Towards 3D registration of fetal brain MRI and ultrasound

We propose a novel method for automatic alignment of fetal brain magnetic resonance image volumes and 3D ultrasound images. Volumetric fetal brain MRI is obtained from reconstructing 2D slices, and is then segmented using a preterm neonatal probabilistic atlas. The segmentation is further refined to extract features that are prominent in fetal brain ultrasound. From the MR segmentation, a pseudo-ultrasound image is simulated, taking into account the direction of the beam in the real ultrasound volume. The simulated and real ultrasound images are then affinely aligned using normalized cross-correlation. We show that the proposed approach performs better than the standard approach for multi-modal registration of the MR and ultrasound volumes driven by normalized mutual information.