Reply to ‘The method matters: free water imaging in Parkinson’s disease is not a binary verdict’

We reply to the Matters Arising on our study comparing neuromelanin and free water imaging for early Parkinson’s disease diagnosis. The authors raised considerations regarding image registration, age imbalance, repetition time, and conceptual framing. We clarify that our registration pipeline is validated for subcortical structures, that age does not account for neuromelanin’s superior performance, and that our acquisition parameters meet recommended standards.

Our findings reflect comparative performance, not a verdict. We thank the authors for their thoughtful engagement with our recent work on the comparative diagnostic performance of neuromelanin (NM) and free water (FW) imaging in early Parkinson’s disease (PD)1 and for raising important methodological considerations regarding FW imaging. We agree that the technical implementation of FW imaging is a critical determinant of its diagnostic sensitivity, and we welcome the opportunity to clarify our methodology and address each of the points raised.

We acknowledge that fractional anisotropy (FA)-based templates can provide improved alignment in white matter tract analyzes where FA contrast is high. However, for subcortical structures such as the substantia nigra (SN), FA contrast is inherently low, and the optimal registration approach for such regions remains debated. A systematic comparison of nonlinear deformation algorithms for subcortical alignment demonstrated that ANTs Symmetric Normalization (SyN) — the registration framework employed in our pipeline — achieved the highest Dice coefficient and reached accuracy comparable to inter-rater manual segmentation2.

We wish to clarify that our diffusion tensor imaging (DTI) data were not directly registered to an MNI template. We employed a two-step ANTs-based registration in which DTI volumes were first aligned to each subject’s native T1-weighted image (resampled to 1.5-mm isotropic resolution), and the concatenated transformations were then inverted to bring the SN mask back into native DTI space. The SN mask was derived from a high-resolution 7T-based ATAG atlas.

Comparable T1-based registration strategies have been previously employed in FW imaging of the SN and successfully detected FW elevations in idiopathic REM sleep behavior disorder3 and PD4, supporting the validity of this approach. Furthermore, the identical pipeline was applied to both FW and NM analyzes. Although two modalities rely on different biophysical principles and have distinct technical sensitivities, the consistent performance of the pipeline in detecting NM differences suggests that registration error may not fully account for the absence of FW group differences.

We additionally note that in our external validation cohort, FW values were elevated in the associative subregion of patients with early PD despite the use of the same registration approach, indicating that the pipeline can detect subregional FW differences when present. The age difference between our early PD and control groups (median 67 vs. 70 years) was statistically significant but clinically modest, with both groups falling within a narrow range near the plateau of the NM signal–age trajectory5. We agree with the authors’ observation that age and PD influence FW in the same direction, whereas their effects on the NM signal in this age range differ.