Insights on Scan‐Specific Deep‐Learning Strategies for Brain MRI Parallel Imaging Reconstruction
NMR in Biomedicine · 2025
DOI: 10.1002/nbm.70079 ↗Kurz gefasst
Fachartikel in der Zeitschrift NMR in Biomedicine, 2025. Verfasst von Swetali Nimje und 3 weiteren Autor:innen. Beteiligte Einrichtung: Aix Marseille Univ, CNRS, CRMBM, Institut Marseille Imaging, Marseille, France. Hauptthemen laut MEDLINE: Deep Learning, Magnetic Resonance Imaging, Brain, Image Processing, Computer-Assisted. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: French National Research Agency (ANR), AMIDEX, France Life Imaging.
Abstract (Originalsprache)
Scan‐specific deep learning strategies have been proposed for parallel imaging reconstruction in which auto‐calibrated signals (ACS) are used for training. Here, we introduce methods to objectively optimize architecture and training details. In addition, we introduce a new metric to better characterize the quality of the reconstructed images. Various brain MRI situations are considered. The evaluated models encompass single‐layer and three‐layer residual CNN architectures with real and complex convolutions. Hyperparameters such as the level of linearity in leaky activation functions, loss function, kernel sizes and depths are optimized using grid‐search with K‐fold cross validation. The performances regarding ACS reference size and mode are also evaluated. An innovative COrrelation‐Based Residual Artifact Index (COBRAI) quantifying the level of structured residual artifacts is proposed. Qualitative and quantitative comparisons are performed both on the FastMRI and in‐house multi‐contrasts 2D data. The proposed objective grid‐search strategy based on ACS successfully provided optimized hyperparameters, retrospectively validated by enhanced image quality metrics. Notably, it is shown that nonlinearities produce structured residual artifacts, and that, among the models tested, a three‐layer residual linear CNN with complex implementation and a reduced number of parameters is more robust, particularly providing less structured artifacts with less training data, leading to larger acceleration rates. Deep‐learning MRI parallel image reconstruction in the scan‐specific approach can be optimized using grid‐search with K‐fold cross validation. It was successfully applied in various 2D brain MRI situations. The quantification of structured residual artifacts with COBRAI is a useful complementary characterization to state‐of‐the‐art metrics, and it can be used to drive model selection.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Aix Marseille Univ, CNRS, CRMBM, Institut Marseille Imaging, Marseille, France.
- Laboratoire d'Informatique et Systèmes UMR 7020, Aix-Marseille Univ., CNRS, Ecole Centrale de Marseille, Marseille, France.
- APHM, Hôpital Universitaire Timone, CEMEREM, Marseille, France.
Förderung
- French National Research Agency (ANR)
- AMIDEX
- France Life Imaging
Zitieren
APA
Nimje, S., Artières, T., Guye, M., & de Rochefort, L. (2025). Insights on Scan‐Specific Deep‐Learning Strategies for Brain MRI Parallel Imaging Reconstruction. NMR in Biomedicine. https://doi.org/10.1002/nbm.70079
Vancouver
Nimje S, Artières T, Guye M, de Rochefort L. Insights on Scan‐Specific Deep‐Learning Strategies for Brain MRI Parallel Imaging Reconstruction. NMR in Biomedicine. 2025. doi:10.1002/nbm.70079
BibTeX
@article{Nimje2025,
author = {Nimje, Swetali and Artières, Thierry and Guye, Maxime and de Rochefort, Ludovic},
title = {Insights on Scan‐Specific Deep‐Learning Strategies for Brain MRI Parallel Imaging Reconstruction},
journal = {NMR in Biomedicine},
year = {2025},
doi = {10.1002/nbm.70079}
}
EUVIMED