entropy

Article
Ultrasound Entropy Imaging of Nonalcoholic Fatty
Liver Disease: Association with Metabolic Syndrome
Ying-Hsiu Lin 1 , Yin-Yin Liao 2 , Chih-Kuang Yeh 3 , Kuen-Cheh Yang 4,5, *
and Po-Hsiang Tsui 1,6,7, *
1 Department of Medical Imaging and Radiological Sciences, College of Medicine, Chang Gung University,
Taoyuan 33302, Taiwan; [email protected]
2 Department of Biomedical Engineering, Hungkuang University, Taichung 43302, Taiwan;
[email protected]
3 Department of Biomedical Engineering and Environmental Sciences, National Tsing Hua University,
Hsinchu 30013, Taiwan; [email protected]
4 Department of Family Medicine, National Taiwan University Hospital, Beihu Branch, Taipei 10800, Taiwan
5 Health Science & Wellness Center, National Taiwan University, Taipei 10617, Taiwan
6 Department of Medical Imaging and Intervention, Chang Gung Memorial Hospital at Linkou,
Taoyuan 33305, Taiwan
7 Medical Imaging Research Center, Institute for Radiological Research, Chang Gung University and Chang
Gung Memorial Hospital at Linkou, Taoyuan 33302, Taiwan
* Correspondence: [email protected] (K.-C.Y.); [email protected] (P.-H.T.);
Tel.: +886-2-2371-7101 (K.-C.Y.); +886-3-211-8800 (ext. 3795) (P.-H.T.)




Received: 17 October 2018; Accepted: 20 November 2018; Published: 22 November 2018


Abstract: Nonalcoholic fatty liver disease (NAFLD) is the leading cause of advanced liver diseases.
Fat accumulation in the liver changes the hepatic microstructure and the corresponding statistics of
ultrasound backscattered signals. Acoustic structure quantification (ASQ) is a typical model-based
method for analyzing backscattered statistics. Shannon entropy, initially proposed in information
theory, has been demonstrated as a more flexible solution for imaging and describing backscattered
statistics without considering data distribution. NAFLD is a hepatic manifestation of metabolic
syndrome (MetS). Therefore, we investigated the association between ultrasound entropy imaging
of NAFLD and MetS for comparison with that obtained from ASQ. A total of 394 participants were
recruited to undergo physical examinations and blood tests to diagnose MetS. Then, abdominal
ultrasound screening of the liver was performed to calculate the ultrasonographic fatty liver indicator
(US-FLI) as a measure of NAFLD severity. The ASQ analysis and ultrasound entropy parametric
imaging were further constructed using the raw image data to calculate the focal disturbance (FD)
ratio and entropy value, respectively. Tertiles were used to split the data of the FD ratio and
entropy into three groups for statistical analysis. The correlation coefficient r, probability value p,
and odds ratio (OR) were calculated. With an increase in the US-FLI, the entropy value increased
(r = 0.713; p < 0.0001) and the FD ratio decreased (r = –0.630; p < 0.0001). In addition, the entropy
value and FD ratio correlated with metabolic indices (p < 0.0001). After adjustment for confounding
factors, entropy imaging (OR = 7.91, 95% confidence interval (CI): 0.96–65.18 for the second tertile;
OR = 20.47, 95% CI: 2.48–168.67 for the third tertile; p = 0.0021) still provided a more significant link to
the risk of MetS than did the FD ratio obtained from ASQ (OR = 0.55, 95% CI: 0.27–1.14 for the second
tertile; OR = 0.42, 95% CI: 0.15–1.17 for the third tertile; p = 0.13). Thus, ultrasound entropy imaging
can provide information on hepatic steatosis. In particular, ultrasound entropy imaging can describe
the risk of MetS for individuals with NAFLD and is superior to the conventional ASQ technique.

Keywords: ultrasound; hepatic steatosis; Shannon entropy; fatty liver; metabolic syndrome

Entropy 2018, 20, 893; doi:10.3390/e20120893 www.mdpi.com/journal/entropy

Entropy 2018, 20, 893 2 of 16

1. Introduction
Nonalcoholic fatty liver disease (NAFLD) is characterized by excess and abnormal intracellular
accumulation of triglycerides in hepatocytes. Histologically, NAFLD refers to macrovesicular steatosis
and is the leading cause of nonalcoholic steatohepatitis, fibrosis, cirrhosis, and hepatocellular
carcinoma [1,2]. Therefore, NAFLD may be considered a critical health problem, and its early detection,
follow-up, and management can help arrest the progression of advanced liver diseases [3,4].
Currently, liver biopsy is the gold standard for diagnosing NAFLD [5]. However, liver biopsy is
an invasive procedure and can lead to serious complications (e.g., bleeding), and its diagnosis may
be inconsistent between pathologists [6,7]. Moreover, sampling errors limit the use of liver biopsy
in clinical practice. Additionally, most patients with NAFLD have no significant clinical symptoms,
and performing liver biopsies on such patients is ethically controversial. To resolve this dilemma,
noninvasive imaging modalities such as ultrasound, computed tomography, magnetic resonance
imaging, and magnetic resonance spectroscopy (MRS) are commonly used for the assessment of hepatic
steatosis [8]. Ultrasound imaging provides several advantages, including ease of routine examination,
cost-effectiveness, portability, and nonionizing imaging principles, and thus it is currently the first-line
modality for assessing hepatic steatosis and evaluating NAFLD.
Ultrasound performs well in detecting moderate to severe hepatic steatosis [9,10]. However,
its diagnostic accuracy for detecting mild hepatic steatosis is limited. Furthermore, qualitative
descriptions, operator experience, and interobserver and intraobserver variability degrade the
sonographic assessment of fatty liver [11,12]. Quantitative analysis of ultrasound images may provide
additional clues to improve the diagnosis of mild NAFLD. Essentially, liver parenchyma can be
modeled as a scattering medium consisting of numerous acoustic scatterers [13,14] that interact with
the incident wave to form ultrasound backscattered signals. Different scatterer properties result in
different waveforms of backscattered signals, and thus the corresponding statistical properties may
depend on information associated with changes in liver microstructures [13].
Considering the randomness of ultrasound backscattering, statistical distributions are widely used
to model backscattered statistics for tissue characterization [15]. Nakagami [16,17] and homodyned-K
distributions [18] have been applied to model ultrasound backscattered statistics for the assessment of
hepatic steatosis. However, acoustic structure quantification (ASQ) based on Chi-squared testing of
backscattered envelopes is the only technique that has been commercialized in ultrasound scanners
(Toshiba machine) by using the concept of statistical distribution. Initially, ASQ was developed to
quantify the difference between backscattered statistics and Rayleigh distribution [19]. ASQ has been
validated as having high performance in evaluating NAFLD because fat accumulation in the liver
tends to make the statistics of backscattered data follow the Rayleigh distribution [20–24].
When using ASQ or model-based methods to characterize tissue, the data used to estimate
the parameters must conform to the used statistical distribution [25,26]. This requirement may
not always be satisfied, because adjusting the settings in an ultrasound system or using nonlinear
signal-processing approaches (e.g., logarithmic compression) may alter the statistical distribution of
raw data. This limitation has motivated researchers to consider non-model-based statistical approaches.
Among all possible approaches, Shannon entropy—an estimate of signal uncertainty and complexity
proposed in information theory [27]—has the highest potential and flexibility for analyzing ultrasound
backscattering. Hughes first proposed using information (Shannon) entropy to analyze ultrasound
signals, indicating that entropy can be used to quantitatively depict changes in the microstructures
of scattering media [28,29]. In particular, one report demonstrated that information entropy can
describe ultrasound backscattered statistics without considering the statistical properties of ultrasound
data [30]. Recent studies have further indicated that entropy parametric imaging enables visualization
and characterization of hepatic steatosis, thereby making it possible to implement non-model-based
structure quantification of NAFLD [31–33].
While non-model-based entropy imaging plays an increasingly key role in physically describing
changes in the microstructures of fatty liver, its meanings require further biological explanation.
Entropy 2018, 20, 893 3 of 16

