Enhancing Corpus Callosum Segmentation in Fetal MRI via Pathology-Informed Domain Randomization

2025-08-28

AI Summary

Accurate segmentation of the corpus callosum (CC) in fetal MRI is hindered by the rarity of corpus callosum dysplasia (CCD) cases and severe scarcity of annotated pathological data, leading to poor generalization of deep learning models. Method: We propose a pathology-guided domain randomization synthesis framework that embeds anatomically grounded morphological priors of CCD—such as shortened length and structural distortion—into the generative process of healthy fetal MRI scans, producing synthetic yet anatomically plausible CC lesions without requiring real pathological annotations. Contribution/Results: This approach significantly improves model robustness to structural variations. In CC length estimation, mean absolute error decreases from 1.89 mm to 0.80 mm on healthy fetuses and dramatically from 10.9 mm to 0.7 mm on CCD cases. Segmentation topological consistency is substantially enhanced. Our method establishes an interpretable, generalizable paradigm for few-shot rare-disease medical image segmentation.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Segmentation
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Accurate fetal brain segmentation is crucial for extracting biomarkers and assessing neurodevelopment, especially in conditions such as corpus callosum dysgenesis (CCD), which can induce drastic anatomical changes. However, the rarity of CCD severely limits annotated data, hindering the generalization of deep learning models. To address this, we propose a pathology-informed domain randomization strategy that embeds prior knowledge of CCD manifestations into a synthetic data generation pipeline. By simulating diverse brain alterations from healthy data alone, our approach enables robust segmentation without requiring pathological annotations. We validate our method on a cohort comprising 248 healthy fetuses, 26 with CCD, and 47 with other brain pathologies, achieving substantial improvements on CCD cases while maintaining performance on both healthy fetuses and those with other pathologies. From the predicted segmentations, we derive clinically relevant biomarkers, such as corpus callosum length (LCC) and volume, and show their utility in distinguishing CCD subtypes. Our pathology-informed augmentation reduces the LCC estimation error from 1.89 mm to 0.80 mm in healthy cases and from 10.9 mm to 0.7 mm in CCD cases. Beyond these quantitative gains, our approach yields segmentations with improved topological consistency relative to available ground truth, enabling more reliable shape-based analyses. Overall, this work demonstrates that incorporating domain-specific anatomical priors into synthetic data pipelines can effectively mitigate data scarcity and enhance analysis of rare but clinically significant malformations.

Problem

• Improving corpus callosum segmentation in fetal MRI scans
• Addressing data scarcity for rare corpus callosum dysgenesis cases
• Enhancing biomarker extraction without pathological annotations

Innovation

• Pathology-informed domain randomization for data synthesis
• Simulating brain alterations without pathological annotations
• Improving segmentation accuracy and topological consistency