Automatic segmentation of eight tissue classes in neonatal brain MRI

Area of Science:

• Medical imaging
• Neuroscience
• Biomarkers

Background:

• Neonatal brain tissue volumetric measurements can predict neurodevelopmental outcomes.
• Accurate segmentation of neonatal brain MRI is crucial for research and clinical applications.

Purpose of the Study:

• To develop an automatic method for probabilistic segmentation of neonatal brain MR images.
• To enable accurate volumetric measurements of neonatal brain tissues for neurodevelopmental assessment.

Main Methods:

• Acquired T1- and T2-weighted MRI scans from 108 preterm neonates.
• Developed a supervised pixel classification method using voxel intensity and spatial information.
• Segmented images into eight tissue classes: grey matter, white matter, CSF, brainstem, and cerebellum.

Main Results:

• Achieved Dice similarity coefficients ranging from 0.75 to 0.92 for most tissue classes.
• Demonstrated accurate segmentation of eight tissue classes, with challenges in segmenting myelinated white matter (Dice coefficient 0.47).
• Probabilistic segmentation volumes compared favorably with manual reference standards.

Conclusions:

• The proposed method accurately segments neonatal brain MR images into specified tissue classes, excluding myelinated white matter.
• This method uniquely distinguishes between ventricular and extracerebral cerebrospinal fluid.
• The technique shows potential for predicting neurodevelopmental outcomes and evaluating neonatal neuroprotective trials.