Value of diffusion-weighted magnetic resonance images for discrimination of focal benign and malignant hepatic lesions: A meta-analysis

Authors


Abstract

Purpose:

To evaluate the ability of DW-MRI in differentiating malignant hepatic tumors from benign lesions.

Materials and Methods:

Meta-analysis of 14 diagnostic studies was used. A systematic search in Medline, Embase, Web of Science (from January, 1966, to October, 2009), and Cochrane Controlled Clinical Trials Register Database (through third Quarter 2009) was used with screening of the literature.

Results:

A meta-analysis of all 95 published studies was performed. Fourteen studies fulfilled the inclusion criteria (804 patients with 1665 hepatic lesions). The global sensitivity was 0.91 (95% confidence interval [CI], 0.86–0.94), the specificity was 0.93 (95% CI, 0.86–0.97), the positive likelihood ratio (PLR) was 13.10 (95% CI, 6.30–27.26), the negative likelihood ratio (NLR) was 0.10 (95% CI, 0.06–0.15), and the diagnostic odds ratio (DOR) was 133.76 (95% CI, 49.77–359.45). The area under the curve of the summary receiver operator characteristic (SROC) was 0.96 (95% CI 0.94–0.98).

Conclusion:

Diffusion-weighted magnetic resonance imaging is potential technically feasible to differentiate malignant from benign focal liver lesions. Apparent diffusion coefficient measurements can be useful in providing rapid quantifiable information. J. Magn. Reson. Imaging 2010;32:130–137. © 2010 Wiley-Liss, Inc.

EARLY AND ACCURATE detection and characterization of focal liver lesions are important for treatment planning for patients with liver malignant neoplasms such as hepatocellular carcinoma (HCC), metastases, high-grade dysplastic nodules (HDN), and cholangiocarcinomas (CC) (1). The tumor size, number of lesions, and intrahepatic metastases are not only as important negative prognostic factors but can affect therapy (1, 2).

Magnetic resonance imaging (MRI) currently yields the highest accuracy for detection of HCC but has not improved the early detection of small HCC all that much compared to computed tomography (CT) or ultrasound (3). Diffusion is the random, thermally induced movement of water molecules in biologic tissues, called Brownian motion (4). Diffusion-weighted (DW) magnetic resonance imaging (MRI) is sensitive to molecular diffusion and allows for tissue characterization by probing tissue microstructural changes, quantified as the apparent diffusion coefficient (ADC) (5). DW images has been reported to be useful for differentiating malignant and benign lesions and for characterization of malignant and benign lesions through quantification of ADC (6–9).

We performed the present meta-analysis to assess the diagnostic use of DW-MRI and to establish the overall accuracy of DW-MRI measurement for characterization of malignant hepatic lesions.

MATERIALS AND METHODS

Search Strategy and Study Selection

We searched the following electronic databases: Medline, Embase, Web of Science (from January, 1966, to October, 2009), and Cochrane Controlled Clinical Trials Register Database (through third Quarter 2009) for all studies examining the diagnostic accuracy of DW-MRI for detection of malignant hepatic lesions. For the electronic search we used the following terms or MeSH subject headings: “Diffusion Magnetic Resonance Imaging,” “Diffusion-weighted magnetic resonance images,” “DW-MRI,” “DW magnetic resonance images”; “Carcinoma, Hepatocellular,” “Hepatocellular Carcinoma,” “liver cancer”; “metastases,” “cholangiocarcinomas” and liver; and “diagnosis,” “sensitivity,” “specificity,” “predictive value,” “likelihood ratio,” “false positive,” “false negative,” and “review,” “meta-analysis.”

We contacted the authors for further study details if needed and searched the reference lists from primary and review articles. No language restriction was used, and all foreign-language publications were translated. Further searches were performed by manually reviewing abstract booklets, conference proceedings, and review articles.

