Expansion of Applicable Sequences of Precise IQ Engine (PIQE)

Kensuke Shinoda
Senior Engineer
MR Clinical Application Group, MRI Systems Development Department
CT-MR Division, Canon Medical Systems Corporation
Figure 1: PIQE reconstruction pipeline
In magnetic resonance imaging (MRI), there is an inherent trade-off between signal-to-noise ratio (SNR), acquisition time, and spatial resolution. Precise IQ Engine (PIQE) has been developed as a novel reconstruction technology to overcome the trade-off. Utilizing two dedicated neural networks, PIQE enables up-sampling the acquired MR images, resulting in sharper images while keeping the truncation artifacts (aka Gibbs’ ringing artifacts) suppressed.

When PIQE was first commercialized, it was solely available with 2D Fast Spin Echo (FSE2D) sequences. However now PIQE can be used with other sequences as well. The expansion of the applicable sequences of PIQE was achieved through its original design. PIQE consists of two separated neural networks for denoising and up-scaling as its key components. Although the first release of PIQE was focused on FSE2D, both of the two neural networks had been trained with the data sets which cover a wider range of contrast (PDw, T1w, T2w, FLAIR, T2*w, diffusion weighted, and so on), body parts (brain, spine, and MSK regions), and magnetic field strengths (3T and 1.5T). This training strategy has resulted in making PIQE very robust and reliable for various contrasts from many kinds of pulse sequences.

Some previous studies 1, 2, 3 demonstrate better SNR and sharpness with reduced Gibbs ringing artifacts versus standard reconstruction used in routine clinical practices. This article demonstrates further test results that PIQE enables high-resolution imaging or short acquisition time preserving structural detail and reducing Gibbs ringing artifact on typical brain scan protocols of T1w with 2D Spin Echo (SE2D), T2*w with 2D Field Echo (FE2D), and EPI-DWI.
Figure 2: Brain Axial T1w with 2D Spin Echo (SE2D) sequence
The standard protocol was acquired with 288 PE matrix in 2 minutes 44 seconds, while the fast protocol was acquired with 224 PE matrix in 2 minutes 10 seconds.
Figure 3: Brain Axial T2*w with 2D Field Echo (FE2D) sequence
The standard protocol was acquired with 320 PE matrix in 1 minute 56 seconds, while the fast protocol was acquired with 224 PE matrix in 1 minute 24 seconds. FineRecon, so called ZIP, was applied in cases of NONE and DSD by 2.
Figures 2 and 3 show brain T1w images with SE2D and T2*w images with FE2D sequences respectively. The fast protocol was acquired with less PE matrix size in a shorter scan time than the standard protocol. In both cases, NONE × 2 of the fast protocol makes Gibbs ringing artifacts more noticeable due to the reduced matrix size. Conventional filter DSD of the standard protocol (Standard-DSD × 2) and the fast protocol (Fast-DSD × 2) could mitigate Gibbs ringing artifacts at the expense of blurring. PIQE × 3 with standard protocol provides the highest sharpness and the highest SNR of the other methods. Also, Fast-PIQE × 3 reduces Gibbs ringing artifacts while providing sharpness and noise reduction compared to Fast-DSD × 2 and Standard-DSD × 2 as well, which requires a longer scan time.
Figure 4 shows Brain Axial b0 images and isoDWI of Spin Echo EPI-DWI sequences. The standard protocol was acquired with TR/TE = 5700/75 ms, echo train spacing (ETS) = 0.9 ms, Bandwidth = 1302 Hz, b-value = 0 and 1000 in 3-axis, acquisition matrix =160, and the number of acquisitions (NAQ) = 3, while the fast protocol was scanned with TR/TE = 4603/75 ms, ETS = 0.7 ms, Bandwidth = 1953 Hz, b-value = 0 and 1000 in 3-axis, acquisition matrix = 128, and NAQ = 1. Therefore, the scan time is 92 s for the standard protocol, and 42 s for the fast protocol by reducing matrix size and NAQ.

In both standard and fast of NONE × 2, Gibbs ringing artifacts can be observed because single shot EPI-DWI is acquired with small matrix size to shorten the scan time and/or EPI readout duration. In LPF × 2, the Gibbs ringing artifacts are reduced, but image blurring occurs. Standard PIQE × 3 shows the highest sharpness and denoising effect, while suppressing Gibbs artifacts better than the others. Fast PIQE × 3 also shows better or equivalent image quality compared to Standard LPF × 2 in terms of sharpness, SNR and the ringing artifacts.

These results show that PIQE can successfully improve image quality in terms of sharpness, denoising, and artifact reduction for Spin Echo, Field Echo and EPI as well. Also, these results demonstrate that PIQE can be used for both achieving high-resolution reconstruction and short time acquisition with preserving structural details and reducing noise and Gibbs artifacts.
Figure 4: Brain Axial EPI-DWI of b0 and isoDWI

References
1. Kutsuna H, et al. High Resolution MR Reconstruction with Functionally Separate Neural Networks. Proc. ISMRM 2023 p.2292
2. Prevost V, et al. Deep learning-based pipeline to improve sharpness in knee imaging at both 1.5T and 3T: a clinical evaluation. Proc. ISMRM 2023 p.4923
3. Matsuo K, et al. Feasibility study of super-resolution deep learning-based reconstruction using k-space data in brain diffusion-weighted images. Neuroradiology 65, 1619–1629 (2023). doi:10.1007/s00234-023-03212-y

Disclaimer
Some features presented in this article may not be commercially available on all systems shown or may require the purchase of additional options. Due to local regulatory processes, some commercial features included in this publication may not be available in some countries. Please contact your local representative from Canon Medical Systems for details and the most current information.

Subscribe to Canon MRI Latest Information

Sign up to receive update. Be the first to know about our products and publications.

Contact Us