Abstract
The corpus callosum (CC) plays a fundamental role in inter-hemispheric integration, supporting motor coordination, language, and sensory processing. The ability to accurately assess the morphology of CC prenatally would allow the establishment of normative neurodevelopmental trajectories. Automated fetal brain tissue segmentations typically rely on T2-weighted (T2w) MR images, however limited contrast between the developing CC and adjacent structures hampers precise delineation. We proposed a multi-modal deep learning model that combines anatomical (T2w) and microstructural (fractional anisotropy, FA) information to address this limitation. Leveraging 192 fetal brain scans of the developing Human Connectome Project from 21 to 38 gestational weeks, we generated high-fidelity ground-truth annotations using an initialization pipeline that combines morphological filtering of FA and color FA images. We then systematically evaluated three segmentation models based on T2w, FA, and combined (T2w + FA) inputs. The dual-channel (T2w + FA) model significantly outperformed state-of-the-art methods, achieving a Dice score of 0.74 ± 0.07, volume similarity of 0.92 ± 0.06, and perfect topological consistency (Euler Difference = 0). Furthermore, performance remained quite stable under a simulated reduced-acquisition protocol with fewer diffusion-weighted directions, suggesting potential applicability to lower-resolution clinical diffusion MRI, though further validation on external sites and routine clinical-quality acquisitions is needed. Morphometric analysis revealed robust gestational age-dependent growth trajectories for most of the extracted parameters, while FA remained stable throughout gestation (Spearman ρ = −0.02), suggesting that the microstructural properties underlying callosal anisotropy change little over this developmental period. This multi-modal model offers a robust tool for prenatal CC assessment, enabling quantitative characterization of typical develop ment and facilitating the identification of derailments from the expected developmental trajectory. Code and trained models are publicly available at https://github.com/tommaso-ciceri/Multi-modal-deep-learning-for-fetal-MRI-corpus-callosum-segmentation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PIPPI_014.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{DiMar_Multimodal_MICCAISAT2026,
author = { Di Stefano, Marina AND Peruzzo, Denis AND Montano, Florian AND Licandro, Roxane AND De Luca, Alberto AND De Zwarte, Sonja M. C. AND Leemans, Alexander AND Ciceri, Tommaso},
title = { { Multi-modal deep learning for fetal corpus callosum segmentation and characterization in MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17257},
month = {pending},
page = {pending}
}