Learning to combine complementary segmentation methods for fetal and 6-month infant brain MRI segmentation

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Medical Image Analysis

Background:

• Accurate segmentation of brain structures is crucial for understanding prenatal and postnatal brain development.
• Existing segmentation methods fall into registration-based and learning-based families, each with unique strengths.
• Combining these complementary approaches can enhance segmentation performance.

Purpose of the Study:

• To explore ensembling strategies (stacking and cascading) for combining registration-based and learning-based brain segmentation techniques.
• To evaluate the performance of these ensembling strategies on segmenting 6-month infant brains and fetuses with isolated non-severe ventriculomegaly (INSVM).
• To identify potential neurodevelopmental markers by analyzing cortical folding in INSVM fetuses using improved segmentations.

Main Methods:

• Employed stacking and cascading ensembling strategies to integrate distinct brain segmentation method families.
• Conducted experiments on datasets including 6-month infant brains and fetuses diagnosed with INSVM.
• Utilized segmentation results to investigate cortical folding patterns in INSVM fetuses.

Main Results:

• Both ensembling strategies (stacking and cascading) significantly outperformed individual segmentation methods.
• The cascading strategy achieved superior results, securing top rankings in the iSeg2017 MICCAI Segmentation Challenge for white matter, gray matter, and cerebro-spinal fluid segmentation.
• Segmentation analysis revealed that INSVM fetuses exhibit reduced cortical convolution compared to controls.

Conclusions:

• Ensembling segmentation techniques effectively combines complementary approaches for improved brain structure analysis.
• Cascading offers a powerful strategy for enhancing segmentation accuracy in pediatric neuroimaging.
• Cortical folding abnormalities may serve as potential biomarkers for neurodevelopmental outcomes in fetuses with INSVM.