Automatic cortical surface parcellation in the fetal brain using attention-gated spherical U-net

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Artificial Intelligence in Medicine

Background:

• Accurate cortical surface parcellation is crucial for understanding fetal neurodevelopment.
• Existing methods for fetal brain parcellation have limitations in precision and sensitivity.

Purpose of the Study:

• To develop and validate a novel deep-learning model for automatic cortical surface parcellation of the fetal brain.
• To compare the performance of the proposed model against existing methods.

Main Methods:

• Development of an attention-gated spherical U-net model.
• Training and validation using MRI data from 55 typically developing fetuses.
• Comparative analysis with surface registration-based methods and the original spherical U-net.

Main Results:

• The proposed model achieved a significantly higher Dice coefficient (0.899 ± 0.020) compared to previous methods.
• Demonstrated the lowest median boundary distance (2.47 ± 1.322 mm) and mean absolute percent error in surface area measurement (10.40 ± 2.64%).
• Attention gates were effective in capturing subtle but important cortical surface information.

Conclusions:

• The attention-gated spherical U-net provides a precise and accurate method for fetal brain cortical surface parcellation.
• This model can enhance the detection of regional cortical anomalies.
• Potential for earlier identification of neurodevelopmental disorders in fetuses.