Premium
Classifying white matter hyperintensities according to intensity and spatial localisation reveals specific association with cognition
Author(s) -
Melazzini Luca,
Bordin Valentina,
Suri Sana,
Zsoldos Enikő,
Ebmeier Klaus P,
Jenkinson Mark,
Mackay Clare,
Sardanelli Francesco,
Griffanti Ludovica
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.042751
Subject(s) - hyperintensity , cognition , neuropsychology , association (psychology) , white matter , audiology , medicine , psychology , magnetic resonance imaging , radiology , neuroscience , psychotherapist
Background White matter hyperintensities (WMH) on T2‐weighted images are imaging biomarkers of brain small vessel disease. When classified according to location (periventricular/deep), they have shown different associations with cognition. WMH can also appear hypointense on T1‐weighted (T1w) images as a possible sign of irreversible tissue damage. We hypothesise that sub‐classifying WMH combining intensity information and spatial localisation may provide better insight into the association with cognition, not detectable for the total WMH burden. Method We analysed data from 684 subjects of the Whitehall II imaging sub‐study. A supervised machine learning method (BIANCA) was used to segment WMH. An automatic method based on cluster localisation and image intensity was then applied to classify WMH into 4 categories according to adjacency to the ventricles (periventricular/deep) and appearance on T1w images (either T1w‐hypointense or not) (Figure 1). Derived volumes were entered into a general linear model as predictors of the participants’ cognitive scores on neuropsychological tests. Result Periventricular T1w‐hypointense WMH were significantly related to worse performance in the trail‐making test A (p = 0.011), digit‐symbol (p = 0.028), and digit‐coding (p = 0.009) tests. When including only the total WMH burden in the model, we could not find any associations between WMH and cognition. Age, gender, years of education, systolic and diastolic blood pressure were used as covarietes in the statistical models. Conclusion Sub‐classifying WMH according to both location and appearance on T1w images provided added value compared to total WMH burden alone. These are promising findings for WMH interpretation in the clinical practice and for the development of methods for analysing imaging biomarkers related to cognition.