A framework for quality control of corpus callosum segmentation in large-scale studies

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• The corpus callosum (CC) is crucial for brain hemisphere interconnection.
• Accurate CC segmentation is vital for downstream neuroimaging analyses like parcellation and registration.
• Automated quality control (QC) is essential for large-scale neuroimaging studies.

Purpose of the Study:

• To introduce a novel framework for automated quality control (QC) of corpus callosum (CC) segmentation.
• To enable reliable analysis of large neuroimaging datasets without manual intervention.
• To assess the quality of CC segmentations without requiring ground-truth data.

Main Methods:

• A framework utilizing shape signatures computed at 49 resolutions was developed.
• Support vector machine (SVM) classifiers were trained at each resolution.
• A disagreement metric clustered classifiers, and an ensemble was formed by selecting one from each cluster.

Main Results:

• The proposed framework achieved a 98.25% area under the curve (AUC) on the test set.
• An ensemble of 12 components demonstrated superior performance compared to individual classifiers.
• The method successfully assessed CC segmentation quality on 207 subjects.

Conclusions:

• The shape descriptor is robust and versatile across multiple resolutions.
• The ensemble approach yields high-quality, heterogeneous classifiers for optimal performance.
• This method provides an effective trade-off between ensemble size and high AUC for CC segmentation QC.