Edinburgh Imaging

MSc projects 1819 004

Variability in longitudinal White Matter Hyperintensity (WMH) volume measurement in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) provided by the University of California, Davis group – factors affecting stability & reproducibility in comparison to other publicly available datasets.

  • White matter hyperintensities (WMH) are an ageing and age-related phenomenon, and are also often associated with various diseases such as Alzheimer’s disease and vascular cognitive impairment and are thought to be a sign of small-vessel disease. While the development and etiology of WMH are fairly well understood, there is still some debate as to whether or not the process is always progressive towards larger volumes, or whether or not volumes may fluctuate and decrease over time. The ADNI project has generated data on hundreds of Alzheimer’s Disease (AD), mild cognitive impairment (MCI) and normal control (NC) subjects. In preliminary examinations of the WMH volume data provided by the UCD group (recontributed to ADNI), there seem to be substantial longitudinal fluctuations in WMH volumes over time, often with large decreases from one time point to the next. Investigation seemed to indicate that there was substantial variability in the ADNI1 dataset, particularly at intermediate measurement points, which cannot be accounted by either biological variation or acceptable measurement error. This was further substantiated by examining subjects who carried over from ADNI (PD/T2-based measures) to ADNI2 (FLAIR based measures) where many sizable differences were observed in the same subjects. ADNI2 on the contrary was remarkably stable, as was the Edinburgh groups’ data, which showed very good internal measurement consistency, but still some inter-method differences. Mixed-effect growth curve modeling displayed only a linear model fit in the ADNI1 data, and displayed fits with linear, quadratic and cubic models for ADNI2, while latent growth curve analysis did not demonstrate even a significant linear model fit for ADNI1 but did show significant linear and quadratic model fitting for ADNI2, also suggesting that the ADNI1 UCD data is inconsistent and highly variable, while the ADNI2 data is more consistent with the literature on WMH change over time. “Caveat Emptor” should be exercised when using data produced by other groups and in particular, great caution and perhaps apprehension and skepticism should be used when considering using the UCD ADNI1 WMH data.
Project type:
  • Analysis of existing data
Imaging keywords:
Application / disease keywords:
  • 18-19