2596

Quantitative Susceptibility Mapping from Deep-Learning Based Reconstruction of Undersampled Gradient-Recalled Echo Data

Ramin Jafari^1,2, Pascal Spincemaille², Thanh D. Nguyen², Junghun Cho^1,2, Martin R. Prince², and Yi Wang^1,2
¹Cornell University, Ithaca, NY, United States, ²Weill Cornell Medicine, New York, NY, United States

In this work we show how compressed sensing along with Deep Learning can be used to shorten acquisition time, perform image reconstruction, and generate QSM maps

Figure1. Comparison of fully sampled and undersampled (density masks; S1=14%, S2=28.9%, S3=50.7%) reconstruction of magnitude/phase images, water/fat separation results (PDFF(%), R2*(Hz), field (Hz)), and QSM (ppm).

Figure 2. ROI analysis comparing PDFF, R2*, field, and QSM in the liver and subcutaneous fat calculated from reconstructed undersampled data with different density masks (S1=14%, S2=28.9%, S3=50.7%).