We included all studies that met the following criteria: assessing the diagnostic accuracy of DW-MRI for malignant hepatic lesions; providing both sensitivity (true-positive rate) and specificity (true-negative rate) of DW-MRI for the diagnosis of malignant hepatic lesions; providing sufficient information to construct the 2 × 2 contingency table for individual study subjects; and stating a test method for DW-MRI in the methods. We excluded conference abstracts, abstracts, and letters because of limited information. Two reviewers independently judged study eligibility while screening the citations. Disagreements were resolved by consensus.

Data Extraction

The final set of articles was assessed independently by two reviewers. The reviewers independently abstracted data from each study to obtain information on the year of publication, country of origin, number of patients, types of malignant, types of benign control, confirmation of liver lesions, DW-MRI test methods, diagnostic cutoff points, number of lesions, sensitivity and specificity of the data, and methodological quality. Each reviewer extracted the data to construct a 2 × 2 table. Any disagreements were resolved by consensus.

Quality Assessment

The methodological quality of each study was assessed using a checklist based on criteria adapted from the Cochrane Collaboration guidelines (10, 11) and the quality assessment for studies of diagnostic accuracy (QUADAS) tool (maximum score, 14) (12).

Statistical Analysis

We used standard methods recommended for meta-analyses of diagnostic test evaluations (10, 11). For each study the sensitivity, specificity, positive and negative likelihood ratios, and a diagnostic odds ratio (DOR) were calculated. The DOR is the ratio of the odds of a positive result in malignant hepatic lesions compared with benign hepatic lesions: [sensitivity/(1-sensitivity)]/[(1-specificity)/specificity]. Each study was weighted using an inverse variance method. We constructed summary receiver operator characteristic (SROC) curves to summarize the study results. A smoothed curve was then fitted across the studies to represent the relationship between sensitivity and the proportion of false positives (1-specificity). The sensitivity and specificity for the single test threshold identified for each study were used to plot an SROC curve. Pooling of the summary indices was performed using the bivariate mixed-effects binary regression model (13).

To detect heterogeneity, the likelihood ratios and DORs were graphically displayed using forest plots and analyzed using Cochran's Q test. A P-value of less than 0.05 by Cochran's Q test indicated significant heterogeneity. To quantify the extent of heterogeneity the I2 statistic was used to measure the percentage of variability among summary indices that were caused by heterogeneity rather than chance. A study with an I2 greater than 50% indicated substantial heterogeneity.

To explore sources of heterogeneity among studies, univariate meta-regression analysis (inverse variance weighted) was used. The covariates included spectrum characteristics (eg, study setting, prevalence, type of bacterial infection), quality of the study (QUADAS scores), and methodological features (eg, sample size).

Publication bias was examined visually by inspecting funnel plots and statistically by using Egger's regression model (14). If publication bias was present, the effect of such a bias on the final summary estimate was assessed using the trim and fill method (15). This method imputes the missing studies and recalculates a new summary estimate. The difference between the calculated and observed values was then used to determine the effect of bias on the diagnostic performance of the test. Analyses were performed using Stata (v. 10.0; StataCorp, College Station, TX).

RESULTS

We retrieved 95 potentially eligible reports and 22 publications dealing with DW-MRI for the diagnosis of malignant hepatic lesions considered as potentially suitable for inclusion in the analysis. After full-text review, eight studies were excluded (Fig. 1). In total, 14 studies (8, 9, 16–27) including 804 patients with 1665 hepatic lesions were available for the final analysis. The trials identified were from Japan, France, Netherlands, Germany, Turkey, Greece, UK, Belgium, and the United States. All studies included in the analysis used 1.5 T scanner systems. The b-values of the DW-MRI sequence varied between different studies. Of these studies, seven studies used the sensitivity encoding (SENSE) technique. The size of lesions ranged from 0.3–17.8 cm. Of these 1665 hepatic lesions, 943 hepatic lesions were malignant; malignant hepatic lesions included hepatocellular carcinomas, liver metastases, cholangiocarcinomas, and high-grade dysplastic nodules. There were 741 benign hepatic lesions including hemangiomas, cysts, abscess, adenomas, focal nodular hyperplasia, intrahepatic hematoma, and low-grade dysplastic nodules. In five studies benign lesions included solid lesions. Eight studies reported a cutoff of ADCs. The threshold ranged from 1.47–5.5 × 103 mm/s2. The details of all 14 studies are shown in Table 1.

