Estimated Brain Age in Healthy Aging and Across Multiple Neurological Disorders

Li Chai, Jun Sun, Zhizheng Zhuo, Wei Ren, Xiaolu Xu, Yunyun Duan, De‐Cai Tian, Yutong Bai, Ningnannan Zhang, Haiqing Li, Yuxin Li, Yongmei Li, Fuqing Zhou, Jun Xu, James H. Cole, Frederik Barkhof, Jianguo Zhang, Huaguang Zheng, Yaou LiuJournal of Magnetic Resonance Imaging 2024 · DOI

This study utilizes deep learning to quantify the "biological age" of the brain, revealing that neurological disorders—particularly Multiple Sclerosis—significantly accelerate brain aging. By calculating the "Brain Age Gap" (the difference between chronological age and AI-estimated brain age), the researchers demonstrate that this metric is a powerful indicator of both cognitive decline and physical disability across diverse pathologies.

TL;DR

Researchers used a 3D convolutional neural network (CNN) to estimate "brain age" from MRI scans of over 4,200 individuals, finding that neurological disorders significantly accelerate biological brain aging. Multiple Sclerosis (MS) showed the most dramatic aging gap (~10 years), and across all conditions, a higher "brain age" correlated strongly with cognitive impairment and physical disability.

What Problem Does This Paper Address?

Chronological age is a blunt instrument in neurology. Two 60-year-olds can have vastly different brain health depending on genetics, lifestyle, and disease status. While we know that diseases like Alzheimer’s or Parkinson’s cause structural changes, we have lacked a unified, quantitative metric to compare the "rate of aging" across different types of neurological conditions (inflammatory, neurodegenerative, and vascular).

The authors address this by using the Brain Age Gap (BAG). If a 40-year-old’s brain looks like a 50-year-old’s on an MRI, their BAG is +10 years. This study aims to characterize these gaps across five major disorders to see if BAG can serve as a universal biomarker for brain health and clinical progression.

Methods

The researchers employed a multi-step pipeline centered on deep learning:

  • The Model: A 3D Convolutional Neural Network (CNN) previously trained on a massive dataset of 9,794 healthy individuals. This model learns to "read" 3D T1-weighted MRI scans to predict chronological age.
  • Bias Correction: AI models often overestimate the age of young people and underestimate the age of the elderly. The authors applied a statistical adjustment to remove this dependency, ensuring the BAG was a pure reflection of biological deviation.
  • Segmentation: They used FreeSurfer and the Lesion Segmentation Tool to quantify regional brain volumes and White Matter Hyperintensities (WMH).
  • Clinical Correlation: BAG was correlated against standard clinical scores like the MoCA (cognition) and EDSS (physical disability).

Datasets

The study involved a large, multi-center cohort:

  • Healthy Controls (HC): 2,913 individuals (from six centers in China and the UK-based Cam-CAN dataset).
  • Neurological Cohorts: 1,341 patients, including:
  • Multiple Sclerosis (MS): 331
    • Neuromyelitis Optica Spectrum Disorder (NMOSD): 189
    • Alzheimer’s Disease (AD): 239
    • Parkinson’s Disease (PD): 244
    • Cerebral Small Vessel Disease (cSVD): 338

Key Insights and Ideas

  • MS is the "Oldest" Brain: Surprisingly, MS patients exhibited the largest BAG (+10.3 years), surpassing even Alzheimer’s (+6.5 years). This highlights that chronic inflammation in MS may be even more "aging" to global brain structure than primary neurodegeneration in the mid-stages of AD.
  • Deep Gray Matter as a "Canary": Across almost all disorders, volume loss in deep gray matter structures—specifically the accumbens and thalamus—was the strongest driver of an increased brain age.
  • The "Accelerated Ager" in Healthy People: Even in the healthy control group, individuals with a BAG ≥ 5 years showed significantly higher white matter hyperintensities (WMH). This suggests that BAG could identify "at-risk" individuals before they show clinical symptoms of a disorder.
  • Cognitive Link: BAG wasn't just a structural curiosity; it was significantly associated with lower scores on the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) across the disease spectrum.

Key Results

  • Ranking the Aging Gap: The mean BAG for each disorder compared to age-matched controls was:
  • MS: +10.30 years (Effect size d = 1.30)
    1. AD: +6.50 years (Effect size d = 1.13)
    2. PD: +4.24 years (Effect size d = 0.90)
    3. cSVD: +3.24 years (Effect size d = 0.60)
    4. NMOSD: +2.96 years (Effect size d = 0.63)
  • Physical Disability: In MS patients, increased BAG was significantly correlated with higher EDSS scores (r = 0.17), meaning "older" brains were directly tied to worse physical mobility.
  • Model Accuracy: The underlying brain-age model achieved a mean absolute error (MAE) of only 2.63 years in its validation set, indicating high precision.

Limitations and Future Work

  • Cross-Sectional Design: The study looks at a snapshot in time. Longitudinal studies are needed to see if a high BAG today predicts a faster rate of decline tomorrow.
  • The "Black Box" Problem: While the CNN is accurate, it is difficult to pinpoint exactly which voxel-level features the AI is using to determine "age."
  • Lack of Lifestyle Data: Factors like smoking, exercise, and genetics—which heavily influence aging—were not accounted for in this analysis.
  • Future Direction: The authors suggest that BAG could eventually be used in clinical trials to see if new treatments can "slow down" the brain's biological clock.

Summary updated

Abstract

4 TECHNICAL EFFICACY: Stage 2.
BibTeX
@article{chai2024estimated,
  author = {Li Chai and Jun Sun and Zhizheng Zhuo and Wei Ren and Xiaolu Xu and Yunyun Duan and De‐Cai Tian and Yutong Bai and Ningnannan Zhang and Haiqing Li and Yuxin Li and Yongmei Li and Fuqing Zhou and Jun Xu and James H. Cole and Frederik Barkhof and Jianguo Zhang and Huaguang Zheng and Yaou Liu},
  title = {{Estimated Brain Age in Healthy Aging and Across Multiple Neurological Disorders}},
  journal = {Journal of Magnetic Resonance Imaging},
  year = {2024},
  doi = {10.1002/jmri.29667},
}
Figure