6-MONTH INFANT BRAIN MRI SEGMENTATION GUIDED BY 24-MONTH DATA USING CYCLE-CONSISTENT ADVERSARIAL NETWORKS.

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• Infant brain MRI segmentation is challenging due to low white matter (WM) and gray matter (GM) contrast at 6 months (isointense phase).
• Limited manual annotations hinder automated segmentation development for this critical developmental stage.
• Higher contrast in early adult brain MRI (e.g., 24 months) allows for reliable segmentation with existing tools.

Purpose of the Study:

• To develop a method for reliable tissue segmentation of 6-month-old infant brain MRI by leveraging data from 24-month-old individuals.
• To address the limitations of manual annotation and improve automated segmentation accuracy in the challenging isointense phase.

Main Methods:

• Proposed a 3D-cycleGAN-Seg architecture to transfer appearances between 6-month and 24-month brain MRI scans, generating synthetic isointense images.
• Employed segmentation feature consistency between time-points to guide generator training.
• Introduced a feature matching loss using cosine distance to enhance synthetic image quality.
• Jointly trained the segmentation model on 6-month-old images using transferred 24-month-old data.

Main Results:

• The proposed method demonstrated superior performance compared to existing deep learning-based segmentation techniques.
• Generated synthetic images effectively captured the characteristics of the isointense phase.
• Achieved reliable tissue segmentation for 6-month-old infant brain MRI.

Conclusions:

• The novel approach successfully utilizes higher-contrast adult MRI data to improve infant brain MRI segmentation.
• This method offers a promising solution for automated segmentation in the challenging isointense phase of early brain development.
• The technique has the potential to advance pediatric neuroimaging research and clinical applications.