The establishment and validation of ultrasound entropy imaging to characterize hepatic steatosis are
based on the association of entropy value with hepatic histological changes [33]. However, NAFLD
is not only a change in liver microstructures caused by fat accumulation but also strongly related
to obesity, hypertension, type 2 diabetes mellitus, and dyslipidemia, all of which are metabolic
abnormalities and can be considered hepatic manifestations of metabolic syndrome (MetS) [34,35].
MetS is typically caused by insulin resistance, and although glucose clamp is the gold standard for
quantifying insulin resistance, it is a complex procedure that is unsuitable for routine use. For this
reason, Matthews et al. developed the homeostatic model assessment for insulin resistance (HOMA-IR)
index, which is calculated using fasting insulin and blood glucose for a general evaluation of MetS [36].
The HOMA-IR index correlates with the conventional ultrasound B-scan image features of hepatic
steatosis [37,38], implying that ultrasound imaging can depict metabolic information. Therefore,
we explored the relationship between MetS and quantitative ultrasound analysis of NAFLD by using
entropy imaging.
This study had two objectives: (i) investigating the association of ultrasound entropy imaging
of NAFLD with MetS to endow entropy images with new biological insights, and (ii) comparing the
performance of entropy imaging in predicting the risks of suffering from MetS with that of conventional
ASQ to determine whether non-model-based approaches are at all superior for evaluating MetS.
The results showed that ultrasound entropy imaging performed well in describing the metabolic
behavior of patients with NAFLD. Moreover, ultrasound entropy imaging was superior to ASQ in risk
evaluation for MetS.

2. Materials and Methods

2.1. Subjects
This study was conducted following approval by the Institutional Review Board of National
Taiwan University Hospital. All participants were asked to complete standardized questionnaires
and provided informed consent. Participants with the following conditions were excluded: excessive
alcohol intake (>20 g/day for women and >30 g/day for men) and chronic liver disease (chronic
hepatitis, autoimmune, drug-induced, vascular, or inherited hemochromatosis or Wilson disease).
A total of 394 patients were recruited.

2.2. Anthropometric Indices and Biochemical Analyses

Routine physical examinations and blood tests were conducted for each participant. Body mass
index (BMI) was calculated as weight divided by height squared. Waist circumference (WC) was
measured at the middle between the costal margin and iliac crest. Systolic blood pressure (SBP) and
diastolic blood pressure (DBP) were recorded. Fasting plasma glucose (FPG), total cholesterol (TCH),
triglycerides (TG), high-density lipoprotein (HDL-C), low-density lipoprotein (LDL-C), aspartate
aminotransferase (AST), alanine aminotransferase (ALT), and insulin were measured after 8 hours
of overnight fasting. Using FPG and insulin, the HOMA-IR index was calculated to examine insulin
resistance [36].

2.3. Diagnosis of MetS

Data obtained from anthropometric and blood examinations were further used to identify MetS.
According to the modified National Cholesterol Education Program Adult Treatment Panel III Criteria
(NCEP-ATP III), MetS (for the Taiwanese population) is diagnosed when at least three of the following
criteria are satisfied [39]: (i) WC ≥ 90 cm in men and ≥ 80 cm in women; (ii) SBP ≥ 130 mmHg or
DBP ≥ 85 mmHg or use medication for hypertension; (iii) hyperglycemia (FPG ≥ 100 mg/dL) or the
use of medication for diabetes; (iv) hypertriglyceridemia (TG ≥ 150 mg/dL) or use of medication for
hyperlipidemia; and (v) low HDL-C (≤40 mg/dL in men and ≤50 mg/dL in women).
Entropy 2018, 20, 893 4 of 16

2.4. Ultrasound Examinations for NAFLD Evaluation

After blood withdrawal, standard abdominal ultrasound screening of the liver was performed
immediately by three physicians, each with more than 20 years’ experience. A clinical ultrasound
scanner (Model 3000; Terason, Burlington, MA, USA) equipped with a convex transducer (Model
5C2A; Terason) of 3 MHz was used; the transducer had 128 elements and the pulse length of the
incident wave was approximately 2.3 mm. For each participant, the ultrasonographic fatty liver
indicator (US-FLI) was used as a semiquantitative measure of severity of NAFLD [40]. Specifically,
the US-FLI was calculated using the following criteria: (i) presence of liver/kidney contrast graded as
mild/moderate (score 2) or severe (score 3); (ii) presence (score 1 each) or absence (score 0 each) of
posterior attenuation of ultrasound beam, vessel blurring, difficult visualization of the gallbladder
wall, difficult visualization of the diaphragm, and areas of focal sparing. NAFLD was diagnosed if the
score ≥2 [40].

2.5. Quantitative Analysis using ASQ and Entropy Imaging

Except for the standard abdominal scans, all physicians followed the same protocols and system
settings for data acquisition and quantitative analysis. For each patient, the same scanner was used
to scan the liver through the subcostal scanning approach. It has been shown that a signal-to-noise
ratio (SNR) > 11 dB allows reliable descriptions of ultrasound backscattered statistics [41]. For this
consideration, the system gain index was set at 6, corresponding to a SNR of approximately 30 dB,
which was obtained from the calibrations in the previous study [42]. Such a high SNR implies that
no significant noise components exist in the backscattered signals, ensuring the quality of parameter
estimation in the ASQ analysis. The imaging depth was 16 cm and the focal zone corresponded to the
central part of the liver to reduce the effect of beam diffraction. Raw image data consisting of 128 scan
lines of backscattered radio frequency signals at a sampling rate of 30 MHz were obtained using the
software kit provided by Terason. The envelope image of each raw image raw datum was constructed
by taking the absolute value of the Hilbert transform of each scan line. The grayscale B-mode image
was formed based on the logarithm-compressed envelope data at a dynamic range of 40 dB.
In ASQ, the Chi-squared test is used to evaluate the difference between the sample and the
population data. The following equation is used to define parameter Cm 2 [19]:

σm2 
π
 2
σm
2
Cm = = (1)
σR2 (µm ) 4 − π µ2m

where µm and σm 2 are the average and variance of the measured backscattered envelopes, respectively.

The value of σR2 (µm ) indicates the variance of the Rayleigh-distributed data estimated using µm .
In this study, the sliding window technique was used to obtain a Cm 2 parametric map. In brief,

a window was created to move across the entire envelope image in steps representing the number of
pixels corresponding to the window overlap ratio (WOR); during this process, local parameters were
successively estimated using local envelope data within the window so that a parametric map could
eventually be constructed. The window side length (WSL) was three times the pulse length, which is
an appropriate size for stably estimating ultrasound statistical parameters [17]. The WOR was 50% to
provide a tradeoff between the parametric image resolution and computational time [43]. A region of
interest (ROI) manually outlined on the B-mode image of the liver parenchyma was used for analysis
of the Cm2 parametric map. Some basic criteria suggested previously were used for determining the

ROI [17]: (i) visible blood vessels were excluded in the ROI to reduce the bias of characterizing liver
parenchyma. The size of the ROI was set 3 × 3 cm2 ; (ii) the ROI was located at the focal zone, reducing
the effects of attenuation and diffraction on the backscattered signals.
Referring to a previous study [44], the histogram of Cm 2 in the ROI revealed a narrow

distribution when the tissue was homogeneous. A relatively broad distribution represented either
diffusely inhomogeneous (consisting of microstructures) or focally inhomogeneous (consisting of
Entropy 2018, 20, 893 5 of 16

