UNet-based self-attention network for accurate T1/T2 mapping by magnetic resonance fingerprinting

Motivation: Deep learning (DL)-assisted magnetic resonance fingerprinting (MRF) provides the opportunity for accurate parameter mapping with highly accelerated data acquisition. Goal(s): To develop a DL-assisted approach for high-resolution T1 and T2 mapping of the entire rodent brain using 3D MRF. Approach: A U-Net-based network was developed and trained on synthetic data generated from a mouse brain atlas. Its performance was evaluated with both phantom and in vivo experiments. Results: The network demonstrated robust, accurate T1 and T2 mapping at high undersampling rates, suggesting that whole-brain MRF can be achieved at reduced acquisition time without compromising accuracy. Impact: We present a novel deep learning-based 3D MRF method for accurate T1 and T2 mapping of the entire rodent brain using highly undersampled data, enabling dynamic MRF acquisition at higher temporal resolution.