An ultrasound multiparametric method to quantify liver fat using magnetic resonance as standard reference

Area of Science:

• Medical Imaging
• Hepatology
• Quantitative Ultrasound

Background:

• Non-invasive fatty liver content (FLC) measurement is needed for steatotic liver disease monitoring.
• Current ultrasound (US) methods are qualitative and operator-dependent.
• Existing single-parameter US algorithms have limited quantification range.

Purpose of the Study:

• Develop and validate a novel multiparametric algorithm for FLC quantification.
• Utilize B-mode ultrasound images for FLC assessment.
• Establish Magnetic Resonance (MR) imaging as the reference standard.

Main Methods:

• Combined five US-derived quantitative features (AR, HR, DV, HPV, PVW) into a multiparametric model (Steatoscore2.0) using regression.
• Trained the model on 134 subjects and validated independently on 61 subjects.
• Assessed inter-operator reproducibility of the score computation.

Main Results:

• Steatoscore2.0, a mixed linear/exponential model of AR, HR, and DV, showed high correlation with MR-FLC (R=0.92 in training, R=0.86 in validation).
• The model achieved low root mean square errors (RMSE=5.15/4.62) and demonstrated excellent diagnostic performance for identifying FLC thresholds (AUROC=0.958/0.975).
• Steatoscore2.0 significantly outperformed conventional US methods like CAP (R=0.73).

Conclusions:

• The multiparametric Steatoscore2.0 provides FLC values highly comparable to MR.
• This tool is reliable, inexpensive, and user-friendly with any US equipment.
• Steatoscore2.0 is suitable for further evaluation in prospective studies for non-invasive FLC screening and monitoring.