Reinforcing the Weakest Links: Modernizing SIENA with Targeted Deep Learning Integration

Abstract

Percentage Brain Volume Change (PBVC) derived from Magnetic Resonance Imaging (MRI) is a widely used biomarker of brain atrophy, with SIENA among the most established methods for its estimation. However, SIENA relies on classical image processing steps, particularly skull stripping and tissue segmentation, whose failures can propagate through the pipeline and bias atrophy estimates. In this work, we examine whether targeted deep learning substitutions can improve SIENA while preserving its established and interpretable framework. To this end, we integrate SynthStrip and SynthSeg into SIENA and evaluate three pipeline variants on the ADNI and PPMI longitudinal cohorts. Performance is assessed using three complementary criteria: correlation with longitudinal clinical and structural decline, scan-order consistency, and end-to-end runtime. Replacing the skull-stripping module yields the most consistent gains: in ADNI, it substantially strengthens associations between PBVC and multiple measures of disease progression relative to the standard SIENA pipeline, while across both datasets it markedly improves robustness under scan reversal. The fully integrated pipeline achieves the strongest scan-order consistency, reducing the error by up to 99.1%. In addition, GPU-enabled variants reduce execution time by up to 46% while maintaining CPU runtimes comparable to standard SIENA. Overall, these findings show that deep learning can meaningfully strengthen established longitudinal atrophy pipelines when used to reinforce their weakest image processing steps. More broadly, this study highlights the value of modularly modernizing clinically trusted neuroimaging tools without sacrificing their interpretability. Code is publicly available at https://github.com/Raciti/Enhanced-SIENA.git.

Figures (3)

• Figure 1: Qualitative comparison of traditional and deep-learning frameworks for brain extraction and segmentation. (a) Brain extraction results using BET2. (b) Brain extraction results using SynthStrip. (c) Tissue segmentation results using FAST, performed on the BET2 extraction output. (d) Tissue segmentation results using SynthSeg, performed on the BET2 extraction output (labels aggregated for comparison as described in Section \ref{['sec:synthseg-integ']}). Compared to BET2 (a), SynthStrip (b) provides more anatomically consistent and robust exclusion of non-brain structures. Furthermore, SynthSeg (d) demonstrates greater robustness than FAST (c) when processing a suboptimal initial brain extraction.
• Figure 2: Overview of the four evaluated pipelines. The rows represent SIENA Vanilla (baseline), SIENA-SS (modified skull stripping), SIENA-SEG (modified segmentation) and SIENA-SS-SEG (fully integrated) workflows. Input data consists of baseline and follow-up T1-weighted MRI scans (MRI A and MRI B). The skull stripping step shows the brain extraction results, with orange overlays indicating the segmented brain volume and cyan contours marking the skull-mask boundaries. The FLIRT step shows the affinely co-registered scans in a common halfway space. The segmentation step depicts tissue classification into CSF, GM, and WM using FAST or SynthSeg. All workflows conclude with the estimation of PBVC.
• Figure 3: Scan-order consistency of PBVC estimates across pipelines and datasets. Scatter plots compare the estimated PBVC for the original temporal scan order, $\mathrm{PBVC}(A,B)$ (x-axis), with that for the reversed scan order, $\mathrm{PBVC}(B,A)$ (y-axis), in the ADNI (top row) and PPMI (bottom row) datasets. A perfectly scan-order-invariant pipeline would yield points lying exactly on the antisymmetry diagonal. The MFRR and its standard deviation are reported for each configuration.