Automated quantification of white matter lesion confluence on T2 MRI scans
Poster No:
288
Submission Type:
Abstract Submission
Authors:
Tatjana Schmidt1, Robert Salzmann1, Marcella Montagnese2, Timothy Rittman3
Institutions:
1University of Cambridge, Cambridge, United Kingdom, 2Cambridge University, Cambridge, United Kingdom, 3Department of Clinical Neurosciences, University of Cambridge, Cambridge, Cambridgeshire
First Author:
Co-Author(s):
Timothy Rittman
Department of Clinical Neurosciences, University of Cambridge
Cambridge, Cambridgeshire
Department of Clinical Neurosciences, University of Cambridge
Cambridge, Cambridgeshire
Introduction:
White matter hyperintensities (WMH) are a histopathologically heterogeneous entity on FLAIR images and are the consequences of small vessel disease (Prins & Scheltens, 2015). They have been linked to an increased risk of stroke, dementia and death (Debette et al., 2018). The conglomeration of discrete WMHs is referred to as confluence and evidence suggests that it is a clinically useful concept since it reflects the severity of white matter disease across aetiologies (Fazezkas et al., 1993). WMH confluence is thus of great interest for neurodegenerative diseases and is of practical relevance as patients with confluent WMH are often excluded from clinical trials for AD immunotherapy due to an increased adverse effect risk (Rollin-Sillaire et al., 2013).
There is currently no method to automatically quantify WMH confluence. Clinical trials rely on manual scoring with the Fazekas scale (Fazekas et al., 1987) which is time-consuming and subjective. Here we propose an algorithm to quantify the degree of confluence and express it as a value between 0 and 1. The algorithm was applied to data from "Quantitative MRI in the NHS–Memory Clinics", a real world memory clinic study, to examine the relationship between WMH confluence and clinical measures.
There is currently no method to automatically quantify WMH confluence. Clinical trials rely on manual scoring with the Fazekas scale (Fazekas et al., 1987) which is time-consuming and subjective. Here we propose an algorithm to quantify the degree of confluence and express it as a value between 0 and 1. The algorithm was applied to data from "Quantitative MRI in the NHS–Memory Clinics", a real world memory clinic study, to examine the relationship between WMH confluence and clinical measures.
Methods:
Participants were patients recruited from neurology-/psychiatry-led NHS memory clinics (n=350, 166 female, mean age=72) with various neurodegenerative, cognitive or psychiatric diagnoses such as Alzheimer's, vascular, or frontotemporal dementia, dementia with Lewy bodies, depression, and functional memory symptoms. Clinical data included age, diagnosis and scores from the Revised Addenbrookes Cognitive Examination (ACE-R) cognitive test.
MR images were acquired on a 3T MRI system (Magnetom Prisma, Siemens Medical Systems, Germany) and included a 3D MP-RAGE and a FLAIR acquisition. WMH were automatically segmented with FSL's BIANCA toolbox (Griffanti et al., 2016) using FLAIR and T1w images and 20 training images on which WMH have been manually identified. This resulted in a probability map for each subject, indicating for each voxel its probability of belonging to a WMH.
On the basis of these probability maps, the confluence algorithm was run for each subject (Fig. 1). Confluence scores subsequently entered further analysis (Fig. 2). A piecewise regression of confluence against age with a breakpoint at age 70 was calculated. A one-way ANOVA was performed to test whether there was a difference in confluence scores between diagnoses. Linear regressions of all ACE-R subtests against confluence while controlling for age were calculated in order to test whether confluence can explain cognitive performance.
MR images were acquired on a 3T MRI system (Magnetom Prisma, Siemens Medical Systems, Germany) and included a 3D MP-RAGE and a FLAIR acquisition. WMH were automatically segmented with FSL's BIANCA toolbox (Griffanti et al., 2016) using FLAIR and T1w images and 20 training images on which WMH have been manually identified. This resulted in a probability map for each subject, indicating for each voxel its probability of belonging to a WMH.
On the basis of these probability maps, the confluence algorithm was run for each subject (Fig. 1). Confluence scores subsequently entered further analysis (Fig. 2). A piecewise regression of confluence against age with a breakpoint at age 70 was calculated. A one-way ANOVA was performed to test whether there was a difference in confluence scores between diagnoses. Linear regressions of all ACE-R subtests against confluence while controlling for age were calculated in order to test whether confluence can explain cognitive performance.
Results:
The piecewise regression of confluence against age showed a significant association above age 70 (β=0.0017, p=0.014), but not below. Results of the ANOVA indicated no significant difference in confluence between different diagnoses (F=2.39, p=0.069). However, visual inspection of data showed that while confluence scores of patients with mild cognitive impairment or functional memory symptoms had a compact distribution, the scores of patients with dementia (particularly Alzheimer's) showed a wider range. The regression of ACE-R subtests against confluence showed that the fluency subtest had a significant association with confluence (β=-26.2, p=0.001) after controlling for the effect of age.
Conclusions:
The proposed algorithm determining the degree of confluence of WMH successfully quantifies a concept that has previously been rated only manually. Preliminary results indicate that the confluence score increases with age, but does not differ significantly between different kinds of neurodegenerative diseases. In keeping with previous findings (Kaskikallio et al., 2021), it is selectively sensitive to changes in fluency assessed with a subtest of the ACE-R. Our new quantification of white matter lesion confluence opens up a new approach to quantifying an important aspect of cerebrovascular neuropathology.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Lifespan Development:
Aging
Modeling and Analysis Methods:
Image Registration and Computational Anatomy
Methods Development 2
Neuroinformatics and Data Sharing:
Informatics Other
Keywords:
Cerebrovascular Disease
Cognition
Computational Neuroscience
Degenerative Disease
DISORDERS
Informatics
MRI
Open-Source Code
STRUCTURAL MRI
White Matter
1|2Indicates the priority used for review
Provide references using author date format
Fazekas, F. (1987), 'MR signal abnormalities at 1.5 t in Alzheimer’s dementia and normal aging', American Journal of Roentgenology, 149(2), pp. 351–356.
Fazekas, F. (1993), 'Pathologic correlates of incidental MRI white matter signal hyperintensities', Neurology, 43(9), pp. 1683–1683.
Griffanti, L. (2016), 'Bianca (brain intensity abnormality classification algorithm): A new tool for automated segmentation of white matter hyperintensities', NeuroImage, 141, pp. 191–205.
Kaskikallio, A. (2021), 'Effects of white matter hyperintensities on verbal fluency in healthy older adults and MCI/AD', Frontiers in Aging Neuroscience, 13.
Prins, N. D. (2015), 'White matter hyperintensities, cognitive impairment and dementia: An update', Nature Reviews Neurology, 11(3), pp. 157–165.
Rollin‐Sillaire, A. (2013), 'Reasons that prevent the inclusion of Alzheimer’s disease patients in clinical trials', British Journal of Clinical Pharmacology, 75(4), pp. 1089–1097.