Fast  T1, T2  evaluation  with  machine  learning  for  quantitative cardiac  MRI

Quantitative T1/T2 mapping provides important cardiovascular prognostic value. Conventional dictionary-matching based methods are time consuming for cardiac T1/T2 mapping as the dictionary need to be generated on-line. In this work, we propose to use machine learning algorithms for faster T1/T2 prediction. Bloch equation simulation was used to generate training data. The XGBoost and DNN models were evaluated and compared based on simulation, phantom and in vivo studies. Results demonstrated that using the machine learning approach can generate cardiac T1 and T2 maps much faster while generating similar T1 and T2 values compared to the conventional dictionary-matching approach.