Development and blinded test of a software tool for ultrasound-based hepatic fat fraction estimation

• Published in: The Journal of the Acoustical Society of America 2019-10, Vol.146 (4), p.2864-2864
• Main Authors: Han, Aiguo, Boehringer, Andrew S., Montes, Vivian, Andre, Michael P., Erdman, John W., Loomba, Rohit, Valasek, Mark A., Sirlin, Claude B., O'Brien, W. D.
• Format: Article
• Language: English
• ISSN: 0001-4966; 1520-8524
• DOI: 10.1121/1.5136938

We have developed quantitative ultrasound (QUS) and deep learning algorithms to estimate hepatic fat fraction from radiofrequency (RF) ultrasound data backscattered by the liver. To facilitate the translation of such algorithms for clinical care and research, we developed a standalone software tool that can automatically generate fat fraction estimates using each of four separate algorithms based on ultrasonic attenuation coefficient (AC) (Algorithm 1), backscatter coefficient (BSC) (Algorithm 2), both AC and BSC (Algorithm 3), and deep learning with uncalibrated raw RF data (Algorithm 4). Reference phantom data and sonographer-drawn fields of interest (FOIs) outlining the liver boundary were used for Algorithms 1–3 but not in Algorithm 4. A pre-determined, fixed FOI was used in Algorithm 4. All four algorithms were developed using contemporaneous MRI-PDFF and ultrasound RF liver data acquired from 144 adult participants with and without nonalcoholic fatty liver disease. The software is now being tested on an independent cohort of participants with contemporaneous MRI-PDFF as the reference standard. 26 participants have been enrolled to date. Preliminary results show good agreement among the four algorithms, and good correlation between the ultrasound fat fraction estimates and MRI-PDFF (Person’s r = 0.61, 0.63, 0.66, and 0.76 for Algorithms 1–4, respectively). [No. R01DK106419]