obesity | VOLUME 17 NUMBER 12 | dEcEMBER 2009 2239
nature publishing group articles
Methods and techniques
IntroductIon
A continuous accumulation of lipids in the liver may result
in a broad hepatological spectrum, usually referred to as fatty
liver disease (FLD) (1,2). An abundance of fat within the liver,
or steatosis, can progress to steatohepatitis (fat and inflam-
mation, with or without fibrosis) and cirrhosis (maximum
fibrosis score) (3), and has also been associated with hepato-
carcinoma (4). Additionally, FLD, particularly nonalcoholic
FLD, is an underlying condition for cardiovascular disease
(5,6). As alcoholic FLD and nonalcoholic FLD are histologi-
cally indistinct (7), distinction between both is neither possible
nor relevant in relation to measurement of hepatic fat content.
The estimated FLD prevalence is one-third of the general adult
Western population (8–10), and may have been increasing in
parallel with the global increase of obesity (11).
Hepatic fat content can be determined by histological (2) or
biochemical (12,13) analysis of liver tissue by biopsy, magnetic
resonance techniques (14), computed tomography (15), and
ultrasonography (16,17). Ultrasonography is, in contrast to
other diagnostic modalities, an appealing method for large
population studies on FLD, as it is noninvasive (painless, no
harmful radiation), portable, and relatively inexpensive. In
the case of parenchymal liver disease, reflections of liver tissue
by ultrasonography are altered (16,17). In clinical care, ultra-
sonography is the most often used diagnostic modality, but
in a qualitative way. Steatosis can be qualitatively assessed by:
(i) hyperechogenity of liver tissue (“bright liver”) as often com-
pared to hypoechogenity of the kidney cortex, (ii) fine, tightly
packed echoes, (iii) fall of echo amplitude with depth (poste-
rior beam attenuation), and (iv) loss of echoes from the walls
of the portal veins (featureless appearance) (16,17). As this is
a qualitative scoring method and also subjective (18), quanti-
tative approaches for identification of liver disease have been
suggested. However, these methods have never been validated
Ultrasonography to Quantify Hepatic Fat
content: Validation by 1H Magnetic
Resonance Spectroscopy
Mireille A. Edens1, Peter M.A. van Ooijen2, Wendy J. Post1, Mark J.F. Haagmans2,
Wisnumurti Kristanto2, Paul E. Sijens2, Erik J. van der Jagt2 and Ronald P. Stolk1
An abundance of fat stored within the liver, or steatosis, is the beginning of a broad hepatological spectrum, usually
referred to as fatty liver disease (FLD). For studies on FLD, quantitative hepatic fat ultrasonography would be an
appealing study modality. Objective of this study was to develop a technique for quantifying hepatic fat content
by ultrasonography and validate this using proton magnetic resonance spectroscopy (1H MRS) as gold standard.
Eighteen white volunteers (BMI range 21.0–42.9) were scanned by both ultrasonography and 1H MRS. Altered
ultrasound characteristics, present in the case of FLD, were assessed using a specially developed software program.
Various attenuation and textural based indices of FLD were extracted from ultrasound images. Using linear regression
analysis, the predictive power of several models (consisting of both attenuation and textural based measures) on
log 10–transformed hepatic fat content by 1H MRS were investigated. The best quantitative model was compared
with a qualitative ultrasonography method, as used in clinical care. A model with four ultrasound characteristics
could modestly predict the amount of liver fat (adjusted explained variance 43.2%, P = 0.021). Expanding the
model to seven ultrasound characteristics increased adjusted explained variance to 60% (P = 0.015), with r = 0.789
(P < 0.001). Comparing this quantitative model with qualitative ultrasonography revealed a significant advantage of
the quantitative model in predicting hepatic fat content (P < 0.001). This validation study shows that a combination
of computer-assessed ultrasound measures from routine ultrasound images can be used to quantitatively assess
hepatic fat content.
Obesity (2009) 17, 2239–2244. doi:10.1038/oby.2009.154
1Department of Epidemiology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands; 2Department of Radiology, University Medical
Center Groningen, University of Groningen, Groningen, The Netherlands. Correspondence: Mireille A. Edens (m.a.edens@epi.umcg.nl)
Received 5 February 2009; accepted 19 April 2009; published online 21 May 2009. doi:10.1038/oby.2009.154

2240 VOLUME 17 NUMBER 12 | dEcEMBER 2009 | www.obesityjournal.org
articles
Methods and techniques
by an appropriate quantitative gold standard. The purpose of
this study was to develop and validate quantitative analysis of
ultrasonography images, for assessment of hepatic fat content,
using proton magnetic resonance spectroscopy (1H MRS) as
gold standard.
Methods and Procedures
Volunteers
Volunteers were recruited by advertisement, and a heterogenic study
population was strived after. Exclusion criteria were current presence
of hepatic pathology, previous hepatic or renal surgery, and stand-
ard MR-contraindications. The volunteers underwent both hepatic
ultrasonography and 1H MRS, and a short physical examination. All
volunteers gave written informed consent. This study was approved
by the Medical Ethics Committee of the University Medical Center
Groningen.
ultrasonography
Quantitative ultrasonography.
