Tissue Characterization

1343408 - Reducing Bias and Noise in Myocardial T2* Mapping for Intramyocardial Hemorrhage: An Investigation Using Phantoms and Explanted Heart Tissue

Wednesday, January 25, 2023

Location: E-POSTER

Presenting Author(s)

GA

Gregory J. Anthony, PhD

Assistant Professor
Indiana University, Indiana, United States

Primary Author(s)

GA

Gregory J. Anthony, PhD

Assistant Professor
Indiana University, Indiana, United States

Co-Author(s)

Xinheng Zhang, MSc

Ph.D. Candidate
Indiana University School of Medicine
Indianapolis, Indiana, United States
XG

Xingmin Guan, PhD

Postdoctoral Fellow
Indiana University School of Medicine
San Diego, California, United States
HY

Hsin-Jung Yang, PhD

Assistant Professor
Cedars-Sinai Medical Center
Los Angeles, California, United States
Rohan Dharmakumar, PhD

Professor of Medicine, Radiology & Imaging Sciences, Anatomy, Cell Biology & Physiology
Krannert Cardiovascular Research Center, Indiana University School of Medicine
Indianapolis, Indiana, United States

Background: To date, optimal fitting strategies to minimize bias and noise in T₂* measurements of intramyocardial hemorrhage have not been investigated. This is particularly important for quantification of iron in hemorrhagic myocardial infarction, as T₂* can be confounded by low signal-to-noise ratios and off-resonance effects near tissue interfaces in the heart. We studied several fitting methods with the goal of reducing bias and uncertainty in T₂* mapping of phantoms and ex-vivo heart tissue under low-SNR and off-resonance conditions.

Methods: Phantoms composed of 2% agar and ferumoxytol (0–60 μg Fe/mL) and a whole canine ex-vivo heart with hemorrhagic infarction fixed in 10% formalin were imaged in a 3T MR scanner (Siemens MAGNETOM Prisma-Fit) using an 18-channel flex body coil. Multi-gradient-echo (mGRE) sequences were prescribed (TE = 2.4 – 37.4 ms, 8 echoes) along the short axis of the phantom tubes and the heart. Image SNR was modulated via the number of signal averages (NEX = 1–16). Off-resonance conditions were simulated by incrementally applying linear gradients (0–900 μT/m) along one dimension of the image volume during acquisition. T₂* maps were reconstructed off-line by fitting the mGRE data to 2- and 3-parameter exponential and logarithmic models. 2-parameter exponential fitting using 3–7 echoes (truncated fitting) was also performed and compared with 8-echo fits. For each SNR and gradient level, the bias (change in mean T₂* value from the 16-NEX or 0-μT/m acquisition) and uncertainty (standard deviation in T₂*) were computed within each phantom and within infarcted and normal myocardium in the ex-vivo canine heart.

Results: Compared with 2-parameter exponential fitting, 3-parameter exponential fits slightly underestimated T₂* but remained more stable over the SNR range tested. The 3-parameter exponential model exhibited the smallest bias and uncertainty for SNR levels of ~5 or less (Figure 1). Truncation of data points with SNR < 3 minimized T₂* uncertainty while maintaining a bias < 1.5 ms. All fitting models produced negative biases and decreasing noise with increasing applied linear gradients. The 2-parameter exponential model produced the best combination of precision and accuracy across the various off-resonance conditions. For intermediate gradient strengths (100–300 μT/m), truncated fitting mitigated the negative bias induced by the gradient at some cost of increased noise (Figure 2). In the ex-vivo heart, truncated fits were beneficial for reducing off-resonance bias with some noise penalty (Figure 3).

Conclusion: For SNR < 5, using a 3-parameter model or truncated fitting reduces both bias and uncertainty in T₂* measurements. Accurate T₂* values can also be recovered with modest noise penalties in the presence of off-resonant conditions by using truncated fitting with 4 to 6 echoes. The proposed strategy can improve T₂* estimates in the presence of the high susceptibility differences and low-SNR conditions characteristic of hemorrhagic myocardial infarction.