A Deep Attentive Convolutional Neural Network for Automatic Cortical Plate Segmentation in Fetal MRI

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Deep Learning

Background:

• Fetal cortical plate segmentation is crucial for assessing fetal brain maturation and folding patterns.
• Manual segmentation is time-consuming and prone to errors, while automatic methods struggle with low MRI resolution and anatomical variations.
• Accurate segmentation is vital for quantitative analysis of neurodevelopmental trajectories.

Purpose of the Study:

• To develop an advanced deep learning model for automated fetal cortical plate segmentation.
• To overcome limitations of existing methods in segmenting thin cortical structures in low-resolution fetal brain MRI.
• To provide a robust and efficient tool for large-scale studies on fetal brain development.

Main Methods:

• A fully convolutional neural network incorporating deep attentive modules and mixed kernel convolutions was developed.
• The architecture utilized deep supervision and residual connections to enhance segmentation accuracy.
• The method was trained and evaluated on reconstructed fetal brain MRI scans across a wide gestational age range (16-39 weeks).

Main Results:

• The deep learning method achieved superior performance compared to state-of-the-art deep models and multi-atlas techniques.
• Quantitative metrics included an average Dice similarity coefficient of 0.87, Hausdorff distance of 0.96 mm, and symmetric surface distance of 0.28 mm.
• Segmentation was completed in under one minute per fetal brain, demonstrating high computational efficiency.

Conclusions:

• The developed deep learning method offers a powerful and efficient solution for automated fetal cortical plate segmentation.
• This technique can significantly accelerate research into normal and abnormal fetal brain maturation and cortical folding.
• The method's accuracy and speed facilitate large-scale neuroimaging studies, aiding in the understanding of fetal neurodevelopment.