Feasibility study on using the diffusion-weighted imaging liver pulsation artifact as marker of liver fibrosis

Technical Summary: Feasibility Study on Using Diffusion-Weighted Imaging Liver Pulsation Artifact as a Marker of Liver Fibrosis

Problem Statement
Diffusion-weighted imaging (DWI) is a standard tool for detecting abdominal lesions, particularly in the liver. However, cardiac pulsation waves often propagate into the liver, causing phase dispersion and signal loss (the "pulsation artifact") in diffusion-weighted images. This artifact is most pronounced in the left liver lobe due to its proximity to the heart and is typically considered a nuisance that reduces sensitivity for detecting left-liver metastases or necessitates excluding the left lobe from analysis.

Currently, the gold standard for diagnosing liver fibrosis is invasive biopsy, which carries risks of complications and sampling error. While Magnetic Resonance Elastography (MRE) is the most accurate non-invasive alternative, it requires additional hardware (actuators) and, in the case of intrinsic MRE methods, additional scan sequences. This study hypothesizes that the severity of the DWI pulsation artifact is mechanically linked to tissue stiffness. Consequently, the "bug" of the pulsation artifact could be repurposed as a "feature" to serve as an intrinsic, non-invasive marker for liver cirrhosis without requiring extra equipment or scan time.

Methodology
The study employed a mixed retrospective and prospective design involving 13 healthy volunteers and 19 patients with histopathologically confirmed liver status (16 with cirrhosis, Ishak scores 5–6; 3 without cirrhosis, Ishak scores 0–2).

• Imaging Protocol: Data were acquired on 1.5 T and 3 T MRI systems using echo-planar DWI sequences with two b-values: $b=50$ s/mm² and $b=800$ s/mm².
• Segmentation: The left (segments 1–4) and right (segments 5–8) liver lobes were manually segmented on $b=50$ s/mm² trace-weighted images, using vascular landmarks to ensure accurate boundaries.
• Feature Extraction: 18 first-order statistical features (including percentiles, mean absolute deviation, standard deviation, skewness, and kurtosis) were extracted from the segmented regions on both $b=50$ and $b=800$ images.
• Normalization Strategy: To account for global intensity variations between lobes (e.g., surface coil flare), the authors computed a ratio of features between the left and right lobes ($F_{ratio}$) for both b-values. A final normalized feature ($F_{final}$) was derived by dividing the $b=800$ ratio by the $b=50$ ratio:
  $$F_{final} = \frac{F_{ratio, b800}}{F_{ratio, b50}}$$
  This approach isolates the signal loss specific to the high b-value pulsation artifact in the left lobe.
• Statistical Analysis: The study utilized Receiver Operating Characteristic (ROC) analysis to differentiate cirrhotic from non-cirrhotic livers. Given the small sample size, a leave-one-out bootstrapping (LOOBT) approach with Firth penalized logistic regression was used to estimate Area Under the Curve (AUC) values and minimize overfitting.

Key Results

• Artifact Behavior: The study observed that pulsation artifacts (signal voids) were prominent in the left liver lobe of non-cirrhotic livers but were largely absent in cirrhotic livers. This suggests that increased tissue stiffness in cirrhosis dampens the propagation of cardiac pulsation waves, thereby reducing the artifact.
• Best Performing Feature: The first quartile ($q1$, the 25th percentile) of the normalized signal intensity in the left lobe was the most effective discriminator.
  • Whole-dataset analysis: $q1$ achieved an AUC of 0.95 (95% CI: 0.87–1.00).
  • Leave-one-out analysis: $q1$ maintained a high AUC of 0.92 (95% CI: 0.80–1.00).
• Bivariate Analysis: Combining $q1$ with skewness or kurtosis yielded slightly higher AUCs (0.93) in the leave-one-out analysis, though the authors note this improvement may be a statistical artifact given the poor univariate performance of skewness and kurtosis.
• Differentiation: The $q1$ values for cirrhotic livers were generally larger than those for non-cirrhotic livers. Notably, patients with early fibrosis (Ishak 1–2) exhibited higher $q1$ values than healthy volunteers, suggesting the metric may be sensitive to stiffness changes even before full cirrhosis develops.

Significance and Claims
The paper claims to demonstrate that the DWI pulsation artifact, traditionally viewed as a technical limitation, can be quantitatively analyzed to serve as a simple, intrinsic marker for liver cirrhosis.

• Clinical Utility: The primary significance lies in the potential to assess liver stiffness using standard DWI sequences already present in most liver MRI protocols, without the need for additional hardware (as in extrinsic MRE) or extended scan times (as in intrinsic MRE).
• Mechanism: The findings support the hypothesis that the magnitude of the pulsation artifact is inversely related to liver stiffness; stiffer (cirrhotic) tissue dampens the mechanical waves, reducing the artifact.
• Limitations and Modesty: The authors explicitly state that the study is limited by a small sample size, potential selection bias (retrospective HCC cohort), and the lack of age-matched controls. They acknowledge that the method is currently blind to fibrosis in the right liver lobe. The authors conclude that while the quantitative results are promising, the suitability of this approach for routine clinical practice (which relies on qualitative assessment) requires validation in prospective clinical studies. They do not claim the method replaces MRE or biopsy but rather offers a potential supplementary tool derived from existing data.