Brain Age Estimation From MRI Using Cascade Networks With Ranking Loss.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomarker Discovery

Background:

• Accurate prediction of chronological age from neuroimaging data using deep neural networks is feasible.
• Predicted brain age can serve as a biomarker for aging-related diseases.

Purpose of the Study:

• To propose a novel 3D convolutional network, TSAN, for accurate brain age estimation from T1-weighted MRI data.
• To evaluate TSAN's performance and its potential as a biomarker for neurodegenerative diseases.

Main Methods:

• A two-stage cascade network architecture (TSAN) was developed for brain age estimation.
• Novel ranking losses combined with Mean Square Error (MSE) loss were applied.
• Densely connected paths were utilized to integrate multi-scale feature maps.

Main Results:

• TSAN achieved high accuracy in brain age estimation (MAE: 2.428, PCC: 0.985) on 6586 MRIs.
• The brain age gap effectively distinguished Alzheimer's disease (AD) and Mild Cognitive Impairment (MCI) from healthy controls (HC).
• Classification AUC for AD/HC and MCI/HC were 0.904 and 0.823, respectively.

Conclusions:

• TSAN provides accurate brain age estimation.
• The brain age gap is a potent biomarker for dementia risk.
• Brain age gap analysis shows potential for early-stage dementia risk screening.