Fetal MRI by Robust Deep Generative Prior Reconstruction and Diffeomorphic Registration

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Radiology

Background:

• Fetal MRI is challenged by motion artifacts and non-uniform sampling across slices.
• Current volumetric reconstruction methods require regularization to handle non-homogeneous and non-isotropic data.

Purpose of the Study:

• To develop a robust volumetric reconstruction method for fetal MRI using a deep generative prior.
• To improve image quality and gestational age prediction accuracy in fetal MRI.

Main Methods:

• Proposed a deep generative prior integrated with a diffeomorphic volume-to-slice registration.
• Validated the method on 72 fetal datasets (20-36 weeks gestational age).
• Compared results against a state-of-the-art reconstruction technique.

Main Results:

• The proposed method demonstrated improved image resolution and quality compared to existing methods.
• Achieved more accurate gestational age prediction, outperforming previous approaches.
• Combining fetal brain and trunk information yielded a mean absolute error of 0.618 weeks (R²=0.958).

Conclusions:

• The deep generative prior offers robust volumetric reconstruction for fetal MRI.
• Enhanced image quality and gestational age prediction accuracy were observed.
• Integrating multi-organ information improves diagnostic capabilities in fetal MRI.