Myocardial T2* mapping: influence of noise on accuracy and precision

Christopher M. Sandino, Peter Kellman, Andrew E. Arai, Michael S. Hansen, Hui XueJournal of Cardiovascular Magnetic Resonance 2015 · DOI

Abstract

A method for applying an automatic truncation method to pixel-wise T2 mapping that reduces T2 overestimation due to noise bias was proposed. A formulation for estimating pixel-wise standard deviation (SD) maps for T2* that can serve as a quality map for interpreting images and for comparison of imaging protocols was also proposed and validated.

Introduction

T2 mapping is a cornerstone of MRI-based quantification of iron concentration in organs like the heart, liver, and brain. In myocardial tissue, iron concentration is inversely proportional to the T2 relaxation time; faster signal decay indicates higher iron levels. While traditional region-of-interest (ROI) based mapping provides stability by averaging signals, the clinical trend is shifting toward pixel-wise mapping to capture spatial heterogeneity.

However, pixel-wise mapping is inherently limited by lower Signal-to-Noise Ratios (SNR). In magnitude MRI, noise does not merely add variance; it introduces a "Rician bias" that prevents the signal from decaying to zero. This creates a "tail" in the decay curve, leading to systematic overestimation of T2* values, which can mask severe iron overload.

Noise-induced bias in the T2* decay curve. The observed data (blue dots) flattens at low SNR, leading to an overestimated T2* fit (black line) compared to the ground truth (red line).

Methodology

The researchers developed a two-pronged framework to address these challenges:

  1. Automatic Truncation for Accuracy: To mitigate the upward bias caused by the noise floor, an SNR-scaled reconstruction was implemented. Data points in the decay curve falling below a threshold (2.5 standard deviations of the noise) are automatically truncated before performing a non-linear regression fit using a 2-parameter mono-exponential model.
  2. Analytical SD Maps for Precision: A novel formulation was derived to estimate the standard deviation of T2 at each pixel. By inverting a first-order approximation of the signal model's Hessian matrix, the system can predict the "error" of the fit, generating a spatial "quality map" alongside the T2 map.

The framework was validated using Monte-Carlo simulations (32,768 trials), agar gel phantoms with varying iron concentrations, and in-vivo scans of 20 human subjects to assess performance in real-world biological environments and identify motion artifacts.

Results and Discussion

The study demonstrated that the analytical SD maps are highly reliable, showing a >99.9% correlation with measured SD in phantom trials. This provides clinicians with a concrete metric to judge the reliability of a specific scan.

Multi-echo phantom images and validation of analytical SD maps. The estimated SD maps (right) show high correlation with the measured SD maps, providing a reliable metric for pixel-wise precision.

Key findings include:

  • Bias Mitigation: In simulations (T2 = 2ms, SNR = 15), standard fitting overestimated T2 by 3.6ms. Automatic truncation reduced this bias to 0.6ms.
  • Artifact Detection: SD maps proved sensitive to respiratory motion. While a T2* map might appear normal, a corresponding SD map with elevated values (e.g., 30ms vs. the expected 3ms) warns of underlying data corruption.
  • Protocol Optimization: The authors quantified the precision cost of speed; using parallel imaging to accelerate scans was shown to double the pixel-wise SD, allowing for informed trade-offs between scan time and data quality.

Conclusion

This work introduces a practical "safety net" for myocardial T2* mapping. By combining automatic truncation to improve accuracy and analytical SD maps to quantify precision, the proposed framework enhances the clinical utility of pixel-wise mapping for high-stakes decision-making in iron-overload management.

Summary updated

Abstract

A method for applying an automatic truncation method to pixel-wise T2* mapping that reduces T2* overestimation due to noise bias was proposed. A formulation for estimating pixel-wise standard deviation (SD) maps for T2* that can serve as a quality map for interpreting images and for comparison of imaging protocols was also proposed and validated.
BibTeX
@article{sandino2015myocardial,
  author = {Christopher M. Sandino and Peter Kellman and Andrew E. Arai and Michael S. Hansen and Hui Xue},
  title = {{Myocardial T2* mapping: influence of noise on accuracy and precision}},
  journal = {Journal of Cardiovascular Magnetic Resonance},
  year = {2015},
  doi = {10.1186/s12968-015-0115-3},
}
Figure