Entropy 2018, 20, x 5 of 16

macrostructures such as vessels) tissue. To eliminate macrostructural information, 2 sliding window

processing of the envelope image was performed again (generating a second 2C m map denoted as 2
processing of the envelope image was performed again (generating a second Cm map denoted as rCm
map),rC m2 where
map), local
wheredata
localindata
theinwindow
the window werewere excluded
excluded if the
if the amplitude>>(µ
amplitude (μ + + ασ)
ασ) (μ:
(µ:mean
mean value
value
of of envelope
envelope dataininthe
data theROI,
ROI,σ:σ:standard
standard deviation
deviation of envelope
envelopedata dataininthe
theROI,
ROI,and andα:α:a aremoval
removal
of CCm 22/rC22 was lower than the threshold k,
coefficient).
coefficient). ForForeach
eachpixel
pixellocation
locationin inthe
theROI,
ROI, if
if the
the ratio of m / rCm m was lower than threshold
2
Cmk,was 2considered to exhibit no significant changes after rejecting the outliers of the envelope signals.
C m was considered to exhibit no significant changes after rejecting the outliers of the envelope
In this condition, Cm 2 was assigned to the pixel location. If C2 /rC2 was greater than k, rC2 was used.
2 m m 2 2 m
signals. In this condition,
Finally, the values of Cm 2 and C rCm 2 was assigned to the pixel location. If C m / rC m was greater than k,
m in the ROI were separated to construct two histograms to represent
rC m2 was used.
microstructure Finally,
(diffuse the values of or
inhomogeneity C m2homogeneity)
2
and rC m in and the ROI were separated
macrostructure to inhomogeneity)
(focal construct two
histograms
curves. to represent
The focal microstructure
disturbance (FD) ratio(diffuse inhomogeneity
was defined as the ratio or homogeneity)
of the area under and macrostructure
the curve for Cm 2

and (focal
rCm 2 inhomogeneity)
histograms, expressed curves. as The focal disturbance (FD) ratio was defined as the ratio of the area
2 2
under the curve for C m and rCm histograms, expressed as
2 histogram


AUC rCm
FD − ratio = ( 2 ℎ ) (2)
− = AUC (Cm histogram) (2)
( ℎ )
WhenWhen thetheresolution
resolutioncellcellofofthe
thetransducer
transducer contains
contains aa large
large number
numberofofrandomly
randomlydistributed
distributed
scatterers, the statistical distribution of ultrasound backscattered envelopes
scatterers, the statistical distribution of ultrasound backscattered envelopes exhibits the Rayleigh exhibits the Rayleigh
distribution
distribution [13–15].
[13–15].This
Thiscondition
conditionrepresents
represents that
that no macrostructuresexist
no macrostructures existininthe
thetissue
tissue toto generate
generate
information of rC 2 , and
2 thus the FD ratio is theoretically equal to zero. On the contrary, the FD ratio
information of mrCm , and thus the FD ratio is theoretically equal to zero. On the contrary, the FD ratio
increases with the degree of deviation from Rayleigh statistics [44]. Please note that the removal
increases with the degree of deviation from Rayleigh statistics [44]. Please note that the removal
coefficient α = 7 [19] and the threshold k = 1.2 [44] were suggested previously but could be empirically
coefficient α = 7 [19] and the threshold k = 1.2 [44] were suggested previously but could be empirically
determined [19,45]. Because the initial equipment (Toshiba system with software package) was
determined [19,45]. Because the initial equipment (Toshiba system with software package) was
unavailable
unavailable in in this study,
this study,wewefine-tuned
fine-tunedthe theparameters
parameters for the used
for the usedTerason
Terasonsystem.
system.The The values
values of of
αα
andandk were set at
k were set3 and
at 3 1.1,
andrespectively. The algorithmic
1.1, respectively. schemescheme
The algorithmic of estimating the FD ratio
of estimating the FD is illustrated
ratio is
in Figure 1.
illustrated in Figure 1.

Figure
Figure 1. 1.Computational
Computational flowchart
flowchart for
for ultrasound
ultrasound acoustic
acoustic structure
structurequantification
quantification(ASQ)
(ASQ)and
and
entropy
entropy estimations.
estimations.
Entropy 2018, 20, 893 6 of 16

The algorithm for ultrasound entropy imaging is also based on the sliding window technique to
process the envelope image and is illustrated in Figure 1. Because the acquired ultrasound backscattered
signals digitalized by the imaging system belong to discrete signals, the Shannon entropy of a discrete
random variable Y with possible values {y1 , y2 , . . . ., yn } (i.e., the envelope data points included within
the sliding window) was calculated using the following discrete form:
n
Hc ≡ − ∑ w(yi ) log2 [w(yi )] (3)
i =1

where w(·) represents the function of probability distribution. In this study, the statistical histogram of
the data (bins = 200) was used as an alternative w(·) for estimation [31,32]. To compare the results of
entropy with those of ASQ, ultrasound entropy imaging was constructed using the same WSL (6.9 mm)
and WOR (50%). The ROI used in the ASQ analysis was directly applied to the entropy parametric
image to calculate the average entropy value.

2.6. Statistical Analysis

The Kolmogorov-Smirnov, Anderson-Darling, Cramer-Von Mises, and Shapiro-Wilk tests of the
data (the US-FLI, FD ratio, and entropy) were used for normality testing. Tertiles were used to split the
data of the FD ratio and entropy into three groups. For each group, the categorical data were presented
as percentages and the continuous variables were expressed as mean ± standard deviation. Initially,
the interrelationships between the US-FLI, FD ratio, and entropy value were plotted to calculate the
Pearson correlation coefficient r and probability value p. Then, the categorical data were analyzed
using the Chi-squared test. The continuous variables in each group were compared using analyses of
variance. The Cochran-Armitage trend test was conducted to test for trends in the anthropometric and
metabolic factors by using tertiles of the FD ratio and entropy value. The associations of the FD ratio
and entropy value with MetS were assessed using a multiple logistic regression model adjusted for
age, sex, alcohol consumption, smoking, betel nut chewing, hours of exercise per week, menopause
status (women only), BMI, and HOMA-IR. To further compare the abilities of the FD ratio and entropy
in predicting the risk of suffering from MetS, the odds ratio (OR) and 95% confidence interval (CI)
were calculated. The significant difference was set at p < 0.05. All statistical analyses were conducted
using SAS version 9.3 (SAS Inc., Cary, NC, USA).

3. Results

3.1. Baseline Characteristics of the Participants

The baseline characteristics of the participants are shown in Table 1. A total of 394 participants
was recruited, comprising 151 (38.3%) men and 243 (61.7%) women (mean age: 40.5 ± 11.3 years).
According to information obtained from questionnaires, anthropometric examinations, blood tests,
and ultrasound evaluations of NAFLD, the overall prevalence of MetS was 19.3% and the US-FLI, FD
ratio, and entropy value of the participants were 2.22 ± 2.25, 0.96 ± 0.44, and 3.99 ± 0.06, respectively.
To observe how the statistical properties of backscattered signal varied with the severity of NAFLD, dot
and box plots of the FD ratio and entropy value corresponding to each US-FLI were plotted (Figure 2).
Based on observations of the data, exponential increasing and decreasing functions were used for
fitting dot plots of the entropy and FD ratio, respectively. With an increase in the US-FLI, the FD ratio
decreased (r = –0.630; p < 0.0001) and the entropy value monotonically increased (r = 0.713; p < 0.0001).
Box plots further identified outliers for the entropy and FD ratio. Some outliers were found to exist in
the data distributions of the entropy and FD ratio. This is acceptable and reasonable, especially for
a large amount of biodata (total n = 394). The US-FLI underestimating the extent of NAFLD [16] is
another possible reason for the outliers of entropy and FD ratio. On the other hand, the normality tests
based on four kinds of methods (as described in Section 2.6) indicated that the data of the US-FLI, FD
ratio, and entropy did not follow the normal distribution (p < 0.0001). However, the data distribution
Entropy 2018, 20, 893 7 of 16

does not affect the subsequent analysis (using the OR to evaluate the risk of MetS) because the OR
estimations were based on the tertiles of entropy and FD ratio.

Table 1. Patient characteristics (n = 394).

