Different neural network architectures for predicting 9T CEST contrasts from 3T spectral data are investigated as well as the influence of different training data sets on the quality of resulting predictions. Although optimized convolutional neural network (CNN) architectures perform well, the best results were reached with a simpler feedforward neural network (FFNN). As CNNs have many hyperparameters to tune, this work forms a basis for CNN architecture optimization for the proposed super-resolution CEST application.
This abstract and the presentation materials are available to members only; a login is required.