OR WAIT null SECS
In a new prospective study, an emerging deep learning model, which incorporates parametric mapping with quantitative ultrasound to estimate liver fat fraction, demonstrated a 90 percent sensitivity rate and a 91 percent specificity rate for diagnosing hepatic steatosis in patients with non-alcoholic fatty liver disease (NAFLD).
Emerging research suggests the combination of deep learning and quantitative ultrasound (QUS) provides robust accuracy in the diagnosis of hepatic steatosis, a prominent feature of non-alcoholic fatty liver disease (NAFLD), which has an estimated worldwide prevalence of 32.4 percent.1,2
Employing magnetic resonance imaging (MRI)-derived proton density fat fraction (PDFF) as a reference standard, researchers assessed the use of a convolutional neural network model to estimate hepatic fat fraction and diagnose hepatic steatosis in a prospective study of 173 people with known or suspected NAFLD.
The deep learning-based algorithm, which incorporates radiofrequency data analysis, assesses QUS parametric imaging via tissue attenuation imaging (TAI) and tissue scatter-distribution imaging (TSI), and brightness-mode (B-mode) ultrasound images to estimate ultrasound fat fraction (USFF), according to the recently published study in Radiology.
In addition to demonstrating a strong correlation with MRI-PDFF (Pearson r=0.86), the study authors found that the deep learning algorithm estimate of USFF had an area under the curve (AUC) of 97 percent, a sensitivity rate of 90 percent and a specificity rate of 91 percent in diagnosing hepatic steatosis.1
The study authors said the use of QUS parametric mapping was key to the strong diagnostic performance of the USFF approach to diagnosing hepatic steatosis.
“(Quantitative ultrasound) parametric maps derived from radiofrequency data analysis can provide more information about liver tissue composition, which can be lost during B-mode image generation,” wrote lead study author Sun Kyung Jeon, M.D., Ph.D., who is affiliated with the Department of Radiology at Seoul National University Hospital and the Seoul National University College of Medicine, and colleagues. “Furthermore, as radiofrequency data are not affected by dynamic range and filtering settings, using the QUS parametric map as the input data of the 2D (convolutional neural network) algorithm may be valuable for developing robust algorithms for assessing hepatic steatosis.”
(Editor’s note: For a related article, see “MRI Study Shows Higher Fibrosis Risk Among First-Degree Relatives of People with NAFLD and Advanced Fibrosis.”)
Noting challenges and limitations with other diagnostic techniques for hepatic steatosis, ranging from the error rate for the invasive liver biopsy to the high cost and low accessibility of MRI, and the modest accuracy rate of conventional B-mode ultrasound, Jeon and colleagues suggested that the non-invasive USFF model may be a viable screening alternative for NAFLD with potential for treatment monitoring as well.
In an accompanying editorial, Paul S. Sidhu, MRCP, FRCR and Cheng Fang, FRCR pointed out that in relation to MRI PDFF values > 5 percent, > 15 percent and > 25 percent, the USFF algorithm had respective AUCs of 97 percent, 96 percent, and 95 percent for the detection of hepatic steatosis.3
While they caution about signal saturation that can occur in cases of severe hepatic steatosis with the USFF model and other QUS parameters, Drs. Sidhu and Fang maintained that ultrasound fat quantification has considerable promise in the imaging realm for liver disease.
“(Ultrasound) fat quantification is an emerging technology with the capability of transforming liver disease care. It is perhaps the only all-encompassing imaging tool in the assessment and follow-up of chronic liver conditions,” noted Dr. Sidhu, a professor of Imaging Sciences at King’s College London and consultant radiologist at King’s College Hospital, and Dr. Fang, a consultant radiologist at King’s College Hospital.
In regard to study limitations, Jeon and colleagues acknowledged that the ultrasound data came from a single scanner. The study authors noted they did not assess the algorithm using B-images alone nor did they evaluate the possible impact of body mass index and skin to liver capsule distance in relation to the algorithm. Jeon and colleagues also pointed out that inflammation and fibrosis, possible confounding factors, were not evaluated in the study. Conceding that radiofrequency data is not readily available in healthcare systems at the present time, the researchers suggested access to this data will be more commonplace in the near future.
1. Jeon SK, Lee JM, Joo I, Yoon JH, Lee G. Two-dimensional convolutional neural network using quantitative US for noninvasive assessment of hepatic steatosis in NAFLD. Radiology. 2023 Jan 3: 221510. doi: 10.1148/radiol.221510. Online ahead of print.
2. Riazi K, Azhari H, Charette JH, et al. The prevalence and incidence of NAFLD worldwide: a systematic review and meta-analysis. Lancet Gastroenterol Hepatol. 2022;7(9):851-861.
3. Sidhu PS, Fang C. US-based hepatic fat quantification: an emerging technique and game changer? Radiology. 2023 Jan 3: 223002. doi: 10:1148/radiol.223002. Online ahead of print.