Infant Brain Age Estimation With T1w/T2w Ratio MRI: A Myelination-Aware Deep Learning Approach.

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Artificial Intelligence

Background:

• Brain age estimation is a noninvasive MRI biomarker for neurodevelopment.
• Early-life brain age estimation is underexplored, especially using myelination-sensitive MRI and biologically informed models.

Purpose of the Study:

• To develop and evaluate a deep learning framework for infant brain age estimation.
• Utilize T1w/T2w ratio MRI for enhanced accuracy and biological relevance.

Main Methods:

• Developed a 3D convolutional neural network framework incorporating multi-task learning.
• Trained models using T1w, T2w, and T1w/T2w ratio inputs with biologically defined age labels.
• Evaluated performance using cross-validation and external validation on infant cohorts.

Main Results:

• T1w/T2w ratio models achieved the best performance (MAE: 1.489 ± 0.302 months, r = 0.966 ± 0.012).
• Auxiliary-task and multi-scale modeling further improved accuracy (MAE: 1.203 months, r = 0.979).
• External validation confirmed the T1w/T2w ratio model's effectiveness (MAE: 1.16 months), with Grad-CAM highlighting myelination-relevant white matter.

Conclusions:

• Biologically informed deep learning with T1w/T2w ratio MRI enables accurate and interpretable infant brain age estimation.
• The framework demonstrates promising cross-scanner performance.
• This approach may advance MRI-based assessment of early brain maturation.