US Attenuation Coefficient for Liver Fat Quantification: A Need for Standardization

See also the article by Ferraioli et al in this issue.

Aiguo Han is an assistant professor in the Department of Biomedical Engineering and Mechanics at Virginia Polytechnic Institute and State University. His research interests focus on the development and clinical translation of novel US imaging technologies, such as quantitative US and artificial intelligence methods for noninvasive liver fat quantification. Dr Han is a fellow of the American Institute of Ultrasound in Medicine.

Abnormal fat accumulation in the liver, known as steatotic liver disease, affects over 30% of the global population and has substantial clinical implications. Fat accumulation in the liver may lead to liver inflammation and hepatocyte necrosis, potentially causing liver fibrosis, cirrhosis, and even hepatocellular carcinoma. Steatotic liver disease can be caused by various etiologic factors, with the predominant type being metabolic dysfunction–associated steatotic liver disease. Metabolic dysfunction–associated steatotic liver disease can be managed with lifestyle changes. Recently, pharmacologic treatment has become available following Food and Drug Administration approval of the first medication for the treatment of metabolic dysfunction–associated steatohepatitis with moderate to advanced fibrosis, an advanced stage of metabolic dysfunction–associated steatotic liver disease (1). Other drugs are undergoing clinical trials and have shown promising results for the treatment of metabolic dysfunction–associated steatohepatitis (1). Given the prevalence of steatotic liver disease and the availability of treatment options, accurate and precise liver fat quantification is crucial for disease diagnosis and treatment monitoring.

Existing methods of liver fat quantification have well-known limitations. Liver biopsy is used to assess liver fat content semiquantitatively in four grades. However, its limitations include invasiveness, sampling error, and high cost. MRI techniques, including confounder-corrected chemical shift–encoded MRI proton density fat fraction and confounder-corrected MR spectroscopy, provide accurate and precise noninvasive quantification of liver fat. Despite this, these methods are expensive and not routinely available (2). CT is important in assessing incidentally detected steatosis, but is unsuitable for primary diagnosis or monitoring of hepatic steatosis because of the potential risks of ionizing radiation (2). US is noninvasive, safe, widely available, and cost-effective, thus avoiding many of the aforementioned limitations. However, conventional US is qualitative and has limitations such as operator and system dependence of image quality and subjective interpretation.

Quantitative US has emerged as a promising approach for liver fat quantification, wherein quantitative US parameters (eg, attenuation coefficient [3], backscatter coefficient [4], and speed of sound [5]) are derived from the raw radiofrequency ultrasound signals backscattered from the liver with procedures designed to remove or reduce system dependence from the raw signals. The resulting quantitative US parameters describe intrinsic tissue properties correlated with liver fat. The US attenuation coefficient is the most frequently used quantitative US parameter for liver fat quantification in published studies and has been shown to be positively correlated with liver fat content. This parameter is a quantitative measure of the decrease in the amplitude of ultrasound waves as they travel through tissue. Various attenuation estimation methods have been proposed, including spectral shift, spectral difference, spectral log difference, and hybrid methods (6).

As attenuation estimation methods become commercially available in many clinical US systems, a critical question arises regarding the clinical adoption of these methods: Are the liver attenuation coefficient values reproducible between operators and across systems from different vendors? The answer to this question is important because it has important implications for eventual clinical adoption. For example, high interoperator reproducibility is essential when using the attenuation coefficient for patient follow-up. Low intersystem reproducibility may have multiple negative consequences that could hinder clinical adoption (6). Although prior studies have shown good interoperator reproducibility (7) and good intersystem reproducibility (8) for investigator-implemented attenuation estimation methods with a limited number of systems, data on interoperator and intersystem agreement remain limited for commercially available attenuation estimation methods across a large number of systems (9).

