Optimization of sparse saturation transfer MR fingerprinting (ST-MRF) by ranking the importance of saturation transfer contrast features

Motivation: Optimizing ST-MRF sequence design is critical to accelerate image acquisition and improve reconstruction accuracy. Goal(s): To develop a deep-learning framework that can optimize MRF acquisition for tissue parameter determination with a minimal number of scan parameter settings. Approach: An interpretable neural network was designed to optimize MRF sequences by ranking the importance of saturation contrast features and evaluated using numerical phantoms and in vivo experiments at 3T. Results: Importance-ranking network-based sequence optimization demonstrated its ability to improve the choice of scan parameter values for quantification of tissue parameters. Sequence optimization achieved 1.7-fold acquisition acceleration without compromising the fidelity of the tissue parameter quantification. Impact: Ranking the importance of saturation transfer contrast features facilitates choosing the best combination of sequence parameters for tissue quantification with fingerprinting (ST-MRF). An interpretable network based on importance ranking can significantly accelerate data acquisition for ST-MRF and conventional Z-spectral acquisition.