Neuromelanin-sensitive MRI (NM‑MRI) allows in vivo assessment of brainstem nuclei implicated in Parkinson's disease, notably the substantia nigra (SN) and locus coeruleus (LC). Longitudinal NM‑MRI studies typically apply cross-sectional processing to each timepoint, which can introduce processing-related variability and confound estimates of within-subject change. The authors developed and validated a purpose-built longitudinal neuromelanin-MRI processing framework intended to improve measurement stability across repeated scans and reduce technical variability that might obscure true biological change.
The analysis used imaging and clinical data from the Quebec Parkinson Network. The full dataset comprised 268 participants: 199 with Parkinson's disease and 69 controls. A longitudinal subset included 74 participants (49 PD, 25 controls) who were re-scanned approximately one year later. Primary measurement outcomes included contrast‑ratio and volume metrics for the SN (reported as SN_CR and SN volume metrics) and contrast‑ratio metrics for the LC (LC_CR).
Validation experiments focused on multiple potential sources of technical variability. First, slice-by-slice intensity normalization was evaluated to correct slice-dependent intensity variation across brainstem NM‑MRI acquisitions. Second, conventional bias field correction (N4) was applied and assessed specifically for its effect on LC signal asymmetry. Third, the study compared conventional cross-sectional processing to a longitudinal registration approach to test whether within-subject alignment across timepoints improves measurement precision.
Implementing longitudinal registration materially reduced variability in annualized change estimates for control participants. For SN_CR metrics, annualized change variability decreased by 25–36%; for LC_CR metrics, decreases were 27–34%. In addition, residual variability for contrast-based measures was reduced by up to 28%. These reductions indicate improved within-subject measurement stability when longitudinal registration is applied, supporting the framework's ability to lower technical noise in repeated NM‑MRI measures.
Across baseline and follow-up timepoints, longitudinal registration generally produced larger PD–control effect sizes, with the most consistent improvements observed for SN volume metrics. Despite improvements in measurement precision and larger effect sizes in many cases, the authors report that there were no significant method × group × time interactions. This indicates that, over the approximately one-year follow-up evaluated here, estimated longitudinal trajectories of NM‑MRI metrics did not differ significantly between the longitudinal processing pipeline and conventional cross-sectional methods.
Beyond annualized change variability, the longitudinal framework also reduced residual variability for contrast-based metrics. The reported maximum reduction in residual variability was up to 28%, demonstrating that longitudinal registration and the associated processing steps yield more precise contrast measures for SN and LC signals. The slice-by-slice intensity normalization additionally reduced brainstem intensity variability by 3.6%, indicating a modest but measurable improvement in slice-level uniformity.
A systematic leftward signal asymmetry was observed in the LC across the dataset. This asymmetry persisted even after application of N4 bias field correction, suggesting that it is likely attributable to scanner-related effects that are not fully modeled by conventional bias-field correction methods. The persistence of leftward LC asymmetry highlights a technical limitation that the authors could not fully resolve within the tested bias correction approach and that may require alternative correction strategies or scanner-specific calibration.
The longitudinal NM‑MRI processing pipeline developed in this study is openly available at the project GitHub repository: https://github.com/NIST-MNI/nm-longitudinal-pipeline. The authors report that ethical approval was granted by the Research Ethics Board of the McGill University Health Centre and that participant consent procedures were followed. The study is presented as a medRxiv preprint and has not been peer reviewed; the authors note that longitudinal registration improved technical precision but did not produce statistically significant improvements in detected longitudinal PD–control differences within the follow-up interval analyzed. This implies that the pipeline may be most valuable for enhancing sensitivity to subtler biological effects or in studies with longer follow-up or larger samples.
The study demonstrates that a dedicated longitudinal neuromelanin‑MRI processing framework, incorporating slice-by-slice intensity normalization and within-subject longitudinal registration, reduces technical variability and improves precision of SN and LC measurements. While the pipeline did not change the overall longitudinal trajectories detected between PD and control groups over the one-year interval studied, it increased measurement stability and effect sizes for some metrics, particularly SN volume. The persistent LC asymmetry after conventional bias correction highlights the need to consider scanner-related effects in multicenter and longitudinal NM‑MRI studies. The authors provide their processing pipeline openly to support replication and broader adoption, with the goal of improving sensitivity to subtle longitudinal changes in future PD imaging research.