In this issue of Radiology, Ferraioli et al (10) assessed the interoperator and intersystem agreement of attenuation coefficients for liver fat quantification in participants with varying degrees of liver steatosis. This well-controlled single-center same-day prospective study included 26 participants, two experienced operators, and eight US systems from eight different vendors. Each system was equipped with a proprietary attenuation coefficient algorithm. Each participant underwent US examination by both operators on all eight systems, yielding a total of 416 attenuation coefficient observations. Additionally, MRI proton density fat fraction was measured in each participant and used as the reference standard for liver fat content. The authors found a high correlation between attenuation coefficient and MRI proton density fat fraction for five of the eight systems (r range, 0.70–0.86), a moderate correlation for two (r values of 0.62 and 0.62), and a poor correlation for one (r value of 0.47). These results are consistent with the literature and support the promising potential of the attenuation coefficient as an accurate biomarker for liver steatosis. Interoperator agreement was found to be excellent for one system (intraclass correlation coefficient, 0.90), good for two systems (0.76 and 0.76), and moderate for five systems (0.50–0.74). No statistically significant bias was observed between the two operators for seven of the eight systems. In terms of intersystem agreement, none of the system pairs achieved excellent agreement. Specifically, one pair showed good agreement (intraclass correlation coefficient, 0.79), eight pairs showed moderate agreement (0.50–0.73), and the remaining 19 pairs showed poor agreement (0.11–0.48).

Ferraioli et al (10) performed a subgroup analysis to explore how various clinical factors affected overall interoperator and intersystem agreement. They studied three factors: body mass index (<30 vs ≥30 [calculated as patient weight in kilograms divided by patient height in meters squared]), skin-to–liver capsule distance (<2.5 vs ≥2.5 cm), and liver steatosis (present vs absent). These factors were found to affect intersystem agreement but not interoperator agreement. A lower body mass index (<30), a shorter skin-to–liver capsule distance (<2.5 cm), and the presence of steatosis were individually associated with higher intersystem agreement. Although the authors warned that these findings should be interpreted cautiously due to the sample size, the thorough analysis of various factors is a strength of the study, as it offers important insights and directions to future investigators who might be interested in elucidating the sources of intersystem variability.

Another strength of the study is the standardization of the region of interest. The depth and size of the region of interest are known sources of variability for commercial attenuation coefficient algorithms. To minimize the impact of this, Ferraioli et al standardized the depth (2 cm below the liver capsule) and size (3 cm from near to far field) of the region of interest for all systems except two that had fixed regions of interest disallowing user adjustment. The authors repeated the analyses excluding these two systems and found improved overall intersystem agreement compared with including all eight systems (intraclass correlation coefficient, 0.42 vs 0.33), which was expected because the region of interest is a known source of variability. Conversely, overall interoperator agreement remained essentially unchanged after excluding the two systems compared with including all eight systems (intraclass correlation coefficient, 0.82 vs 0.84). This result was also expected because the regions of interest in the two systems were fixed, although they differed from the standardized region of interest.

The authors’ efforts in standardizing the US imaging protocol extended beyond the region of interest. The intercostal space in the right liver lobe was used for attenuation coefficient measurements in all participants. Five consecutive readings were taken in each participant, with the median used as the observation for statistical analysis. All systems were operated at 3.5 MHz. These efforts in standardization facilitated the interpretation of the results. The observed differences between systems were likely due to differences in how attenuation coefficients were calculated in each system, as pointed out by the authors. Because the commercial attenuation estimation methods are proprietary, it is conceivable that the underlying attenuation coefficient algorithms may differ markedly from one system to another.

Although this prospective study is of high quality, timely, and much needed, limitations include (a) the large 95% CIs of the reported intraclass correlation coefficients in most cases and (b) the high prevalence of obesity in the study sample (mean body mass index, 34.5), as also noted by the authors. Despite these limitations, the study provides important clinical data that fill a substantial knowledge gap in the literature. The takeaway is clear and convincing: The substantial variability in US attenuation coefficient measurements with commercial algorithms from different systems precludes the interchangeability of these systems for diagnosing and monitoring liver steatosis.

Based on the study by Ferraioli et al (10), improving intersystem agreement may require standardizing the algorithms among vendors. This could be challenging due to the potential high cost for vendors to make changes to their existing commercial implementations (6). However, the cost could be even greater to patient care and the industry if measurements are not reproducible across systems (6). It is worth noting that there is an ongoing effort led by the American Institute of Ultrasound in Medicine–RSNA Quantitative Imaging Biomarkers Alliance Pulse-Echo Quantitative Ultrasound Biomarker Committee to help understand the sources of variability and standardize quantitative US biomarker measurements and implementation. The continued focus on this standardization is crucial for ensuring that the US attenuation coefficient can be reliably used in clinical practice.

In conclusion, the study by Ferraioli et al (10) supports the promising potential of the attenuation coefficient as an accurate biomarker for liver steatosis, enhances our understanding of potential sources of variability, and suggests potential directions for minimizing variance. Studies such as this drive us forward toward the much-needed standardization and eventual clinical adoption of the US attenuation coefficient for liver fat quantification.