Region-informed machine learning model for choroid plexus segmentation in Alzheimer's disease

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Alzheimer's Disease Research

Background:

• The choroid plexus (CP) is crucial for cerebrospinal fluid production and implicated in Alzheimer's disease (AD).
• Accurate segmentation of the CP in MRI is challenging due to its complex shape and variable signal intensity.
• Existing methods like FreeSurfer (FS) and previous Gaussian Mixture Models (GMMs) have limitations in CP segmentation accuracy.

Purpose of the Study:

• To develop and evaluate a novel region-informed GMM (One-GMM) for automated choroid plexus segmentation.
• To compare the performance of the One-GMM method against manual segmentation, FreeSurfer (FS), and a two-component GMM (Two-GMM).

Main Methods:

• T1w and FLAIR MRI scans from 38 participants (cognitively normal, MCI, AD) were analyzed.
• A region-informed GMM (One-GMM) was applied using anatomical priors from FreeSurfer.
• Segmentation accuracy was quantified using Dice Similarity Coefficient (DSC), Hausdorff distance (HD95), and volume difference (VD%), compared to manual segmentation.

Main Results:

• The One-GMM achieved significantly higher accuracy (DSC 0.82) than FS (DSC 0.24) and Two-GMM (DSC 0.66).
• One-GMM demonstrated lower Hausdorff distance (HD95: 6.06 mm) compared to FS (26.21 mm) and Two-GMM (10.58 mm).
• Segmentation accuracy was consistent across diagnostic groups, with no significant differences in CP volume observed between groups.

Conclusions:

• The region-informed One-GMM offers a substantial improvement in choroid plexus segmentation accuracy over existing methods.
• This method provides a practical and accessible tool for CP quantification in AD research and other studies.
• An aging effect on CP volume was observed in cognitively normal individuals.