Reliability of ultrasound hepatorenal index and magnetic resonance imaging proton density fat fraction techniques in the diagnosis of hepatic steatosis, with magnetic resonance spectroscopy as the reference standard

Abstract

Purpose To evaluate the reliability of ultrasound hepatorenal index (US-HRI) and magnetic resonance imaging proton density fat fraction (MRI-PDFF) techniques in the diagnosis of hepatic steatosis, with magnetic resonance spectroscopy proton density fat fraction (MRS-PDFF) as the reference standard. Materials and methods Fifty-two adult volunteers (30 men, 22 women; age, 31.5 ± 6.5 years) who had no history of kidney disease or viral/alcoholic hepatitis were recruited to undergo abdominal US, MRI, and MRS examinations. US-HRI was calculated from the average of three pairs of regions of interest (ROIs) measurements placed in the liver parenchyma and right renal cortex. On MRI, the six-point Dixon technique was employed for calculating proton density fat fraction (MRI-PDFF). An MRS sequence with a typical voxel size of 27 ml was chosen to estimate MRS-PDFF as the gold standard. The data were evaluated using Pearson’s correlation coefficient and receiver operating characteristic (ROC) curves. Results The Pearson correlation coefficients of US-HRI and MRI-PDFF with MRS-PDFF were 0.38 (p = 0.005) and 0.95 (p<0.001), respectively. If MRS-PDFF ≥5.56% was defined as the gold standard of fatty liver disease, the areas under the curve (AUCs), cut-off values, sensitivities and specificities of US-HRI and MRI-PDFF were 0.74, 1.54, 50%, 91.7% and 0.99, 2.75%, 100%, 88.9%, respectively. The intraclass correlation coefficients (ICCs) of US-HRI and MRI-PDFF were 0.70 and 0.85. Conclusion MRI-PDFF was more reliable than US-HRI in diagnosing hepatic steatosis.

Evaluation of Artificial Intelligence-Calculated Hepatorenal Index for Diagnosing Mild and Moderate Hepatic Steatosis in Non-Alcoholic Fatty Liver Disease

“…The mean AI-HRI (2.27) in the study cohort was higher than the mean HRI (1.4–1.56) in previous studies. This can be mainly attributed to the lower percentage of patients without significant steatosis (27.7%) in our study cohort compared to patient populations in previous studies where the percentage of negative cases (47.8–69.2%) was much higher [ 8 , 9 , 10 ]. We also found significant differences between AI-HRI measured in normal liver, mild steatosis, and mild and moderate steatosis, indicating robust diagnostic performance.…”

“…In the present study, we have demonstrated that AI-HRI is a reliable method for the diagnosis and classification of hepatic steatosis in NAFLD. In previous studies, which evaluated HRI, the liver and kidney ROIs were manually selected, which caused significant differences between the diagnostic protocols [ 8 , 9 , 10 , 23 , 24 ].…”

“…The accuracy of manually labeled HRI in the detection of mild steatosis showed considerable variability between different studies, with AUC values ranging between 0.68 and 0.92 [ 8 , 9 , 10 ]. The performance of AI-HRI was within the above range; however, its AUC of 0.76 was superior to some recently published results obtained with high-end ultrasound systems, which measured HRI manually [ 8 ].…”

“…The correlation between the two observers’ measurements was weaker with AI-HRI than with conventional HRI (r s =0.57 vs. Pearson’s r =0.70). Meanwhile, the mean interobserver bias of −10% and the LOA of −78% and 58% in a Bland–Altman analysis were all higher compared to the bias of 2% and LOA of −47% and 51% reported for HRI [ 10 ]. The disagreement between the results can be partly explained by the differences in the study protocol.…”

“…Another semi-quantitative metric is the hepatorenal index (HRI), which is the ratio between the brightness of the liver and the right renal cortex on grayscale US. HRI is less operator-dependent, and it has been shown to have a good detection rate of mild and even better detection of moderate and severe steatosis [ 8 , 9 , 10 ]. Meanwhile, the calculation of HRI can be time-consuming, and selecting a region of interest (ROI) can be subjective, weakening the measurement’s reproducibility.…”