FetalSR: Super-resolving High-isotropic-resolution Image Volume from Single Thick-slice Stack with Deep Learning for Fetal Brain Morphometry

Motivation: High-resolution volume reconstruction is crucial for fetal brain MRI studies. Current methods, such as NiftyMIC, require thick-slice stacks acquired in at least three different orientations, which is time-consuming and prone to failure due to motion artifact. Goal(s): To reconstruct high-isotropic-resolution fetal brain MRI volume from a single thick-slice stack acquired in any axial/sagittal/coronal orientation. Approach: FetalSR pipeline consisting of an efficient slice-to-template registration workflow and a deep learning-based super-resolution reconstruction approach is proposed. Results: FetalSR provides high-quality high-resolution images highly similar to reference images, accurate results in downstream brain tissue segmentation tasks, and robustness to cases that conventional NiftyMIC method fails to reconstruct. Impact: FetalSR minimizes data needed for high-isotropic-resolution fetal brain volume reconstruction, reduces scan time and increases reconstruction robustness. It enables quantification of brain morphological features of developmental and abnormal fetuses in large-scale and a wider range of clinical and neuroscientific studies.