Figure 1.

Study identification, inclusion, and exclusion for meta-analysis.

Table 1. Summary of Included Studies
Study yearCountryStudy designPatients, No.Types of malignantTypes of benign controlConfirmation of liver lesionsDiffusion-weighted MRI methodSensitivity encoding (SENSE)Cutoff (ADC×103 mm /s2)Lesions, No.TPFPFNTNSens. %Speci. %Quality Score
  1. TP, true-positive; FP, false-positive; FN, false-negative; TN, true-negative; HCC, hepatocellular carcinoma; CC, cholangiocarcinomas; HDN, high-grade dysplastic nodules; FNH, focal nodular hyperplasia; RN, regenerative nodules; LDN, low-grade dysplastic nodules; ADC, apparent diffusion coefficients; NA, not available.

Ichikawa et al, Ref.9JapanCR46HCC, metastases ID:0.7–10 cmHemangiomasSurgical specimens, biopsy, follow-up imaging1.5-T system, Single shot echo planar DW-MRI, b-values (1.6,16, and 55 s/mm2)NO5.574590411  8
Kim et al, Ref.16JapanCR70HCC, metastases, CC ID:0.7–10 cmHemangiomas, cysts, angiomyolipoma ID:0.8–7 cmSurgical specimens, biopsy, follow-up imaging1.5 T system, Single shot echo-planar DW-MRI, b-values (3, 57,192, 408,517,705 and 846 s/mm2)NO1.6 (b = 850)79486124  8
Taouli et al, Ref.8FranceCR43HCC, metastases ID: 1–8.7 cmHemangiomas, cysts, adenoma ID:2–15.5 cmSurgical specimens, biopsy, follow-up imaging1.5-T system, two breath-hold DW-MRI, b-values (0,134, 267,400, and 500 s/mm2),NO1.552213424848911
Nasu et al, Ref.17JapanCR62Metastases ID:0.5–6 cmNodules, cystsSurgical resection, follow-up imaging1.5 T system, DW SENSE MRI, b-values (0, and 500 s/mm2)YESNA95334751  9
Coenegrachts et al, Ref.18NetherlandsCR24Metastases ID:0.4–2.6 cmHemangiomas, cysts ID:0.3–7.2 cmSurgical resection1.5-T system, breath-hold DW-MRI, b values (b50, b520, b5300, b5800 s/ mm2 )YESNA12940008910010010
Bruegel et al, Ref.19GermanyCR52Metastases ID:0.3–8.4 cmHemangiomas, cysts ID:0.3–4.8 cmsurgical specimens, biopsy, follow-up imaging1.5-T system, breath-hold DW-MRI, b-values (50, 300, and 600 s/mm2)YESNA22410681298  11
Bruegel et al, Ref.20GermanyCR102HCC, metastasesFNH, hemangiomas, cystssurgical specimens, biopsy1.5-T system, breath-hold DW-MRI, b-values (50, 300, and 600 s/mm2)NO1.632048416995  12
Erturk et al, Ref.21TurkeyCR78HCC, metastases ID:1–4.2 cmHemangiomas, cysts ID:1.4–4.8 cmBiopsy, follow-up imaging1.5 T system, breath-hold DW SENSE MRI, b-values (400 and 1,000 s/mm2)YES1.6386402440  11
Goshima et al, Ref.22JapanCR37HCC, metastasesHemangiomas, cystsSurgical specimens, biopsy, follow-up imaging1.5 T system, breath-hold DWI, b-values (100, 200, 400, and 800 s/mm2)NONA55269218  9
Gourtsoyianni et al, Ref.23GreeceCR27HCC, metastasesHemangiomas, cystsMRI, clinical follow-up1.5-T system, Single shot echo planar DW-MRI, b-values (0, 50, 500 and 1,000 s/mm2)NO1.47371500221001008
Koh et al, Ref.24UKCR38Metastases ID:0.5–9.5 cmHemangiomas, cysts ID:0.5–9.5 cmPathology, follow-up MRI1.5-T system, DW-MRI, b-values (0,150 and 500 s/mm2)YESNA1336521848  9
Parikh et al, Ref.25USACR53HCC, metastases ID:1–15.1 cmHemangiomas, cysts, abscess, adenomas, FNH, intrahepatic hematomaFollow-up imaging, clinical follow-up, surgical specimens1.5 T system, breath-hold DW SENSE MRI, b-values (0, and 50 s/mm2)YES1.6211101173558  12
Vossen et al, Ref.26USACR117HCC, metastases ID:1–17.8 cmHemangiomas, FNH ID:1–17.8 cmBiopsy, follow-up MRI1.5 T system, breath-hold DWI, b value, 500 s/mm2YES2.317212201743  10
Vandecaveye et al, Ref.27BelgiumCR55HCC, HDN, CC ID:0.7–14 cmRN, LDN, FNH, inflammatory pseudo-tumour ID:0.7–14 cmSurgical specimens1.5-T system, Single shot echo planar DW-MRI, b-values (b = 0,100, 600, 1,000 s/mm2 )NONA11459934395.282.712

