ORIGINAL ARTICLE

Characterization of Liver Tumors by Diffusion-Weighted Imaging: Comparison of Diagnostic Performance Using the Mean and Minimum Apparent Diffusion Coefficient Tomohiro Namimoto, MD, PhD, Masataka Nakagawa, MD, Yuuki Kizaki, MD, Ryo Itatani, MD, Masafumi Kidoh, MD, Daisuke Utsunomiya, MD, PhD, Seitaro Oda, MD, PhD, and Yasuyuki Yamashita, MD, PhD Purpose: To determine the minimum apparent diffusion coefficient (ADCmin) values of benign and malignant hepatic lesions based on diffusion-weighted imaging and to compare the diagnostic performance of ADCmin and mean ADC (ADCmean) values for differentiating between benign and malignant tumors of the liver. Materials and Methods: We retrospectively subjected 240 patients with 195 malignant (hepatocellular carcinoma [HCC], n = 137; metastases, n = 44; cholangiocellular carcinoma [CCC], n = 14) and 45 benign tumors (hemangiomas, n = 37; focal nodular hyperplasia [FNH], n = 8). Both ADCmean and ADCmin were evaluated independently by 2 readers, the sensitivity and specificity for the detection of malignancy were calculated, and receiver operating characteristic (ROC) curves were generated. To determine interobserver agreement, we calculated the Pearson correlation coefficient. Results: Mean ADC (10−3 mm2/s) was 1.19 for malignant (HCC, 1.15; metastasis, 1.23; CCC, 1.51) and 2.01 for benign tumors (hemangioma, 2.09; FNH, 1.52; P < 0.001). Minimum ADC was 0.81 for malignant (HCC, 0.79; metastasis, 0.81; CCC, 0.91) and 1.62 for benign tumors (hemangioma, 1.66; FNH, 1.28; P < 0.001). The sensitivity, specificity, and the calculated area under the ROC curve for diagnosing malignant lesions were 86.2%, 86.7%, and 0.942 (reader 1) and 88.7%, 88.9%, and 0.939 (reader 2) for ADCmean; they were of 92.3%, 97.8%, and 0.984 (reader 1) and 94.9%, 97.8%, and 0.983 (reader 2) for ADCmin. Conclusions: Mean ADC and ADCmin were valuable for differentiating between malignant and benign hepatic lesions. The area under the ROC curve of ADCmin was significant higher than that of ADCmean.

and this is a major limitation that hampers the widespread use of DWI to diagnose liver lesions. Significantly higher ADC values have been demonstrated for benign than malignant hepatic lesions and overlap varied.1–8 Based on their meta-analysis, Xia et al12 reported an area under the curve (AUC) of the summary receiver operator characteristic (ROC) of 0.96; sensitivity ranged from 0.74 to 1.0 (mean, 0.91) and specificity ranged from 0.77 to 1.00 (mean, 0.93). Most of earlier studies focused on the mean ADC (ADCmean) values, and to our knowledge, there has been no comprehensive evaluation of the minimum ADC (ADCmin). The ADC value is inversely related to tumor cellularity, and it has been suggested that ADCmin, the lowest ADC value within lesions, facilitates the accurate grading of astrocytic brain tumors because the regions with ADCmin correspond to the sites of highest cellularity within heterogeneous tumors.13–15 Malignant liver lesions frequently show pathologic heterogeneity, including cancer nests, ductal components, intratumoral necrosis, or fibrosis and are characterized by radiologically heterogeneous features. Based on these considerations, we postulated that areas with the ADCmin reflect the sites of highest cellularity within heterogeneous liver tumors and that these sites are of importance for their differential diagnosis. The purposes of our study were to determine the ADCmin of benign and malignant hepatic lesions based on DWI findings and to compare the diagnostic performance of ADCmin and ADCmean in their differentiation.

