We present an attention-based 3D convolutional encoder-decoder network to synthetize PET Cerebral Blood Flow (CBF) maps from multi-parametric MRI images without using radioactive tracers. Inputs to the prediction model are structural MRI (T1 and T2 fluid-attenuated inversion recovery [FLAIR]), and arterial spin labeling (ASL) perfusion MRI images. Results show that encoder-decoder networks, with attention mechanisms and customized loss functions, can adequately combine multiple MRI image types and predict the gold-standard oxygen-15-water PET CBF maps. Adequate quantification of PET from MRI has a great potential for increasing the accessibility of cerebrovascular diseases assessment for underserved populations, underprivileged communities, and developing nations.
This abstract and the presentation materials are available to members only; a login is required.