GLIDE-Reg: Global-to-Local Deformable Registration Using Co-Optimized Foundation and Handcrafted Features

Yunzheng Zhu; Aichi Chien; Kimaya kulkarni; Luoting Zhuang; Stephen Park; Ricky Savjani; Daniel Low; William Hsu

eess.IVFeb 27, 2026

GLIDE-Reg: Global-to-Local Deformable Registration Using Co-Optimized Foundation and Handcrafted Features

Authors: Yunzheng Zhu, Aichi Chien, Kimaya kulkarni, Luoting Zhuang, Stephen Park, Ricky Savjani, Daniel Low, William Hsu

Organizations: Medical & Imaging Informatics, Department of Radiological Sciences, David Geffen School of Medicine at UCLA, USA · Magnetic Resonance Research Labs, Department of Radiological Sciences, David Geffen School of Medicine at UCLA, USA · Department of Radiation Oncology, David Geffen School of Medicine at UCLA, Los Angeles, USA

Abstract

Deformable registration is crucial in medical imaging. Several existing applications include lesion tracking, probabilistic atlas generation, and treatment response evaluation. However, current methods often lack robustness and generalizability across two key factors: spatial resolution and differences in anatomical coverage. We jointly optimize a registration field and a learnable dimensionality reduction module so that compressed VFM embeddings remain registration-relevant, and fuse these global semantic cues with MIND local descriptors. GLIDE-Reg achieves average dice similarity coefficients (DSC) across 6 anatomical structures of 0.859, 0.862, and 0.901 in two public cohorts (Lung250M and NLST) and one institution cohort (UCLA5DCT), and outperforms the state-of-the-art DEEDS (0.834, 0.858, 0.900) with relative improvements of 3.0%, 0.5%, and 0.1%. For target registration errors, GLIDE-Reg achieves 1.58 mm on Lung250M landmarks (compared to 1.25 mm on corrField and 1.91 mm on DEEDS) and 1.11 mm on NLST nodule centers (compared to 1.11 mm on DEEDS). The substantiated performance on the nodule centers also demonstrates its robustness across challenging downstream tasks, such as nodule tracking, which is an essential prior step for early-stage lung cancer diagnosis.

Figures & tables

Dataset	NLST	Lung250M	UCLA5DCT
Study Disease	Indeterminate Pulmonary Nodule	COPD	$N/A$
Study Age	$55$ to $74$	$48$ to $80$	$N/A$
Collection Date	$1999$ to $2001$	After $2007$ (Phase I or II)	$N/A$
Task (Time Diff.)	Longitudinal ( $1$ to $2$ years)	Insp. Exp. ( $\sim$ $30$ mins)	Free-Breathing ( $<$ $5$ mins)
Data Type	Low-Dose CT	Breath-Hold CT	Free-Breath CT
Resolution	$512\times 512\times[67,196]$	$512\times 512\times[112,135]$	$500\times 500\times[342,411]$

Table 1: Specifications of Datasets used for this study.

Figure 1: GLIDE-Reg Pipeline. A pair of 3D images $I_{fix}$ and $I_{mov}$ are passed to a global feature extractor and a local feature extractor (MIND) for extracting the optimal global feature representations $GF_{fix}^{*}$ , $GF_{mov}^{*}$ and local feature representations $LF_{fix}$ , $LF_{mov}$ , respectively. Then the features are registered with 1) feature-independent coupled convex optimization and 2) fused Adam optimization in Local-Global Registration and output the $I_{warp}$ . DR stands for dimensionality reduction.

	Lung250M				NLST				[blinded]5DCT
Method	TRE $\downarrow$	DSC $\uparrow$	$\mathbf{\%\lvert J\rvert_{<0}}$ $\downarrow$	Runtime	TRE $\downarrow$	DSC $\uparrow$	$\mathbf{\%\lvert J\rvert_{<0}}$ $\downarrow$	Runtime	DSC $\uparrow$	$\mathbf{\%\lvert J\rvert_{<0}}$ $\downarrow$
	Landmarks	Mean6			Nodule	Mean6			Mean6
No Reg.	$16.25\pm 3.84$ *	$0.517\pm 0.080$ *	$-$	$-$	$18.93\pm 10.27$ *	$0.472\pm 0.100$ *	$-$	$-$	$0.767\pm 0.049$ *	$-$
Affine	$30.36\pm 12.12$ *	$0.488\pm 0.106$ *	$-$	$<5$ sec	$30.36\pm 23.88$ *	$0.726\pm 0.045$ *	$-$	$<10$ sec	$0.751\pm 0.050$ *	$-$
SyN [ 1 ]	$13.95\pm 3.63$ *	$0.634\pm 0.085$ *	$0.0$	$<1$ min	$17.65\pm 10.24$ *	$0.534\pm 0.125$ *	$<0.001$	$<2$ min	$0.838\pm 0.036$ *	$<0.001$
NiftyReg [ 21 ]	$14.15\pm 3.93$ *	$0.563\pm 0.086$ *	$0.565\pm 0.006$	$<2.5$ min	$26.93\pm 14.25$ *	$0.396\pm 0.088$ *	$1.846\pm 1.992$	$<4$ min	$0.807\pm 0.032$ *	$0.010\pm 0.013$