Key Words: liver, neoplasm, magnetic resonance imaging, diffusion (J Comput Assist Tomogr 2015;39: 453–461)

D

iffusion-weighted imaging (DWI) has been reported to be useful for the detection and characterization of focal liver lesions.1–8 It can facilitate the acquisition of criteria for lesion characterization independent of the T1- and T2-relaxation times. It does not require the administration of contrast agents because it quantifies diffusion effects via apparent diffusion coefficient (ADC) measurements. The diffusion coefficient is related to the molecular mobility of water molecules and reflects tissue properties such as the size of the extracellular space, viscosity, and cellularity.9–11 Quantitative ADC threshold values of variable accuracy, depending on the patient population and lesion type, have been reported for lesion characterization. The proposed ADC cutoffs depend also on the b values used for acquisition, From the Department of Diagnostic Radiology, Graduate School of Life Sciences, Kumamoto University, Honjo, Kumamoto, Japan. Received for publication October 9, 2014; accepted January 7, 2015. Reprints: Tomohiro Namimoto, MD, PhD, Department of Diagnostic Radiology, Graduate School of Life Sciences, Kumamoto University, 1–1–1, Honjo, Kumamoto 860-8556, Japan (e‐mail: [email protected]). The authors declare no conflict of interest. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

MATERIALS AND METHODS Study Population This retrospective study was approved by our institutional review board; informed patient consent was waived. To protect patient privacy, we removed all identifiers from our records at the completion of our analyses. We queried our magnetic resonance imaging (MRI) database to identify patients who underwent studies from October 2010 to September 2011 to evaluate focal liver lesions and patients in whom such lesions were detected incidentally (Fig. 1). We identified 311 patients with focal liver tumors; 49 were initially excluded because the nature of their focal lesions was not confirmed. Of these, 35 had suspected liver tumors without follow-up studies, and in the other 14, the focal lesion was smaller than or equal to 10 mm. We subsequently excluded another 22 patients because they had undergone antineoplastic chemotherapy within 6 months of the MRI studies. The final study population was composed of 240 patients (154 were male, 86 were female), ranging in age from 12 to 87 years (mean [SD], 65.55 [12.10] years). Of these, 195 had malignant tumors (hepatocellular carcinoma [HCC], n = 137; metastases, n = 44; cholangiocellular carcinoma [CCC], n = 14) and 45 were found

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

www.jcat.org

453

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Namimoto et al

FIGURE 1. Flowchart of patient selection.

to have benign lesions (hepatic cavernous hemangioma, n = 37; focal nodular hyperplasia [FNH], n = 8). The diagnosis of HCC was based on surgical (n = 53) or percutaneous biopsy findings (n = 8). Another 76 patients presented with a typical clinical history of cirrhosis or chronic hepatitis, elevated α-fetoprotein levels and lipiodol uptake after transhepatic arterial chemoembolization, or disease progression on follow-up computed tomographic or MRI scans acquired at least 1 year after the initial imaging studies. In 28 of the 44 patients with metastases, the nature of the lesion was confirmed histopathologically after biopsy and/or surgery. In the remaining 16, the diagnosis was based on tracer uptake by the lesions on positron emission tomography–computed tomographic scans and disease progression or a decrease in the lesion size on serial cross-sectional

imaging studies performed after the start of chemotherapy in patients with known extrahepatic primary malignancies. The primary tumors in the 44 patients with metastases were colorectal (n = 27), gastric (n = 5), pancreatic (n = 5), gall bladder (n = 2), breast (n = 1), esophageal (n = 1), ovarian (n = 1), neuroendocrine (n = 1), and urothelial carcinoma (n = 1). All 14 CCCs were histopathologically verified by biopsy and/or surgery. The diagnosis of hemangioma and FNH was findings of histopathologic analysis of biopsy or surgical specimens or typical patterns on MR images without change over time, with a follow-up of at least 1 year. Hemangiomas were diagnosed on MRI scans based on the combination of high signal intensity on T2-weighted imaging and typical postcontrast discontinuous peripheral nodular and centripetal enhancement in conjunction with a stable lesion size on serial

FIGURE 2. Box and whisker plots of the ADCmean (A) and ADCmin (B) values of focal hepatic lesions.

