Estimating liver fat fraction from ultrasonic attenuation and backscatter coefficient measures in adults with nonalcoholic fatty liver disease

Abstract

Between 80 and 100 million people in the United States suffer from nonalcoholic fatty liver disease (NAFLD). It is currently being diagnosed via MRI and biopsies; however, these tend to be expensive and painful. In this study, a Siemens 3000 ultrasound scanner was used to obtain the quantitative ultrasound (QUS) parameters [backscatter coefficient, BSC, and attenuation coefficient, AC] of 84 participants with known or suspected NAFLD (thus few uncompromised liver samples expected). Participant recruitment is continuing. Also obtained was the liver proton density fat fraction (PDFF) via MRI. AC and BSC individually were linearly correlated (~0.6) to PDFF. The data were then divided into two equally sized sets, training and test, in order to calculate PDFF from a function of AC and BSC. Using the training set a polynomial equation was created where AC and BSC were used to estimate PDFF. It was found that a polynomial 1,2 [AC,BSC] yielded an improved correlation (0.76) when evaluated using the test set. Significant improvement is obtained to estimate PDFF using both AC and BSC as opposed to using either of these QUS parameters separately. [Support: NIH R01DK106419 and NIH R37EB002641.]Between 80 and 100 million people in the United States suffer from nonalcoholic fatty liver disease (NAFLD). It is currently being diagnosed via MRI and biopsies; however, these tend to be expensive and painful. In this study, a Siemens 3000 ultrasound scanner was used to obtain the quantitative ultrasound (QUS) parameters [backscatter coefficient, BSC, and attenuation coefficient, AC] of 84 participants with known or suspected NAFLD (thus few uncompromised liver samples expected). Participant recruitment is continuing. Also obtained was the liver proton density fat fraction (PDFF) via MRI. AC and BSC individually were linearly correlated (~0.6) to PDFF. The data were then divided into two equally sized sets, training and test, in order to calculate PDFF from a function of AC and BSC. Using the training set a polynomial equation was created where AC and BSC were used to estimate PDFF. It was found that a po...