| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Gastrointestinal Imaging |
1 From the Department of Radiology (S.H.K., J.M.L., J.H.K., J.K.H., K.H.L., S.H.P., S.K.A., Y.J.K., K.R.S., B.I.C.), Institute of Radiation Medicine (J.M.L., J.H.K., J.K.H., B.I.C.), Department of Medical Engineering (J.H.K., K.G.K.), Department of Surgery (N.J.Y., K.S.S.), and Department of Pathology (H.S.L.), Seoul National University College of Medicine, 28 Yongon-dong, Chongno-gu, Seoul 110-744, Korea; and Department of Radiology, Seoul National University Bundang Hospital, Seoul, Korea (K.H.L.). Received January 25, 2004; revision requested April 1; revision received May 12; accepted June 15. Address correspondence to J.M.L. (e-mail: leejm@radcom.snu.ac.kr).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Institutional ethics committee approved study; written informed consent was obtained. ANNs, constructed with three-layered 15-neuron back-propagation algorithm, were trained to predict appropriateness of a donor liver with respect to macrosteatosis by using statistically significant laboratory and US parameters derived from univariate analyses, together with correct diagnosis. Input variables for ANNs were alkaline phosphatase, glutamic oxaloacetic transaminase, glutamic pyruvate transaminase, 
-glutamyltransferase, hepatorenal ratio of echogenicity, and tail area ratio and tail length of portal vein wall echogenicity. Three radiologists graded US images in 94 potential donors (71 men and 23 women) on the basis of four degrees of hepatic steatosis. After training and testing of ANNs, performance of ANNs and radiologists in predicting appropriateness of potential donors was evaluated with receiver operating characteristic (ROC) analysis and compared by means of univariate z score test.
RESULTS: Among 94 potential donor livers, 76 were normal or had mild steatosis, and 18 had moderate or severe macrosteatosis at histopathologic examination. Area under ROC curve (Az) of ANNs (Az = 0.9673) was significantly greater than that of radiologists (faculty, Az = 0.9106, P = .048; fellow, Az = 0.9038, P = .044; resident, Az = 0.8931, P = .038). No statistically significant difference in sensitivity for predicting appropriateness as a liver donor with respect to macrosteatosis was found between ANNs (88.9%) and radiologists (P > .05). However, specificity of ANNs (96.1%) was significantly better than that of radiologists (P < .003).
CONCLUSION: ANNs might be a useful tool to categorize whether a donor liver is appropriate for transplantation with respect to macrosteatosis on the basis of multiple variables related to laboratory and US features. Further study is needed.
© RSNA, 2005
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Hepatic steatosis can be characterized quantitatively (mild, moderate, or severe) and qualitatively (macro- or microvesicular) by means of histopathologic examination. Although the impact of macro- or microvesicular hepatic steatosis on liver injury remains controversial, there is currently a general consensus in the hepatic transplant community that moderate to severe macrosteatosis (>30%) increases the risk of postoperative complications and patient death after liver transplantation (7–11). Therefore, a clinical consensus exists that predicting the appropriateness of a donor liver with respect to macrosteatosis preoperatively is necessary.
In the diagnosis of hepatic steatosis, ultrasonography (US) is simple and provides useful information (12–14). Computed tomography (CT) and magnetic resonance (MR) imaging are excellent for detection of fat deposition and are useful imaging methods for the assessment of hepatic steatosis (15–20). To date, many studies have focused on whether these imaging modalities could allow quantification of hepatic steatosis. However, only one study has focused on predicting the appropriateness of a donor liver with respect to hepatic steatosis, to our knowledge (16). In addition, there are no reports regarding whether these imaging methods can reflect hepatic steatosis qualitatively—that is, macro- or microvesicular steatosis.
Artificial neural networks (ANNs) represent a computer-based method that has shown high performance in various fields of clinical medicine (21–24). Neural networks undergo training sessions in which a number of measurements for each example of a training set and the desired classification are fed to the network. The networks learn to associate the training examples with the given classification for each case. When used for characterization of US images of experimental diffuse liver disease, networks have demonstrated more than 80% separation accuracy between normal tissue and steatosis and between normal tissue and steatotic cirrhosis (25). However, there are no reports on the comparison of performance between ANNs and radiologists for evaluation of hepatic steatosis.
Thus, the purpose of this study was to retrospectively compare the performance of ANNs applied to US images with that of radiologists for prediction of the appropriateness of a donor liver with respect to macrosteatosis prior to liver transplantation.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
The remaining 94 donors (age range, 18–52 years; mean age, 32 years) (71 men [age range, 18–52 years; mean age, 31 years] and 23 women [age range, 21–46 years; mean age, 34 years]) with available pathologic specimens were included in our study. Seventy-four of 94 patients underwent hepatic resection for liver donation (right hemihepatectomy in 38, left lateral segmentectomy in 20, right-sided extended hepatectomy in 14, and left hemihepatectomy in two). In the remaining 20 of 94 patients who did not undergo surgery, 13 had macrosteatosis of moderate degree, five had macrosteatosis of severe degree, and two had complex hepatic vascular anatomy. Mean interval between US and biopsy was 39.3 days (range, 0–140 days). Approval for this study was obtained from the appropriate ethics committee at our institution. Written informed consent was obtained from all patients.
Liver Biopsy and Evaluation
Liver biopsy was performed in all 94 patients. Among them, surgical biopsy hepatic specimens with a size of 1 cm3 were obtained from 74 patients who underwent hepatic resection and two patients who had complex hepatic vascular anatomy. Hepatic specimens were obtained at the area close to the resection margin in the 74 patients and at the surface of the right lobe in the other two patients. Thirty-two patients underwent inter- or subcostal liver biopsy of the right lobe with US guidance by using an 18-gauge automatic biopsy gun (ACECUT; TSK laboratory, Tochigi, Japan). On average, 2.2 biopsies (range, 1–3 biopsies) were performed per liver. Specimens 2 cm in length or longer were fixed in formalin and stained with hematoxylin-eosin. Among these 32 patients, 14 donors underwent both percutaneous and surgical biopsy, while 18 underwent only percutaneous biopsy. The interval between the two biopsies ranged from 2 to 72 days (mean, 20.3 days).
One hundred eight biopsy slides in 94 patients were examined retrospectively. They were assessed in blinded fashion by an experienced pathologist (H.S.L., with 8 years of experience) for the particular type of steatosis (macro- and microvesicular) and its extent. When there were hepatocytes containing one large vacuole of fat that displaced the nuclei to the periphery of the cell, it was considered to be macrovesicular fat deposit. On the contrary, when the cytoplasm contained many small fatty inclusions and the nuclei remained in the center of the cell, it was considered microvesicular steatosis.
To assess the qualities of hepatic fat, it was determined whether deposited fat was macro- or microvesicular fat. Macro- and microsteatosis were evaluated qualitatively in 10 consecutive fields (objective, 25x). In addition to qualitative analysis, steatosis was classified into four groups quantitatively—that is, normal is less than 5%, mild is between 5% and 29%, moderate is between 30% and 59%, and severe is if more than 60% of the hepatocytes have fat vacuoles within the cytoplasm (26). Biopsy findings of macrovesicular steatosis were divided into two groups—appropriate or inappropriate for transplantation—according to the 30% cutoff established by several authors (7–11).
To evaluate temporal variation in the amount of fatty infiltration, the degrees of steatosis between percutaneous and surgical biopsy in the 14 patients who underwent both biopsies were compared, as the time between these biopsies ranged from 2 to 72 days.
US Image Acquisition
Abdominal US examination was performed by one of three abdominal radiologists (Y.J.K., with 10 years of experience; S.H.K., with 8 years of experience; and S.K.A., with 4 years of training) by using an SSD 5500 system (Aloka, Wallingford, Conn) and a Sequoia 512 system (Acuson, Mountain View, Calif) with a convex 3–5-MHz transducer. The time gain control was set at its central position, and the power was set at a constant level. US was performed and included one set of two images that was obtained constantly from each patient. To begin, the probe was positioned in a right coronal subcostal position in each patient so that stable parenchymal images of the liver and the right kidney were obtained simultaneously with no large vessels in the images (Fig 1).
|
|
Subjective Evaluation by Radiologists
For subjective evaluation, US images from all 94 patients were evaluated retrospectively and independently by three radiologists, who were given the results of the laboratory data (fasting blood glucose, uric acid, cholesterol, alkaline phosphatase, glutamic oxaloacetic transaminase, glutamic pyruvate transaminase, and -glutamyltransferase levels) but were blinded to the histopathologic findings and the results from the neural networks at the classification procedure. The three radiologists included one faculty radiologist (J.M.L., with 13 years of experience), one abdominal imaging fellow (S.K.A., with 4 years of training), and one junior resident (K.R.S., with 2 years of training). The radiologist graded each US study according to the presence and severity of fatty infiltration by using the criteria in Table 1 (12). The grade of fatty liver was classified as absent (score of 1), mildly present (score of 2), moderately present (score of 3), or severely present (score of 4) by each reader. Patients with a score of 1 or 2 were classified into an appropriate group for transplantation; patients with a score of 3 or 4 were placed in a group inappropriate for transplantation.
|
Regions of interest (ROIs) with a size of 1 x 1 cm were selected by an abdominal radiologist (Y.J.K.) who did not participate in subjective US analysis and who blinded the histopathologic results so as to contain only hepatic parenchyma and renal cortex with no large vessels, renal sinus, or medulla. Two ROIs were selected along the focusing area of the image as was best possible so as to be applied at the same distance from the probe and near the center line of the image to avoid distorting effects in ultrasonic wave patterns. The liver-to-kidney ratio was assessed from the results of hepatic intensity divided by renal intensity (Fig 2). Therefore, the more severe the hepatic steatosis, the higher the liver-to-kidney ratio.
|
|
In the case of a normal liver, hyperechoic components derived from the portal vein border will be relatively higher than in hepatic steatosis, and the distance to the most hyperechoic spot from the baseline will also be longer than in hepatic steatosis on the histogram. To quantify them, the points corresponding to 20% of the value of the highest point were determined. We assessed the degree of the portal vein wall echogenicity from the result of dividing the area of the left side of the predetermined point with that of the right side and named it the "tail area ratio of portal vein echogenicity." We also measured the distance from the predetermined point to the highest echoic point and named it the "tail length of portal vein echogenicity" (Fig 3).
|
|
-glutamyltransferase levels were also analyzed. Cutoff values for these findings were set by using standard criteria for categoric values—that is, fasting glucose value more than 110 mg/dL (6.1 mmol/L), uric acid value more than 7.0 mg/dL (420 µmol/L), cholesterol value more than 240 mg/dL (6.2 mmol/L), alkaline phosphatase value more than 115 IU/L (115 U/L), glutamic oxaloacetic transaminase and glutamic pyruvate transaminase values more than 40 IU/L, and -glutamyltransferase value more than 63 IU/L were considered abnormal. The interval between laboratory testing and biopsy ranged from 0 to 71 days (mean, 11.4 days).
Construction of ANNs
ANNs were constructed by using commercially available software (NeuroSolutions version 4.0; Neurodimension, Gainesville, Fla) and conformed with a three-layered perceptron architecture. A more general description of ANNs can be found elsewhere (28). The ANNs consisted of one input layer, one hidden layer, and one output layer. Statistically significant laboratory and US parameters, as determined by means of univariate analysis, were used as input variables for an input layer of ANNs. The hidden layer, which was connected to the input layer, consisted of 15 neurons. A nonlinear sigmoid function was used as a transfer function for each of the neurons in the hidden and output layers of the networks.
During the training process, the connection weights between the neurons were adjusted by using a back-propagation updating algorithm (29). The hidden layer was connected to an output layer of a single neuron, which encoded whether the US and laboratory findings were classified as appropriate as a liver donor with respect to macrosteatosis (Fig 4).
|
In the final training procedure, a fourfold cross-validation procedure was used to maximize performance reliability. Each of the four different networks was trained until the error in the training set reached the stopping criterion. The test results of the four different networks were combined to calculate ANN performance in terms of predicting the appropriateness of a liver donor with respect to macrosteatosis. The output values for the test were in the range of 0 to 1, which could be viewed as representing the likelihood of moderate to severe macrosteatosis.
Statistical Analysis
Pathologic results were correlated with US grades by radiologists with the use of 
statistics, weighted statistics, and Spearman rank correlation. Interobserver agreement between readers was also evaluated by using statistics and weighted statistics. We considered a value of more than 0.81 to represent almost perfect agreement and values of 0.61–0.80 and 0.41–0.60 to represent substantial or moderate agreement, respectively. Values less than 0.40 were considered to represent fair agreement (30).
In addition, to determine whether US has the capability to reflect hepatic steatosis qualitatively (macro- or microvesicular steatosis), the correlation between the two independent variables—that is, the pathologic and US degree of steatosis as determined semiquantitatively by a pathologist and by each of the three radiologists—was evaluated by using ordinal logistic regression analysis. Both macro- and microvesicular steatosis, as determined by the pathologist, were also used as a categoric variable with four degrees in the logistic models, because US diagnoses for fatty liver as determined by the radiologists were classified by using categoric variables.
The distributions of laboratory findings and US characteristics derived from the computer-aided diagnosis program were compared between appropriate and inappropriate liver donors with respect to macrosteatosis by using the Fisher exact test or the 
2 test for categoric variables and the Mann-Whitney U test for continuous variables. Statistical analyses were performed by using SPSS version 11.0 software for Windows (SPSS, Chicago, Ill). A P value less than .05 was considered to indicate a statistically significant difference.
The individual performances of radiologists and ANNs were evaluated by means of receiver operating characteristic (ROC) analysis; areas were calculated by using a parametric method (31,32). Binormal ROC curves were estimated by using the ROCKIT algorithm (available through the Internet from C. E. Metz, University of Chicago, Ill), which was used to obtain maximum-likelihood estimates of binormal ROC curves from the continuous ordinal-scale rating data. The area under the ROC curve (Az) was calculated to summarize the performance of each radiologist and ANN in the task of classification between appropriate and inappropriate status for the potential liver donor. The univariate z score test was used to assess the significance of differences in Az values under the two estimated binormal ROC curves. In addition, the sensitivity and specificity were calculated by using only those patients deemed inappropriate by each radiologist and those with an ANN output level of 0.5 or higher. Sensitivity and specificity were presented with a 95% confidence interval (CI) and were compared by using the McNemar test with SPSS software (SPSS), which is a nonparametric test for two related dichotomous variables (33). A P value less than .05 was considered to indicate a statistically significant difference.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Histopathologic Findings
Forty-two of 94 patients were classified as having normal findings, 34 had mild macrosteatosis, 13 had moderate macrosteatosis, and five had severe macrosteatosis. For microsteatosis, 36 patients were classified as having normal findings, 55 had mild microsteatosis, two had moderate microsteatosis, and one had severe microsteatosis. Thus, 76 patients were categorized as appropriate and 18 as inappropriate for liver donation with respect to macrosteatosis by means of histopathologic findings.
Patient Age and Sex
No significant difference in patient age or sex was evident between the two groups. The mean patient age in the appropriate group was 31.4 years ± 10.1 (standard deviation), compared with 30.7 years ± 13.9 in the inappropriate group (P = .81). Of the 76 patients in the appropriate group, 56 (74%) were men and 20 (26%) were women, while in the inappropriate group of 18 patients, 15 (83%) were men and three (17%) were women (P = .55).
US Findings
US grades as interpreted by the radiologists are presented with the grades of macrosteatosis according to the pathologist in Table 2. The faculty radiologist classified 39 patients as having normal findings, 21 as having a mildly fatty liver, 15 as having a moderately fatty liver, and 19 as having a severely fatty liver on the basis of US images. Thus, 60 patients were categorized as the appropriate group, and 34 were placed in an inappropriate group for liver donation. There was relatively close correlation between the pathologic grades of macrosteatosis and the US grade, as determined by radiologists (Table 3).
|
|
statistics and weighted statistics indicated a moderate strength of agreement between the pathologist and the radiologists both in terms of grading (four grades) and grouping (two groups). The Spearman rank correlation between the two series was 0.729 between the pathologist and the faculty radiologist, 0.725 between the pathologist and the fellow, and 0.668 between the pathologist and the resident (P < .001). As was expected, a general trend was observed that readers with more years of experience showed better agreement with the pathologist. Figure 5 shows a regression plot that demonstrates the correlation between US and pathologic grades of macrosteatosis.
|
|
|
= 0.823) between the faculty radiologist and the fellow in terms of classifying the fatty liver into the four grades, and there was substantial agreement between the faculty radiologist and the resident ( = 0.648) and between the fellow and the resident ( = 0.636). In terms of the "appropriate" or "inappropriate" classifications, substantial agreement (0.622–0.770) was achieved between radiologists.
|
|
2 test, and Mann-Whitney U test). Significant differences were found for all three US parameters between the two groups (normal findings or mild macrosteatosis [appropriate group] versus moderate or severe macrosteatosis [inappropriate group]). In addition, moderate or severe macrosteatosis was significantly associated with abnormal alkaline phosphatase, glutamic oxaloacetic transaminase, glutamic pyruvate transaminase, and -glutamyltransferase values. The largest odds ratio (17.160) was shown by glutamic pyruvate transaminase.
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Another important result is that the specificity of ANNs was significantly better than that of radiologists. Because liver transplantation has become the standard treatment for patients with various liver diseases and its success rate has increased dramatically, the demand for liver transplantation has also increased greatly in most countries. For example, about 13 000 patients are currently waiting for a liver transplantation in the United States, while only 4000 organs are available each year (35). In such a situation, higher specificity is crucial to enable expansion of the pool of organ donors. In this context, the results obtained during the present study, particularly in terms of the high specificity of the developed ANNs, are both meaningful and extremely encouraging.
To detect and quantify the fat content of a liver in vivo, various imaging modalities, including US, CT, and MR imaging, have been used (12–20,25–27,36). They provide reproducible noninvasive measures of global hepatic fat content, with no risk of intrahepatic sampling error. At US, fatty droplets within hepatocytes scatter US beams to produce the phenomenon of a bright echotexture, in which the liver is more echogenic than the adjacent kidney.
Freese and Lyons (13) postulated that US is a simple, noninvasive, and quantitative test for fatty infiltration of the liver and found a linear relationship between the total lipid concentration and US backscatter coefficient. However, the dependence of US on subjective interpretation is a major limitation in terms of assessing the degree of hepatic steatosis. The various levels of interobserver agreement (0.636–0.823) between readers observed in the present study testify to this shortcoming.
The quantitative evaluation of echogenicity on the basis of histograms has been proposed as a means of overcoming variability due to subjective evaluation (27,36). In these studies, however, other important factors such as laboratory data were not considered, and only the acoustic intensity of the liver (36) or ratios and differences of liver and renal cortical echo amplitude (27) were analyzed. In the present study, to measure the brightness of hepatic parenchyma objectively with US, we estimated the ratio of portal vein wall intensity and liver-to-kidney intensity ratios. Given that impaired visualization of the intrahepatic portal vein borders and a diffuse increase in hepatic echogenicity are important diagnostic criteria of hepatic steatosis at US (12), we believe that the quantification of the impaired visualization of the intrahepatic portal vein border by using tail area ratio and tail length of portal vein echogenicity could add confidence to US diagnoses of hepatic steatosis, as opposed to measuring the brightness of hepatic parenchyma only. As was expected, the values of all US parameters were significantly different between the two groups of normal findings or mild macrosteatosis (appropriate group) and moderate or severe macrosteatosis (inappropriate group). Therefore, we suggested that US parameters such as portal vein wall intensity should be used as additional tools for quantifying hepatic steatosis, together with liver-to-kidney ratio.
Increased hepatic echogenicity, the so-called bright liver, is also found in patients with increased fibrous content, and it is not possible to differentiate between fatty infiltration and fibrosis by means of hepatic intensity alone (14,37). However, because most living liver donor candidates are healthy and usually have no other medical problems, increased hepatic echogenicity in a potential liver donor can be assumed to be caused by fatty infiltration.
One of the interesting findings in the present study is that US grades for fatty liver, as determined by radiologists, are significantly correlated only with the grade of macrovesicular fat accumulation in hepatocytes and not with microvesicular fat accumulation. This result is in line with a report by Pamilo et al (38). They found that increased hepatic echogenicity is caused only by large fat droplets exceeding 100 µm in diameter. Thus, it is likely that microvesicular fat accumulation has little effect on increased hepatic echogenicity; however, it has not been established how microsteatosis of moderate or severe degree could influence US findings. Further prospective study of a larger population with profound microsteatosis is needed.
The ability of ANNs to learn specific patterns between input and output data strongly depends on the quality of input data. It should be noted that in the present study, we used only objective laboratory and US data as input variables and not subjective values determined by radiologists. Therefore, the quality of input data for ANNs can be constant, regardless of other subjective factors. When compared with other studies in which many subjective ratings by radiologists or other physicians were used, our methods and results can be applied more easily and objectively to computer-aided diagnosis for the characterization of diffuse liver disease at US. We appreciate that ANNs are not a cure-all for complex data analysis, and several criticisms may be encountered with these techniques. These include the empirical nature of choosing network parameters and the complexity of the inner processes, which are not explained easily. Despite such limitations, ANNs provide us with a predictive ability that can be demonstrated clearly and understood with a thorough understanding of the processes involved.
Some limitations of our study should be mentioned, which are derived primarily from fundamental problems of US studies. Although many confounding factors, including the effect of focusing and distance from the transducer, can make echogenicity difficult to quantify objectively, we tried to minimize these factors by performing US with constant settings and by applying ROIs to the liver and right kidney at the same distance from the transducer. However, problems due to the use of different US equipment remain to be overcome. In addition, biopsies of the liver are subjective to a well-recognized sampling error due to the inhomogeneous distribution of fat; moreover, the process of fat infiltration is continuous, and there may be temporal variations in fat content. The results of the 14 patients who underwent both percutaneous and surgical biopsy indicated that temporal variability was not a serious problem. However, the small number of cases and the short interval (mean, 20.3 days) between the two biopsies prevent generalization of this result.
Finally, we used the same data set for both training and testing of the ANNs. This limitation is also related to the small number of cases deemed inappropriate as donor livers. Although the validity of training and testing by using the same data set is widely accepted in medical and medical engineering fields (21–24,34), it could inflate the performance of the ANNs on the test set. Accordingly, further prospective studies with a larger number of cases, allowing adequate training and independent testing, are needed to assess the performance of ANNs for prediction of the appropriateness of a donor liver. If such a study produces encouraging results, the possibility of applying this method to clinical practice should be explored.
In conclusion, ANNs might be a useful tool to categorize a donor liver as either appropriate or inappropriate for transplantation with respect to macrosteatosis on the basis of multiple variables related to laboratory and US features. Although ANNs showed better performance than that of radiologists, further comparative studies with larger cases are needed before clinical application.
|
FOOTNOTES |
|---|
Authors stated no financial relationship to disclose.
Author contributions: Guarantors of integrity of entire study, S.H.K., J.M.L., J.H.K., K.S.S.; study concepts, S.H.K., J.M.L., J.H.K., K.G.K.; study design, S.H.K., J.M.L., J.H.K., K.G.K., J.H.K.; literature research, S.H.K., K.H.L., K.G.K., K.R.S.; clinical studies, S.H.K., S.K.A., Y.J.K., N.J.Y., K.S.S.; data acquisition, S.H.K., S.K.A., Y.J.K., N.J.Y., K.S.S.; data analysis/interpretation, S.H.K., J.M.L., Y.J.K., H.S.L., B.I.C.; statistical analysis, S.H.K., K.G.K., S.H.P.; manuscript preparation, S.H.K.; manuscript definition of intellectual content, S.H.K., J.M.L., J.K.H., B.I.C.; manuscript editing and revision/review, all authors; manuscript final version approval, S.H.K., J.M.L., B.I.C.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. W. Lee, S. H. Park, K. W. Kim, E. K. Choi, Y. M. Shin, P. N. Kim, K. H. Lee, E. S. Yu, S. Hwang, and S.-G. Lee Unenhanced CT for Assessment of Macrovesicular Hepatic Steatosis in Living Liver Donors: Comparison of Visual Grading with Liver Attenuation Index Radiology, August 1, 2007; 244(2): 479 - 485. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. K. An, J. M. Lee, K.-S. Suh, N. J. Lee, S. H. Kim, Y. J. Kim, J. K. Han, and B. I. Choi Gadobenate dimeglumine-enhanced liver MRI as the sole preoperative imaging technique: a prospective study of living liver donors. Am. J. Roentgenol., November 1, 2006; 187(5): 1223 - 1233. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. H. Kim, J. M. Lee, J. K. Han, J. Y. Lee, K. H. Lee, C. J. Han, J. Y. Jo, N.-J. Yi, K.-S. Suh, K.-S. Shin, et al. Hepatic Macrosteatosis: Predicting Appropriateness of Liver Donation by Using MR Imaging--Correlation with Histopathologic Findings Radiology, July 1, 2006; 240(1): 116 - 129. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||