454

www.jcat.org

© 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Liver Tumor Characterization With Minimum ADC

FIGURE 3. A 73-year-old woman with HCC. A, DWI at b = 0 s/mm2. The tumor (arrow) is high signal intense in DWI at b = 0 s/mm2. B, DWI at b = 800 s/mm2. C, ADC map. D, Dynamic gadoxetic acid–enhanced MRI in the arterial phase. There is signal loss between b = 0 s/mm2 (A) and b = 800 s/mm2 (B). The ADC map shows intratumoral heterogeneity and the coexistence of high- and low-value areas. The ADCmean is 1.52  10−3 mm2/s, and the ADCmin is 0.93  10−3 mm2/s. The ADCmean value wrongly classifies this as a benign lesion. Dynamic-enhanced MRI (arterial phase) demonstrates early lesion enhancement.

cross-sectional imaging studies. Of the 8 FNH cases, 2 were confirmed histopathologically after surgery. The other 6 were diagnosed based on typical MRI features and upon fulfilling the following criteria: (a) isointense or slightly hypointense compared with the liver on T1-weighted images; (b) isointense or slightly hyperintense on T2-weighted imaging; (c) intense, homogeneous enhancement during the hepatic arterial phase; and (d) isointense or hyperintense in relation to the adjacent liver parenchyma during hepatic venous and hepatobilliary phases. In patients with multiple lesions, only the largest lesion was selected for further analysis by 1 of the 2 radiologists who established the final diagnosis.

MR Imaging A 3.0-T whole-body MRI scanner (Intera Achieva 3.0 T TX; Philips Healthcare, Best, the Netherlands) and a 32-channel sensitivity-encoding (SENSE) cardiac phased-array coil were used. The liver was imaged in the axial plane in all patients both before and after the administration of 0.1 mL/kg (0.025 mmol/ mL) gadoxetic acid (Primovist; Bayer Schering Pharma, Berlin, Germany). The contrast agent was intravenously administered automatically at a rate of 1 mL/s with a power injector, followed by a 30-mL saline flush. The MRI protocol included a chemical shift (in-phase/opposed-phase) sequence (repetition time [TR]/ echo time [TE], 144/1.2 and 2.3 ms; flip angle, 55°; matrix size, 237  176; number of signal averaging [NSA] 1; and SENSE 2), a breath-hold single-shot fat-suppressed T2-weighted sequence (TR/TE, 751/80 ms; flip angle, 90°; matrix size, 288  174; NSA 2; and SENSE 2), a breath-hold single-shot fat-suppressed heavily T2-weighted sequence (TR/TE, 1130/200 ms; flip angle, 90°; matrix size, 320  239; NSA 1; and SENSE 2), a 7-mm section thickness and a 1-mm intersection gap, and a field of view (FOV) of 35 to 40 cm. For gadoxetic acid–enhanced MRI-unenhanced, arterial phase (30-45 seconds), portal phase (80 seconds), late (240 seconds), and 20-minute delayed

hepatobiliary phase images were obtained using a T1-weighted 3D turbo-field-echo sequence (T1-weighted high-resolution isotropic volume examination; Philips Healthcare; TR/TE, 3.1/ 1.5 ms; flip angle, 10°; matrix size, 304  243; NSA 1; and SENSE 2) with a 4-mm section thickness and an FOV of 35 to 40 cm. Diffusion-weighted single-shot echo-planar imaging with simultaneous respiratory triggering was performed using a TR/ TE of 2100/60 ms. The TR was matched to each patient's length of the respiratory cycle. The scanning parameters were a b value of 0 and 800 s/mm2, spectral presaturation with inversion recovery for fat suppression; matrix size, 128  112; SENSE, 2; FOV, 38  38 cm; NSA 4; slice thickness, 7 mm; slice gap, 1 mm; and 25 axial slices.16 Because of no significant variation of diffusion-weighted parameters after gadoxetic acid administration, we performed DWI after contrast-enhanced dynamic images.17 An ADC map was generated from 0 and 800 s/mm2b values.

