Ultrafast submillimeter model-based NN quantification of whole-brain T1 and T2 using phase-cycled bSSFP

The bSSFP sequence is intrinsically sensitive to T1 and T2, motion robust, and allows highly efficient data acquisition. Slow convergence in qMRI parameter fitting can potentially be mitigated by machine learning, which benefits greatly from the availability of accurate ground truth data. This work presents an unsupervised model-based NN that incorporates the analytical bSSFP signal equation into the training loop, thus avoiding the need for ground truth relaxometry measurements and enabling instantaneous multi-parametric submillimeter whole-brain mapping of T1 and T2. NN performance was compared to MIRACLE quantitatively for in silico noise corrupted data and qualitatively for in vivo data.