A network model that can segment parasaggital dura from multi-center high-resolution FLAIR images.
The parasagittal dura (PSD) refers to the dura adjacent to the parasagittal sinus. Previous research utilizing dynamic contrast-enhanced imaging to study the clearance of cerebrospinal fluid (CSF) tracers has found that the PSD is where the tracer exits the brain. The PSD may also play a role in monitoring and modulating inflammatory responses in intracranial spaces. The PSD may become hypertrophy in certain disease conditions, which remains to be elucidated.
The model was trained using data from ADNI and local multi-center cohorts, covering different scanner vendors and imaging protocols. It is designed to support rapid and stable quantitative analysis of PSD structures in large-scale neuroimaging studies.
Independent testing showed that the model can reliably identify the core PSD regions and generate segmentation masks suitable for volumetric measurements. The model is primarily intended for research use, particularly for studies investigating the relationships between PSD imaging features and brain diseases. Users are advised to interpret the results in combination with image quality assessment and, when necessary, manual quality control.
The mcPSD-Net can be easily used in the nnU-Net environment. Please check https://github.com/MIC-DKFZ/nnUNet for detailed instructions. The trained model can be downloaded from (https://1drv.ms/f/c/36bb922ed9e9f74e/IgCtnhjjUAayRruoAcSwc9JbAY8a9VzuRBk36IiFdpEbrK8?e=aHPruO). A 3D FLAIR image is needed as the input.
The command for prediction is: nnUNetv2_predict -d DatasetPSD -i INPUT_FOLDER -o OUTPUT_FOLDER -f 0 1 2 3 4 -tr nnUNetTrainer -c 3d_fullres -p nnUNetPlans
Guan L, Lin M, Zhang R, Shao X, Chen B, Yang J, Zhao Z, Huang P. Segmentation of the parasagittal dura mater on multi-center 3D-FLAIR MRI. NeuroImage. 2026 Jul 3:122099.
Peiyu Huang, huangpy@zju.edu.cn