Back to the list
Congress: ECR25
Poster Number: C-12662
Type: Poster: EPOS Radiologist (scientific)
Authorblock: H. Velling, M. Brix, M. Bode, E. Lammentausta; Oulu/FI
Disclosures:
Heidi Velling: Nothing to disclose
Mikael Brix: Nothing to disclose
Michaela Bode: Nothing to disclose
Eveliina Lammentausta: Nothing to disclose
Keywords: Artificial Intelligence, MR physics, MR, Physics, Technology assessment, Artifacts, Quality assurance
Purpose Deep learning reconstruction (DLR) algorithms have been brought into use in magnetic resonance imaging (MRI). In DLR algorithms, the acquisition is sped up by collecting fewer k-space samples or signal averages and compensating inferior data quality with artificial intelligence-based reconstruction. DLR methods have been demonstrated to enable significant reductions in acquisition times and improve signal-to-noise ratio (SNR) [1-5].Since DLR generates the diagnostic images, special attention is needed to identify the limits of optimal operation and to ensure diagnostic image quality....
Read more Methods and materials When MRI sequences are accelerated, the parallel imaging acceleration (PAT) factor, phase oversampling (PO) and the number of averages (AVE) are often modified. In this study, the impact of changing these parameters on quantitative and subjective image quality was assessed. A cylindrical Magphan MRI phantom (SMR 170, Phantom Laboratory, Greenwich, NY, USA) and a volunteer were scanned with a 3T scanner using both conventional and DRB sequences. A spine coil and a body matrix coil were used. A T1-weighted turbo...
Read more Results Signal intensity - PhantomWithout DRB enabled, no parameter (PAT, PO, AVE) had a significant effect on signal intensity at ROIs near the edges of the phantom. However, increasing PO resulted in a minimal increasement of signal at ROI in the middle of the phantom.With DRB enabled, increasing the PAT factor resulted in a significant increasement in signal intensity. Conversely, increasing the number of averages resulted in a signal decrease, whenever the number of averages was greater or equal to the...
Read more Conclusion When applying DRB, the number of averages did not improve image quality. In some cases, it could have the opposite effect. To avoid the signal reduction phenomenon, the number of averages should be smaller than the PAT factor. Instead, increasing phase oversampling to the range of 150%-200% will resolve artifacts. It is also notable that DRB may have an impact on the general texture of the image.
Read more References Almansour H, Herrmann J, Gassenmaier S, et al. Deep Learning Reconstruction for Accelerated Spine MRI: Prospective Analysis of Interchangeability. Radiology. 2023;306(3):e212922. doi:10.1148/radiol.212922 Johnson PM, Lin DJ, Zbontar J, et al. Deep Learning Reconstruction Enables Prospectively Accelerated Clinical Knee MRI. Radiology. 2023;307(2):e220425. doi:10.1148/radiol.220425 Kiryu S, Akai H, Yasaka K, et al. Clinical Impact of Deep Learning Reconstruction in MRI. Radiographics. 2023;43(6):e220133. doi:10.1148/rg.220133 Rastogi A, Brugnara G, Foltyn-Dumitru M, et al. Deep-learning-based reconstruction of undersampled MRI to reduce scan times: a multicentre, retrospective, cohort study. Lancet Oncol. 2024;25(3):400-410. doi:10.1016/S1470-2045(23)00641-1 Yoo...
Read more
GALLERY