Quantitative Data Synthesis

Figure 2 shows the forest plot for the sensitivity and specificity of 14 DW-MRI used for the diagnosis of malignant hepatic lesions. The sensitivity ranged from 0.74–1.0 (mean, 0.91; 95% confidence interval [CI], 0.86–0.94), while the specificity ranged from 0.77–1.00 (mean, 0.93; 95% CI, 0.86–0.97). We also found that the positive likelihood ratio (PLR) was 13.10 (95% CI, 6.30–27.26), the negative likelihood ratio (NLR) was 0.10 (95% CI, 0.06–0.15; Fig. 3), and the DOR was 133.76 (95% CI, 49.77–359.45).

Figure 2.

Forest plot of the sensitivity and specificity of DW-MRI in the diagnosis of malignant hepatic lesions with corresponding heterogeneity statistics. Pooled estimates for the DW-MRI were as follows: sensitivity, 0.91 (95% CI, 0.86–0.94); specificity, 0.93 (95% CI, 0.86–0.97).

Figure 3.

Scattergram of the positive likelihood ratio and negative likelihood ratio. Pooled estimates for the DW-MRI test were as follows: PLR 13.10 (95% CI, 6.30–27.26), NLR 0.10 (95% CI, 0.06–0.15).

The SROC curve presents a global summary of test performance, and it shows the tradeoff between sensitivity and specificity. A graph of the SROC curve for DW-MRI showing true-positive rates vs. false-positive rates from individual studies is shown in Fig. 4. The SROC curve (Fig. 4) yielded a maximum joint sensitivity and specificity of 0.93 (95% CI, 0.86–0.97), an area under the curve of 0.96 (95% CI, 0.94–0.98), indicating a high level of overall accuracy.

Figure 4.

Summary receiver operating characteristics curve with confidence and predictive ellipses for DW-MRI test. Circles indicate 95% confidence and predictive ellipses; n = 14 studies.

Cochran's Q for sensitivity, specificity, PLR, NLR, DOR, and SROC were 71.33 (P < 0.001), 73.04 (P < 0.001), 74.67 (P < 0.001), 77.59 (P < 0.001), 230 (P < 0.001), and 14.13 (P < 0.001), respectively, and I2 for sensitivity, specificity, PLR, NLR, and DOR was 81.77, 82.20, 74.33, 83.25, 99, and 85.85, respectively, indicating significant heterogeneity and inconsistency between studies.

Multiple Regression Analysis and Publication Bias

Study country, number of patients, spectrum characteristics, methodological features, ADC cutoff, and the quality of the study were used in the meta-regression analysis to assess the source of variability among studies. As shown in Tables 2 and 3, higher quality studies (QUADAS score, 10) produced sensitivity and specificity values that were not significantly higher than lower-quality studies. There were significant differences for sensitivity encoding (sensitivity, P = 0.03; specificity, P = 0.08) indicating that the SENSE technique may affect diagnostic accuracy. Study country, number of patients, and spectrum characteristics did not affect the diagnostic accuracy, except for the observation that the United States of America (USA) potentially affected diagnostic sensitivity (P = 0.06) and solid benign lesion included potentially affected diagnostic specificity (P = 0.08).

Table 2. Meta-Regression of the Effects of Study Country, Number of Patients, Spectrum Characteristics, Methodological Features, ADC Cutoff, and the Quality of the Study on the Diagnostic Sensitivity of DW-MRI
CovariatesStudies, no.CoefficientEstimate (95% CI)P
  1. DW-MRI, diffusion-weighted magnetic resonance images; ADC, apparent diffusion coefficient.

QUADAS≥1082.260.91 (0.84–0.95)0.82
Country    
 Japan42.550.93 (0.84–0.97)0.50
 Germany22.240.90 (0.77–0.96)0.91
 USA21.520.82 (0.67–0.91)0.06
Patient no.142.330.91 (0.86–0.94)0.98
Lesions ≥10072.090.89 (0.82–0.93)0.30
Types of malignant
 Metastases only42.120.89 (0.79–0.95)0.62
Types of benign
 Solid lesion included51.810.86 (0.69–0.94)0.08
Sensitivity encoding DW-MRI71.890.87 (0.81–0.91)0.03
ADC cutoff82.290.91 (0.84–0.95)0.98
Table 3. Meta-Regression of the Effects of Study Country, Number of Patients, Spectrum Characteristics, Methodological Features, ADC Cutoff, and the Quality of the Study on the Diagnostic Specificity of DW-MRI
CovariatesStudies, no.CoefficientEstimate (95% CI)P
  1. DW-MRI, diffusion-weighted magnetic resonance images; ADC, apparent diffusion coefficient.

QUADAS≥1082.620.93 (0.84–0.97)0.92
Country    
 Japan41.940.87 (0.66–0.96)0.27
 Germany22.170.90 (0.61–0.98)0.60
 USA22.770.94 (0.67–0.99)0.88
Patient no.142.60.93 (0.86–0.97)0.99
Lesions ≥10072.740.94 (0.85–0.98)0.65
Types of malignant
 Metastases only43.310.96 (0.90–0.99)0.10
Types of benign    
 Solid lesion included51.810.86 (0.69–0.94)0.08
Sensitivity encoding DW-MRI73.000.95 (0.89–0.98)0.08
ADC cutoff80.070.93 (0.83–0.98)0.94

Publication bias was detected using Egger's regression model (P = 0.001). These results indicated a potential publication bias. Visual inspection of the funnel plot suggested that missing studies were likely to fall below the summary estimate. These studies were then imputed to calculate a new summary estimate (Fig. 5). The new DOR was a little lower than the observed DOR visually by inspecting funnel plots. Therefore, the existing studies could have overestimated the diagnostic performance of DW-MRI.

Figure 5.

Funnel graph for the assessment of potential publication bias in DW-MRI test. The funnel graph plots the log of the diagnostic odds ratio (DOR) against the standard error (SE) of the log of the DOR. A indicates the observed summary estimate. B indicates the new summary estimate if all imputed studies were included.

