Comparison of Automated Algorithms for White Matter Lesion Segmentation in Parkinson’s Disease
Poster No:
95
Submission Type:
Abstract Submission
Authors:
So Hoon Yoon1, Aswin Abraham2, Hamied Haroon3, Sarah Al-Bachari2
Institutions:
1Catholic Kwandong University International St. Mary's Hospital, Incheon, Incheon, 2University College London, London, United Kingdom, 3The University of Manchester, Manchester, United Kingdom
First Author:
Co-Author(s):
Introduction:
Parkinson's disease (PD) is a neurodegenerative disorder with diverse motor and non-motor symptoms. White matter lesions (WMLs) are an accepted surrogate marker of small vessel disease and have been linked to cognitive and motor impairments in PD (Bohnen, 2011; Malek, 2016). Due to the heterogeneity of PD, large-data analyses are required to identify subgroups within PD that are most affected by vascular changes and their clinical implications for PD. While manual segmentation remains the gold standard for WML quantification, it is impractical for large-scale studies, necessitating reliable automated segmentation techniques (Hotz, 2022). Accurate volumetric and spatial agreement is required, as lesion location in clinically eloquent regions has been associated with disease pathology and clinical features in PD. This study aimed to evaluate automated WML segmentation algorithms to identify the most accurate and reliable method for large-scale analyses in PD.
Methods:
We evaluated four WML segmentation algorithms: UNet-pgs (Park, 2021), Lesion Segmentation Tool Lesion Prediction Algorithm (LST-LPA) (Schmidt, 2012), FreeSurfer (Fischl, 2012), and Brain Intensity Abnormality Classification Algorithm (BIANCA) (Griffanti, 2016). The dataset included 204 PD patients (mean age 66.5 ± 7.8) and 68 healthy controls (mean age 66.2 ± 8.4), from freely available datasets of Parkinson's Progression Markers Initiative, the University of Pennsylvania, and the Montreal Neurological Institute Biobank. MRI scans included FLAIR and T1-weighted sequences across diverse scanner parameters and lesion loads. Manual segmentation performed by trained team members served as the gold standard for comparison. BIANCA was tested using the leave-one-out (BIANCA-LOO) and predefined training set (BIANCA-trained) methods, at thresholds of 0.9 and 0.99. LST-LPA was assessed with thresholds of 0.4 and 0.5, with and without bias field correction. FreeSurfer was also applied with and without bias field correction. Algorithm performance was assessed using Dice score, Hausdorff distance, log-transformed absolute volume difference (LAVD), and intraclass correlation coefficients (ICC). Subgroup analyses stratified participants by lesion load (low, medium, high) and brain regions (frontal, temporal, parietal, occipital, cingulate, insular, centrum semiovale & corpus callosum). Validation was conducted by correlating lesion volumes with age, visual rating scales, and clinical features.
Results:
UNet-pgs achieved the highest Dice score (0.47 ± 0.21 in PD, 0.40 ± 0.20 in controls), followed by BIANCA, LST-LPA, and FreeSurfer. It also exhibited the lowest Hausdorff distance (8.86 ± 3.94 in PD, 6.32 ± 2.76 in controls) and the highest ICC values (0.965 in PD, 0.967 in controls). However, the lowest LAVD was observed with BIANCA-LOO at a threshold of 0.99 (0.30 ± 0.18 in PD, 0.28 ± 0.15 in controls). For lesion load subgroups, UNet-pgs maintained superior performance, with Dice scores of 0.39 ± 0.17 for low loads, 0.50 ± 0.20 for medium loads, and 0.58 ± 0.18 for high loads in PD. Regionally, UNet-pgs demonstrated the highest accuracy across all brain regions in both groups, except for the occipital lobe in the control group. Validation showed significant correlations between lesion volumes and age (r = 0.56, p < 0.001) and visual rating scales, including Fazekas (r = 0.62, p < 0.001) and Wahlund scores (r = 0.58, p < 0.001). However, no significant correlations were found between lesion volumes and clinical scales, such as the Unified Parkinson's Disease Rating Scale, Hoehn & Yahr stage, or Montreal Cognitive Assessment scores.
·Figure. A sample of segmentations of UNet-pgs and BIANCA-LOO with a threshold of 0.99
Conclusions:
UNet-pgs emerged as the most effective automated method for WML segmentation in PD, consistently outperforming other algorithms across lesion loads and brain regions. Its accuracy and reliability make it a promising tool for large-scale analyses, facilitating future research into vascular contributions to PD pathology.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Segmentation and Parcellation 2
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity
Keywords:
Machine Learning
Movement Disorder
MRI
White Matter
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Fischl, B. (2012). FreeSurfer. Neuroimage, 62(2), 774-781.
Griffanti, L. (2016). BIANCA (Brain Intensity AbNormality Classification Algorithm): A new tool for automated segmentation of white matter hyperintensities. Neuroimage, 141, 191-205.
Hotz, I. (2022). Performance of three freely available methods for extracting white matter hyperintensities: FreeSurfer, UBO Detector, and BIANCA (Vol. 43, No. 5, pp. 1481-1500). Hoboken, USA: John Wiley & Sons, Inc..
Malek, N. (2016). Vascular disease and vascular risk factors in relation to motor features and cognition in early Parkinson's disease. Movement Disorders, 31(10), 1518-1526.
Park, G. (2021). White matter hyperintensities segmentation using the ensemble U-Net with multi-scale highlighting foregrounds. Neuroimage, 237, 118140.
Schmidt, P. (2012). An automated tool for detection of FLAIR-hyperintense white-matter lesions in multiple sclerosis. Neuroimage, 59(4), 3774-3783.