Predicting Age from White Matter Diffusivity with Residual Learning.

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Computational Neuroscience

Background:

• Brain age estimation from MRI is crucial for identifying neurological disorders.
• Diffusion Tensor Imaging (DTI) offers insights into white matter microstructural changes.
• Distinguishing DTI's microstructural contributions from macrostructural ones is key for accurate brain age prediction.

Purpose of the Study:

• To develop white-matter-specific brain age estimation models.
• To isolate and leverage microstructural DTI features for improved age prediction.
• To assess the efficacy of feature extraction from ROIs versus deep learning (ResNet) for age estimation.

Main Methods:

• Two methods were used to predict age from DTI data, excluding macrostructural information.
• Method 1: Extracted microstructural features from Regions of Interest (ROIs).
• Method 2: Applied 3D Residual Neural Networks (ResNets) to learn features directly from registered DTI images, minimizing macrostructural variations.

Main Results:

• The ResNet method achieved a Mean Absolute Error (MAE) of 4.69 years for cognitively normal and 4.96 years for impaired participants.
• The ROI-based feature extraction method yielded higher MAEs (6.11 and 6.62 years, respectively).
• The ResNet model demonstrated superior ability in capturing subtle, non-macrostructural features for brain age prediction.

Conclusions:

• White-matter-specific age estimation using DTI microstructural features is feasible and effective.
• Deep learning approaches like ResNet significantly enhance the accuracy of brain age prediction by leveraging subtle features.
• Accurate brain age prediction from microstructural DTI data holds promise for early detection of neurological deviations.