The most important findings of this study are threefold: First, there was a moderate scan–rescan reliability of the TSE scan in visualizing the LC; second, the LC volume estimated with the TSE scan appears to be smaller than volumes reported in ex vivo studies; and third, we observed a lateralization effect in terms of LC volume and intensity.
Scan–rescan reliability
There was a moderate scan–rescan reliability of the LC. Taking into consideration the challenges of imaging the LC due to its location and small volume and the fact that these reliability indexes are similar to other, bigger structures located in less susceptible parts of the brain (e.g., the amygdala, reliability of 0.67–0.89 for automated segmentation and 0.75 for manual; Bartzokis et al.
1993; Morey et al.
2010), we conclude that localization and segmentation of the LC in vivo are a challenging but reliable enterprise.
The moderate inter- and intra-rater reliability (as assessed with the Dice coefficient) shows moderate reproducibility of the TSE scan in terms of LC visualization. This reliability was stable across the two raters, the two scan sessions, the two segmentation sessions, and the two hemispheres. A stable inter-rater and inter-segmentation session reliability is an indication that the raters performed the segmentation in a reliable manner. The moderately stable scan-to-scan reliability has implications for longitudinal studies and suggests that this scan can be applied to the same participant more than once with a moderate confidence that it will lead to the same result. Our evaluations are limited to two scanning sessions, but future research can investigate the reliability of the TSE scan in multiple sessions.
This is the first study that was designed to assess TSE scan reliability of the LC, but there are two other studies of which the results are pertinent to this topic. The intra-rater values reported in these studies are higher than those reported here (0.89–0.94 and 0.98–0.99 for Ohtsuka et al.
2013 and Takahashi et al.
2015, respectively, and 0.65–0.74 for our study). This discrepancy can be explained by methodological differences. More concretely, we assessed intra-rater agreement using Dice coefficients and masks that were manually segmented in each individual’s native space, whereas Ohtsuka et al. (
2013) and Takahashi et al. (
2015) report intra-observer agreement using an ICC approach (instead of Dice coefficient) and a fixed 1- or 2-mm-diameter circle for LC segmentation. The approach of employing fixed diameter for the ROI segmentation is not optimal for assessing reliability because it entails the risk of losing part of the LC or of misattributing surrounding tissues to the LC. Indeed, as already mentioned, although histological studies show that the LC is 2.0–2.5 mm wide, there is a substantial variability in the LC shape. Additionally, this approach utilizes a fixed circle that is smaller than the actual size of the LC; thus, it might capture a region where the LC signal is at its maximum and bias the intra-rater values toward the high end of the scale. Finally, in Takahashi et al. (
2015), one rater performed the segmentation three times and the in between interval was shorter than in this study (1 week vs. at least 2 weeks), while in Ohtsuka et al. (
2013) the segmentation interval is not mentioned.
Regarding the scan-to-scan reproducibility, a third study should be mentioned: Langley et al. (
2016) report higher reproducibility values for the scan–rescan magnetization transfer contrast (ICC = 0.76) and a mean Dice coefficient of 0.63 for the delineation of the LC scan-to-scan volumes. However, our findings cannot be directly compared with the results of this study, because Langley and colleagues utilized a different MRI sequence: a gradient echo pulse scan. It has been argued that this sequence, similar to the TSE sequence, is sensitive to the presence of neuromelanin (Chen et al.
2014; Langley et al.
2016). In addition, there are also methodological differences between the two studies in terms of: (a) segmentation procedure (no manual segmentation of the mask), (b) ROI definition (LC contrast extraction based on a fixed 3-mm-diameter circle placed over the left and the right LC, and consecutive exclusion of the voxels that were four standard deviations above the mean intensity of the reference ROI), (c) definition of LC intensity assessment, and (d) scan-to-scan session interval (both scanning sessions were on the same day).
LC volume
The volume of the individual-rater LC masks was 9.53 mm
3 on average (SD 3.83) and ranged between 0.82 and 25.29 mm
3 (per hemisphere). There is a discrepancy in the postmortem literature regarding the exact size and location of the LC, and there seem to be large inter-individual differences in LC cell distribution (Afshar et al.
1978; Fernandes et al.
2012; German et al.
1988; see Table
2). However, the volume found in our study is smaller than one would expect based on postmortem studies (see Table
2). A similar LC volume was reported with another type of neuromelanin MRI sequence, the gradient echo pulse scan (Chen et al.
2014). The reason why MRI scans lead to decreased LC volume estimates compared to postmortem estimates is not clear, but we speculate that the discrepancy might be due to the following reasons: (a) methodological MRI factors, such as the possibility that current neuromelanin MRI scans might not be very sensitive, and an improvement of these scan sequences might lead to better volume estimations; (b) the homogeneity of the sample in terms of age span (e.g., young/homogenous vs. old/non-homogeneous population); and (c) partial volume effects. We will discuss each of these factors in turn.
