Influence of visible white matter lesions on volumes from automated parcellations.

Abstract

AbstractBackgroundWhite matter hyperintensities (WMH) are a common occurrence in ageing populations. Their hypointense appearance on T1‐weighted scans is known to affect automated processing and subsequent extraction of volumetric measurements. While lesion infilling with healthy appearing tissue is a common practice in the context of multiple sclerosis, this is not yet the case in studies of neurodegenerative diseases. We investigated the impact of WMH infilling on the volumetric outputs of two automated parcellation pipelines.MethodT1‐weighted and T2‐FLAIR scans were acquired for 462 individuals in the Insight46 neuroimaging substudy of the MRC National Survey for health and Development (NSHD). WMH were automatically segmented and T1‐weighted scans were subsequently infilled over the lesion masks. Two automated parcellation methods Geodesic Information Flows (GIF) and FastSurfer (FS) were run on the two T1‐weighted scans (original and infilled). Cortex, lateral ventricles, basal ganglia (BG), whole brain and total intracranial volumes (TIV) were derived from the four resulting parcellations. Spearman correlation was used to assess the relationship within and across methods according to infilling status. Correlations were also calculated between the observed volumes or percentage difference (original vs infilled volumes) and the WMH burden expressed as percentage of TIV.ResultThe relationship between parcellated volumes and WMH burden varied across methods and tissues/brain areas. While there was little observable relationship for GIF between brain volume or TIV difference and WMH, this relationship was positive for FS. The relationship with cortical volume difference was in opposite directions for GIF (negative relationship) and FS (positive relationship), indicating a tendency to decrease the estimated volume when WMH were visible for GIF and increase it for FS. For lateral ventricles and basal ganglia, volumetric measurements were higher in the presence of visible WMH for GIF and FS (Figure1). Correlations between FS and GIF were slightly stronger after infilling but correlation with WMH burden were attenuated (Table2).ConclusionThe impact of presence of visible WMH on T1‐weighted scan varies according to the chosen technique of analysis and strong correlations were observed between the difference in measured volumes and the WMH burden. Infilling should be considered when T1‐weighted derived volumes are studied.