In vivo visualization of the locus coeruleus in humans: quantifying the test–retest reliability

Abstract

The locus coeruleus (LC) is a brainstem nucleus involved in important cognitive functions. Recent developments in neuroimaging methods and scanning protocols have made it possible to visualize the human LC in vivo by utilizing a T1-weighted turbo spin echo (TSE) scan. Despite its frequent use and its application as a biomarker for tracking the progress of monoaminergic-related neurodegenerative diseases, no study to date has investigated the reproducibility and inter-observer variability of LC identification using this TSE scan sequence. In this paper, we aim to quantify the test–retest reliability of LC imaging by assessing stability of the TSE contrast of the LC across two independent scan sessions and by quantifying the intra- and inter-rater reliability of the TSE scan. Additionally, we created a probabilistic LC atlas which can facilitate the spatial localization of the LC in standardized (MNI) space. Seventeen healthy volunteers participated in two scanning sessions with a mean intersession interval of 2.8 months. We found that for intra-rater reliability the mean Dice coefficient ranged between 0.65 and 0.74, and inter-rater reliability ranged between 0.54 and 0.64, showing moderate reproducibility. The mean LC contrast was 13.9% (SD 3.8) and showed scan–rescan stability (ROI approach: ICC = 0.63; maximum intensity approach: ICC = 0.53). We conclude that localization and segmentation of the LC in vivo are a challenging but reliable enterprise although clinical or longitudinal studies should be carried out carefully.

Higher rostral locus coeruleus integrity is associated with better memory performance in older adults

“…Repeated measurements of an independent younger adult sample further confirmed high reproducibility of the intensity assessment (see Supplemental results). However, as reported earlier for neuromelanin-sensitive sequences, our analyses do not capture the complete LC extent 47,66 leading to lower volume estimates than those obtained post-mortem studies 15 . In their histological analyses, Fernandes and colleagues describe a common, central LC zone which is shared across all subjects 14 .…”

“…Accordingly, the segments with the most dispersed noradrenergic cells in the caudal LC show less overlap on a group level and are thus arguably more difficult to detect using a semi-automatic approach with a conservative intensity threshold (i.e., 4SD above mean intensity of a reference region 64,65 ). This circumstance is likely exacerbated by partial volume effects due to the relatively thick slices in most neuromelanin-sensitive sequences 66 . However, manual 58 Figure 2d).…”

“…We established a multi-step analysis procedure to extract individual LC intensity values that addressed a series of methodological concerns raised in previous research: First, TSE scans of the LC typically demonstrate a low signal-to-noise ratio due to the high spatial resolution needed to image the small nucleus 64 . In addition, the LC shows amorphous borders, complicating the manual segmentation of the structure 41,66 . Thus, the use of semi-automatic threshold-based approaches was recently suggested to circumvent manual tracing of the nucleus and associated errors as well as potential biases (e.g., non-blinded raters 47,64,65 ).…”

“…The cut-off of 4 standard deviations above the mean reference intensity was selected on the basis of previous research 64,65 and was confirmed visually as generating a volume that matched prior mapping studies of the LC 41,58 . A lateralized threshold was computed for each slice to counteract previously reported unilateral biases in brainstem signal intensity 58,66 and intensity fluctuations along the rostrocaudal axis. Next, we generated a volume of interest (VOI), subsequently referred to as search space, that included only those areas that survived thresholding.…”

“…The aggregated maximum intensity plot was then converted to express the likelihood to find the LC for any given voxel (i.e., a LC location probability map; cf. 41,66 ). We thresholded the image to show only locations that were shared across participants (i.e., the upper 90 percentiles) and spatially standardized the probability map in a two-step procedure.…”