Variables Value *
Questionnaires
Gender F/M 243/151
Age (yrs) 40.5 ± 11.3 (20–72)
Menopause 25 (6.4)
Smoking
Never 336 (85.3)
Current 42 (10.7)
Previous 16 (4.1)
Alcohol
Never 322 (81.7)
Current 64 (16.2)
Previous 8 (2)
Betel Nuts
Never 375 (95.2)
Current 19 (4.8)
Exercise time (mins/per week) 99.6 ± 189.4 (0–1500)
Anthropometric variable
BMI (kg/m2 ) 24.1 ± 4.6 (14.8–43.7)
Waist (cm) 81.9 ± 11.3 (55–123)
SBP (mmHg) 122.5 ± 16.3 (86–180)
DBP (mmHg) 77.9 ± 11.9 (50–133)
Biochemistry parameters
FPG (mg/dL) 87.7 ± 17.6 (58–272)
TCH (mg/dL) 192.9 ± 35.5 (101–320)
TG (mg/dL) 112.4 ± 90.3 (25–888)
HDL-C (mg/dL) 57.3 ± 15.8 (25–120)
LDL-C (mg/dL) 120.8 ± 32.5 (47–238)
AST (U/L) 22.9 ± 8.9 (11–68)
ALT (U/L) 26.5 ± 21.4 (2–151)
Insulin (µU/mL) 9.1 ± 8.2 (2–84.4)
HOMA-IR 1.17 ± 1.03 (0.26–10.2)
MetS (%) 76 (19.3)
Ultrasound parameters
US-FLI Score 2.22 ± 2.25 (0–8)
ASQ FD-ratio 0.96 ± 0.44 (0.21–2.89)
Entropy 3.99 ± 0.06 (3.80–4.07)
* Categorical data are expressed as numbers (percentage); continuous variables are expressed as mean ± SD
(range). BMI: body mass index; SBP: systolic blood pressure; DBP: diastolic blood pressure; FPG: fasting plasma
glucose; TCH: total cholesterol; TG: triglycerides; HDL-C: high-density lipoprotein cholesterol; LDL-C: low-density
lipoprotein cholesterol; AST: aspartate aminotransferase; ALT: alanine aminotransferase; HOMA-IR: homeostasis
model assessment for insulin resistance; MetS: metabolic syndrome.
 Entropy 2018, 20, x 8 of 16

homeostasis
Entropy 2018, 20,model
893 assessment for insulin resistance; MetS: metabolic syndrome. 8 of 16

Figure2. (a)
Figure and and
2. (a) (b) Dot
(b)plots
Dot ofplots
entropy
of value (left)/FD
entropy value (focal
(left) disturbance)
/ FD (focalratio (right) corresponding
disturbance) ratio (right)
tocorresponding
each US-FLI. to (c) each
and (d) Box plots
US-FLI. of entropy
(c) and (d) Box value (left)/FD
plots of entropyratio
value(right)
(left) corresponding to
/ FD ratio (right)
each US-FLI.
corresponding to each US-FLI.

3.2. Characteristics of Participants in Different Tertiles

3.2. Characteristics of Participants in Different Tertiles
Table 2 shows the characteristics of participants in different tertiles of the FD ratio and entropy.
Table 2 shows
No significant the characteristics
difference in age was found of participants
(p = 0.2602).in different tertiles of
The percentage in the
menFD ratio andwith
decreased entropy.
an
increase in the FD ratio (p < 0.0001). Compared with the patients in higher tertiles, those in loweran
No significant difference in age was found (p = 0.2602). The percentage in men decreased with
increase
tertiles in the
(lower FDFD ratio exhibited
ratios) (p < 0.0001). Compared
lower HDL-C with the patients
(p < 0.0001) in higher
and higher WC,tertiles, thosefat,
BMI, body in lower
SBP,
DBP, FPG, TCH, TG, LDL-C, MetS, insulin, HOMA-IR, and abnormal liver function (p = 0.0353 SBP,
tertiles (lower FD ratios) exhibited lower HDL-C (p < 0.0001) and higher WC, BMI, body fat, for
DBP,pFPG,
TCH; TCH,
= 0.0022 forTG, LDL-C,
LDL-C; p <MetS,
0.0001insulin,
for theHOMA-IR, and abnormal
others). Similar results wereliverfound
function (p =tertiles
in the 0.0353 offor
TCH; pWith
entropy. = 0.0022 for LDL-C;
an increase p < 0.0001
in entropy (fromfor theto
lower others).
higher Similar
tertiles),results were
WC, BMI, found
body fat, in theDBP,
SBP, tertiles
FPG,of
entropy.
TCH, TG, With
LDL-C, an MetS,
increase in entropy
insulin, (from and
HOMA-IR, lower to higherliver
abnormal tertiles), WC,increased
function BMI, body fat,
(all p <SBP, DBP,
0.0001),
FPG,
and TCH,decreased
HDL-C TG, LDL-C, (p <MetS, insulin,
0.0001). These HOMA-IR, and abnormal
results revealed liver function
that the ultrasound increased
entropy value and(allFD
p<
0.0001), and HDL-C
ratio correlate with MetS. decreased (p < 0.0001). These results revealed that the ultrasound entropy value
and FD ratio correlate with MetS.
Entropy 2018, 20, 893 9 of 16

Table 2. Characteristics of participants in different tertiles of ultrasound quantitative parameters.

Entropy ASQ FD-ratio

Variables 1st tertile 2nd tertile 3rd tertile p-value 1st tertile 2nd tertile 3rd tertile p-value
No. of participants 131 131 132 131 131 132
Gender F/M 113/18 76/55 54/78 <0.0001 65/66 77/54 101/31 <0.0001
Age (yrs) 38.33 ± 9.87 41.6 ± 11.86 41.7 ± 11.81 0.009 41.40 ± 11.79 40.79 ± 11.02 39.40 ± 11.07 0.2602
Waist (cm) 73.31 ± 7.28 81.96 ± 9.9 90.43 ± 9.18 <0.0001 88.24 ± 10.98 82.61 ± 9.99 74.79 ± 8.32 <0.0001
BMI * (kg/m2 ) 20.83 ± 2.35 24.2 ± 4.05 27.29 ± 4.41 <0.0001 26.35 ± 4.47 24.61 ± 4.70 21.40 ± 2.78 <0.0001
SBP (mmHg) 114.88 ± 13.1 121.98 ± 16.97 130.58 ± 14.81 <0.0001 128.76 ± 16.74 122.99 ± 15.48 115.76 ± 14.06 <0.0001
DBP (mmHg) 73.58 ± 9.93 76.81 ± 12.44 83.36 ± 11.27 <0.0001 81.56 ± 11.95 77.82 ± 12.00 74.42 ± 10.86 <0.0001
FPG (mg/dL) 81.64 ± 8.36 86.95 ± 11.45 94.47 ± 25.39 <0.0001 92.85 ± 24.77 87.59 ± 13.60 82.70 ± 9.13 <0.0001
TCH (mg/dL) 181.29 ± 31.12 194.95 ± 35.77 202.33 ± 36.34 <0.0001 198.20 ± 36.30 195.49 ± 36.13 185.01 ± 32.87 0.0353
TG (mg/dL) 65.44 ± 28.39 106.46 ± 67.7 164.86 ± 118.71 <0.0001 151.11 ± 121.47 111.71 ± 72.79 74.63 ± 40.21 <0.0001
HDL-C (mg/dL) 65.66 ± 13.78 57.03 ± 14.59 49.17 ± 14.63 <0.0001 50.66 ± 12.33 56.26 ± 17.57 64.82 ± 13.83 <0.0001
LDL-C (mg/dL) 106.73 ± 27.43 123.6 ± 31.9 131.98 ± 32.79 <0.0001 127.57 ± 32.65 124.27 ± 31.24 110.64 ± 31.23 0.0022
AST (U/L) 19.60 ± 5.92 22.59 ± 8.48 26.37 ± 10.42 <0.0001 26.68 ± 10.72 22.49 ± 8.50 19.45 ± 5.06 <0.0001
ALT (U/L) 15.84 ± 7.53 25.01 ± 18.65 38.67 ± 26.67 <0.0001 39.51 ± 28.02 24.93 ± 16.31 15.26 ± 5.99 <0.0001
Insulin (µU/mL) 5.52 ± 3.4 9.06 ± 9.29 12.05 ± 8.7 <0.0001 10.71 ± 7.54 10.44 ± 10.63 5.7 ± 3.49 <0.0001
HOMA-IR 0.71 ± 0.44 1.16 ± 1.14 1.56 ± 1.10 <0.0001 1.39 ± 0.98 1.33 ± 1.31 0.73 ± 0.45 <0.0001
MetS (%) 1 (0.8%) 20 (15.3%) 55 (41.7%) <0.0001 44 (33.6%) 24 (18.3%) 8 (6.1%) <0.0001
US-FLI Score 0.53 ± 0.9 1.71 ± 1.55 4.42 ± 2.0 <0.0001 3.82 ± 2.40 2.18 ± 1.82 0.67 ± 1.09 <0.0001
* BMI: body mass index; SBP: systolic blood pressure; DBP: diastolic blood pressure; FPG: fasting plasma glucose; TCH: total cholesterol; TG: triglycerides; HDL-C: high-density lipoprotein
cholesterol; LDL-C: low-density lipoprotein cholesterol; AST: aspartate aminotransferase; ALT: alanine aminotransferase; HOMA-IR: homeostasis model assessment for insulin resistance;
MetS: metabolic syndrome.
Entropy 2018, 20, 893 10 of 16

