The ro-ALPS index exhibited a significantly smaller variance than the original ALPS index. In the evaluation of intra- and inter-reliability, the reorientation technique showed good-to-excellent reproducibility in calculating the ALPS index even in subjects with head rotation. The wider range of the 95% limit of agreement of the Bland–Altman plot for subjects with x axis rotation indicated that x axis rotation may considerably affect the calculation of the original ALPS index.

Compared with conventional MR sequences, DTI data are complicated because they contain both signal intensity information and vector information [14]; therefore, DTI analyses, especially tractography and connectome evaluation, may be recommended to be performed on the original DTI space. In contrast, this study calculated the ALPS index on the reoriented space. One reason is that the ALPS index is a simple metric and is not influenced by continuity of the voxels. The vector directions along the x-, y-, and z axes in relation to the brain are crucial for accurate calculation of the ALPS index. A previous study also reported that alterations in the imaging plane and head position largely influenced reproducibility of the ALPS index [11], and that it may be appropriate to determine the imaging plane according to the direction of the anterior commissure–posterior commissure line. The technique used in this study enabled easy creation of diffusivity maps with the reoriented x-, y-, and z axes, and the imaging plane of the reoriented diffusivity maps appeared to show an appropriate direction in relation to the head position. Hence, the reoriented head position and fitted diffusivity maps may lead to a smaller variance in the ALPS index.

For reproducibility analyses, the ro-ALPS index showed good-to-excellent intra- and inter-rater reliability, even when the head position was moderately rotated. Using the reorientation technique, the best regions for ROI placement at the projection and association fibers can be easily identified, leading to very good intra- and inter-rater reliabilities. In this study, we manually placed ROIs to evaluate the reproducibility of this technique, whereas previous studies introduced other methods to automatically or semi-automatically place ROIs [9,10,11, 15]. A combination of these methods may further improve the reproducibility of calculating the ALPS index, although our method has already demonstrated good-to-excellent intra- and inter-reliability. Another advantage of our technique is that a reoriented diffusivity map, which corrects the x-, y-, and z directions of diffusivity fitted to each brain, can be created. Zhang et al. [10] registered the original diffusivity map to the brain template, but did not register vector information; hence, their technique could not be applied to subjects with head rotation. The FSL function “vecreg” is a command-line tool that allows the registration of vector data. It does not calculate a transformation but simply applies a given transformation to the input vector field. This function can be applied to the V1 (1st eigenvector), V2 (2nd eigenvector), V3 (3rd eigenvector), and diffusivity maps. Therefore, this technique may possess sufficient function for calculating the ro-ALPS index. When the original DTI with multiple MPG is registered to a target image and the reoriented b-vector information could be calculated using other methods, reoriented diffusivity maps could also be created; however, this might be complicated. In contrast, our technique could create reoriented diffusivity maps using only the FSL function; therefore, this technique could be easily used to calculate the ALPS index even when DTI data were not obtained with an appropriate head position. To our knowledge, this technique has been used in a few recent studies [16, 17]; however, reproducibility of the ALPS index has not been evaluated. We believe this technique would be useful for improving intra- and inter-reliability and will be beneficial in future studies to estimate glymphatic function of the brain.

The Bland–Altman plot between the ALPS and ro-ALPS showed a wider range of the 95% limit of agreement for subjects with x axis rotation (all cases with chin-up) than for subjects with an appropriate head position, indicating that x axis rotation may considerably affect the calculation of the original ALPS index. This finding was in agreement with that of a previous study that reported a relatively lower ALPS index for subjects with chin-up [11]. As shown in Supplemental Table 1, although subjects with x axis head rotation tended to show high mean(Dx,proj, Dx,assoc) and high mean(Dy,proj, Dz,assoc), the high mean(Dy,proj, Dz,assoc) may have derived in relatively lower values for the ALPS index. Meanwhile, z axis rotation (neck rotation) may have little effect on the calculation of the original ALPS index, partly because this study placed ROIs on both hemispheres and included subjects with neck rotations to both sides. Nonetheless, alteration in the head position and imaging plane had a remarkable impact on the calculation of the ALPS index.

Although the sample size was not large, the test–retest evaluation confirmed that the reorientation technique shows high reproducibility for the calculation of the ALPS index.

The major limitation of this study was that we used open-source DTI data, and that most DTI datasets were obtained once, with a single head position. The DTI data of the OASIS-3 were obtained with consistent acquisition parameters and the same number of MPG directions. When we evaluated the correlation between the ALPS index and age, a significant association was identified, which was compatible with the results of previous studies [10, 11, 18, 19]. Hence, the OASIS-3 data seemed reliable to some extent. To determine whether the ro-ALPS index is truly reliable, further validation is required using a large sample of subjects who underwent MR examination with a rigid neutral head position and with a rotated head position. In contrast, the current study would be a good simulation of retrospective studies because the head position could not be corrected for previously obtained DTI data. Second, the imaging resolution and size of the ROIs used in the calculation of the original and reoriented ALPS indices were different, which may have affected their values. In contrast, a mean value within the ROIs was mathematically used for calculating the ALPS index, which may have mitigated the effect of the difference in resolution or size. Third, although this study evaluated the degree of head rotation by visual inspection, it may be appropriate to decide the severity of head rotation by an objective method based on the transform matrix or the angle of the imaging plane in relation to the anterior commissure–posterior commissure line. However, the weighted κ coefficient of the severity of head rotation showed good agreement, which suggests that our method was acceptable to certain degree.

In conclusion, the registration technique in this study enabled creation of reoriented diffusivity maps along the x-, y-, and z axes in relation to the reoriented head position and improved reproducibility in calculating the DTI-ALPS index even when the head position or imaging plane was inappropriate. This technique has potential to be widely applied in future prospective and retrospective studies to evaluate the DTI-ALPS index.