Automatic brain tissue segmentation in fetal MRI using convolutional neural networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Fetal MRI enables early detection of brain abnormalities.
• Accurate segmentation of fetal brain tissues is crucial for analysis.
• Manual segmentation is time-consuming; automatic methods face challenges like intensity inhomogeneity due to fetal movement.

Purpose of the Study:

• To develop an automatic fetal brain tissue segmentation method robust to intensity inhomogeneity artifacts.
• To improve the accuracy and efficiency of fetal brain segmentation in MRI scans.

Main Methods:

• Utilized a convolutional neural network (CNN) architecture for segmentation.
• Employed data augmentation by synthetically introducing intensity inhomogeneity into training data.
• Two CNNs were used: one for intracranial volume extraction, another for segmenting seven brain tissue classes.

Main Results:

• The proposed method achieved an average Dice coefficient (DC) of 0.88, an improvement from 0.77 without augmentation.
• Mean surface distance (MSD) decreased from 0.78 mm to 0.37 mm with augmented data.
• The approach demonstrated enhanced segmentation performance, particularly in the presence of artifacts.

Conclusions:

• Synthetic data augmentation with intensity inhomogeneity effectively improves CNN-based fetal brain segmentation.
• The method shows potential to replace or supplement traditional preprocessing steps like bias field correction.
• This automated approach can enhance the accuracy and efficiency of fetal neuroimaging analysis.