3.3. The Risks of Metabolic Syndrome in Different Tertiles for the FD Ratio and the Entropy Value
The risk of metabolic syndrome in each tertile are compared in Table 3. For the FD ratio, the second
tertile (lower FD ratios) exhibited a higher risk of MetS (OR = 0.48; 95% CI: 0.26–0.89) than did the
third tertile (OR = 0.04; 95% CI: 0.01–0.14) after use of model 1 adjusted for age, sex, smoking, alcohol
consumption, betel nut chewing, hours of exercise per week, and menopause status (p < 0.0001).
Following further adjustment for BMI (model 2), the ORs in the second and third tertiles were
0.59 (95% CI: 0.30–1.18) and 0.41 (95% CI: 0.16–1.05), respectively (p = 0.1144). After use of HOMA-IR
to further adjust the OR (model 3), the ORs in the second and third tertile were 0.55 (95% CI: 0.27–1.14)
and 0.42 (95% CI: 0.15–1.17), respectively (p = 0.13). Notably, entropy improved the performance
of predicting the risk of MetS. Through use of model 1, the OR of entropy in the third tertile (85.57;
95% CI: 11.25–650.56) was larger than that in the second tertile (51.29; 95% CI: 2.76–164.43) (p < 0.0001).
After adjustment using model 2, the OR of entropy in the third tertile (26.84; 95% CI: 3.34–215.4) was
higher than that in the second tertile (10.27; 95% CI: 1.29–82.14) (p = 0.0007), as in model 3 (OR = 7.91,
95% CI: 0.96–65.18 for the second tertile; OR = 20.47, 95% CI: 2.48–168.67 for the third tertile; p = 0.0021).
The results indicated that non-model-based entropy provides a stronger link to biologically metabolic
information than does conventional ASQ.

Table 3. ORs in each tertile of entropy and the FD (focal disturbance) ratio for evaluating the risk
of MetS.

Entropy ASQ FD-ratio

1st tertile 2nd tertile 3rd tertile 1st tertile 2nd tertile 3rd tertile
p-value p-value
(n = 131) (n = 131) (n = 132) (n = 131) (n = 131) (n = 132)
51.29 85.57 0.48 0.04
Model 1 * ref <0.0001 ref <0.0001
(2.76–164.43) (11.25–650.56) (0.26–0.89) (0.01–0.14)
10.27 26.84 0.59 0.41
Model 2 ref 0.0007 ref 0.1144
(1.29–82.14) (3.34–215.4) (0.30–1.18) (0.16–1.05)
7.91 20.47 0.55 0.42
Model 3 ref 0.0021 ref 0.13
(0.96–65.18) (2.48–168.67) (0.27–1.14) (0.15–1.17)
* Model 1: adjusted for age, gender, smoking, alcohol consumption, betel nut chewing, hours of exercise per week,
and menopause status (women only). Model 2: same as model 1 plus further adjustment for BMI. Model 3: model 2
plus further adjustment for HOMA-IR.

4. Discussion

4.1. Significance of This Study

With the development and commercialization of ultrasound statistical models and parametric
imaging, physicians gradually have a new choice for diagnosing NAFLD. As stated in the Introduction,
more than one statistical distribution can be used to assess hepatic steatosis, and the ASQ technique
has the clinical benefit of using the model-based method to analyze the statistical properties of
backscattered signals from fatty liver. The best statistical distribution for modeling the backscattered
statistics of NAFLD is yet to be determined. However, ultrasound entropy imaging based on
information theory is more adaptive to various signal characteristics because the calculation of entropy
does not need to consider the statistical properties of the signal itself. Therefore, when viewing entropy
imaging as a new approach for NAFLD diagnosis, it is necessary to not only perform pathological
validations but also to explore the metabolic meanings of entropy. Studies have confirmed that the
value of ultrasonic entropy is closely related to the pathological changes of hepatic steatosis [32,33].
However, the present study expands our understanding and domain knowledge of ultrasound entropy
imaging; we demonstrated that ultrasound entropy imaging can describe the risk of MetS for those
with NAFLD and is superior to the conventional ASQ technique.
Entropy 2018, 20, 893 11 of 16

4.2. Effects of NAFLD on FD Ratio and Entropy

The US-FLI was used as a semiquantitative measure of NAFLD in this study. Our results
showed that both the FD ratio obtained from ASQ analysis and the entropy value of ultrasound
entropy imaging correlated with the US-FLI, indicating that these two parameters vary with the
progress of NAFLD because macrovesicular steatosis is the major pathological change of NAFLD.
Macrovesicular steatosis refers to the presence of a single large fat droplet in a hepatocyte that pushes
the nucleus to the periphery. In this scenario, the number of acoustic scatterers (fat droplets) increases
equivalently in the scattering medium (liver parenchyma), and the enhancement of constructive wave
interference results in changes in the waveforms of the backscattered signals, making the corresponding
backscattered statistics vary from pre-Rayleigh (backscattered statistics for healthy livers in practice)
to Rayleigh distribution (hepatic steatosis) [16,17,21]. This explains why the FD ratio ASQ parameter
monotonically decreases with an increase in the degree of hepatic steatosis. Concurrently, the effect of
constructive wave interference leads to increases in signal uncertainty and complexity, making the
entropy value [31,32].

4.3. Insulin Resistance: Bidirectional Link between MetS and NAFLD

In general, the increased prevalence of MetS is primarily a result of overnutrition and a sedentary
lifestyle. MetS is a key risk factor for cardiovascular disease incidence and mortality, as well as for
all-cause mortality [46]. The central etiological cause of MetS is commonly considered to be insulin
resistance, which is defined as the failure of insulin to stimulate glucose transport to its target cells [47].
Insulin is a pleiotropic hormone that regulates several cell functions, including stimulation of glucose
transport, cell growth, energy balance, and regulation of gene expression [48]. The functions of insulin
are associated with two signal pathways: the phosphatidylinositol 3-kinase-protein kinase B pathway
and mitogen-activated protein kinase pathway [49].
Once these signal pathways have been altered, insulin resistance is initialized. Free fatty acids
(FFAs) play a key role in the development of insulin resistance [49]. As insulin resistance develops,
a large quantity of plasma FFAs are released by white adipose tissues into the liver, leading to hepatic
fat accumulation [50]. At the same time, overnutrition and a sedentary lifestyle closely correlate with
the occurrence of NAFLD. In those who suffer from NAFLD, hepatic fat accumulation can result in
hepatic insulin resistance to strengthen the behavior of MetS [49]. Therefore, insulin resistance could
be considered the bidirectional link between MetS and NAFLD [49,51].