Table 2
Estimation of human LC volume based on prior postmortem literature
| 13–17 | 2.5 | 2.5 | 17.2–32.8 | 3.14 × (1.25)2 × 15 = 73.59 | Entire LC |
7.2 | 2.5 | 2.5 | | 35.26 | “Core” LC only |
| 14.5 | 2.5 | 2 | | 3.14 × 1.56 × 14.5 = 71 | Entire LC |
11 (80% of cases) | 2.5 | 2 | | 3.14 × 1.56 × 11 = 53.88 | “Core” LC only |
10 (90% of cases) | 2.5 | 2 | | 3.14 × 1.56 × 10 = 48.98 | “Core” LC only |
7.5 (100% of cases) | 2.5 | 2 | | 3.14 × 1.56 × 7.5 = 36.74 | “Core” LC only |
| 10 | 1.28 | 1.23 | | 3.14 × 1.63 × 10 = 51.44 | Entire LC |
6 (100% of cases) | 1.04 | 1.10 | | 3.14 × 1.21 × 6 = 22.81 | “Core LC” only |
Regarding the first point, it has been argued that the TSE scan can visualize the LC because, similar to histological methods, it is sensitive to the neuromelanin pigments that exist in the LC cells (Keren et al.
2009,
2015; Sasaki et al.
2006). Histological and MRI studies show that neuromelanin concentration is highly dense in the center (“core”) of the LC and more spread in the rostral and caudal extremities. For Keren et al., the elevated signal in the (in vivo) TSE scan corresponded to the location of greatest LC neuron density as reported in the postmortem LC study by German et al. (
1988) and Keren et al. (
2009,
2015). For Fernandes et al. (
2012), and for Afshar et al. (
1978), this area corresponds to the part of the LC that is common for every case (present and shared by the 100% of the cases; see Table
2). This might mean that the TSE scan captures mainly the “core” region of the LC or cannot fully capture the part where the LC cell distribution is less dense. If the TSE scan cannot capture the entire size of the LC, it will substantially reduce the volume of the LC compared to the size reported in histological studies. Although the exact volume of this highly dense, “core” region of the LC is not mentioned in prior studies, it can be estimated based on the information provided in the papers. Based on this information, we estimate that the core region of the LC is approximately 35 mm
3 for German et al. 37 mm
3 for Fernandes et al. and 23 mm
3 for Afshar et al. (see Table
2). These core LC volume values are closer to the LC volume reported in our study, although still a factor three larger than the measured volumes.
As far as age is concerned, although not all studies support this finding (Fernandes et al.
2012; Mouton et al.
1994; Takahashi et al.
2015), postmortem and in vivo MRI studies show that changes in size or intensity occur to the LC structure with age (Clewett et al.
2016; German et al.
1988; Keren et al.
2009; Lohr and Jeste
1988; Manaye et al.
1995; Ohtsuka et al.
2013; Shibata et al.
2006; Vijayashankar and Brody
1979; Zecca et al.
2004). It has also been argued that neuromelanin concentrations increase with age (Mann and Yates
1974; Zecca et al.
2004). If that is the case, the inclusion of young participants in our study might have resulted in smaller LC volumes due to lower levels of neuromelanin. Future research concentrating on reproducibility of the TSE scan in elder participants, employing similar methods as in the current study, can help address this question.
Finally, partial volume effects might play a role too. Indeed, when imaging a small and thin brain structure like the LC, the volume can be underestimated, for example due to loss of visualization of the upper or lower part of the LC (Hoffman et al.
1979; Vos et al.
2011). Yet, the use of high contrast, high spatial resolution sequence, similar to the one used here, decreases these effects, leading to increased visualization of the tissue, less mixing of signals coming from different regions, and sharper definition of the individual tissue (Kneeland et al.
1986).
LC contrast
The range in LC
contrast ratio (4.5–32.4%) was wide, suggesting a large inter-subject variation in visualization of the LC (Fig.
4a). Our results are similar to Takahashi et al. (
2015), who, by using a TSE sequence, report an LC contrast range of 6.24–20.94% (median 14.35%) for healthy volunteers and a significant drop of LC contrast in patients with mild cognitive impairment and Alzheimer’s disease. The LC
contrast ratio did not differ between scan sessions 1 and 2, suggesting that the scan is reliable and can be used in longitudinal studies. Yet, the fact that the reliability is moderate and that a high correlation was observed between the LC
contrast ratio of the right and left LC only for scan session 1 but not for session 2 (Fig.
4b) suggests that changes in signal intensities over time should be interpreted with caution. The mean LC
contrast ratio for the peak voxel analysis (14.4%) was similar to the mean LC
contrast ratio of the ROI analysis (13.9%). However, similar to Keren et al. (
2009), and contrary to the ROI approach, we found no significant lateralization effect in the peak voxel approach. This suggests that the peak approach might not be sensitive enough to detect the effect due to its limited coverage and decreased robustness.