Table 2: Comparison of registration performance on three datasets evaluated by DSC(

\uparrow

) and TRE(

\downarrow

) (in

mm

), and nonpositive Jacobian determinant

\%\lvert J\rvert_{<0}

(

\downarrow

). No Reg. represents the baseline performance for the image pairs. Methods with top two performance for TRE and DSC in each column are bolded and underlined , respectively. Statistical testings are computed per pair on the TRE and average DSC using Wilcoxon signed-rank test and adjusted by Benjamini-Hochberg correction with false discovery rate of

0.05

. Methods that are outperformed by or outperform GLIDE-Reg with statistical significance (

p<0.05

) are marked with

*

and †, respectively.

Figure 2: Comparison of the best performing 9 algorithms (best four in the learning-based and all five feature-based instance-optimization algorithms) on the NLST dataset.

Figure 3: Comparison of the feature-based instance-optimization methods on one challenging case from Lung250M, where all learning-based methods failed.

Table 6

Explore similar work

Architectural and Regularization Components in Deep Learning Medical Image Registration: Systematic Ablation Study
Aug 23, 2026Nabira RashidMedical Image Registration Deformable Image Registration
Carnegie Mellon University · Harvard Medical School
Search-MIND: Training-Free Multi-Modal Medical Image Registration
Apr 10, 2026Boya Wang, Ruizhe Li, Chao Chen +1Medical Image Registration Cross-Modal Image Registration
Intelligent Modelling & Analysis Group (IMA), School of Computer Science, University of Nottingham, UK · Lab for Uncertainty in Data and Decision Making (LUCID), Nottingham Biomedical Research Centre (BRC), School of Medicine, University of Nottingham, UK
An Accessible Solution for Deformable Image Registration Compared with Learning-Based Approaches
Aug 3, 2026Onur Ali Zeybekoglu, David Tilly, Orcun GokselMedical Image Registration Deformable Image Registration
Department of Information Technology, Uppsala University, Uppsala, Sweden · Department of Medical Physics, Uppsala University Hospital, Uppsala, Sweden · Department of Immunology, Genetics and Pathology, Uppsala University, Uppsala, Sweden

Aug 23, 2026J/K move · Enter open · S save

Architectural and Regularization Components in Deep Learning Medical Image Registration: Systematic Ablation Study

Nabira Rashid

Carnegie Mellon University · Harvard Medical School

Medical Image Registration Deformable Image Registration

Open paper AskPDF arXiv

Deep learning registration methods routinely stack two kinds of enhancement on a base network: architectural additions such as affine pre-alignment stages, and training-objective additions such as regularization losses. Papers tend to adopt both at once, so it is unclear which is doing the work. I ran a controlled ablation to separate them. Using the OASIS brain MRI dataset (394 training subjects, 20 test subjects), I trained four variants of the same registration pipeline: a baseline 3D U-Net with basic similarity losses, the same U-Net with a full regularization suite, an affine-plus-deformable architecture with basic losses, and the affine architecture with the full suite. I evaluated registration accuracy (MSE, NCC, SSIM), deformation quality (Jacobian determinant preservation, displacement statistics, an anatomical plausibility score), and computational cost. Regularization alone accounted for most of the gain: a 21.3% relative gain on the MSE-improvement metric (1.78% to 2.16%, P<.001) and a 21.8% relative gain in NCC improvement, while cutting maximum deformation from 53.1 to 0.51 units, a 99.0% reduction, at essentially no computational cost (-0.06% inference time). The combined model produced the largest accuracy gain, 25.8% (1.78% to 2.24%), and raised anatomical plausibility from 0.596 to 0.930, at a moderate +9.8% inference-time cost. Gradient correlation rose from 0.742 at baseline to 0.980 for the fully enhanced model. All enhanced variants reached sub-voxel accuracy under plausible deformation constraints. Regularization losses are the primary driver in this setting, delivering the accuracy gains and almost all of the deformation control for free at inference time, while the affine architecture adds a smaller complementary benefit at acceptable cost. The 99% reduction in unrealistic deformations addresses a known barrier to clinical deployment.

Abstract

Explore similar work

Architectural and Regularization Components in Deep Learning Medical Image Registration: Systematic Ablation Study

Search-MIND: Training-Free Multi-Modal Medical Image Registration

An Accessible Solution for Deformable Image Registration Compared with Learning-Based Approaches