FetGEs: a Deep Learning Approach for Fetal MRI Ganglionic Eminence Segmentation

The Ganglionic Eminence (GE) plays a pivotal role in neural migration during fetal brain development. Early detection of GE anomalies is crucial for identifying migration deficiencies that may lead to neurological or psychiatric disorders postnatally.
Here, we propose an automated GE segmentation method for fetal Magnetic Resonance Imaging (MRI) data by extending 3D UNets and exploiting a registration–driven generative data augmentation technique to increase the number of scans from 138 to more than 2,500 with manually defined labels by an expert neuroradiologist. Our solution spans 19 to 38 weeks of gestation, achieving a mean Dice score of 0.79 ± 0.04 in the 2nd trimester and 0.74 ± 0.05 in the 3rd trimester. Overall, the GE volume decreases throughout pregnancy (R² = 0.77, ranged 31–668 mm³), highlighting an inverse relationship to the whole brain volume, which continues to grow (R² = 0.93).

Network configuration

3D UNet MONAI-based architecture: a single input channel following 5 layers, with a number of filters equal to 64 for the first layer and doubled for the following ones. Data were downsampled/upsampled in the encoder/decoder part using 2‒strided convolutions residual units with a down/up‒sampling kernel size of 3x3x3 voxels, followed by instance normalization and PReLU activation blocks.

Input Data

3D fetal brain reconstructions meeting the following criteria:

• Adequate signal-to-noise ratio (SNR) and overall image quality
• Full coverage of the region of interest
• Good-quality 3D reconstruction
• Aligned to the standard radiological atlas space
• Gestational age: 19–38 weeks
• No significant shading artifacts
• No severe structural anomalies
• Acquired using 1.5T or 3T MRI
• Echo time (TE): 100–140 ms
• Slice thickness: 2.0–4.5 mm
• Field of view (FOV): 200–230 mm
• In-plane resolution: 0.62 × 0.62 mm to 1.17 × 1.17 mm

Instructions - How to use

source activate ge_monai  
INPUT_DIR=/PATH/TO/NIFTY_MIC/DIRECTORY  
OUTPUT_DIR=/PATH/TO/OUTPUT/DIRECTORY  
python union_unet.py -i ${INPUT_DIR} -o ${OUTPUT_DIR}  

How to cite

@inproceedings{your_paper,
title={FetGEs: a Deep Learning Approach for Fetal MRI Ganglionic Eminence Segmentation},
author={T. Ciceri, M. Stuempflen, J. Tischer, G. Kasprian, D. Peruzzo, and R. Licandro},
booktitle={MICCAI PIPPI},
year={2025},
doi={}
}