A Differentiable and Parallel Registration Framework for 3D Reconstruction From Serial Metal Mapping Images
IET Image Processing · 2026-01 · 20 (1)
DOI: 10.1049/ipr2.70452 ↗Kurz gefasst
Fachartikel in der Zeitschrift IET Image Processing, online erschienen am 23.08.2026. Verfasst von Pengyuan Qi und 4 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
In this paper, a differentiable and parallel registration framework that integrates both feature‐based and intensity‐based registration for 3D reconstruction from serial MMIs is proposed. To make full use of both spatial and intensity information, feature points and mapping images are utilised to perform feature‐based and intensity‐based registration. Both contribute to the final reconstruction loss. To enable efficient gradient‐based optimisation, the transformed discrete feature points are modelled using a Gaussian mixture. This allows for the development of a differentiable loss. Additionally, the resampling of mapping images during the intensity‐based registration process is performed using a differentiable bilinear sampling module. To achieve simultaneous updating of all transformation parameters, a novel non‐source‐to‐target registration concept is proposed and two symmetric losses are designed for feature‐based and intensity‐based registration accordingly. The proposed registration method is found to be effective in reconstructing a smooth 3D entity. For registered slices, the mean intersection over union is 90.98% and the mean Dice score is 95.24%. Compared to other well‐performing methods, our method demonstrates top‐level accuracy for 3D reconstruction with better time efficiency. The reconstruction results show that our proposed framework can achieve efficient registration of MMIs and precise reconstruction of 3D models, indicating that the method presents a promising alternative to time‐consuming sequential registration methods.
Zitieren
APA
Qi, P., Huang, C., Yuan, D., Zhang, J., & Wang, J. (2026). A Differentiable and Parallel Registration Framework for 3D Reconstruction From Serial Metal Mapping Images. IET Image Processing, 20(1). https://doi.org/10.1049/ipr2.70452
Vancouver
Qi P, Huang C, Yuan D, Zhang J, Wang J. A Differentiable and Parallel Registration Framework for 3D Reconstruction From Serial Metal Mapping Images. IET Image Processing. 2026;20(1). doi:10.1049/ipr2.70452
BibTeX
@article{Qi2026,
author = {Qi, Pengyuan and Huang, Changheng and Yuan, Dalong and Zhang, Junyu and Wang, Jike},
title = {A Differentiable and Parallel Registration Framework for 3D Reconstruction From Serial Metal Mapping Images},
journal = {IET Image Processing},
year = {2026},
volume = {20},
number = {1},
doi = {10.1049/ipr2.70452}
}
EUVIMED