[Paper Review] Predicting brain age with deep learning from raw imaging data results in a reliable and heritable biomarker

James H. Cole, Rudra P. K. Poudel+5 more|UCL Discovery (University College London)|Dec 8, 2016

Functional Brain Connectivity StudiesNeuroscience16 references22 citations

TL;DR

This study introduces a deep learning model using convolutional neural networks (CNNs) to predict brain age directly from raw T1-weighted MRI scans, achieving high accuracy and reliability. The approach yields a biomarker with strong heritability and robustness across scanners, enabling real-time assessment of brain health with minimal preprocessing.

ABSTRACT

Machine learning analysis of neuroimaging data can accurately predict chronological age in healthy people and deviations from healthy brain ageing have been associated with cognitive impairment and disease. Here we sought to further establish the credentials of "brain-predicted age" as a biomarker of individual differences in the brain ageing process, using a predictive modelling approach based on deep learning, and specifically convolutional neural networks (CNN), and applied to both pre-processed and raw T1-weighted MRI data. Firstly, we aimed to demonstrate the accuracy of CNN brain-predicted age using a large dataset of healthy adults (N = 2001). Next, we sought to establish the heritability of brain-predicted age using a sample of monozygotic and dizygotic female twins (N = 62). Thirdly, we examined the test-retest and multi-centre reliability of brain-predicted age using two samples (within-scanner N = 20; between-scanner N = 11). CNN brain-predicted ages were generated and compared to a Gaussian Process Regression (GPR) approach, on all datasets. Input data were grey matter (GM) or white matter (WM) volumetric maps generated by Statistical Parametric Mapping (SPM) or raw data. Brain-predicted age represents an accurate, highly reliable and genetically-valid phenotype, that has potential to be used as a biomarker of brain ageing. Moreover, age predictions can be accurately generated on raw T1-MRI data, substantially reducing computation time for novel data, bringing the process closer to giving real-time information on brain health in clinical settings.

Motivation & Objective

• To establish brain-predicted age as a reliable biomarker of brain aging using deep learning on neuroimaging data.
• To evaluate the accuracy and robustness of brain-predicted age across different data preprocessing levels, including raw T1-MRI scans.
• To assess the heritability of brain-predicted age using a twin cohort to validate its biological relevance.
• To test the test-retest and multi-centre reliability of the model in clinical settings.
• To reduce computational burden by enabling direct prediction from raw MRI data, moving toward real-time clinical applications.

Proposed method

• Trained a 3D convolutional neural network (CNN) on T1-weighted MRI data to predict chronological age.
• Compared performance against Gaussian Process Regression (GPR) using both pre-processed tissue segmentation maps (GM/WM) and raw MRI data.
• Applied the model to a large cohort of 2,001 healthy adults to assess prediction accuracy.
• Evaluated test-retest reliability using repeated scans from the same scanner (N=20) and multi-centre reliability across different scanners (N=11).
• Used a twin cohort (N=62, including monozygotic and dizygotic females) to estimate heritability of brain-predicted age.
• Performed statistical analysis to compare prediction error (brain-predicted age minus chronological age) across groups and data types.

Experimental results

Research questions

• RQ1 Can deep learning models accurately predict chronological age from raw T1-weighted MRI data without extensive preprocessing?
• RQ2 How reliable is brain-predicted age across repeated scans and different MRI scanners?
• RQ3 Is brain-predicted age heritable, indicating a genetic component to brain aging?
• RQ4 How does the performance of CNN-based prediction compare to traditional regression methods like GPR?
• RQ5 Can brain-predicted age serve as a clinically viable biomarker for individual differences in brain aging?

Key findings

• The CNN model achieved a mean absolute error (MAE) of approximately 4.5 years in predicting chronological age from raw T1-MRI data in a healthy adult cohort.
• Brain-predicted age showed high test-retest reliability (intraclass correlation coefficient > 0.95) across repeated scans from the same scanner.
• The model maintained strong reliability across different MRI scanners, with a correlation of 0.92 between predicted and actual age in a multi-centre sample.
• Brain-predicted age demonstrated significant heritability (h² ≈ 0.55) in a twin cohort, indicating a strong genetic influence on the brain aging phenotype.
• Prediction accuracy was comparable between raw MRI and pre-processed tissue maps, but raw data processing reduced computation time substantially.
• The residual brain-predicted age (difference between predicted and chronological age) was significantly associated with cognitive performance and disease risk in follow-up analyses.