Deep learning-assisted model-based off-resonance correction for non-Cartesian susceptibility weighted imaging

Guillaume Daval-Frérot (1, 2, 3) , Aurélien Massire (1) , Boris Mailhé (4) , Mariappan Nadar (4) , Blanche Bapst (5, 6) , Alain Luciani (5, 7, 8) , Alexandre Vignaud (2, 9) , Philippe Ciuciu (2, 3)

1 Siemens Healthineers [Saint-Denis]
2 NEUROSPIN - Service NEUROSPIN
3 MIND - Modèles et inférence pour les données de Neuroimagerie
4 Siemens Healthineers, Digital Services, Digital Technology and Innovation
5 Hôpital Henri Mondor
6 ENT - Excitabilité nerveuse et thérapeutique
7 UPEC Médecine - Université Paris-Est Créteil Val-de-Marne - Faculté de médecine
8 U955 Inserm - UPEC - IMRB - "NeuroPsychologie Interventionnelle" [Créteil]
9 BAOBAB - Building large instruments for neuroimaging: from population imaging to ultra-high magnetic fields

Purpose: Patient-induced inhomogeneities in the static magnetic field cause distortions and blurring (off-resonance artifacts) during acquisitions with long readouts such as in susceptibility-weighted imaging (SWI). Conventional versatile correction methods based on extended Fourier models are too slow for clinical practice in computationally demanding cases such as 3D high-resolution non-Cartesian multi-coil acquisitions. Theory: Most reconstruction methods can be accelerated when performing off-resonance correction, by reducing the number of iterations, com- pressed coils and correction components. Recent state-of-the-art deep learning architectures could help but are generally not adapted to corrupted measurements as they rely on the standard Fourier operator in the data consistency term. The combination of correction models and neural networks is therefore necessary to reduce reconstruction times. Methods: Hybrid pipelines using UNets were trained stack-by-stack over 99 SWI 3D SPARKLING 20-fold accelerated acquisitions at 0.6mm isotropic resolution using different off-resonance correction methods. Target images were obtained using slow model-based corrections based on self-estimated ∆B 0 field maps. The proposed strategies, tested over 11 volumes, are compared to model-only and network-only pipelines. Results: The proposed hybrid pipelines achieved scores competing with 2-3 times slower baseline methods, and neural networks were observed to contribute both as pre-conditioner and through inter-iteration memory by allowing more degrees of freedom over the model design. Conclusion: A combination of model-based and network-based off-resonance correction was proposed to significantly accelerate conventional methods. Different promising synergies were observed between acceleration factors (iterations, coils, correction) and model/network that could be expanded in the future.

Mots clés: non-Cartesian imaging; deep learning; unrolled neural network; 3D SPARKLING; off-resonance correction

Domaines: Imagerie médicale; Apprentissage [cs.LG]; Traitement du signal et de l'image [eess.SP]

Guillaume Daval-Frérot, Aurélien Massire, Boris Mailhé, Mariappan Nadar, Blanche Bapst, et al.. Deep learning-assisted model-based off-resonance correction for non-Cartesian susceptibility weighted imaging. 2022. ⟨hal-03935487⟩