Brain Age Estimation on T2-FLAIR Scans for Application to Multiple Sclerosis.

Area of Science:

• Neuroimaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Brain-predicted age difference (brain-PAD) shows clinical relevance in multiple sclerosis (MS).
• Existing brain age models often require 3D T1-weighted MRI scans, which are not standard in MS clinical practice, hindering translation.
• T2-FLAIR sequences are fundamental for MS diagnosis and monitoring, presenting an opportunity for novel biomarker development.

Purpose of the Study:

• To develop a brain age prediction model using T2-FLAIR MRI scans.
• To validate brain-PAD derived from T2-FLAIR as a biomarker for MS severity and progression.
• To compare the performance of T2-FLAIR-based models with traditional T1-weighted models.

Main Methods:

• Trained and evaluated 3D convolutional neural network models (Inception-ResNet-V2) using multicentre T2-FLAIR and T1-weighted MRI data.
• Collected data from healthy participants for model training and a cohort of people with MS (pwMS) and healthy controls for external validation.
• Utilized SmoothGrad for saliency mapping to identify key predictive regions and linear models for clinical validation against MS diagnosis, phenotype, duration, and Expanded Disability Status Scale (EDSS).

Main Results:

• T2-FLAIR based models achieved accurate brain age predictions (test set MAE = 3.31 years, R² = 0.944), comparable to T1-weighted models (test set MAE = 3.34 years, R² = 0.942).
• Brain age predictions were primarily influenced by subcortical regions, notably the thalamus.
• T2-FLAIR brain-PAD was significantly higher in pwMS (7.07 years) versus controls (-0.50 years) and correlated with disease duration (R=0.24) and EDSS (R=0.30).

Conclusions:

• Brain age prediction using T2-FLAIR MRI scans is feasible and accurate, matching the performance of T1-weighted scans.
• T2-FLAIR derived brain-PAD serves as a potential, easily accessible biomarker for assessing MS severity and tracking disease progression.
• This approach facilitates the clinical translation of brain age as a valuable tool in managing multiple sclerosis.