Imaging: Ultrasonography was performed using a Philips ATL ultra-
sound machine (Philips, Best, The Netherlands), with a 5–2 MHz curved
array transducer. In one ultrasound image, both liver and right kidney
were visualized (17), as shown in Figure 1. Imaging was performed by
an experienced radiologist. One standard image, with “persist” (med),
“2D opt” (gen), “frame rate” (high), “gain” (40), and “image depth”
(14.7 cm), was used for analysis.
Analysis: Images were analyzed by an operator (operator 1) twice, with
a 1-month interval, and the average values by operator 1 were used in
this study. In order to study interoperator reliability, another operator
(operator 2) analyzed the images, while untrained for the method and
blinded for all study outcomes.
Data extraction and data: Data were extracted from ultrasound
images using a modified version of a specially developed software pro-
gram (Department of BME, Technion IIT, Haifa, Israel) in the MAT-
LAB programming environment, previously described by Gaitini et al.
(19). Figure 1 shows an example of data extraction. According to a
standard protocol, regions of interest and attenuation lines were inter-
actively placed in the liver images in order to calculate several attenu-
ation indices and several textural indices. Figure 2 shows a scheme on
quantitative ultrasonography measures, including the presently vali-
dated indices in the white boxes.
Qualitative ultrasonography. In addition to the quantitative approach,
the radiologist made an ultrasound image with optimum settings, as
used in clinical care. This image was qualitatively scored by the radiolo-
gist, according to standard qualitative criteria (16,17,20), while blinded
for all study outcomes.
Multi-voxel proton Mrs
In general, by means of radiofrequency transmission and reception, a
magnetic resonance scanner detects resonance signals of both hepatic
lipids (mainly methylene, i.e., CH
2
, from fatty acyl chains) and hepatic
water (21). As previously described in detail (14,22), 1H MRS was per-
formed, using a 1.5 Tesla whole-body scanner (MAGNETOM Avanto;
Siemens Medical Solutions, Erlangen, Germany) equipped with gra-
dients of up to 40 mT/m (maximal slew rate = 200 mT/m/ms) and a
six-channel spine array coil. Subjects were in supine position with a
large flex coil placed over the liver, which was simultaneously used with
the spine array coil as receiver. T
1
-weighted gradient-echo images were
recorded to assess the anatomy of liver. Using a field of view of 16 ×
16 cm2 and a volume of interest of 5 × 8 × 4 cm3 positioned within the
liver, hybrid two-dimensional-spectroscopic imaging (chemical shift
imaging), point resolved spectroscopy with a repetition time of 5,000 ms
and an echo time of 30 ms was performed. The chemical shift imaging
measurement lasted 16 × 16 × 5 = 1,280 s, corresponding to ~21 min.
Shimming was automated and water suppression was not applied in
order to be able to calculate the fat–water ratio distributions in the liver
directly. At the used repetition time of 5 s, T
1
saturation of the water and
fat signals is negligible, i.e., repetition time >5T
1
. At the used echo time
of 30 ms, the correction applied to our data, to compensate for the fact
that the fat signal has a longer T
2
(78 ms) than water (60 ms), was 12.2%.
Hepatic fat content was calculated by the peak CH
2
signal (at 1.3 parts/
million) divided by the sum of the peak CH
2
signal and peak H
2
O sig-
nal (at 4.7 parts/million), using water as an internal reference (14,22).
1H MRS has been validated, by comparison with both histological and
biochemical analysis of liver tissue by biopsy (21,23,24).
A hepatic fat content of 5.56% by 1H MRS is used as cutoff value for
diagnosing FLD, based on the 95th percentile hepatic fat distribution of
a low risk group (10).
statistics
Univariate analysis and multiple regression analysis. As dis-
tribution of hepatic fat content by 1H MRS was skewed, values were
log 10–transformed. Plotting and correlation (Pearson) was used to
explore univariate concordance with log 10 1H MRS. The classifica-
tion of variables in Figure 2 (white boxes), followed by “backward
selection,” was used for variable selection in a linear regression model.
First, the variables from separate boxes of Figure 2 were assessed, i.e.,
separate ultrasonography aspects. Second, variables from combina-
tions of boxes of Figure 2 were assessed, i.e., information from several
ultrasonography aspects.
Evaluation and bootstrap. Models were evaluated on adjusted
explained variance (adj. R2) and explained variance (R2). By means of
bootstrap, 95% confidence intervals were estimated for regression coef-
ficients, and adj. R2 and R2. Moreover, a 95% prediction interval was
calculated.
Quantitative vs. qualitative ultrasonography. The χ2-test was
used to test the differences between the two methods. Addition-
Figure 1 Ultrasonography image analysis. Both a region of interest
(quadrangle) and an attenuation line (closed line) were placed according
to a standard protocol. The region of interest had to be placed in a bright
area, while avoiding large artifacts like rib shadows and large blood
vessels, at a depth of 4–6 cm. The attenuation line had to be placed in a
bright pathway, while avoiding large artifacts, at a straight line from the
“origin” of ultrasound (striped line). The region of interest served for the
determination of several textural indices. The attenuation line was used
for determining attenuation estimates (19).