A fully-automated statistical method for characterization of flow artifact presence in cardiac MRI

Sotirios A. Tsaftaris, Xiangzhi Zhou, Rohan DharmakumarJournal of Cardiovascular Magnetic Resonance 2011 · DOI

Abstract

Flow artifacts in MR images can appear as ghosts within and outside the body cavity. Current approaches for optimizing sequences for suppressing such artifacts rely on expert scoring or on semi-automated methods for evaluation.

Introduction

In cardiac MRI, blood flow frequently generates "ghost" artifacts—faint, repeating streaks that manifest both inside and outside the body. These artifacts can obscure critical anatomy and compromise diagnostic accuracy. Traditionally, optimizing imaging sequences to minimize these ghosts has required subjective expert scoring by radiologists or semi-automated tools necessitating manual input. This study introduces a fully-automated statistical method to quantify ghosting by analyzing the "noise" in the air surrounding the subject, providing a robust, scalable alternative to human-in-the-loop evaluation.

Methodology

The proposed method leverages the fact that flow artifacts fluctuate over the cardiac cycle, unlike static background noise. The algorithm focuses on the air region outside the body to detect these fluctuations.

  1. Segmentation: A "level-set" approach automatically identifies the boundary between the body and the surrounding air.
  2. Statistical Analysis: For each pixel in the air region, the mean ($I_A$) and variance ($I_V$) are calculated across the time-series (cine) frames.
  3. Normalization: To ensure robustness against coil bias (uneven signal intensity from MRI receiver coils), the system calculates a ratio: $I_R = I_V / (I_A)^2$.
  4. Kurtosis Metric ($Q_K$): The distribution of $I_R$ values in the air is analyzed. While clean images show a predictable distribution, flow artifacts appear as statistical outliers. The method calculates the excess kurtosis ($\gamma$)—a measure of the distribution's "heavy-tailedness"—to derive the quality metric $Q_K$. Higher kurtosis directly correlates with more significant ghosting.
Statistical analysis workflow: (left) original cine image $I(t)$, (middle) normalized variance image $I_R$ highlighting artifacts in the air region, and (right) histogram $H$ of $I_R$ values where artifacts appear as outliers.

Experimental Validation

The method was validated using 16 studies from six healthy dogs on a Siemens 1.5T scanner. Three 2D SSFP cine sequences were employed to generate varying artifact levels:

  • Sequence A: Minimal artifacts (negligible BOLD sensitivity).
  • Sequence B: Significant artifacts (optimal BOLD sensitivity, no flow compensation).
  • Sequence C: Reduced artifacts (optimal BOLD sensitivity with flow compensation).

Ground truth was established by three expert reviewers who rated the studies on a scale of 1 (least ghosts) to 5 (most ghosts) while blinded to the sequence type.

Results and Key Insights

The automated $Q_K$ metric demonstrated a strong correlation ($R = 0.7, P < 0.01$) with the median expert scores ($Q_H$). Furthermore, the method successfully discriminated between the three sequences, correctly identifying Sequence B as having the highest artifact presence and Sequence A the lowest.

Key insights from the study include:

  • Background as a Proxy: The "empty" space around the body provides a cleaner environment for statistical outlier detection than the complex moving anatomy of the heart.
  • Higher-Order Statistics: Kurtosis is more effective than simple signal-to-noise ratios at capturing the "spiky" nature of ghosting artifacts.
  • Temporal Analysis: Analyzing image stacks over time captures artifact variability across the cardiac cycle that static frames might miss.

Discussion

While promising, the study was limited to a small sample of healthy dogs. Future work must validate the method on larger, diverse human populations where varying body habitus or pathology might impact segmentation accuracy. Additionally, while the normalization step provides hardware robustness, the reliance on precise body-air segmentation remains a potential point of failure. Future clinical integration will also need to address the computational overhead of the processing algorithm.

Summary updated

Abstract

Flow artifacts in MR images can appear as ghosts within and outside the body cavity. Current approaches for optimizing sequences for suppressing such artifacts rely on expert scoring or on semi-automated methods for evaluation.
BibTeX
@article{tsaftaris2011fully,
  author = {Sotirios A. Tsaftaris and Xiangzhi Zhou and Rohan Dharmakumar},
  title = {{A fully-automated statistical method for characterization of flow artifact presence in cardiac MRI}},
  journal = {Journal of Cardiovascular Magnetic Resonance},
  year = {2011},
  doi = {10.1186/1532-429x-13-s1-p45},
}
Figure