Fuzzy connectedness image segmentation for newborn brain extraction in MR images

Area of Science:

• Medical Imaging Analysis
• Neuroscience
• Biomedical Engineering

Background:

• Neonatal brain shape is dynamic and susceptible to developmental changes and diseases.
• Accurate brain segmentation in MR images is crucial for quantifying shape and size.
• Limited research exists on neonatal brain MR image analysis and segmentation.

Purpose of the Study:

• To introduce a novel fuzzy connectedness (FC) method integrated with a fuzzy object model (FOM) for neonatal brain MR image segmentation.
• To accurately segment white matter and surrounding cortex in infant brains.
• To provide a quantitative tool for analyzing neonatal brain morphology.

Main Methods:

• Development of a fuzzy object model (FOM) using a training dataset to define fuzzy degrees of belonging for parenchyma based on location and intensity.
• Calculation of fuzzy connectedness (FC) using object and homogeneous affinities, with object affinity derived from the FOM.
• Sequential segmentation of white matter followed by the surrounding cortex.

Main Results:

• The proposed method was applied to 10 newborn subjects (revised age: -1 to +2 months).
• Leave-one-out cross-validation (LOOCV) demonstrated high accuracy.
• Mean false-positive volume fraction: 1.33%; Mean false-negative volume fraction: 2.90%; Geometric mean: 1.42%.

Conclusions:

• The novel fuzzy connectedness with fuzzy object model method provides accurate and reliable segmentation of neonatal brain MR images.
• This technique is effective for quantifying brain structures in early development.
• The method holds promise for advancing research in neonatal brain development and pathology.