Segmentation of the parasagittal dura mater on multi-center 3D-FLAIR MRI.

The parasagittal dura (PSD), located bilaterally alongside the superior sagittal sinus (SSS), has been increasingly implicated in cerebrospinal fluid-meningeal lymphatic communication, contributing to CSF waste clearance and immune surveillance. We assembled a training set of 55 3D-Fluid-Attenuated Inversion Recovery (3D-FLAIR) scans from Alzheimer's Disease Neuroimaging Initiative(ADNI) and local datasets to train an nnU-Net-based model for automated PSD segmentation (mcPSD-Net). Segmentation performance was assessed against manual ground truth in an independent test set (N=25) containing images acquired from various scanners. Three voxel overlap metrics (DICE, Precision and Recall) and three surface distance metrics (HD, HD95, ASSD) were used to evaluate the accuracy of the mcPSD-Net. Multiple linear regression models were performed to evaluate the associations of PSD volume (normalized by intracranial volume) with age, and sex in the whole ADNI3 dataset and a local community dataset. In the testing dataset, mcPSD-Net achieved good performance (Dice coefficient=0.76; precision=0.84; recall=0.73; HD=21.42mm; HD95=4.00mm; ASSD=0.53mm). Regression analyses demonstrated that PSD volume increased significantly with age in both the ADNI3 (standardized β = 0. 390, p<0.001) and local community dataset (standardized β = 0. 307, p<0.001). Compared to females, males had a significantly larger PSD volume in both datasets (p<0.001 for all). Furthermore, including scanner vendor as a covariate did not improve model fitting in the ADNI3 dataset. In summary, we developed a deep learning model that segments PSD from 3D-FLAIR images acquired using different vendors and imaging protocols, providing a tool for related clinical imaging studies.