<p>Segmenting MRI images of infant brains presents unique challenges compared to adult brain segmentation, primarily due to low tissue contrast and the rapid development of immature brain structures. This study proposes an enhanced U-Net architecture that integrates Guided Filter Modules (GFMs) as edge detection layers to improve the delineation of anatomical boundaries in low-contrast neonatal brain MRIs. The primary objective is to achieve more accurate segmentation of gray matter, white matter, and cerebrospinal fluid in 6-month-old infants. To evaluate the proposed approach, ablation experiments were conducted using the iSeg19 dataset from the Medical Image Computing and Computer-Assisted Intervention (MICCAI) challenge. Three model variants were assessed: (i) the baseline U-Net, (ii) U-Net with a single edge-detection layer, and (iii) U-Net with GFMs integrated at the skip connections between encoder and decoder stages. The GFMs refine the transmitted feature maps by preserving spatial edge information from the encoder while suppressing noise before fusion in the decoder, thereby enhancing tissue boundary representation. Additionally, the model was benchmarked against the latest infant MRI segmentation model, xflz. Quantitative evaluations using Dice coefficient, modified Hausdorff distance (MHD), and average surface distance (ASD) demonstrate that the fully integrated REDGFM-U-Net consistently outperforms both the baseline and benchmark models across all brain regions, with notable improvements in areas exhibiting weak boundaries and subtle contrast differences. These results highlight the effectiveness of GFM-based edge enhancement for neonatal brain MRI segmentation, offering a promising direction for improving accuracy in infant neuroimaging.</p>