Synopsis

Keywords: Dementia, Modelling, Brain Age

A certain subset of the aging population maintains the cognitive abilities with minimal structural changes with aging while another subset of people transforms to mild cognitive impairment and dementia with manifold structural variations. Since a subset maintains the cognitive abilities, brain structural volume with aging while another subset undergoes cognitive-impairment and drastic brain structure alterations, we believe that the brain age pattern is distinct from chronological age. Here, we have used a set of MRI determined brain volumetry to determine the brain age as a measure of normal and pathological aging.

INTRODUCTION

The decline in cognitive abilities with aging may be a manifestation of brain structure atrophy and hypertrophy with aging. While a certain set of aging individuals have an early onset of cognitive disabilities, others manifest cognitive impairments at a very late age. MRI-determined brain structure health serves as a potential imaging biomarker for studying normal and pathological aging as the changes in brain volumetry may happen decades before the manifestation of clinical symptoms of mild cognitive impairment and dementia. It remains poorly understood what are the brain biomarker changes that indicate a transformation of an aging individual into mild cognitive impairment and dementia while a similar-aged individual continues to preserve brain structure and cognitive abilities. Therefore, we hypothesize that understanding the Brain age (BAG) based on the brain region volume may be a potential indicator of the trajectory of normal and pathological aging compared to Chronological age (CAG). Here, we have developed a platform for predicting brain age by employing optimal numbers of MRI-segmented brain structures and covariates.

METHODS

Brain volumes and small vessel disease pathology (White matter hyperintensity) obtained using T1w and T2-FLAIR MRI from cognitive normal (CN) subjects (Nmri = 2038) in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) were used to develop a predictive model for brain age determination. T1w images were segmented to obtain the volume of Gray Matter, White matter, Cerebrum Gray Matter, Cerebrum White matter, Cerebrospinal fluid, Left hippocampus, Right Hippocampus, and Gray-White matter ratio (GM/WM). The volumes were normalized to Intracranial volumes. Total white matter hyperintensity (WMH) volume was determined using the T2-FLAIR image. The entire dataset was normalized and the outliers (> 3 or < -3) were removed. For subjects with multiple scans, a single scan was selected randomly. Various combinations of the MRI-determined brain structure volumes were used to develop a brain age model that employed a “boosting” architecture with generalized additive model-based prediction¹(Fig. 1). The difference between estimated brain age (BAG) and chronological age revealed the individual brain age gap estimation(BrainAGE) score², which indicates brain health and cognitive abilities. For BrainAGE, the positive values indicated advanced structural brain maturation/aging, whereas negative values indicated delayed structural brain maturation/aging. In longitudinal studies, increasing BrainAGE scores were indicative of accelerated brain aging over time. Thus, the individual BrainAGE score directly quantifies the amount of acceleration or deceleration of brain maturation/aging in terms of years. For example, if a 64-year- old individual shows a BrainAGE score of +6 years, the typical brain volumetry of this individual resembles the brain volumetry of a 71- year- old cognitively normal individual (Fig. 2).

RESULTS and DISCUSSION

Neuroanatomic volumes of cerebrum gray matter, cerebrum white matter, cerebrospinal fluid, left and right hippocampus and microvascular pathology such as WMH were used to investigate the pattern of regional brain structure changes with aging across cognitively normal subjects. A significant progressive decline was observed for gray matter, white matter, and hippocampus with age accompanied by a progressive increase in the volume of CSF and WMH (Fig. 3). Upon investigation of the association between estimated brain age and chronological age using the segmented brain volumes, CSF volume exhibited a significantly positive association (+0.65 ) with chronological age in normal subjects followed by cerebrum gray matter (-0.449) and cerebrum white matter (-0.46 ). Upon a ten-fold cross-validation score for various combinations of neuroanatomical volume to understand the brain age, a combination of cerebrum gray matter, cerebrum white matter, total gray matter, total white matter, total hippocampus volume, CSF, and WMH volume showed the highest cross-validation accuracy for brain age determination (Fig.4). The bias in brainAGE was corrected by using a simple linear regression with brainAGE as the independent variable and the chronological age as the dependent variable (Fig. 5B) in which the effect of the chronological age was removed to obtain the corrected brainAGE³. Application of the BrainAge prediction model using the optimized neuroanatomical features developed from the cognitively-normal cohort in a random sample of Mild Cognitively Impaired (N=50) subjects successfully predicted increased brain age in the MCI group compared to the chronological age with significantly higher accuracy (Fig. 5C).

CONCLUSION

We have established a highly precise brain age estimation model using ten MRI-determined neuroanatomical features wrapped in generalized additive models using splines and boosting. We also elucidate the need for correction of biases for accurate determination of the brainAGE delta. The model successfully showed that subjects have a dissociation between brain age and chronological age, where brain age is higher than their chronological age.

Acknowledgements

MRI and Cognitive data were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering