DOI:10.2214/AJR.07.3829
AJR 2008; 190:W372-W373
© American Roentgen Ray Society
Reply
Adrian S
ftoiu,
Dan Ionu
Gheonea and
Tudorel Ciurea
University of Medicine and Pharmacy, Craiova, Romania
WEB—This is a Web exclusive article.
The article "Real-Time Elastography for Noninvasive Assessment of
Liver Fibrosis in Chronic Viral Hepatitis," published by Friedrich-Rust
et al. in 2007 in the American Journal of Roentgenology
[1], induced an ongoing
discussion on the possible role of real-time elastography for the assessment
of liver fibrosis
[2-4].
On the basis of this discussion, we would like to make the following
comments.
We have previously suggested that the region of interest of elastography
should be placed to include the surrounding liver tissues (i.e., fatty tissue,
intercostal muscles, diaphragm, peritoneum, etc.). Although we have not yet
published our results, we have used a hue histogram analysis separately for
the liver and surrounding tissues by using different regions inside the
elastography region of interest (Fig.
1). On the basis of a previously described approach for dynamic
analysis [5], we included in
the analysis all color frames from a 10-second cine-loop elastography in an
attempt to prevent variability and artifacts through averaging. After
inclusion of 97 consecutive patients, the method seems to moderately
differentiate liver steatosis, chronic hepatitis, and advanced liver
cirrhosis, with an average accuracy of approx imately 75%, comparable to the
results of liver biopsy. A separate subgroups analysis of chronic hepatitis
patients indicated that real-time elastography was not able to accurately dif
ferentiate the individual degree of liver fibrosis (F0 to F4) for each
patient.
View larger version (77K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 1 —Sonoelastography of right liver lobe in 58-year-old woman
with advanced liver cirrhosis. Region of interest is placed on edge of liver
parenchyma and on surrounding tissues. Hue histogram analysis was subsequently
performed separately for liver tissue as a semiquantitative method for
assessment of liver elasticity (mean hue = 201.27, indicating the presence of
advanced fibrosis). Skewed distribution of colors toward hard values also
suggests severe fibrosis.
|
|
We certainly agree with Gulizia et al.
[6] that using a relative
comparison between the liver and the surrounding tissues (in a more or less
similar way to the strain ratio proposed by the manufacturers) might yield
better results. We also agree that when significant fibrosis appears, the
normal distribution of the colors is skewed toward hard values
(Fig. 1). Consequently, we
think that other statistical and mathematic tools should be used for the
analysis of elastography images. Such options are represented by the use of
tailored artificial neural networks that might better adapt to a vector type
of information yielded by the use of dynamic hue histogram analysis, whereas
other artificial intelligence techniques, such as evolutionary support vector
machines or evolutionary classifiers, should also be tested in the future,
specifically for liver patients.
We do not agree with Gulizia et al.
[6] that inclusion of the right
branches of the hepatic vein would be very useful because the presence of
vessels induces clear artifacts in the elastography images
(Fig. 2), similar to the
presence of ascites (Fig. 3).
This is easily understandable if we look at the relative values shown by the
real-time elastography software in the region of interest. In the presence of
a very soft (highly elastic) structure such as the hepatic vein or ascites,
the rest of the liver would be depicted as hard, irrespective of its
elasticity (strain).
View larger version (80K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 2 —Sonoelastography of right liver lobe in 47-year-old man with
liver steatosis. Region of interest is placed inside liver to include large
branch of hepatic veins. Liver parenchyma has hard appearance (dark
blue) when extremely soft (red) elastic vessel is interposed in
sonography section.
|
|
View larger version (51K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 3 —Sonoelastography of right liver lobe in 53-year-old man with
advanced liver cirrhosis and small amount of ascites surrounding liver. Due to
presence of very soft (red) and elastic fluid near liver capsule, liver
parenchyma is also depicted as very hard (dark blue). Consequently,
other types of ascites (for example peritoneal carcinomatosis) might possibly
induce same artifact, even in presence of normal liver tissue.
|
|
One limitation of our approach was the examination of the liver with a
linear transducer of 6.5 MHz, which might be too high to examine correctly and
consistently the right liver lobe. A better option might be represented by the
use of a lower frequency linear transducer. Development of a pressure gauge is
certainly necessary because manual application of pressure cannot be
standardized. Usually, a small deformation (below 2%) of the tissues is
needed, and this is very difficult to obtain, even by experienced
sonographers. Moreover, the utility of the compression scale provided by the
manufacturers (with values that should be approximately 3 to 4 on a zero to 6
arbitrary scale) is also questionable.
