Deep Learning Enables Reduced Gadolinium Dose for Contrast-Enhanced Blood-Brain Barrier Opening Quantitative Measurement

Focused ultrasound (FUS) has the capability to open the blood-brain barrier (BBB), and magnetic resonance imaging (MRI) with contrast agents can be employed to detect this opening. However, the repeated use of gadolinium-based contrast agents (GBCAs) raises safety concerns. In this study, we propose an approach to modeling the volume transfer constant (K
^trans
) through deep learning, aiming to reconstruct artificial K
^trans
maps and enhance intensity using T1-weighted MRI scans with low-dose contrast agents. We successfully validated this concept using a state-of-the-art algorithm known as the temporal network, which focuses on extracting time domain features at the voxel level. Subsequently, we introduced the Spatiotemporal Network (ST-Net), a spatiotemporal convolutional neural network (CNN)-based deep learning architecture, incorporating a three-dimensional CNN encoder. The performance of ST-Net was evaluated on nine separate testing datasets, each representing FUS-induced BBB opening in different regions of the brain.