A Label-fusion-aided Convolutional Neural Network for Isointense Infant Brain Tissue Segmentation

Area of Science:

• Medical imaging
• Neuroscience
• Artificial intelligence

Background:

• Infant brain development presents unique challenges for MRI segmentation due to low tissue contrast, particularly during the isointense stage (6-8 months).
• Accurate segmentation of white matter, gray matter, and cerebrospinal fluid is crucial for analyzing pediatric brain development and disorders.

Purpose of the Study:

• To develop an automated deep learning approach for segmenting isointense infant brain MRI scans.
• To improve the accuracy of brain region segmentation in infants using T1- and T2-weighted MRI.

Main Methods:

• A label-fusion-aided deep learning strategy was employed, utilizing fully convolutional neural networks (FCNNs).
• The FCNN was applied to individual brain regions identified by registration-based segmentation, rather than a single whole-brain model.
• T1- and T2-weighted MRI data from the iSEG MICCAI Grand Challenge 2017 dataset were used for training and validation.

Main Results:

• The proposed method demonstrated superior performance compared to traditional joint label fusion and FCNN-only techniques.
• Segmentation accuracy was quantitatively assessed using Dice coefficients, indicating improved results.
• Region-specific FCNN application captured more detailed features, leading to refined segmentation outcomes.

Conclusions:

• The label-fusion-aided deep learning approach effectively addresses the challenges of segmenting isointense infant brain MRI.
• This method offers a more accurate and refined tool for pediatric neuroimaging analysis.
• The findings suggest a promising direction for automated analysis in developmental neuroscience research.