4.4. Superiority of Entropy in the Assessment of NAFLD and MetS

Several studies have clearly indicated that NAFLD is not only a cause of liver disease but also a key
risk indicator of cardiovascular disease [52–54]. Patients with both NAFLD and MetS have an increased
risk of cardiovascular disease [55]. For these reasons, a quantitative ultrasound parameter used for
evaluating NAFLD should satisfy two requirements: (i) changes in liver microstructures during fatty
infiltration can be described and explained from a histological viewpoint, and (ii) significant metabolic
information can be reflected to satisfy a variety of clinical applications. In this study, both the ASQ
and entropy imaging were shown to able to characterize NAFLD and MetS. Compared with the ASQ,
however, entropy imaging better fulfills the above two requirements, as supported by the current
results. First, the entropy value of ultrasound entropy imaging is more relevant than the FD ratio of
the ASQ to the US-FLI (Figure 2), representing that the entropy image characterizes NAFLD more
effectively. Second, the entropy value better predicted the risk of MetS than the FD ratio did (Table 3),
demonstrating that entropy imaging links metabolic information more strongly.
Possible mechanisms for why ultrasound entropy imaging provides improved performances
in evaluating NAFLD and MetS are discussed below. As mentioned in Section 4.3, NAFLD and
MetS interact with each other. Consequently, as long as ultrasound parameters can robustly and
precisely describe changes in the backscattered statistics during the process of fatty infiltration in
Entropy 2018, 20, 893 12 of 16

the liver, the opportunities to show more metabolic information increase. From this viewpoint,
ultrasound entropy imaging is superior to the ASQ technique. As reviewed in the Introduction,
the ASQ technique based on the analysis of ultrasound backscattered statistics is gaining attention
for the diagnosis of NAFLD. Some animal studies have revealed that ASQ has a high ability to detect
hepatic steatosis [20,24,45]; however, its value in quantifying the degree of hepatic steatosis in human
liver remains in dispute because of inconsistent findings. For example, Son et al. demonstrated that the
FD ratio correlated with the hepatic fat fraction (HFF) measured using MRS (r = −0.87; p < 0.001) [21],
whereas Karlas et al. found that the FD ratio did not significantly correlate with the HFF (r = −0.43;
p = 0.004) [22]. Failure to use the same procedures and settings for ASQ measurements may be one
cause for inconsistent findings. The criteria used for rejecting envelope signals and comparing Cm 2

and rCm 2 in the ASQ algorithm may also result in uncertainty in ASQ analysis [33] because these

criteria are empirically determined in practical applications [19,45]. By contrast, ultrasound entropy
imaging does not require additional signal rejection criteria, and thus it is less influenced by the
effects of computational settings and parameter tuning. In addition, the advantages of information
entropy lie in entropy estimation no longer being limited by the statistical properties of signals [30,32],
implying that entropy is a data-adaptive parameter for ultrasound tissue characterization. A relatively
simple but more adaptive computational scheme enables ultrasound entropy imaging to robustly and
stably perform NAFLD evaluations, as supported by histopathological validations of both the animal
model [32] and clinical trials [31,33]. These reasons explain why ultrasound entropy imaging correlates
with MetS more significantly than does ASQ. In other words, when entropy works for characterizing
NAFLD, it simultaneously provides significantly metabolic meanings that benefit evaluations in
various aspects.

4.5. Comparison with Related Studies

A novel parameter named the controlled attenuation parameter (CAP) has been developed based
on the properties of ultrasonic signals acquired by transient elastography (Fibroscan® ). The CAP was
demonstrated to correlate with fat accumulation in the liver [56,57] and facilitate the diagnosis of
hepatic steatosis [58,59]. Furthermore, one study found that the CAP correlated with several MetS
components [60]. However, the question of whether the CAP can perform well in NAFLD diagnosis
remains unanswered because unfavorable diagnoses have been reported [61–63]. This is likely because
the meaning of the CAP corresponds to the viscoelastic properties of the liver but does not provide
information on changes in the microstructure, which is crucial in the clinical evaluation of hepatic
steatosis. In the future, combining entropy imaging with the CAP may be a feasible strategy for a more
complete evaluation of NAFLD and MetS than either one alone.

4.6. Limitations of This Study

This study had two limitations. First, the effect of body habitus on the association of entropy
and ASQ analysis with MetS was not investigated. For instance, obesity may restrain quantitative
measurements of ultrasonography. Second, the original equipment (Toshiba systems and software
packages) was unavailable for ASQ analysis. Therefore, we implemented ASQ analysis by using
backscattered envelope data acquired from the Terason system. We cannot deny that some bias of
estimation accuracy of the FD ratio may have existed among system platforms. However, prior to this
study, we fine-tuned the parameters of the algorithm to mitigate this concern.

5. Conclusions
In this study, we performed ultrasound entropy imaging of NAFLD and studied its association
with MetS through comparisons with biochemical examinations and HOMA-IR (the measure of insulin
resistance). In addition, the dependency of entropy imaging on MetS was compared with that of the
conventional ASQ technique to explore possible strengths. The clinical results showed that the entropy
value was more closely correlated with the US-FLI than the FD ratio of ASQ analysis, indicating that
Entropy 2018, 20, 893 13 of 16

ultrasound entropy imaging improved the performance of NAFLD evaluation. Moreover, the entropy
value and FD ratio correlated with metabolic indices and HOMA-IR, thereby confirming metabolic
meanings of quantitative ultrasound. Notably, after adjustment for confounding factors, ultrasound
entropy still provided a stronger link to the risk of MetS than did the FD ratio according to statistical
OR analysis. Compared with model-based ASQ, ultrasound entropy imaging not only characterizes
liver microstructures but also reflects metabolic information of NAFLD more significantly. Thus,
ultrasound entropy imaging—a data-adaptive approach for tissue characterization—may play a key
role in the evaluation of NAFLD and MetS.

Author Contributions: Experimental design: K.-C.Y. and C.-K.Y.; methodology: P.-H.T.; software: Y.-H.L.; data
collection: Y.-Y.L.; data analysis: K.-C.Y.; manuscript preparation: Y.-H.L. and P.-H.T. All the authors have read
and approved the final manuscript.
Funding: This work was supported by the Ministry of Science and Technology in Taiwan (Grant Nos. MOST
106-2221-E-182-023-MY3, 106-2314-B-002-237-MY3, and 107-2218-E-182-004) and Chang Gung Memorial Hospital,
Linkou, Taiwan (Grant No. CMRPD1H0381).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Yu, A.S.; Keeffe, E.B. Nonalcoholic fatty liver disease. Rev. Gastroenterol. Disord. 2002, 2, 11–19. [CrossRef]
[PubMed]
2. Loomba, R.; Abraham, M.; Unalp, A.; Wilson, L.; Lavine, J.; Doo, E.; Bass, N.M. Association between diabetes,
family history of diabetes, and risk of nonalcoholic steatohepatitis and fibrosis. Hepatology 2012, 56, 943–951.
[CrossRef] [PubMed]
3. Rezvani, M.; Shaaban, A.M. Patterns of fatty liver disease. Curr. Radiol. Rep. 2016, 4, 26. [CrossRef]
4. Beeman, S.C.; Garbow, J.R. Fatty Liver Disease. In Imaging and Metabolism; Springer International Publishing
AG: Cham, Switzerland, 2018; pp. 223–241.
5. Bravo, A.A.; Sheth, S.G.; Chopra, S. Liver biopsy. N. Engl. J. Med. 2001, 344, 495–500. [CrossRef] [PubMed]
6. Sumida, Y.; Nakajima, A.; Itoh, Y. Limitations of liver biopsy and non-invasive diagnostic tests for the
diagnosis of nonalcoholic fatty liver disease/nonalcoholic steatohepatitis. World J. Gastroenterol. 2014,
20, 475–485. [CrossRef] [PubMed]
7. Nalbantoglu, I.L.; Brunt, E.M. Role of liver biopsy in nonalcoholic fatty liver disease. World J. Gastroenterol.
2014, 20, 9026–9037. [CrossRef] [PubMed]
8. Ma, X.Z.; Holalkere, N.S.; Kambadakone, R.A.; Mari, M.K.; Hahn, P.F.; Sahani, D.V. Imaging-based
quantification of hepatic fat: Methods and clinical applications. Radiographics 2009, 29, 1253–1280. [CrossRef]
[PubMed]
9. Saadeh, S.; Younossi, Z.M.; Remer, E.M.; Gramlich, T.; Ong, J.P.; Hurley, M. The utility of radiological imaging
in nonalcoholic fatty liver disease. Gastroenterology 2002, 123, 745–750. [CrossRef] [PubMed]
10. Lee, S.S.; Park, S.H.; Kim, H.J.; Kim, S.Y.; Kim, M.Y.; Kim, D.Y. Non-inva¬sive assessment of hepatic steatosis:
Prospective comparison of the accuracy of imaging examinations. J. Hepatol. 2010, 52, 579–585. [CrossRef]
[PubMed]
11. Strauss, S.; Gavish, E.; Gottlieb, P.; Katsnelson, L. Interobserver and intraobserver variability in the
sonographic assessment of fatty liver. AJR 2007, 189, 320–323. [CrossRef] [PubMed]
12. Cengiz, M.; Sentürk, S.; Cetin, B.; Bayrak, A.H.; Bilek, S.U. Sonographic assessment of fatty liver:
Intraobserver and interobserver variability. Int. J. Clin. Exp. Med. 2014, 7, 5453–5460. [PubMed]
13. Mamou, J.; Oelze, M.L. Quantitative Ultrasound in Soft Tissues; Springer: Dordrecht, The Netherlands;
New York, NY, USA, 2013.
14. Tsui, P.H.; Zhou, Z.; Lin, Y.H.; Hung, C.M.; Chung, S.J.; Wan, Y.L. Effect of ultrasound frequency on the
Nakagami statistics of human liver tissues. PLoS ONE 2017, 12, e0181789. [CrossRef] [PubMed]
15. Destrempes, F.; Cloutier, G. A critical review and uniformized representation of statistical distributions
modeling the ultrasound echo envelope. Ultrasound Med. Biol. 2010, 36, 1037–1051. [CrossRef] [PubMed]
Entropy 2018, 20, 893 14 of 16