Lateralization effect
Our results of the LC volume and ROI intensity analysis suggest an LC lateralization with the right LC being larger and of higher intensity than the left LC. This lateralization effect was not reported before and the majority of the LC studies highlight its bilateral hemispheric symmetry (Chan-Palay and Asan
1989a,
b; Fernandes et al.
2012; German et al.
1988; Keren et al.
2009; Ohm et al.
1997; Vijayashankar and Brody
1979). However, German et al. (
1988) mention that “although there is a bilateral symmetry, the two sides do not appear identical” and report that the total horizontal area of the left LC is smaller than that of the right LC for one of the five cases. Keren et al. (
2009) found that “the LCs are not perfectly symmetrical in peak or in the variance of the peak location.” When the same authors employed 7 T MRI (using a RARE-INV MR scanning sequence), the asymmetry became more obvious (note the hemispheric asymmetry in size and shape of the putative LC contrast through slices 5–7 in Fig.
4, p. 6; Keren et al.
2015). In line with our study, Keren et al. (
2015) show elevated contrast in the right LC in comparison with the left side at least for one subject (see Fig.
5; Keren et al.
2015).
It is important to note that lateralization in the brainstem has not been investigated in detail for two reasons. First, until the discovery of the ability of the TSE scan to generate LC-specific contrast, it was not possible to image the monoamine brainstem nuclei in vivo. Second, it has been a common approach in MRI methods to investigate lateralization effects in the cortex, but to perceive the brainstem and the LC as one single midline structure (e.g., Morey et al.
2010; Ohtsuka et al.
2013; Takahashi et al.
2015). However, lateralization effects have been reported for other brain structures that exist in pairs (e.g., the amygdalae and the hippocampi; Baas et al.
2004; Cahill et al.
2004; Frings et al.
2006; Iglói et al.
2010).
Finally, technical explanations of the observed lateralization effects, such as RF asymmetry, cannot be ruled out. For example, Zwanenburg et al. reported signal asymmetries in FLAIR scans due to RF inhomogeneities (Zwanenburg et al.
2013). Taking into consideration that lateralization effects play an important role in brain function, future studies should further investigate whether our finding of LC lateralization can be replicated, and whether this lateralization also exists for LC function.
The LC probability atlas
Our results show substantial variability in the spatial location of the LC, given that the maximum percentage overlap was only 36%.
There is only one in vivo atlas of the human LC published to date (Keren et al.
2009). The atlas described in this study differs on three crucial aspects from that atlas: segmentation method, sample type, and information. Contrary to the atlas by Keren et al. (
2009), the entire visible LC was segmented, providing a more extensive coverage of the LC. This aspect of our approach is more relevant for fMRI studies in which the extent of activation refers to multiple voxels instead of peak coordinates; an fMRI study that uses a peak approach atlas entails the risk that the cluster of activation extending outside the LC map is missed. Additionally, in the current atlas we adopted a quantification approach and we provide the probabilistic information on where the LC is located. This information can, for instance, be used to weigh the measured fMRI signal with the probability of it originating from the LC. Finally, our LC atlas is based on a homogeneous sample of young participants, which is more representative of and relevant for most experimental studies in psychology and neuroscience, given that the majority of the (fMRI) studies in cognitive neuroscience are based on healthy young volunteers (Chiao
2009; Henrich et al.
2010).
Although the probability LC atlas can be used as an ROI for the LC in future studies, it should be noted that the use of an atlas is always less anatomically precise than the individually determined masks. Given that our TSE scanning protocol is relatively short (7 min), and covers a large region in the brainstem, with a relatively high spatial resolution (0.34 × 0.34 × 1.5 mm), we recommend to include such a structural scan during the data acquisition phase (in this study we also provide a relevant segmentation protocol to assist in the creation of individual LC masks, see “
Appendix 1”). If this is, however, not feasible, one could consider using the probability atlas.
A strong aspect of the LC atlas, as mentioned above, is the homogeneous sample on which it was based. But one limitation is the small size of this sample.
Another limitation refers to the TSE scan which has a limited coverage of the brainstem due to the compromise between signal-to-noise ratio and increased resolution. Although our study has a larger coverage than other studies, it still does not provide full coverage, making planning of the imaging volume somewhat troublesome during the acquisition. By planning the volume perpendicular to the brainstem, by utilizing anatomical landmarks such as the fourth ventricle and the inferior colliculus, we were successful in always including the LC into the imaged volume.
Finally, an additional limitation of the TSE scan is the voxel size of 0.35 × 0.35 × 1.5 mm which might be considered relatively big for such a small structure as the LC. Initial pilot scans with a smaller voxel size were tested but showed substantial loss of image quality. A possible explanation for this is that the increased acquisition time resulted in more motion artifacts.