Related Experiment Video
Updated: Oct 10, 2026
Lesion Explorer: A Video-guided, Standardized Protocol for Accurate and Reliable MRI-derived Volumetrics in Alzheimer's Disease and Normal Elderly
Published on: April 14, 2014
Comparison of automated white matter lesion segmentation approaches for use in large, multi-site data analyses in
Sarah Al-Bachari1,2, So Hoon Yoon3,4, Phoebe Emson2
1Department of Clinical and Movement Neurosciences, UCL Queen Square Institute of Neurology, University College London, London, England, United Kingdom.
Background:
The vascular role in Parkinson's disease (PD) is emerging, yet the literature remains conflicted, motivating large-data analyses with greater statistical power. White matter lesions (WML) are an accepted imaging marker of small vessel disease. Accurate automated WML segmentation techniques are crucial for large-scale studies; however, evaluation of the optimum approach in PD is lacking. This study evaluated automated WML segmentation algorithms to determine the most accurate and reliable method, among those selected, for multi-site large data analysis in PD.
Methods:
We assessed whole-brain volumetric T1-weighted and FLAIR images from 201 PD patients (mean age, 66.6 ± 7.86 years) and 64 healthy controls (HC; mean age, 66.3 ± 8.67) across three datasets composed of different scanners, imaging parameters and lesion loads. WML were manually segmented to provide the gold standard, and four freely available automated algorithms were evaluated, FSL's BIANCA, FreeSurfer, SPM's LST-LPA and U-Net-pgs, using the performance metrics: Dice score, Hausdorff distance, recall, precision, F1 score, log absolute volume difference (LOGAVD) and intraclass correlation coefficient (ICC). Sub-analyses were performed based on lesion load, lobar regions and acquisition parameters.
Results:
U-Net-pgs produced the highest Dice score (PD: 0.46 ± 0.21; HC: 0.39 ± 0.21), recall (PD: 0.75 ± 0.24; HC: 0.58 ± 0.26), precision (PD: 0.48 ± 0.24; HC: 0.62 ± 0.25), F1 score (PD: 0.53 ± 0.21; HC: 0.54 ± 0.20) and ICC (PD: 0.87; HC: 0.89) and lowest Hausdorff distance (PD: 8.89 ± 3.96; HC: 6.33 ± 2.91) and LOGAVD (PD: 0.31 ± 0.31; HC 0.27 ± 0.30) across PD and HC. U-Net-pgs also showed overall superior performance in all lesion loads for PD and across brain regions in both PD and HC.
Conclusion:
Overall, of those we evaluated, U-Net-pgs emerged as the highest performing automated method across lesion loads and brain regions, for WML segmentation in PD and HC. The accuracy and reliability of U-Net-pgs, across various scanner and image acquisition parameters, make it a promising tool for large-scale analyses, facilitating future research investigating WML in PD.
