Quasi-steady-state CEST prediction based on TCN-LSTM

An important topic in chemical exchange saturation transfer (CEST)-magnetic resonance imaging (MRI) is that certain CEST effects (such as amide proton effects) require sufficiently long time to reach steady state, which becomes a constraint the clinical application. To address this issue, we develop deep learning-based model predict quasi-steady-state (QUASS) from images with short times obtained actual experiments. We outline study follows: (1) A Bloch-McConnell equation designed obtain simulated Z-spectra data training dataset, all possible parameters of were optimized automatically acquire large amount reflecting metabolite combinations. (2) tumor-bearing rat was established on 7T horizontal diameter small animal MRI scanner, allowing ground-truth generation. (3) By combining advantages temporal convolutional network (TCN) and short-term memory (LSTM) modelling, TCN-LSTM developed QUASS data. (4) evaluate performance TCN-LSTM, multilayer perceptron (MLP), recurrent neural (RNN), LSTM, gated unit (GRU), BiLSTM TCN are included comparison experiment. In terms mean absolute error, mutual information (MI), structural similarity (SSIM) feature (FSIM), results show provides better prediction than its counterparts.