Automatic segmentation and reconstruction of the cortex from neonatal MRI

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Developmental Neuroscience

Background:

• Neonatal brain MRI segmentation is challenging due to evolving gray matter (GM) and white matter (WM) contrast.
• T2-weighted images often show higher WM signal than GM, leading to segmentation errors, especially near the gray matter-cerebrospinal fluid interface.

Purpose of the Study:

• To develop an automatic algorithm for accurate segmentation and reconstruction of cortical surfaces in neonatal brains.
• To address and correct mislabeled voxels and partial volume effects in neonatal brain MRI.

Main Methods:

• An automatic segmentation algorithm was developed to detect and correct mislabeled voxels.
• Implicit surface evolution was used for reconstructing inner, central, and outer cortical surfaces.
• The algorithm was validated against manual segmentation and landmark studies on 25 neonates (27-45 weeks gestational age).

Main Results:

• The proposed algorithm significantly improved segmentation accuracy for GM and WM compared to the expectation maximization (EM) scheme (mean Dice: 0.758 for GM, 0.794 for WM).
• Cortical surface reconstruction achieved high accuracy (mean error: 0.73 mm for inner, 0.63 mm for outer surface).
• Confirmed age-related increases in cortical surface area and curvature, with surface area scaling to cerebral volume via a power law.

Conclusions:

• The developed algorithm provides accurate segmentation and reconstruction of neonatal cortical surfaces.
• Findings support established relationships between brain development metrics and gestational age in neonates.
• The method offers a valuable tool for analyzing infant brain development using MRI.