with a temporal time course.
Although imaging-based surrogate measures of aging have been provided by different methodological studies (Davatzikos et al., 2009, Franke et al., 2010, Konukoglu et al., 2013), the idea of separately investigating aging and residual morphological changes has not been proposed before.
The objective of this work is to introduce a framework to identify and disentangle the brain anatomical changes related to normal aging from those related to other biological processes, such as AD. In particular, our framework is based on the hypothesis that relates the development of AD to the abnormal accumulation of beta-amyloid (Aβ) peptide in the brain (Jack et al., 2010). We thus define “normal aging” as the morphological brain evolution which is not caused by Aβ. This evolution is modeled by nonlinear registration and is used as a reference to characterize observed anatomy as a contribution from normal morphological aging (normal aging process) plus a specific morphological process that encodes the subject’s specific variability such as pathological traits. We test our framework on healthy participants positive to the cerebrospinal fluid (CSF) Aβ42 marker, in participants affected by mild cognitive impairment (MCI) and in AD patients.
The method is based on diffeomorphic nonlinear registration and is detailed in Section 2. In Section 3, we show that such a framework provides a meaningful and accurate description of anatomical brain changes across the stages of AD, characterized by increased morphological aging plus specific and local atrophy features.
Section snippets
Methods
The proposed method relies on specific modeling assumptions which are summarized here:
- •
The model of normal aging is derived from imaging data by applying a registration-based protocol detailed in Section 2.1. In particular, we assume that normal aging can be modeled through nonlinear registration as a smooth and continuous process that can be extrapolated backward and forward in time beyond the observed imaging follow-up time. Moreover, we assume that normal aging is a constant process in time,
Experimental data
Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, the Food and Drug Administration, private pharmaceutical companies, and nonprofit organizations, as a $60 million, 5-year public-private partnership. The Principal Investigator of this initiative is Michael W. Weiner,
Conclusions
We proposed a method to describe brain anatomy as contributions of 2 independent processes: morphological aging and a specific component. These components identify different clinical stages, and are compatible with the hypothesis that points to the abnormal levels of CSF Aβ42 as a presymptomatic marker of AD in the early stages.
We showed that more advanced AD stages (from Aβ+ to MCI converters, and finally to AD) are associated with both “virtually older” brains, and with increased specific
Disclosure statement
The authors have actual or potential conflicts of interest.
Acknowledgements
This work was partially funded by the European Research Council (ERC advanced Grant MedYMA 2011-291080), ANR blanc Karametria and the EU project Care4Me. Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI; National Institutes of Health Grant U01 AG024904). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following:
References (42)
- et al.
Local MRI analysis approach in the diagnosis of early and prodromal Alzheimer’s disease
Neuroimage
(2011) - et al.
Automatic classification of patients with Alzheimers disease from structural MRI: a comparison of ten methods using the ADNI database
Neuroimage
(2011) - et al.
Imaging cerebral atrophy: normal ageing to Alzheimer’s disease
Lancet
(2004) - et al.
Estimating the age of healthy subjects from T1-weighted MRI scans using kernel methods: exploring the influence of various parameters
Neuroimage
(2010) - et al.
Automatic differentiation of anatomical patterns in the human brain: validation with studies of degenerative dementias
Neuroimage
(2002) - et al.
Average brain models: a convergence study
Computer Vis. Image Understanding
(2000) - et al.
3D characterization of brain atrophy in Alzheimer's disease and mild cognitive impairment using tensor-based morphometry
Neuromage
(2008) - et al.
Hypothetical model of dynamic biomarkers of the Alzheimer's pathological cascade
Lancet Neurol.
(2010) - et al.
Neighbourhood approximation using randomized forests
Med. Image Anal.
(2013) - et al.
Healthy aging: an automatic analysis of global and regional morphological alterations of human brain
Acad. Radiol.
(2012)
LCC-Demons: a robust and accurate diffeomorphic registration algorithm
Neuroimage
Anterior and posterior cingulate cortex volume in healthy adults: effects of aging and gender differences
Brain Res.
Tracking atrophy progression in familial Alzheimer's disease: a serial MRI study
Lancet Neurol.
Longitudinal MRI atrophy biomarkers: relationship to conversion in the ADNI cohort
Neurobiol. Aging
Atrophy of the cerebellar vermis in ageing: a morphometric and histologic study
J. Neurol. Sci.
Relations between brain tissue loss, CSF biomarkers, and the ApoE genetic profile: a longitudinal MRI study
Neurobiol. Aging
A log-Euclidean framework for statistics on diffeomorphisms
Med. Image. Comput. Comput. Assist. Interv.
Secrets of aging: what does a normally aging brain look like?
F1000 Biol. Rep.
Neural plasticity in the ageing brain
Nat. Rev. Neurosci.
Sex differences in brain aging: a quantitative magnetic resonance imaging study
Arch. Neurol.
Relation of education to brain size in normal aging: implications for the reserve hypothesis
Neurology
Cited by (59)
Degenerative adversarial neuroimage nets for brain scan simulations: Application in ageing and dementia
2022, Medical Image AnalysisCitation Excerpt :Briefly, the input to many disease progression models (Fonteijn et al., 2012; Young et al., 2014; Jedynak et al., 2012; Donohue et al., 2014; Lorenzi et al., 2019; Oxtoby et al., 2018; Young et al., 2018) is unstructured data such as scalar biomarkers, including those extracted from MRI for assessing neurodegeneration. Spatiotemporal models, e.g., Lorenzi et al. (2015), Durrleman et al. (2013), attempt to incorporate structural information from the MRI themselves. All these models aim to produce quantitative templates of disease progression that promise utility for, e.g., recruiting the right patients at the right time into clinical trials.
DeepAtrophy: Teaching a neural network to detect progressive changes in longitudinal MRI of the hippocampal region in Alzheimer's disease
2021, NeuroImageCitation Excerpt :Compared to clinical cognitive tests, MRI-derived biomarkers are more sensitive to change over time, particularly in early stages of AD progression, therefore requiring a smaller cohort and/or shorter trial duration to detect a significant change due to treatment (Ard and Edland, 2011; Jack et al., 2010; Sperling et al., 2011; Weiner et al., 2015). While there is little debate that longitudinal structural MRI is a critical biomarker for AD clinical trials and disease development estimations (Cullen et al., 2020; Lawrence et al., 2017; Lorenzi et al., 2015b), it remains an open question on how to optimally extract measures of change from MRI scans. The straightforward approach of measuring the volume of the hippocampus (or other structure of interest) at multiple time points independently and then comparing them longitudinally suffers from relatively high coefficient of variability in these measurements (Leow et al., 2006; Schuff et al., 2009).
Learning to synthesise the ageing brain without longitudinal data
2021, Medical Image Analysis
- †
Data used in preparation of this article were obtained from the Alzheimers Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.