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

Riccardo Raciti; Lemuel Puglisi; Francesco Guarnera; Daniele Ravì; Sebastiano Battiato

eess.IVMar 13, 2026

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

Authors: Riccardo Raciti, Lemuel Puglisi, Francesco Guarnera, Daniele Ravì, Sebastiano Battiato

Organizations: Dept. of Math and Computer Science, University of Catania, Viale Andrea Doria, 6, Catania, Italy · MIFT Department, University of Messina, Viale Ferdinando Stagno d’Alcontres 31, Messina, Italy

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 & tables

Panel of Figure 1 — 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 4.2 ). 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.

Dataset	$N$	Female (%)	Age Mean
ADNI	1006	445 (44%)	73.18 $\pm$ 7.07
PPMI	310	133 (43%)	63.0 $\pm$ 8.90

Table 1: Summary of participants demographics across the evaluated cohorts.

$\Delta$ Calc.	Original Score	Included Measures
$t_{0}-t_{1}$	Decrease indicates worsening	MMSE, MoCA, GM Volume, BPF, MSEADLG
$t_{1}-t_{0}$	Increase indicates worsening	ADAS-13, CDR-SB, FAQ

Table 2: Standardization of the Progression Index (

\Delta

). Measures are grouped by the calculation required to ensure that a positive

\Delta

consistently reflects disease progression, indicating either clinical decline or structural atrophy.

Test	Metric	SIENA Vanilla	SIENA-SS	SIENA-SEG	SIENA-SS-SEG
$\Delta$ MMSE	$r$ [95% CI]	-0.226 [-0.284, -0.167]	-0.497 [-0.542, -0.449]	-0.252 [-0.309, -0.193]	-0.384 [-0.436, -0.331]
( $N=1006$ )	Steiger’s $Z$	Baseline	-8.84 ( $p<0.001$ )	-1.14 ( $p=0.255$ )	-4.55 ( $p<0.001$ )
$\Delta$ CDR-SB	$r$ [95% CI]	-0.258 [-0.315, -0.199]	-0.608 [-0.646, -0.567]	-0.290 [-0.346, -0.232]	-0.453 [-0.501, -0.402]
( $N=985$ )	Steiger’s $Z$	Baseline	-11.98 ( $p<0.001$ )	-1.41 ( $p=0.158$ )	-5.70 ( $p<0.001$ )
$\Delta$ ADAS-13	$r$ [95% CI]	-0.254 [-0.311, -0.195]	-0.524 [-0.567, -0.477]	-0.271 [-0.328, -0.213]	-0.405 [-0.455, -0.352]
( $N=1006$ )	Steiger’s $Z$	Baseline	-8.97 ( $p<0.001$ )	-0.75 ( $p=0.453$ )	-4.39 ( $p<0.001$ )

Table 3: Comparative analysis of correlations between PBVC and longitudinal clinical changes in the ADNI cohort. The sample size (

N

) varies per clinical test and is denoted in the first column. The 95% confidence intervals (CIs) were calculated using Fisher’s

r

-to-

z

transformation. Steiger’s

Z

test was utilized to compare the correlation coefficient of each modified pipeline against the SIENA Vanilla baseline.

Test	Metric	SIENA Vanilla	SIENA-SS	SIENA-SEG	SIENA-SS-SEG
$\Delta$ MSEADLG	$r$ [95% CI]	-0.058 [-0.171, 0.057]	-0.126 [-0.238, -0.012]	-0.041 [-0.156, 0.074]	-0.116 [-0.228, -0.001]
( $N=291$ )	Steiger’s $Z$	Baseline	-0.64 ( $p=0.520$ )	0.93 ( $p=0.352$ )	-0.41 ( $p=0.681$ )
$\Delta$ MoCA	$r$ [95% CI]	0.021 [-0.163, 0.205]	-0.122 [-0.299, 0.064]	-0.066 [-0.247, 0.119]	-0.117 [-0.294, 0.069]
( $N=114$ )	Steiger’s $Z$	Baseline	-1.73 ( $p=0.253$ )	-0.84 ( $p=1.000$ )	-1.54 ( $p=0.368$ )
$\Delta$ GM	$r$ [95% CI]	-0.020 [-0.199, 0.160]	-0.219 [-0.384, -0.042]	-0.151 [-0.321, 0.029]	-0.216 [-0.381, -0.038]
( $N=120$ )	Steiger’s $Z$	Baseline	-2.19 ( $p=0.087$ )	-1.21 ( $p=0.677$ )	-2.05 ( $p=0.120$ )

Table 4: Comparative analysis of correlations between PBVC and longitudinal clinical changes in the PPMI cohort. The sample size (

N

) varies per clinical test and is denoted in the first column. The 95% confidence intervals (CIs) were calculated using Fisher’s

r

-to-

z

transformation. Steiger’s

Z

test was utilized to compare the correlation coefficient of each modified pipeline against the SIENA Vanilla baseline, with

p

-values Bonferroni-corrected for three comparisons per test.

