Active dictionary learning: fast and adaptive parameter mapping for dynamic MRI

A new parameter mapping method, active dictionary learning, for dynamic MRI is proposed. This method trains a neural network adaptively by AI-guided MR signal simulation. For an MRF sequence with M₀, T₁, T₂, B₁, and ΔB₀, our method successfully estimates the parameters much faster than conventional methods (ours: 30 min for whole process; dictionary methods: 6 hours for generation, 36 hours or 3 hours for matching). AI-guided active dictionary learning enables adaptive quantification of out-of-range parameters and efficient computation, suggesting the usefulness of the method not only in dynamic imaging but also in applications where adaptation to parameters is necessary.