Family of Deep Image Prior Networks for Accelerated 3D LGE-MRI Acquisition with Enhanced Reconstruction

Md Hasibul Husain Hisham, Shireen Elhabian, Ganesh Adluru, Andrew Arai, Eugene Kholmovski, Ravi Ranjan, Edward Dibella

Keywords: unsupervised network, mri reconstruction, deep image prior, unet, image reconstruction

TL;DR: Family of Deep Image Prior Networks for Accelerated 3D LGE-MRI Acquisition with Enhanced Reconstruction

Abstract:

Late Gadolinium Enhancement (LGE) MRI is essential for visualizing and treating left atrial fibrosis, but current protocols require lengthy acquisition times (7-20 minutes) and often produce suboptimal image quality. While recent advances in isotropic imaging have shown promise, scan times of 12-15 minutes still present clinical challenges. This study evaluates the efficacy of existing Deep Image Prior (DIP) frameworks for accelerated 3D LGE-MRI reconstruction. We comprehensively assess multiple DIP variants - vanilla DIP, reference-guided DIP, DIP with Total Variation, and self-guided DIP - on their ability to reconstruct high-quality isotropic (1.25mm$^3$) images from highly undersampled k-space data. Using data from 10 subjects, we demonstrate that self-guided DIP achieves superior reconstruction quality (PSNR: 32.8±1.2 dB, SSIM: 0.891±0.015 at 1/4th of acquisition time) compared to traditional compressed sensing and other DIP variants. Our evaluation shows that DIP-based reconstruction can maintain diagnostic quality with acquisition times reduced to 2-4 minutes, particularly in preserving thin left atrial wall details. These findings suggest that DIP-based methods could improve clinical workflow efficiency and patient comfort in high-resolution 3D LGE studies for atrial fibrillation patients.

Primary Subject Area: Image Acquisition and Reconstruction

Secondary Subject Area: Unsupervised Learning and Representation Learning

Paper Type: Validation or Application

Registration Requirement: Yes

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: 253