Direct deformation estimation with recurrent inference machines for longitudinal MRI

Publication Type:

Conference Abstract

Authors:

Jonas Maes; Jana Osstyn; Matthan Caan; Arnold Jan den Dekker; Jan Sijbers

Source:

2026 ISMRM-ISMRT Annual Meeting and Exhibition, International Society for Magnetic Resonance in Medicine (2026)

Abstract:

Recently, Delta-MRI was proposed to estimate longitudinal anatomical changes from accelerated follow-up MRI scans. Delta-MRI, however, suffers from long computation times and requires user-defined regularisation. To develop a time-efficient Delta-MRI framework without user-defined regularisation. Delta-MRI is augmented with recurrent inference machines (RIMs) that incorporate a physics-based forward model and learn to solve the inverse problem of estimating anatomical changes directly from a reference image and undersampled follow-up scans, while jointly learning an implicit prior. RIM-augmented Delta-MRI outperforms conventional Delta-MRI in terms of follow-up image quality (PSNR gain: 2 dB), reconstruction speed (100x), and user-friendliness (no user-defined regularisation).

Research area: 
Magnetic Resonance Imaging