DISCUSSION

The present meta-analysis complied with the recommendations for reporting meta-analyses of diagnostic tests (28). This systematic review identified 14 eligible diagnostic trials that assessed the diagnostic accuracy of DW-MRI for malignant hepatic lesions. The pooled DOR of 14 studies (1665 hepatic lesions) was 133.76. Unlike the traditional ROC plot that explores the effect of varying thresholds (ie, cutpoints for determining positive tests) on sensitivity and specificity in a single study, each data point in the SROC plot represents a separate study. The SROC curve presents a global summary of test performance and shows the trade-off between sensitivity and specificity. We found the area under the SROC to be 0.96, with a lower limit 95% CI of 0.94. The results of this systematic review and meta-analysis indicated that DW-MRI could be used as a helpful diagnostic criterion for malignant hepatic lesions. Figure 6 shows that using a DW-MRI test would raise the posttest probability to 81% when pretest positive from 25% with a PLR of 13 and would reduce the posttest probability as low as 3% when negative with a NLR of 0.1. This indicates that using DW-MRI was helpful for increasing accuracy for detection of the malignant hepatic lesions. It was suggested that ADC was useful in the characterization of focal hepatic lesions. However, we found the threshold value of ADCs had a large variability depending on the studies. Taouli et al (8) found that the ADCs varied by b-values. In our meta-analysis the b-values of included studies also varied.

Figure 6.

Fagan plot of the probability for DW-MRI test in the diagnosis of malignant hepatic lesions, prior probability (0.25).

An exploration of the reasons for heterogeneity rather than the computation of a single summary measure is an important goal of meta-analysis (29). We found significant heterogeneity with regard to sensitivity, specificity, PLR, NLR, DOR, and SROC among the studies analyzed. Our meta-analysis suggested that the SENSE technique may affect diagnostic accuracy. A possible explanation is that in parallel imaging techniques such as SENSE the quality of DW images of the hepatic lesions has markedly improved through the improvement of the signal-to-noise ratio (30, 31). Although the tests did not reach significance in specificity (0.08), solid benign lesion included potentially affected diagnostic specificity. A possible explanation is that DC values of benign solid lesions such as focal nodular hyperplasia and adenoma were similar to those of hepatocellular cancer and metastases (32). This indicated that the diagnostic accuracy of DW-MRI test were underpowered when benign control including solid lesions.

Publication bias is common in diagnostic studies, possibly more so than in studies of randomized controlled trials (33). We detected publication bias in our review. As expected, the missing studies fell below the summary estimate. With imputed values, the recalculated DOR was only a little lower, but it was still close to the observed value, which indicates the true diagnostic performance of DW-MRI. However, the statistical methods used to assess publication bias have limitations (34–37). Therefore, the above findings must be interpreted in this context.

Our meta-analysis had several limitations. First, the exclusion of conference abstracts and letters to the editors may have led to the publication bias that was observed in the present meta-analysis. Second, the number of eligible studies was small, and these studies dealt with very different things (such as the type of benign and malignant hepatic lesions, DW-MRI method), potentially not reflecting the broader experience. An inflation of accuracy estimates could occur due to preferential acceptance of articles reporting favorable results. Third, some of the studies were small in number, especially number of patients. Studies based on small samples sizes may have allowed for a Type II error. Because of the lack of required data reported in the original publications and the small number of eligible studies, we could not analyze the ADC cutoff of the corresponding type of malignant hepatic lesions. Finally, there was a limitation of DW-MRI that solid benign lesions (such as FNH and adenomas) have ADC values similar to those of malignant tumors. Moreover, cystic lesions and hemangiomas are easily characterized by other MR features.

In conclusion, DW-MRI may provide useful information for the diagnosis of malignant hepatic lesions. The evidence presented in this review supports the usefulness of measuring DW-MRI in clinical practice. But DW-MRI may not be a sufficient test for solid lesions due to its low diagnostic specificity. Further studies are required in order to analyze ADC cutoff of the corresponding type of malignant hepatic lesions.

Acknowledgements

Conception and design: Haiping Shen; Analysis and interpretation of the data: Jiyong Jing; Drafting of the article: Dong Xia, Jiyong Jing; Critical revision of the article for important intellectual content: Haiping Shen; Final approval of the article: Dong Xia, Jiyong Jing, Haiping Shen, Jianjun Wu; Provision of study materials or patients: Haiping Shen, Jianjun Wu; Statistical expertise: Jiyong Jing; Administrative, technical, or logistic support: Jiyong Jing; Collection and assembly of data: Dong Xia, Jiyong Jing.

Ancillary