Ventriculomegaly is a condition characterized by abnormal enlargement of the brain’s lateral ventricles, often indicating underlying neurological abnormalities detectable during prenatal development. It poses significant challenges for automated segmentation in fetal MRI due to data scarcity and anatomical variability. While AI-based tools perform well on healthy fetal brains, their generalization to pathological cases remains limited. To address this, we propose a novel generative AI framework that synthesizes high-fidelity fetal MRIs with ventriculomegaly by morphologically modifying healthy anatomical label maps. We propose Fetal&Neonatal-DDPM, a conditional diffusion model on semantic label images that generates realistic synthetic pathological fetal MRI images with controlled ventricular dilation. By selectively expanding lateral ventricle labels in healthy scans, we create diverse pathological training data. Evaluations on clinical datasets with severe ventriculomegaly demonstrate that segmentation models trained with our synthetic images outperform those trained solely on real data, achieving a Dice score of 0.8628 (vs. 0.8167) in severe ventriculomegaly cases. Notably, synthetic data improved ventricle segmentation by 26% (Dice: 0.9253 vs. 0.7317) and enhanced anatomical consistency compared to manual annotations. This development represents a significant step towards enhancing the precision of prenatal image analysis and segmentation, while also introducing innovative methods for anonymizing data by creating synthetic pathological imaging.