Image Analysis Apparent diffusion coefficient values were calculated using the formula ADC ¼ð− ln ½Sb=S0Þ=b; where Sb is the signal intensity of the region of interest (ROI) on DWI scans at a b value of 800 s/mm2, S0 is the signal intensity of the ROI on DWI scans at a b value of 0 s/mm2, and b is the gradient b factor with a value of 800 s/mm2. Apparent diffusion coefficient maps were calculated on a pixel-by-pixel basis using the built-in software of the MR unit. Two independent radiologists (reader 1 and reader 2, with 20 and 4 years of experience reading liver MRI, respectively) manually drew ROIs on the DWI scans. They were blinded to the final pathologic results. Initially, a slice from the axial ADC maps representing the mass lesion with the largest diameter was selected. Regions of interest were placed in

© 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

www.jcat.org

455

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Namimoto et al

FIGURE 4. A 62-year-old woman with nodule-in-nodule HCC. DWI at b = 0 s/mm2. B, DWI at b = 800 s/mm2. C, ADC map. D, Dynamic gadoxetic acid–enhanced MRI in the arterial phase. E, hepatobiliary phase. A, The nodule-in-nodule HCC arising from degenerative nodule (arrow) shows central high and peripheral low signal intensity in DWI b = 0. In the central area of the cancerous lesion, a barely signal loss from b = 0 (A) to b = 800 (B) is shown. The ADC map (C) shows hyporintensity in the central area. The central area demonstrates intense arterial enhancement (D) and low signal intensity in the hepatobiliary phase (E). The ADCmean is 1.08  10−3 mm2/s, the ADCtarget is 1.07  10−3 mm2/s, and the ADCmin is 0.93  10−3 mm2/s. All ADC parameters are correctly diagnosed as malignant lesion.

regions with high signal intensity on the DWI (b value of 0 s/mm2) scans; the contrast and morphologic characteristics at the early phase of contrast-enhanced T1- and T2-weighted imaging were used to guide the ROI placements to avoid areas of T2 shinethrough that are usually found in necrotic or cystic parts. T1-weighted imaging was used to avoid areas of hemorrhage with very low ADC values. The ROIs were then copied and pasted onto the corresponding ADC map for quantitative analysis. Mean ADC and ADCmin values were selected for analysis. Magnetic resonance imaging interpretation was on 2 megapixel color liquid crystal displays (RadiForce RX220; Eizo) equipped with image viewer software (ViewR version 1.20; Yokogawa Electric Corporation).

ADCmean and ADCmin cutoff values for the optimal differentiation of benign and malignant liver lesions. The accuracy of each MR technique (ADCmean vs ADCmin) was determined by calculating the AUC. The differences between ROC curves were tested for significance. Then, ADCs from both radiologists were merged, taking the average value for the ADCmean and ADCmin analyses. Apparent diffusion coefficient values were compared among groups using post hoc testing with the Scheffe method. For all tests, P < 0.05 was considered indicative of a statistically significant difference. All statistical analyses were conducted with MedCalc software version 11.3.2.0 (Mariakerke, Belgium).

RESULTS Statistical Analysis To assess the reliability of our multiple-ROI approach, interobserver differences between readers 1 and 2 were evaluated using the Pearson correlation coefficient for the ADCmean and ADCmin values. The level of correlation was defined as very strong (R = 1.0–0.9), strong (R = 0.9–0.7), moderate (R = 0.7–0.5), and weak (R < 0.5). An ROC curve analysis was implemented to define

456

www.jcat.org

As shown in the box and whisker plots of Figure 2, the ADCmean value (10−3 mm2/s) of the HCCs, metastases, CCCs, FNHs, and hemangiomas was 1.15 ± 0.21, 1.23 ± 0.32, 1.51 ± 0.47, 1.52 ± 0.26, and 2.09 ± 0.43, respectively; the corresponding ADCmin values were 0.79 ± 0.20, 0.81 ± 0.26, 0.91 ± 0.39, 1.28 ± 0.10, and 1.66 ± 0.33. Magnetic resonance imaging scans of representative patients are presented in Figures 3 to 7. The ADCmean of hemangiomas was significantly higher than that of HCCs and © 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Liver Tumor Characterization With Minimum ADC

