Accuracy of automated amygdala MRI segmentation approaches in Huntington's disease in the IMAGE‐HD cohort

Abstract

Smaller manually‐segmented amygdala volumes have been associated with poorer motor and cognitive function in Huntington's disease (HD). Manual segmentation is the gold standard in terms of accuracy; however, automated methods may be necessary in large samples. Automated segmentation accuracy has not been determined for the amygdala in HD. We aimed to determine which of three automated approaches would most accurately segment amygdalae in HD: FreeSurfer, FIRST, and ANTS nonlinear registration followed by FIRST segmentation. T1‐weighted images for the IMAGE‐HD cohort including 35 presymptomatic HD (pre‐HD), 36 symptomatic HD (symp‐HD), and 34 healthy controls were segmented using FreeSurfer and FIRST. For the third approach, images were nonlinearly registered to an MNI template using ANTS, then segmented using FIRST. All automated methods overestimated amygdala volumes compared with manual segmentation. Dice overlap scores, indicating segmentation accuracy, were not significantly different between automated approaches. Manually segmented volumes were most statistically differentiable between groups, followed by those segmented by FreeSurfer, then ANTS/FIRST. FIRST‐segmented volumes did not differ between groups. All automated methods produced a bias where volume overestimation was more severe for smaller amygdalae. This bias was subtle for FreeSurfer, but marked for FIRST, and moderate for ANTS/FIRST. Further, FreeSurfer introduced a hemispheric bias not evident with manual segmentation, producing larger right amygdalae by 8%. To assist choice of segmentation approach, we provide sample size estimation graphs based on sample size and other factors. If automated segmentation is employed in samples of the current size, FreeSurfer may effectively distinguish amygdala volume between controls and HD.

Early amygdala and ERC atrophy linked to 3D reconstruction of rostral neurofibrillary tau tangle pathology in Alzheimer’s disease

“…While automatic segmentation protocols of amygdalar or entorhinal atrophy rate, as measured by MRI, would be optimal for measuring large numbers of individuals, existing algorithms have not yet achieved accuracy near that of manual segmentation. Indeed, the amygdala has been cited as a structure harder to segment than others in the MTL ( Jack et al, 1997 ), and recent comparisons of Freesurfer, FIRST, and ANTS segmentations of the amygdala to manual segmentations achieve Dice overlap scores only between 0.6 and 0.7 ( Alexander et al, 2020 ). A potential avenue for future work is the use of our manual segmentations to develop an accurate automatic segmentation algorithm.…”

“…Previous studies have suggested differences of involvement in right versus left amygdalas ( Yue et al, 2016 ). Future work will be needed to confirm or deny any differences in pathology versus atrophy in each hemisphere, using the manual segmentation protocol described here to avoid known biases in automatic segmentation protocols between right and left sided amygdalas, in particular ( Alexander et al, 2020 ).…”

Hippocampal and striatal volumes correlate with spatial memory impairment in Huntington’s disease

“…Our striatal volumetry results are consistent with broad evidence of neuropathological findings seen in HD within the striatum (Tabrizi et al., 2009, 2011). Using manual segmentation, which remains the “gold standard” in terms of volumetric measurement accuracy (Alexander et al., 2020), we found smaller striatal volumes in the HD group compared to healthy controls, and further within the HD group, we found smaller striatal volumes in the manifest HD group compared to the premanifest group. In comparison, our hippocampal volumetry results, which showed no group differences, contradicted previous findings of significant hippocampal degeneration in people with HD (Coppen et al, 2018; Faria et al., 2016; Younes et al., 2014) and in HD mouse models (Petrella et al., 2018; Rattray et al., 2017).…”

Neurite density imaging in amygdala nuclei reveals interindividual differences in neuroticism

“…According to a study by Alexander et al (2020), this also applies to the protocol suggested by Saygin et al (2017). Regarding this point of limitation, it has to be considered that Alexander et al (2020) compared automatic and manual segmentation in MRI data of higher resolution. However, anatomical landmarks for manual parcellation, like those proposed by Entis et al (2012) andA.…”

“…Thus, automated parcellation protocols, like the one proposed by Saygin et al (2017), are currently the most appropriate way to partialize imaging data of low spatial resolution as used in our study. Additionally, as the overestimation caused by automated parcellation seems to be systematic, it mainly affects conclusions about absolute amygdala volume as well as studies comparing absolute volume across several segmentation methods (Alexander et al, 2020).…”