Bias in HD-ISS staging introduced by the FreeSurfer cross-sectional stream: Insights from the Huntington's Disease Young Adult Study (HD-YAS)

Introduction

Huntington's disease (HD) is a devastating progressive neurodegenerative disorder for which there remains no disease-modifying therapy. Inherited CAG repeat expansion undergoes further somatic expansion towards a critical length in vulnerable cell types, particularly the medium-spiny neurons in the striatum, ¹ causing neuronal damage, dysfunction, and death. ² This results in early atrophy within the striatum ³ which extends over time to other subcortical and cortical regions.^4,5

The HD Young Adult Study (HD-YAS) is a deeply phenotyped far-from-onset observational study of closely matched HD gene expanded (HDGE) and control groups, assessed longitudinally at two visits (V1 and V2) approximately 4.5 years apart. ³

The Huntington's Disease Integrated Staging System (HD-ISS) re-defined pre-manifest HD as stage 0 (without detectable biomarkers of pathophysiology) and 1 (caudate and/or putamen atrophy below the 5^th percentile in healthy controls). ⁶

We now enter a pivotal period for HD research, with multiple ongoing clinical trials of disease-modifying therapies. ⁷ Secondary prevention trials aim to target underlying disease pathobiology before significant neurodegeneration, with the goal of preserving the functional integrity of remaining striatal circuitry. Certain experimental therapies are also aiming to interfere with very early pathobiological mechanisms, including somatic expansion.^8,9 HD-ISS stages 0 and 1 are therefore possible inclusion criteria in future secondary prevention trials.

The HD-ISS used the FreeSurfer (FS) longitudinal stream (LG) to define stage 1 cut-offs, with the aim of reducing within-subject variability. ¹⁰ However, clinical trials will likely require application to participants with a single MRI, necessitating cross-sectional stream (CS) processing. MRI-derived volumetric outputs are not robust to pipeline ¹¹ or software version, ¹² and the impact of stream on volumes and staging remains unknown.

This study will explore the relationships and differences between caudate, putamen, and total intracranial volumes (TICV) derived from FS CS and LG in HD-YAS.

Methods

Results

Baseline demographics for participants are displayed in Supplemental Table 1.

All ICC values between CS and LG volumes were > 0.97 and were highly significant at p < 0.0005 (Supplemental Table 2).

Differences between volumes were described using Bland-Altman analysis (plots displayed in Figure 1 and systematic differences displayed in Supplemental Table 3). Compared to LG, CS calculated: i) smaller raw caudate (−5.5% V1, −5.1% V2, both p < 0.00005) and putamen volumes (−3.2% V1, −5.0% V2, both p < 0.00005), with a bias towards greater reduction for larger volumes and ii) similar TICV (−0.2% V1, + 0.3% V2, both p > 0.05). This resulted in smaller adjusted caudate (−5.3% V1, −5.6% V2, both p < 0.00005, without proportional bias) and putamen (−3.0% V1, −5.5% V2, both p < 0.00005, with greater reduction for larger volumes).

OPEN IN VIEWER

Figure 1.

Bland-Altman plots with differences represented as cross-sectional (CS) – longitudinal (LG) stream volumes

These volumetric differences impacted staging, with CS shifting the HDGE group from stage 0 to 1, moving from 9/54 to 15/54 at V1 and 19/54 to 25/54 at V2 (Figure 2).

OPEN IN VIEWER

Figure 2.

HD-ISS staging according to the FreeSurfer cross-sectional (CS) and longitudinal (LG) streams

These findings suggest that CS estimates larger adjusted caudate and putamen volumes which are less likely to reach the stage 1 threshold, according to the existing longitudinally derived cut-offs.

Discussion

This study adds to the growing body of evidence describing the meaningful impact of subtle pipeline differences on striatal segmentations, volumetric outputs, and HD-ISS staging in HDGE individuals. The use of HD-ISS staging as inclusion criteria in interventional trials will likely require application to HDGE individuals with a single MRI brain scan, precluding the use of LG.

Previous studies comparing CS and LG caudate and putamen segmentations are uncommon and have focused on reliability, showing improved test-retest reproducibility.^10,27,28 This is useful for assessing sensitivity to detect subtle change over time, which impacts sample sizes and follow-up periods in interventional trials. However, the impact of stream of staging requires an exploration of measurement bias.

This study has shown that caudate, putamen, and total intracranial volumes varied greatly between CS and LG. The combined effect was that CS estimated smaller adjusted caudate and putamen volumes, with greater effect on larger volumes. This shifted the HDGE group towards stage 1 (Figure 2). In particular, the use of CS at V1, and LG at V2, as might be assumed to be a reasonable methodology to maximize the accuracy of individual segmentations with growing available data, functioned to substantially reduce the staging progression between time-points.

It should be noted, however, that LG may be less suitable for estimating volumes when there is substantial atrophy over time. This limitation arises from a combination of factors: template bias (where a later atrophic scan can distort the segmentation of an earlier healthy scan); registration errors; and non-linear degeneration. ¹⁰ However, with minimal atrophy during the transition from HD-ISS stage 0 and 1, LG remains appropriate for use in this context.

The mechanistic explanation for these differences is challenging to define. LS applies the same affine transform to all time-points, deriving the same TICV, which is approximately the average of the two cross-sectional TICVs (Supplemental Figure 2), explaining the non-significant difference in TICV. Differences in caudate and putamen segmentation likely relate to the creation of a within-subject median template image to initialize the segmentation.

Overall, applying the HD-ISS to interventional trials requires staging with CS. This calculates larger adjusted caudate and putamen volumes than LG, which was used to generate stage 1 cut-offs, shifting from stage 0 to 1 across time-points. The HD-ISS must urgently define cut-offs derived from CS and make them publicly available before widespread application in clinical trials.

Supplemental Material