FIGURE 5. A 55-year-old woman with metastasis from rectal cancer. A, DWI at b = 0 s/mm2. The metastatic lesion (arrow) shows central high and peripheral intermediate high signal intensity in b = 0 s/mm2. B, DWI at b = 800 s/mm2. C, ADC map. D, Fat-suppressed T2-WI. In the central area of the metastatic lesion, there is substantial signal loss, b = 800 s/mm2 (B), and the ADC map (C) shows hyperintensity. In the peripheral area, there is signal loss between b = 0 s/mm2 (A) and b = 800 s/mm2 (B). The ADCmean of the mass is 1.47  10−3 mm2/s, and the ADCmin is 0.88  10−3 mm2/s. Based on the ADCmean value, this is classified as a benign lesion. Fat-suppressed T2-WI showed central high and peripheral intermediate high signal intensity. WI indicates weighted imaging.

metastases (P < 0.01), and that of FNHs was significantly higher than that of HCCs (P < 0.05). The ADCmin of hemangiomas was significantly higher than that of HCCs, metastases, and CCCs (P < 0.01) and this value was significantly higher in FNHs than in HCCs and metastases (P < 0.01). There was no statistically significant difference in ADCmean and ADCmin among HCCs, metastases, and CCCs, or between hemangiomas and FNHs. The results of ROC analysis are shown in Figure 8. The ADCmean values (10−3 mm2/s) of malignant tumors were 1.19 ± 0.31 (reader 1) and 1.19 ± 0.28 (reader 2); for benign tumors, these values were 2.00 ± 0.47 (reader 1) and 1.96 ± 0.46 (reader 2). For malignant tumors, the readers assigned ADCmin values of 0.82 ± 0.27 (reader 1) and 0.83 ± 0.28 (reader 2); these values were 1.62 ± 0.34 (reader 1) and 1.57 ± 0.39 (reader 2) for benign tumors. The difference between malignant and benign tumors in both the ADCmean and the ADCmin was statistically significant (P < 0.001). The AUC, sensitivity, and specificity levels for each reader and each type of ADC are shown in Table 1. The AUC for ADCmean for diagnosing malignant lesions was 0.942 for reader 1 and 0.939 for reader 2. The AUC for ADCmin for diagnosing malignant tumors was 0.984 and 0.983 for readers 1 and 2, respectively. The difference between the ADCmean and ADCmin values for AUC was significant for both readers (P < 0.05). The diagnostic performance of both ADCmean and ADCmin was high for malignant tumors. The use of ADCmin raised sensitivity from 86.2% to 92.3% in reader

1 and from 88.7% to 94.9% in reader 2; specificity increased from 86.7% to 97.8% in reader 1 and from 88.9% to 97.8% in reader 2. Interobserver reproducibility between readers 1 and 2 was very strong (0.965) for ADC mean and strong (0.882) for ADCmin values.

DISCUSSION Diffusion-weighted imaging studies have shown that ADC maps can be used to characterize and distinguish between certain benign and malignant focal liver lesions.1–8 Given the limitations of visual assessments, the ADC-based quantitative analysis of liver tumors has been proposed for their differentiation. However, different levels of sensitivity (74%–100%) and specificity (77%–100%) were reported in studies using ADCmean quantification for the diagnosis of malignant hepatic lesions.12 Moreover, significantly different cutoff thresholds for ADCmean, ranging from 1.47  10−3 to 1.63  10−3 mm2/s, have been proposed for the optimal differentiation between benign and malignant lesions.1–8,12 This discordance may reflect differences in instrumentation, the b values, the methods used for the calculation of ADCmean, and the patient population as well as the absence of standardized image acquisition parameters. The best results were obtained in patients with no or few solid benign tumors, for example, FNH and hepatocellular adenoma, whose ADCmean values are in the range of malignant lesions.4,5 In a group of patients with

© 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

www.jcat.org

457

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Namimoto et al

FIGURE 6. A 62-year-old man with FNH. A, DWI at b = 0 s/mm2. The FNH (arrow) is moderately hyperintense in b = 0 s/mm2. B, DWI at b = 800 s/mm2. C, ADC map. E and E, Dynamic gadoxetic acid–enhanced MRI (arterial phase, D; hepatobiliary phase, E). There is moderate signal loss between b = 0 s/mm2 (A) and b = 800 s/mm2 (B). The ADCmean of the mass is 1.51  10−3 mm2/s, and the ADCmin of this lesion is 1.35  10−3 mm2/s. Based on both ADCmean and ADCmin, a correct diagnosis of benign lesion was returned. D, The FNH demonstrates intense arterial enhancement. E, Hyperintensity persists in the hepatobiliary phase; this is a typical feature of FNH.

