Biomarkers.

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Cerebral Small Vessel Disease (CSVD)

Background:

• Enlarged perivascular spaces (PVS) are key imaging biomarkers for CSVD, linked to aging and neurodegeneration.
• Accurate PVS quantification on T1-weighted MRI is challenging due to imaging variability and low contrast.
• Existing automated methods using convolutional neural networks (CNNs) have limitations in performance and generalizability.

Purpose of the Study:

• To introduce segcsvd_PVS, a novel CNN-based tool for automated PVS segmentation on T1-weighted MRI.
• To evaluate the performance, robustness, and clinical relevance of segcsvd_PVS across diverse datasets and imaging conditions.

Main Methods:

• Developed segcsvd_PVS using a hierarchical CNN framework with anatomical information and robust training strategies.
• Trained and validated the tool on a large cohort (n=1351) using semi-automated RORPO-based ground truth data.
• Assessed performance metrics, robustness to image quality variations, and age-related PVS burden correlations against established methods.

Main Results:

• Segcsvd_PVS demonstrated high sensitivity for basal ganglia PVS (0.81 ± 0.13) and detected significantly larger volumes compared to human tracers.
• The tool showed strong age-related correlations with PVS volumes across three diverse datasets (r ≈ 0.41).
• Segcsvd_PVS exhibited superior robustness to image contrast and noise variations, with minimal impact on age-related correlations.

Conclusions:

• Segcsvd_PVS is a reliable tool for PVS segmentation, especially in basal ganglia regions, offering enhanced sensitivity and robustness.
• The tool effectively detects biologically relevant age-related associations with PVS burden.
• Segcsvd_PVS is suitable for large-scale studies and clinical research to improve understanding of PVS in CSVD and dementia.