# Factored Levenberg-Marquardt for Diffeomorphic Image Registration: An efficient optimizer for FireANTs

Source: [Factored Levenberg-Marquardt for Diffeomorphic Image Registration: An efficient optimizer for FireANTs](<https://www.sota2.com/research/paper/69c0c9050636e5bf4afc3ba4>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-24T05:00:03.763Z

Authors: Rohit Jena, Pratik Chaudhari, James C. Gee

Published: 2026-03-19T18:04:42.000Z

## Abstract

FireANTs introduced a novel Eulerian descent method for plug-and-play behavior with arbitrary optimizers adapted for diffeomorphic image registration as a test-time optimization problem, with a GPU-accelerated implementation. FireANTs uses Adam as its default optimizer for fast and more robust optimization. However, Adam requires storing state variables (i.e. momentum and squared-momentum estimates), each of which can consume significant memory, prohibiting its use for significantly large images. In this work, we propose a modified Levenberg-Marquardt (LM) optimizer that requires only a single scalar damping parameter as optimizer state, that is adaptively tuned using a trust region approach. The resulting optimizer reduces memory by up to 24.6% for large volumes, and retaining performance across all four datasets. A single hyperparameter configuration tuned on brain MRI transfers without modification to lung CT and cross-modal abdominal registration, matching or outperforming Adam on three of four benchmarks. We also perform ablations on the effectiveness of using Metropolis-Hastings style rejection step to prevent updates that worsen the loss function.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Image Registration on LUMIR 38 pairs](<https://www.sota2.com/research/sota/image-registration-on-lumir-38-pairs.md>): Image Registration; LUMIR 38 pairs; Mean Dice: 87.15.

- [Image Registration on OASIS 19 pairs](<https://www.sota2.com/research/sota/image-registration-on-oasis-19-pairs.md>): Image Registration; OASIS 19 pairs; Mean Dice: 82.93.

- [Image Registration on Abdomen-L2R (48 pairs)](<https://www.sota2.com/research/sota/image-registration-on-abdomen-l2r-48-pairs.md>): Image Registration; Abdomen-L2R (48 pairs); Mean Dice Coefficient: 76.15.

- [Image Registration on NLST 10 pairs 2011](<https://www.sota2.com/research/sota/image-registration-on-nlst-10-pairs-2011.md>): Image Registration; NLST 10 pairs 2011; Mean TRE (mm): 1.68.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2603.19371>)

[PDF](<https://arxiv.org/pdf/2603.19371.pdf>)