15 solid malignant liver tumors (13 metastases, 2 HCCs) and 22 nonsolid benign lesions (15 cysts, 7 hemangiomas), sensitivity and specificity were 100% for malignant lesions.5 However, because no solid benign lesions such as FNH or hepatocellular adenoma were included in that study, the probability of a falsepositive malignancy diagnosis was reduced. Among 204 liver lesions studied by Bruegel et al,4 only 4 (3.6%) of 111 benign lesions were FNHs. In their study, the sensitivity and specificity for diagnosing malignant tumors were 90% and 86%, respectively. Realistic results of 74% sensitivity and 77% specificity were obtained by Parikh et al,6 who analyzed 211 liver lesions. In their series, the number and percentage of solid benign lesions (5 adenomas and 4 FNHs; 12%) were higher and could have influenced the diagnostic performance of ADCmean values. Similarly, in our study population, 8 (17.8%) of the 45 benign lesions were solid FNHs. The ADCmean of one of the FNHs was in the range of malignant tumors, resulting in a false-positive diagnosis and, consequently, in a slight decrease in specificity. Other studies also found no statistically significant difference in the ADCmean of FNH, hepatic metastases, and HCC.4,6 However, unlike earlier studies, a recent study of Colagrande et al18 found a statistically significant difference in the ADCmean between FNH and metastases. We also identified a statistically significant difference between the ADCmin value of FNH on the one hand and hepatic metastases and HCC on the other (P < 0.05). We document that the ADCmean and ADCmin values of malignant were lower than those of benign hepatic tumors, and the differences were statistically significant. The lower ADC values of malignant tumors may be attributable to histopathologic

458

www.jcat.org

features of the resected specimens. In malignant tumors, we observed tissue invasion and cancer nests exhibiting increased cellularity and enlarged cells. The diffusion of water molecules in malignant tumor was limited by hypercellularity, enlarged nuclei, hyperchromatism, high nuclear-to-cytoplasmic ratio and reduced extracellular space. These histopathologic characteristics resulted in a decrease in the ADC value. In clinical studies of organs other than the liver, ADCmin was found to be informative. For example, ADCmin but not ADCmean was different in osteosarcoma that did and did not respond to chemotherapy.19 In a DWI study of patients with glioma, ADCmin correlated with cellularity,11 and in a study of patients with astrocytoma, the tumor ADCmin before treatment was identified as a potential marker of survival.15 Although these findings suggest that ADCmin is valuable in a variety of cancers, little is known about its value in patients with liver tumors.20 To the best of our knowledge, ours is the first attempt to evaluate the role of ADCmin in the characterization of liver tumors. We demonstrated that it was very highly accurate in distinguishing between malignant and benign nodules (AUC of 0.980 [reader 1] and 0.983 [reader 2]). We posit that the lower accuracy of ADCmean was due to the unusually high ADC values of some metastases and CCCs in our study population. These malignant tumors tend to exhibit typical malignant features, although metastases with large cystic necrosis that may manifest increased signal intensity and a higher ADCmean may seem to be benign tumors.21 We also found that degenerative nodules in some HCCs were only partly replaced by cancer cells and that such HCCs contained nonmalignant areas © 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Liver Tumor Characterization With Minimum ADC

FIGURE 7. A 68-year-old woman with hemangioma. A, DWI at b = 0 s/mm2. The hemangioma (arrow) shows marked high signal intensity in b = 0 s/mm2. B, DWI at b = 800 s/mm2. There is substantial signal loss in b = 800 s/mm2. C, ADC map. The ADC map shows hyperintensity. D, Fat-suppressed T2-WI. The ADCmean of the mass is 2.08  10−3 mm2/s, and the ADCmin is 1.51  10−3 mm2/s. Based on both ADCmean and ADCmin, the lesion was correctly diagnosed as benign. Fat-suppressed T2-W MRI revealed marked high signal intensity. WI indicates weighted imaging.

