Background In the IMAGE-HD Huntington’s disease (HD) cohort (35 presymptomatic HD [‘preHD’], 36 symptomatic HD [‘symHD’], and 35 controls), Ahveninen et al. (2018) manually segmented amgdalae, finding smaller amygdala volumes associated with poorer motor and cognitive function in HD. Although manual segmentation is the gold standard in terms of accuracy, it is very time consuming. The accuracy of automated methods is undetermined for the amygdala in HD. Ahveninen et al.’s segmentations provide ‘ground truth’ to test these methods against.

Aims We aimed to determine which of three automated approaches would most accurately segment amygdalae in HD: FreeSurfer, FIRST, and ANTS registration with FIRST.

Methods T1-weighted images for the 106 IMAGE-HD participants (23–72 years of age, 59 female) were utilized, and input into the default pipelines of FreeSurfer and FIRST. For the third approach, images were bias corrected, nonlinearly registered to an MNI template using ANTS, then segmented using FIRST.

Results Manual segmentation had the greatest sensitivity, with all groups differing from each other. Amygdala volume was smallest for symHD, preHD, then controls. No automated approach revealed volume differences between preHD and symHD groups. Differences between symHD and controls were detected with ANTS/FIRST. PreHD and symHD differed from controls with FreeSurfer. FreeSurfer introduced a hemispheric bias not evident with manual segmentation, producing larger right amygdalae. In terms of anatomical accuracy, overlap (Dice) scores between manual segmentations and each automated approach were highest for FIRST (0.65), ANTS/FIRST (0.64), then FreeSurfer (0.61).

Conclusions Manual segmentation was more sensitive to group differences in amygdala volume than automated approaches tested, being the only method to detect differences between preHD and symHD groups. If automated methods are used, FreeSurfer may effectively distinguish between controls and both preHD and symHD groups.