16. Liao, Y.Y.; Yang, K.C.; Lee, M.J.; Huang, K.C.; Chen, J.D.; Yeh, C.K. Multifeature analysis of an ultrasound
quantitative diagnostic index for classifying nonalcoholic fatty liver disease. Sci. Rep. 2016, 6, 35083.
[CrossRef] [PubMed]
17. Wan, Y.L.; Tai, D.I.; Ma, H.Y.; Chiang, B.H.; Chen, C.K.; Tsui, P.H. Effects of fatty infiltration in human livers
on the backscattered statistics of ultrasound imaging. Proc. Inst. Mech. Eng. H 2015, 229, 419–428. [CrossRef]
[PubMed]
18. Fang, J.; Zhou, Z.; Chang, N.F.; Wan, Y.L.; Tsui, P.H. Ultrasound parametric imaging of hepatic steatosis
using the homodyned-K distribution: An animal study. Ultrasonics 2018, 87, 91–102. [CrossRef] [PubMed]
19. Toyoda, H.; Kumada, T.; Kamiyama, N.; Shiraki, K.; Takase, K.; Yamaguchi, T.; Hachiya, H. B-mode
ultrasound with algorithm based on statistical analysis of signals: Evaluation of liver fibrosis in patients
with chronic hepatitis C. Am. J. Roentgenol. 2009, 193, 1037–1043. [CrossRef] [PubMed]
20. Kuroda, H.; Kakisaka, K.; Kamiyama, N.; Oikawa, T.; Onodera, M.; Sawara, K.; Oikawa, K.; Endo, R.;
Takikawa, Y.; Suzuki, K.; et al. Non-invasive determination of hepatic steatosis by acoustic structure
quantification from ultrasound echo amplitude. World J. Gastroenterol. 2012, 18, 3889–3895. [CrossRef]
[PubMed]
21. Son, J.Y.; Lee, J.Y.; Yi, N.J.; Lee, K.W.; Suh, K.S.; Kim, K.G.; Lee, J.M.; Han, J.K.; Choi, B.I. Hepatic steatosis:
Assessment with acoustic structure quantification of US imaging. Radiology 2016, 278, 257–264. [CrossRef]
[PubMed]
22. Karlas, T.; Berger, J.; Garnov, N.; Lindner, F.; Busse, H.; Linder, N.; Schaudinn, A.; Relke, B.; Chakaroun, R.;
Tröltzsch, M.; et al. Estimating steatosis and fibrosis: Comparison of acoustic structure quantification with
established techniques. World J. Gastroenterol. 2015, 21, 4894–4902. [CrossRef] [PubMed]
23. Keller, J.; Kaltenbach, T.E.; Haenle, M.M.; Oeztuerk, S.; Graeter, T.; Mason, R.A. Comparison of acoustic
structure quantification (ASQ), shearwave elastography and histology in patients with diffuse hepatopathies.
BMC Med. Imaging 2015, 15, 58. [CrossRef] [PubMed]
24. Lee, D.H.; Lee, J.Y.; Lee, K.B.; Han, J.K. Evaluation of hepatic steatosis by using acoustic structure
quantification US in a rat model: comparison with pathologic examination and MR spectroscopy. Radiology
2017, 285, 445–453. [CrossRef] [PubMed]
25. Shankar, P.M. A general statistical model for ultrasonic backscattering from tissues. IEEE Trans. Ultrason.
Ferroelectr. Freq. Control. 2000, 47, 727–736. [CrossRef] [PubMed]
26. Smolikova, R.; Wachowiak, M.P.; Zurada, J.M. An information-theoretic approach to estimating ultrasound
backscatter characteristics. Comput. Biol. Med. 2004, 34, 355–370. [CrossRef]
27. Shannon, C.E. A Mathematical Theory of Communication. Bell Syst. Tech. J. 1948, 27, 379–423. [CrossRef]
28. Hughes, M.S. Analysis of digitized waveforms using Shannon entropy. J. Acoust. Soc. Am. 1993, 93, 892–906.
[CrossRef]
29. Hughes, M.S.; McCarthy, J.E.; Marsh, J.N.; Arbeit, J.M.; Neumann, R.G.; Fuhrhop, R.W.; Wallace, K.D.;
Znidersic, D.R.; Maurizi, B.N.; Baldwin, S.L.; et al. Properties of an entropy-based signal receiver with an
application to ultrasonic molecular imaging. J. Acoust. Soc. Am. 2007, 121, 3542–3557. [CrossRef] [PubMed]
30. Tsui, P.H. Ultrasound detection of scatterer concentration by weighted entropy. Entropy 2015, 17, 6598–6616.
[CrossRef]
31. Tsui, P.H.; Wan, Y.L. Effects of fatty infiltration of the liver on the Shannon entropy of ultrasound
backscattered signals. Entropy 2016, 18, 341. [CrossRef]
32. Fang, J.; Chang, N.F.; Tsui, P.H. Performance evaluations on using entropy of ultrasound log-compressed
envelope images for hepatic steatosis assessment: An in vivo animal study. Entropy 2018, 20, 120. [CrossRef]
33. Zhou, Z.; Tai, D.I.; Wan, Y.L.; Tseng, J.H.; Lin, Y.R.; Wu, S.; Yang, K.C.; Liao, Y.Y.; Yeh, C.K.; Tsui, P.H.
Hepatic steatosis assessment with ultrasound small-window entropy imaging. Ultrasound Med. Biol. 2018,
44, 1327–1340. [CrossRef] [PubMed]
34. Leite, N.C.; Salles, G.F.; Araujo, A.L.; Cristiane, V.N.; Cardoso, C.R. Prevalence and associated factors
of nonalcoholic fatty liver disease in patients with type-2 diabetes mellitus. Liver Int 2009, 29, 113–119.
[CrossRef] [PubMed]
35. Fabbrini, E.; Sullivan, S.; Klein, S. Obesity and nonalcoholic fatty liver disease: biochemical, metabolic,
and clinical implications. Hepatology 2010, 51, 679–689. [CrossRef] [PubMed]
Entropy 2018, 20, 893 15 of 16