with higher ADC values and malignant areas with lower ADC values. Therefore, we posited that analyzing the ADCmin in “hot spots” was useful for detecting malignant nodules with

only focal infiltration by cancer cell nests. Our hypothesis was confirmed by the finding that the AUC, sensitivity, specificity, and accuracy of ADCmin were higher than those of ADCmean for

FIGURE 8. ROC curves of reader 1 (A) and reader 2 (B) for ADCmean (solid line) and ADCmin (dotted line) values. © 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

www.jcat.org

459

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

Namimoto et al

TABLE 1. ROC Analysis

ADCmean ADCmin

Reader

AUC

SE

95% CI

Cutoff Value (10−3 mm2/s)

Sensitivity, %

Specificity, %

R1 R2 R1 R2

0.942 0.939 0.984 0.983

0.025 0.025 0.013 0.014

0.904–0.968 0.901–0.966 0.959–0.996 0.957–0.995

1.49 1.52 1.14 1.16

86.2 88.7 92.3 94.9

86.7 88.9 97.8 97.8

CI indicates confidence interval.

the differentiation of malignant from benign tumors. Although the difference was not statistically significant, we found that the diagnostic performance of ADCmin was superior and that this value can be used as a parameter at hepatic DWI to differentiate between malignant and benign lesions. The lack of statistical significance between our ADCmin and ADCmean values may also be attributable to our small sample size. Studies with larger patient cohorts are underway to determine the best ADC parameter for the diagnosis of liver masses. Our study has some limitations. First, its design was retrospective and we included patients who had undergone MRI for diagnostic purposes as part of routine clinical care. Although this may have introduced some bias, we think that our results are valid because we included consecutive patients seen in the course of a relatively long period. Second, not all liver tumors were diagnosed histopathologically. Nevertheless, careful consensus readings by experienced abdominal radiologists and follow-up examinations established the final diagnosis in these cases. Third, the number of investigated benign lesions was relatively small (18.8 %) compared with the number of malignant lesions. In addition, the inclusion of some solid benign lesions such as 8 lesions of FNH and no lesion of hepatic adenomas limited the calculated descriptive statistics for these lesions. Fourth, ADC values can change in the presence of hemorrhagic areas and/or artifacts. To avoid the effect of susceptibility artifacts, we did not place ROIs in areas with obvious artifacts or gross lesional hemorrhages. Fifth, ADCmin values at body DWI studies can be subject to noise contamination. However, histogram analysis to assess the spread of ADC values within selected ROIs may overcome this problem. Fifth, the ADCmean, ADCmin, and the threshold values in our study were equipment specific and may not apply to different vendors and b values. Additional studies using different b values and MR units are underway to verify our current findings. In conclusion, our study suggests that the calculation of ADCmean and ADCmin is useful for the differentiation between malignant and benign hepatic lesions. Our comparison of 2 quantitative methods for the characterization of liver lesions demonstrated that ADCmin was more highly sensitive and specific than ADCmean. The difference in AUC between ADCmin and ADCmean was statistically significant. The inclusion of DWI in MR protocols to study the liver, combined with the ADCmin value of the liver lesion, may be an additional valuable tool facilitating the discrimination of malignant from benign hepatic lesions. Prospective trials are underway to fully validate the clinical usefulness of ADCmin in the characterization of focal liver lesions.

REFERENCES 1. Namimoto T, Yamashita Y, Sumi S, et al. Focal liver masses: characterization with diffusion weighted echo-planar MR imaging. Radiology. 1997;204:739–744.

