mlVIRNET: Multilevel Variational Image Registration Network

TL;DR

This paper introduces mlVIRNET, a multilevel deep learning framework for image registration that effectively handles large deformations by progressively refining alignment from coarse to fine scales, demonstrated on lung registration.

Contribution

The paper presents a novel multilevel deep learning approach for image registration, enabling better handling of large deformations compared to existing methods.

Findings

Significantly improved registration accuracy on lung data

Effective multilevel framework for large deformations

Enhanced coarse-to-fine alignment process

Abstract

We present a novel multilevel approach for deep learning based image registration. Recently published deep learning based registration methods have shown promising results for a wide range of tasks. However, these algorithms are still limited to relatively small deformations. Our method addresses this shortcoming by introducing a multilevel framework, which computes deformation fields on different scales, similar to conventional methods. Thereby, a coarse-level alignment is obtained first, which is subsequently improved on finer levels. We demonstrate our method on the complex task of inhale-to-exhale lung registration. We show that the use of a deep learning multilevel approach leads to significantly better registration results.