36. Matthews, D.R.; Hosker, J.P.; Rudenski, A.S.; Naylor, B.A.; Treacher, D.F.; Turner, R.C. Homeostasis model
assessment: Insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations
in man. Diabetologia 1985, 28, 412–419. [CrossRef] [PubMed]
37. Cruz, M.A.F.; Cruz, J.F.; Macena, L.B.; Santana, D.S.; Oliveira, C.C.; Lima, S.O.; Franca, A.V. Association of
the nonalcoholic hepatic steatosis and its degrees with the values of liver enzymes and homeostasis model
assessment-insulin resistance index. Gastroenterol. Res. 2015, 8, 260–264. [CrossRef] [PubMed]
38. Isaksen, V.T.; Larsen, M.A.; Goll, R.; Florholmen, J.R.; Paulssen, E.J. Hepatic steatosis, detected by hepatorenal
index in ultrasonography, as a predictor of insulin resistance in obese subjects. BMC Obes. 2016, 3, 39.
[CrossRef] [PubMed]
39. Yang, K.C.; Hung, H.F.; Lu, C.W.; Chang, H.H.; Lee, L.T.; Huang, K.C. Association of non-alcoholic fatty liver
disease with metabolic syndrome independently of central obesity and insulin resistance. Sci. Rep. 2016,
6, 27034. [CrossRef] [PubMed]
40. Ballestri, S.; Lonardo, A.; Romagnoli, D.; Carulli, L.; Losi, L.; Day, C.P.; Loria, P. Ultrasonographic fatty liver
indicator, a novel score which rules out NASH and is correlated with metabolic parameters in NAFLD.
Liver Int. 2012, 32, 1242–1252. [CrossRef] [PubMed]
41. Tsui, P.H.; Wang, S.H.; Huang, C.C.; Chiu, C.Y. Quantitative analysis of noise influence on the detection of
scatterer concentration by Nakagami parameter. J. Med. Biol. Eng. 2005, 25, 45–51.
42. Zhou, Z.; Wu, W.; Wu, S.; Jia, K.; Tsui, P.H. Empirical mode decomposition of ultrasound imaging for
gain-independent measurement on tissue echogenicity: A feasibility study. Appl. Sci. Basel 2017, 7, 324.
[CrossRef]
43. Tsui, P.H.; Chen, C.K.; Kuo, W.H.; Chang, K.J.; Fang, J.; Ma, H.Y.; Chou, D. Small-window parametric imaging
based on information entropy for ultrasound tissue characterization. Sci. Rep. 2017, 7, 41004. [CrossRef]
[PubMed]
44. Yakoshi, Y.; Kudo, D.; Toyoki, Y.; Isido, K.; Kimura, N.; Wakiya, T.; Sakuraba, S.; Yoshizawa, T.; Sakamoto, Y.;
Kijima, H.; et al. Non-invasive quantification of liver damage by a novel application for statistical analysis of
ultrasound signals. Hirosaki Med. J. 2014, 65, 199–208.
45. Shen, C.C.; Yu, S.C.; Liu, C.Y. Using high-frequency ultrasound statistical scattering model to assess
nonalcoholic fatty liver disease (NAFLD) in mice. IEEE Ultrason. Symp. Proc. 2016, 1, 379–382. [CrossRef]
46. Galassi, A.; Reynolds, K.H.J. Metabolic syndrome and risk of cardiovascular disease: A meta-analysis.
Am. J. Med. 2006, 119, 812–819. [CrossRef] [PubMed]
47. Bugianesi, E.; McCullough, A.J.; Marchesini, G. Insulin resistance: A metabolic pathway to chronic liver
disease. Hepatology 2005, 42, 987e1000. [CrossRef] [PubMed]
48. Carl, L.; Olefsky, J.M. Inflammation and insulin resistance. FEBS Lett. 2008, 582, 97e105. [CrossRef]
49. Asrih, M.; Jornayvaz, F.R. Metabolic syndrome and nonalcoholic fatty liver disease: Is insulin resistance the
link? Mol. Cell Endocrinol. 2015, 418, 55–65. [CrossRef] [PubMed]
50. Donnelly, K.L.; Smith, C.I.; Schwarzenberg, S.J.; Jessurun, J.; Boldt, M.D.; Parks, E.J. Sources of fatty acids
stored in liver and secreted via lipoproteins in when cells fail to respond normally to the hormone insulin in
patients with nonalcoholic fatty liver disease. J. Clin. Invest 2005, 115, 1343–1351. [CrossRef] [PubMed]
51. Bugianesi, E.; Moscatiello, S.; Ciaravella, M.F.; Marchesini, G. Insulin resistance in nonalcoholic fatty liver
disease. Curr. Pharm. Des. 2010, 16, 1941–1951. [CrossRef] [PubMed]
52. Luo, J.; Xu, L.; Li, J.; Zhao, S. Nonalcoholic fatty liver disease as a potential risk factor of cardiovascular
disease. Eur. J. Gastroenterol. Hepatol. 2015, 27, 193–199. [CrossRef] [PubMed]
53. Pisto, P.; Santaniemi, M.; Bloigu, R.; Ukkola, O.; Kesäniemi, Y.A. Fatty liver predicts the risk for cardiovascular
events in middle-aged population: A population-based cohort study. BMJ Open 2014, 4, e004973. [CrossRef]
[PubMed]
54. Motamed, N.; Rabiee, B.; Poustchi, H.; Dehestani, B.; Hemasi, G.R.; Khonsari, M.R.; Maadi, M.; Saeedian, F.S.;
Zamani, F. Non-alcoholic fatty liver disease (NAFLD) and 10-year risk of cardiovascular diseases. Clin. Res.
Hepatol. Gastroenterol. 2017, 41, 31–38. [CrossRef] [PubMed]
55. Misra, V.L.; Khashab, M.; Chalasani, N. Non-alcoholic fatty liver disease and cardiovascular risk.
Curr. Gastroenterol. Rep. 2009, 11, 50–55. [CrossRef] [PubMed]
56. Wang, Y.; Fan, Q.; Wang, T.; Wen, J.; Wang, H.; Zhang, T. Controlled attenuation parameter for assessment of
hepatic steatosis grades: A diagnostic meta-analysis. Int. J. Clin. Exp. Med. 2015, 8, 17654–17663. [PubMed]
Entropy 2018, 20, 893 16 of 16

57. Sasso, M.; Beaugrand, M.; Ledinghen, V.; Douvin, C.; Marcellin, P.; Poupon, R.; Sandrin, L.; Miette, V.
Controlled attenuation parameter (CAP): A novel VCTE guided ultrasonic attenuation measurement for the
evaluation of hepatic steatosis: preliminary study and validation in a cohort of patients with chronic liver
disease from various causes. Ultrasound Med. Biol. 2010, 36, 1825–1835. [CrossRef] [PubMed]
58. Ledinghen, V.; Vergniol, J.; Foucher, J.; Merrouche, W.; Bail, B. Non-invasive diagnosis of liver steatosis using
controlled attenuation parameter (CAP) and transient elastography. Liver Int. 2012, 32, 911–918. [CrossRef]
[PubMed]
59. Mikolasevic, I.; Orlic, L.; Franjic, N.; Hauser, G.; Stimac, D.; Milic, S. Transient elastography (FibroScanR)
with controlled attenuation parameter in the assessment of liver steatosis and fibrosis in patients with
nonalcoholic fatty liver disease: Where do we stand? World J. Gastroenterol. 2016, 22, 7236–7251. [CrossRef]
[PubMed]
60. Mikolasevic, I.; Milic, S.; Orlic, L.; Stimac, D.; Franjic, N.; Targher, G. Factors associated with significant liver
steatosis and fibrosis as assessed by transient elastography in patients with one or more components of the
metabolic syndrome. J. Diabetes Complic. 2016, 30, 1347–1353. [CrossRef] [PubMed]
61. Imajo, K.; Kessoku, T.; Honda, Y.; Tomeno, W.; Ogawa, Y.; Mawatari, H.; Fujita, K.; Yoneda, M.; Taguri, M.;
Hyogo, H.; et al. Magnetic resonance imaging more accurately classifies steatosis and fibrosis in patients with
nonalcoholic fatty liver disease than transient elastography. Gastroenterology 2016, 150, 626–637. [CrossRef]
[PubMed]
62. Myers, R.P.; Pollett, A.; Kirsch, R.; Pomier-Layrargues, G.; Beaton, M.; Levstik, M.; Duarte-Rojo, A.; Wong, D.;
Crotty, P.; Elkashab, M. Controlled attenuation parameter (CAP): A noninvasive method for the detection of
hepatic steatosis based on transient elastography. Liver Int. 2012, 32, 902–910. [CrossRef] [PubMed]
63. Kumar, M.; Rastogi, A.; Singh, T.; Behari, C.; Gupta, E.; Garg, H.; Kumar, R.; Bhatia, V.; Sarin, SK. Controlled
attenuation parameter for non-invasive assessment of hepatic steatosis: Does etiology affect performance?
J. Gastroenterol. Hepatol. 2013, 28, 1194–1201. [CrossRef] [PubMed]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).