460

www.jcat.org

2. Kim T, Murakami T, Takahashi S, et al. Diffusion-weighted single-shot echoplanar MR imaging for liver disease. AJR Am J Roentgenol. 1999;173: 393–398. 3. Taouli B, Vilgrain V, Dumont E, et al. Evaluation of liver diffusion isotropy and characterization of focal hepatic lesions with two single-shot echoplanar MR imaging sequences: prospective study in 66 patients. Radiology. 2003;226:71–78. 4. Bruegel M, Holzapfel K, Gaa J, et al. Characterization of focal liver lesions by ADC measurements using a respiratory triggered diffusion-weighted single-shot echo-planar MR imaging technique. Eur Radiol. 2008;18: 477–485. 5. Gourtsoyianni S, Papanikolaou N, Yarmenitis S, et al. Respiratory gated diffusion-weighted imaging of the liver: value of apparent diffusion coefficient measurements in the differentiation between most common encountered benign and malignant focal liver lesions. Eur Radiol. 2008;18: 486–492. 6. Parikh T, Drew SJ, Lee VS, et al. Focal liver lesion detection and characterization with diffusion-weighted MR imaging: comparison with standard breath-hold T2-weighted imaging. Radiology. 2008;246: 812–822. 7. Goshima S, Kanematsu M, Kondo H, et al. Diffusion-weighted imaging of the liver: optimizing b value for the detection and characterization of benign and malignant hepatic lesions. J Magn Reson Imaging. 2008;28:691–697. 8. Taouli B. Diffusion-weighted MR imaging for liver lesion characterization: a critical look. Radiology. 2012;262:378–380. 9. Le Bihan D, Breton E, Lallemand D, et al. Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging. Radiology. 1998; 168:497–505. 10. Marks MP, de Crespigny A, Lentz D, et al. Acute and chronic stroke: navigated spin-echo diffusion-weighted MR imaging. Radiology. 1996; 199:403–408. 11. Sugahara T, Korogi Y, Kochi M, et al. Usefulness of diffusion weighted MRI with echo-planar technique in the evaluation of cellularity in gliomas. J Magn Reson Imaging. 1999;9:53–60. 12. Xia D, Jing J, Shen H, et al. Value of diffusion-weighted magnetic resonance images for discrimination of focal benign and malignant hepatic lesions: a meta-analysis. J Magn Reson Imaging. 2010;32:130–137. 13. Kitis O, Altay H, Calli C, et al. Minimum apparent diffusion coefficients in the evaluation of brain tumors. Eur J Radiol. 2005;55:393–400. 14. Lee EJ, Lee SK, Agid R, et al. Preoperative grading of presumptive low-grade astrocytomas on MR imaging: diagnostic value of minimum apparent diffusion coefficient. AJNR Am J Neuroradiol. 2008;29: 1872–1877. 15. Murakami R, Hirai T, Sugahara T, et al. Grading astrocytic tumors by using apparent diffusion coefficient parameters: superiority of a one- versus two-parameter pilot method. Radiology. 2009;251:838–845. 16. Kaya B, Koc Z. Diffusion-weighted MRI and optimal b-value for characterization of liver lesions. Acta Radiol. 2014;55:532–542.

© 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

J Comput Assist Tomogr • Volume 39, Number 4, July/August 2015

17. Colagrande S, Mazzoni LN, Mazzoni E, et al. Effects of gadoxetic acid on quantitative diffusion-weighted imaging of the liver. J Magn Reson Imaging. 2013;38:365–370. 18. Colagrande S, Regini F, Pasquinelli F, et al. Focal liver lesion classification and characterization in noncirrhotic liver: a prospective comparison of diffusion-weighted magnetic resonance–related parameters. J Comput Assist Tomogr. 2013;37:560–567. 19. Oka K, Yakushiji T, Sato H, et al. The value of diffusion-weighted imaging for monitoring the chemotherapeutic response of osteosarcoma:

Liver Tumor Characterization With Minimum ADC

a comparison between average apparent diffusion coefficient and minimum apparent diffusion coefficient. Skeletal Radiol. 2010;39:141–146. 20. Nakanishi M, Chuma M, Hige S, et al. Relationship between diffusion-weighted magnetic resonance imaging and histological tumor grading of hepatocellular carcinoma. Ann Surg Oncol. 2012;19: 1302–1309. 21. Chan JH, Tsui EY, Luk SH. Diffusion-weighted MR imaging of the liver: distinguishing hepatic abscess from cystic or necrotic tumor. Abdom Imaging. 2001;26:161–165.

© 2015 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.

www.jcat.org

461

Characterization of Liver Tumors by Diffusion-Weighted Imaging: Comparison of Diagnostic Performance Using the Mean and Minimum Apparent Diffusion Coefficient.

To determine the minimum apparent diffusion coefficient (ADC(min)) values of benign and malignant hepatic lesions based on diffusion-weighted imaging ...
3MB Sizes 0 Downloads 9 Views

Recommend Documents