Deep learning reconstruction refines image assessment in SSFP 5(3)3 MOLLI – a pilot study with normal volunteers at 1.5 T

Presenting Author(s)

• KM

  Kanae Mukai, MD, FSCMR

  Medical Director, Non-invasive Cardiovascular Imaging
  SVH System
  Monterey, California, United States

Primary Author(s)

• KM

  Kanae Mukai, MD, FSCMR

  Medical Director, Non-invasive Cardiovascular Imaging
  SVH System
  Monterey, California, United States

Co-Author(s)

• 

  Carlos Rochitte, MD, PhD, FSCMR

  Associate Professor of Cardiology, Academic Coordinator of Cardiovascular Magnetic Resonance and Computed Tomography sector, and Medical Director of the Research Center at Heart Institute, InCor, University of São Paulo Medical School Cardiovascular Magnetic Resonance and Computed Tomography Director at HCOR Senior Medical Leader at Alta Excelência Diagnóstica, DASA, São Paulo
  InCor - Univ. of São Paulo / HCOR - Heart Hospital
  InCor - Univ. of São Paulo
  São Paulo, Sao Paulo, Brazil
• JL

  Joao A. C Lima, MD

  Professor
  The Johns Hopkins University
  Baltimore, Maryland, United States
• CL

  Chia-Ying Liu, PhD

  Scientist
  Canon Medical Systems Corporation
  B, Maryland, United States
• DB

  Dawn Berkeley, RT

  Senior Manager Medical Affairs - Clinical Development
  Canon Medical Systems USA, Inc, United States
• WA

  Wissam AlGhuraibawi, PhD

  Medical Affairs Manager – Clinical Scientist
  Canon Medical Systems USA, Inc, United States
• YK

  Yoshimori Kassai, MSc

  Chief Specialist
  Canon Medical Systems Corp.
  Otawara, Tochigi, Japan
• ST

  Sho Tanaka, MSc

  Deputy Manager
  Canon Medical Systems Corporation
  Otawara, Tochigi, Japan
• MG

  Mark Golden

  Research coordinator
  Canon Medical Systems, United States
• YY

  Yuichi Yamashita, BSc

  Principal Specialist
  Canon Medical Systems Corporation, Japan
• MK

  Mo Kadbi, PhD

  Leader, Medical Affairs - MRI
  Canon Medical Systems USA, Inc, United States

Background: Standardization with native T1 mapping remains a challenge due to variability in methodology as well as limitations in image quality which affect study repeatability and reproducibility during image acquisition and post-processing.

The study aimed to evaluate the performance of a SSFP 5(3)3 WIP sequence incorporating the vendor-provided deep learning reconstruction (Precision IQ Engine, PIQE, Canon Medical Systems) to reconstruct the original T1 images for the generation of T1 maps. PIQE utilized DLR to reduce noise and applied 3-fold zero-padding interpolation to increase the reconstruction matrix.

We hypothesized that PIQE can refine quantitative image assessment on a cohort of normal volunteers, improve inter-scan repeatability and inter-reader agreement of the T1 values.

Methods: A total of 11 normal healthy male volunteers (mean age 41) with no prior medical history were scanned on a 1.5T MRI system (Orian, Canon Medical Systems, Japan). Participants were scanned in a headfirst supine position and all acquisitions were ECG triggered. SSFP MOLLI 5(3)3 was acquired for 3 slices (base, mid, and apex) with TE/TR =1.6/3.2 ms using a 104 x 192 matrix with an FOV 350 x 350 mm2 and a 30 degree flip angle. Each volunteer was scanned twice during the same study. Visual assessment of the T1 maps (base, mid, and apex) were performed by 3 Level III advanced CMR imagers (visual signal to noise and overall image quality). Grading was based on 5-point Likert scale (≥3 is diagnostic, 4 is diagnostic with artifacts, and 5 is diagnostic without artifacts). SNR and CNR were calculated based on the signal from the LV myocardium and blood pool using the 8th image in the acquisition.

T1 maps were generated using Medis QMap version 8.1. Inter-scan repeatability (between 2 scans of each study) were compared using Wilcoxon signed-rank test based on the septal and global T1 values of the basal slice. Inter-reader reproducibility was assessed on the basal slice of the first scan using the Bland-Altman plots and correlation coefficients. P < 0.05 was used for statistical significance.

Results: Table 1 summarizes the inter-scan comparison. SNR was better with PIQE although it did not appear to alter the mean T1 values. There was a minor difference in the global T1 but no difference in the septal T1 between scans. SNR and CNR were both higher with PIQE. Overall image quality and visual SNR by Likert score were greater than or equal to 3 (diagnostic) for all studies for all imagers (Figure 2). T1 values from PIQE reconstructed images demonstrated better inter-reader correlations with narrower 95% CI (Figure 3).

Conclusion: PIQE improves the consistency of myocardial T1 mapping through noise reduction and a higher reconstruction matrix. Our septal and global native T1 values were comparable to published data at 1.5T ranging from 900-1000ms (Reference 3). As this is a pilot study, a larger dataset (including females) is needed to validate this further.

Figure 1. Summary of myocardial T1 assessment
Figure 2. Case example (left: without PIQE; right: with PIQE)
Figure 3. Inter-reader reproducibility of the global T1 values from the first scan of each volunteer using Bland-Altman plots (a) without PIQE, and (b) with PIQE. (c) T1 values from PIQE reconstruction demonstrated higher correlation (r = 0.97 with PIQE vs r = 0.89 without PIQE).