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.

Pipeline	Device	Mean (s)	SD (s)
SIENA Vanilla	CPU	1855.131	304.985
SIENA Vanilla	GPU	–	–
SIENA-SS	CPU	1827.353	194.975
SIENA-SS	GPU	1790.759	413.627
SIENA-SEG	CPU	1851.025	293.543
SIENA-SEG	GPU	1001.921	153.251

Table 5: End-to-end execution time (in seconds) for each pipeline. The results compare CPU and GPU processing performance between the four examined pipelines.

Neuroimaging

Neuroimaging uses brain measurements and computational methods to map anatomy, activity, connectivity, and disease.

See all 235 papers

Enhancing brain age estimation with structural MRI and synthesized cerebral blood volume maps
Dec 1, 2024Jordan Jomsky, Zongyu Li, Kay C. Igwe +8Alzheimer's Disease Brain Age Prediction
Department of Biomedical Engineering, Columbia University, New York, NY 10027-7041, USA · Zuckerman Institute, Columbia University, New York, NY 10027-7041, USA · Taub Institute for Research on Alzheimer’s Disease and the Aging Brain, Vagelos College of Physicians and Surgeons, Columbia University, New York, NY, USA +5
Brain-PACE: A Deep Siamese MRI Framework for Modelling Longitudinal Brain Acceleration
Sep 10, 2026Samuel Maddox, Jacob Newman, Saber Sami +4Alzheimer's Disease Brain Age Prediction
School of Computing Sciences, University of East Anglia · Norwich Medical School, University of East Anglia
Longitudinal Lesion Inpainting in Brain MRI via 3D Region Aware Diffusion
Mar 5, 2026Zahra Karimaghaloo, Dumitru Fetco, Haz-Edine Assemlal +2Diffusion Models for Image Restoration Diffusion Model Guidance
Clario part of Thermofisher Scientific, Montreal, Canada · Concordia University, Montreal, Canada

Dec 1, 2024J/K move · Enter open · S save

Enhancing brain age estimation with structural MRI and synthesized cerebral blood volume maps

Jordan Jomsky, Zongyu Li, Kay C. Igwe, Yiren Zhang+7

Department of Biomedical Engineering, Columbia University, New York, NY 10027-7041, USA · Zuckerman Institute, Columbia University, New York, NY 10027-7041, USA · Taub Institute for Research on Alzheimer’s Disease and the Aging Brain, Vagelos College of Physicians and Surgeons, Columbia University, New York, NY, USA+5

Alzheimer's Disease Brain Age Prediction Brain MRI Multimodal Learning

Open paper AskPDF arXiv

BrainAGE is a promising imaging-derived biomarker of neurobiological ageing and disease risk, yet current approaches rely predominantly on T1-weighted structural MRI, overlooking functional vascular changes that may precede tissue damage and cognitive decline. DeepCBV maps, synthesized from non-contrast MRI, offer a scalable alternative to contrast-enhanced perfusion imaging by capturing vascular information relevant to early neurodegeneration. We developed a multimodal BrainAGE framework that combines predictions from two separate three-dimensional convolutional neural networks: one trained only on structural MRI scans and another trained only on DeepCBV maps. Each model was trained and validated on 2851 scans from 13 open-source datasets and was evaluated for concordance with MCI and AD. The combined model achieved the most accurate brain age gap for CN controls, with a mean absolute error of 3.95 years, outperforming models trained on MRI or DeepCBV alone. Saliency maps revealed complementary modality contributions: MRI emphasized white matter and cortical atrophy, while DeepCBV highlighted vascular-rich and periventricular regions implicated in hypoperfusion and early cerebrovascular dysfunction, consistent with known patterns of normal ageing. Next, we observed that BrainAGE increased stepwise across diagnostic strata (CN < MCI < AD) and correlated with cognitive impairment. DeepCBV-based BrainAGE showed a particularly strong separation between stable versus progressive MCI, suggesting sensitivity to prodromal vascular changes that precede overt atrophy. Integrating structural MRI with deep learning-derived vascular measures substantially enhances BrainAGE estimation and improves sensitivity to MCI and AD progression, supporting its potential role in risk stratification, early detection and monitoring of therapeutic response.

Abstract

Neuroimaging

Explore similar work

Enhancing brain age estimation with structural MRI and synthesized cerebral blood volume maps

Brain-PACE: A Deep Siamese MRI Framework for Modelling Longitudinal Brain Acceleration

Longitudinal Lesion Inpainting in Brain MRI via 3D Region Aware Diffusion