Another important limitation of our approach was determined by the
examination through the right-side intercostal spaces. This made it impossible
to depict elastography information inside the liver in 42 of the 97 patients
(43.3%), especially in the presence of a large body habitus as well as an
increased width of the thoracic wall (Fig.
4). The penetration of real-time elastography is limited to 3-4
cm, so it is quite difficult to record useful elastography information inside
the liver if the thoracic wall is thicker than 2-3 cm. As suggested by Gulizia
et al. [6], changes in the
elastography parameters (frame rate, elastography-dynamic range, persistence,
etc.) and revision of thresholds that reflect degrees of displacement
representative for the liver might be necessary.
View larger version (51K):
[in this window]
[in a new window]
[as a PowerPoint slide]
|
Fig. 4 —Sonoelastography of right liver lobe in 66-year-old obese
woman with chronic hepatitis with increased distance from skin to liver
capsule (> 2.5 cm). Real-time elastography software was not able to
correctly characterize elasticity inside liver.
|
|
One component of our ongoing study was to test the intra- and interobserver
variability of the method. We have analyzed three in dependent cine-loop
elastography examinations recorded by two separate examiners who were blinded
to each other as well as to the clinical informa tion and the liver biopsy
results. For each patient, we thus recorded six real-time cine-loop
elastography examinations, which were further analyzed through hue histogram
analysis, with averaged values for a 10-second cine loop. For the initial 55
patients with valid real-time elastography recordings inside the liver, we
have obtained good intra- and inter observer variability values, with kappa
values between 0.41 and 0.60, indicating moderate agreement. This was slightly
higher than the values reported by Gulizia et al.
[6], who re ported an
intraobserver variability of about 40% on the basis of single-image analysis.
We previously reported that analysis of single images is biased by the
presence of artifacts, whereas average images are significantly better and
exclude inconsistencies generated by arbitrary selection of the elastography
frames [5].
In accordance with the findings of Gulizia et al.
[6], we also were not able to
distinguish between intermediate degrees of liver fibrosis (F0 to F4).
However, this may not have been the case if the methodology had been improved
through the use of novel transducers and pressure gauge detectors (and
consequent absolute values of elastography information) as well as significant
improvement of the software with real-time calculations of hue histograms. The
addition of artificial intelligence techniques and automated computer-aided
diagnosis might add even more to a method that is extremely difficult to
follow in real-time, especially during dynamic assessment by an untrained
examiner, biased by its own selection of the "best" image.
In conclusion, we would strongly suggest that real-time elastography should
be compared with other noninvasive markers—transient elastography and
liver biopsy results—in a large multicentric study with an improved
methodology that should take into account previous observations made by
different authors (better transducers, im proved software, etc.). Until then,
real-time elastography used for evaluation of liver fibrosis remains an
exciting technique with an incompletely discovered potential.
References
- Friedrich-Rust M, Ong M, Hermann E, et al. Real-time elastography
for noninvasive assessment of liver fibrosis in chronic viral hepatitis.
AJR 2007; 188:758
-764[Abstract/Free Full Text]
- Ferraioli G, Gulizia R, Filice C. Real-time elastography in the
assessment of liver fibrosis. (letter) AJR2007; 189:W170[Free Full Text]
- Sftoiu A, Gheonea DI, Ciurea T. Hue histogram analysis of
real-time elastography images for noninvasive assessment of liver fibrosis.
(letter) AJR 2007;189
: W232-233[Free Full Text]
- Friedrich-Rust M, Hermann E, Zeuzem S, Sarrazin C. Reply to
real-time elastography in the assessment of liver fibrosis. (letter)
AJR 2008;190:W164[Free Full Text]
- Sftoiu A, Vilmann P, Ciurea T, et al. Dynamic analysis of
endoscopic ultrasound (EUS) elastography used for the differentiation of
benign and malignant lymph nodes. Gastrointest Endosc2007; 66:291
-300[CrossRef][Medline]
- Gulizia R, Ferraioli G, Filice C. Open questions in the assessment
of liver fibrosis using real-time elastography. (letter)
AJR 2008; 190:W370
-W371[Free Full Text]
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
Top
References
