STAMM3D: Dataset and Benchmark for Non-Deformable 3D Image Registration in Multiple Modalities

Volumetric image registration is a fundamental computer vision problem with broad applications across fields such as medicine, biological and cellular imaging, geoscience, and cultural heritage preservation. The 3D image registration problem is commonly divided into monomodal and multimodal registration problems, depending on whether the images originate from the same or different imaging modalities. Monomodal 3D image registration is a well-studied problem with established state-of-the-art methods and a range of widely recognized benchmarks. In contrast, multimodal 3D image registration remains an open problem despite numerous developed registration methods. A key challenge is that existing open-access multimodal benchmark datasets are scarce and limited in modality diversity, which constrains registration generalization across modalities. As a result, most multimodal registration methods remain tuned to specific modalities and narrow domains of objects. This leads to the need to redevelop multimodal registration tools for each specific task, and no universal solution exists. To approach the task of universal multimodal image registration, in this paper we introduce STAMM3D (Similarity Transform Alignment of Multimodal 3D images)—a new benchmark for multimodal registration that includes an open-access dataset of 22 3D images of rigid medical, biological, and non-biomedical objects and provides an automated Target Registration Error-based evaluation framework. The publicly availablebenchmark offers a high-resolution dataset with multiple domains of objects and a wide range of different modalities deliberately chosen to test and compare universality and generalization properties of the registration methods. The focus of the dataset is on non-deformable image registration since it directly affects the quality of image fusion, whether used independently or within multi-stage deformable registration pipelines. Using STAMM3D, we benchmark several widespread, multipurpose, open-source non-deformable registration tools (NiftyReg, SimpleITK, SIFT3D, DeedsBCV), establishing a solid performance baseline. While existing tools provide a solid baseline, the comparison results suggest that additional work is needed to improve the robustness of non-deformable registration techniques.

Authors

Institutions

Publication Details

Journal
Journal of Imaging
Published
2026-11-18
DOI
https://doi.org/10.3390/jimaging12110581
Primary Topic
Medical Image Segmentation Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

STAMM3D: Dataset and Benchmark for Non-Deformable 3D Image Registration in Multiple Modalities

Vladimir V. Arlazarov, Mikhail V. Gulyaev, К. М. Подурец, E. S. Kovalenko et al.
1 links
Journal of Imaging
Medical Image Segmentation Techniques
article

STAMM3D: Dataset and Benchmark for Non-Deformable 3D Image Registration in Multiple Modalities

Vladimir V. Arlazarov, Mikhail V. Gulyaev, К. М. Подурец, E. S. Kovalenko, Marina V. Chukalina, Yu. A. Pirogov, Danil Dmitrievich Kazimirov, Nikolay V. Anisimov, Dmitry Nikolaev, Marat I. Gilmanov, Polevoy Dmitry, Oleg Bugai, Leonid Koliaskin, Dmitrii Lebedev, Polina Guryeva
article en

Abstract

Volumetric image registration is a fundamental computer vision problem with broad applications across fields such as medicine, biological and cellular imaging, geoscience, and cultural heritage preservation. The 3D image registration problem is commonly divided into monomodal and multimodal registration problems, depending on whether the images originate from the same or different imaging modalities. Monomodal 3D image registration is a well-studied problem with established state-of-the-art methods and a range of widely recognized benchmarks. In contrast, multimodal 3D image registration remains an open problem despite numerous developed registration methods. A key challenge is that existing open-access multimodal benchmark datasets are scarce and limited in modality diversity, which constrains registration generalization across modalities. As a result, most multimodal registration methods remain tuned to specific modalities and narrow domains of objects. This leads to the need to redevelop multimodal registration tools for each specific task, and no universal solution exists. To approach the task of universal multimodal image registration, in this paper we introduce STAMM3D (Similarity Transform Alignment of Multimodal 3D images)—a new benchmark for multimodal registration that includes an open-access dataset of 22 3D images of rigid medical, biological, and non-biomedical objects and provides an automated Target Registration Error-based evaluation framework. The publicly availablebenchmark offers a high-resolution dataset with multiple domains of objects and a wide range of different modalities deliberately chosen to test and compare universality and generalization properties of the registration methods. The focus of the dataset is on non-deformable image registration since it directly affects the quality of image fusion, whether used independently or within multi-stage deformable registration pipelines. Using STAMM3D, we benchmark several widespread, multipurpose, open-source non-deformable registration tools (NiftyReg, SimpleITK, SIFT3D, DeedsBCV), establishing a solid performance baseline. While existing tools provide a solid baseline, the comparison results suggest that additional work is needed to improve the robustness of non-deformable registration techniques.

Journal of ImagingVol. 12(11)
Russian Academy of Sciences (RU), Lomonosov Moscow State University (RU), Kurchatov Institute (RU), Institute for Information Transmission Problems (RU), Federal Research Center “Computer Science and Control” of the Russian Academy of Sciences (RU)
Medical Image Segmentation Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon