FERN: A Fetal Echocardiography Registration Network for 2D-to-3D Alignment

Paula Ramirez Gilliland, David F A Lloyd, Jacqueline Matthew, Reza Razavi, Milou PM Van Poppel, Andrew P. King, Maria Deprez

Keywords: Fetal echocardiography, 3D Fetal ultrasound, STIC, Congenital Heart Disease, 3D plane localisation, Fetal cardiology, Fetal cardiac imaging

TL;DR: This study uses a Slice-to-Volume Transformer to map 2D standard fetal echocardiography views into a 3D space for antenatal Congenital Heart Disease evaluation.

Abstract: 2D Freehand echocardiography remains the primary imaging modality for routine fetal cardiac care, essential in the antenatal detection of Congenital Heart Disease (CHD). However, there is a lack of spatial context which requires 3D imaging. Current 3D methods, such as Spatio-Temporal Image Correlation (STIC), face limitations in success rate, image quality, and ease of use, and come at the cost of lower spatial and temporal resolution compared to 2D acquisitions. This work studies the feasibility of aligning real high spatial and temporal resolution 2D fetal echocardiography into a reference 3D space defined by lower resolution 3D STIC. FERN, a $\textbf{F}$etal $\textbf{E}$chocardiography $\textbf{R}$egistration $\textbf{N}$etwork, employs transformers for standard fetal echocardiography view alignment. The network is trained on simulated 2D slices derived from 3D volumes at end-diastole, and validated on real 2D acquisitions from fetuses with Coarctation of the Aorta and Right Aortic Arch diagnoses, achieving a mean Euclidean distance of 2.98 $\pm$ 1.27 mm on cardiac region-of-interest points between predicted and manually selected planes. Compared to manually aligned planes, improved image similarity to an average atlas is achieved, confirmed by blinded best plane selection. This work demonstrates that high spatial and temporal resolution 2D fetal echocardiography can be integrated into a 3D context provided by lower-resolution 3D acquisitions or fetal cardiac atlases, potentially resulting in a new 3D visualization tool for enhanced CHD diagnosis.

Primary Subject Area: Image Registration

Secondary Subject Area: Application: Cardiology

Paper Type: Validation or Application

Registration Requirement: Yes

Reproducibility: https://github.com/PaulaRamirezGilliland/FERN

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., All authors and co-authors are correctly listed with proper spelling and avoid Unicode characters., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.

Submission Number: 97