Detection of hepatocellular carcinoma (HCC) using super paramagnetic iron oxide (SPIO)-enhanced MRI: Added value of diffusion-weighted imaging (DWI)

Authors


Abstract

Purpose:

To evaluate whether diffusion-weighted imaging (DWI) improves the detection of hepatocellular carcinoma (HCC) on super paramagnetic iron oxide (SPIO)-enhanced MRI.

Materials and Methods:

This retrospective study group consisted of 30 patients with 50 HCC nodules who underwent MRI at 1.5 Tesla. Two combined MR sequence sets were compared for detecting HCC: SPIO-enhanced MRI (axial T2-weighted fast spin-echo (FSE) and T1-/T2*-weighted fast field echo (FFE) scanned before and after administration of ferucarbotran) and SPIO-enhanced MRI + DWI (SPIO-enhanced MRI with axial DWI scanned before and after administration of ferucarbotran). Three blinded readers independently reviewed for the presence of HCC on a segment-by-segment basis using a four-point confidence scale. The performance of the two combined MR sequence sets was evaluated using receiver operating characteristic (ROC) analysis.

Results:

The average area under the ROC curve (Az) of the three readers for the SPIO-enhanced MRI + DWI set (0.870 ± 0.046) was significantly higher that that for the SPIO-enhanced MRI set (0.820 ± 0.055) (P = .025). The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for detection of HCC were 66.0%, 98.0%, 90.0%, and 91.4%, respectively, for the SPIO-enhanced MRI set, and 70.0%, 98.6%, 92.9%, and 92.4%, respectively, for the SPIO-enhanced MRI + DWI set.

Conclusion:

The SPIO-enhanced MRI + DWI set outperformed the SPIO-enhanced MRI set for depicting HCC. J. Magn. Reson. Imaging 2010; 31: 373–382. © 2010 Wiley-Liss, Inc.

SUPERPARAMAGNETIC IRON OXIDE (SPIO) is a liver-targeting agent taken up by the reticuloendothelial system and is clinically used to improve the detectability of focal liver lesions, mainly by increasing tumor-to-liver contrast (1). The uptake of SPIOs by residual Kupffer cells has been reported in some tumors such as focal nodular hyperplasia (2), hepatic adenoma (2), and well-differentiated hepatocellular carcinoma (HCC) (3). SPIO-MRI thus also contributes to the radiological differentiation of liver tumors.

Diffusion-weighted echo-planar imaging (DWI) of the brain is widely used in the evaluation of acute stroke. However, the application of DWI to the abdomen and pelvis has proved more challenging due to artifacts from susceptibility, respiration, and chemical shift, which cause image distortion and fat misregistration (4). Despite these obstacles, recent reports have demonstrated the feasibility of DWI for the detection and characterization of liver tumors (5–12). This sequence is known as its high-contrast resolution (5) and may contribute to improvement of tumor identification. In fact, Nasu et al (5) have shown that combined reading of DWI with T1- and T2-weighted imaging results in higher accuracy in the detection of hepatic metastases than reading of SPIO-enhanced MR images. In addition, the apparent diffusion coefficient of benign hepatic lesions has recently been reported to be higher than that of malignant lesions, including lesions related to liver metastasis and HCC (7–10). However, to the best of our knowledge, there have been only a few reports focusing on the detectability of HCC on DWI (6, 12).

In this study, we investigated whether DWI can improve the detection of HCC lesions on SPIO-enhanced MRI.

MATERIALS AND METHODS

Patients

This study was approved by the institutional review board at our hospital, and the requirements for informed consent were waived for the retrospective study. The study included 30 consecutive patients who underwent SPIO-enhanced MRI including DWI and combined computed tomography hepatic arteriography (CTHA) and CT arterioportography (CTAP) before surgery for HCC, as indicated by medical data recorded at our hospital between November 2006 and November 2007. These patients, who constituted the study group, included 23 men and seven women (age range: 34–78 years; mean age 66.0 years). The MR and CT studies were performed within one month before surgery for all cases. Three patients had a history of hepatectomy, and three other patients had undergone radio frequentry ablation (RFA) for HCC. The hepatitis B surface antigen was present in five cases and the hepatitis C virus antibody in 21 cases. Neither the hepatitis B surface antigen nor the hepatitis C virus antibody was seen in four cases. The grading of liver dysfunction was pre-operatively evaluated based on the Child-Pugh classification, and 26 and four patients were categorized into Grades A and B, respectively. The surgical methods performed included partial hepatectomy in 19 cases, subsegmentectomy in two cases, segmentectomy in six cases, and lobectomy in one case. Only RFA after opened laparotomy was performed in the remaining two cases.

Tumor Confirmation and Characterization

A total of 50 nodules were identified in 30 patients. Their size ranged from 0.5 cm to 6.5 cm, with a mean of 2.0 (±1.39) cm. Thirty-two nodules were more than 1 cm in maximal diameter, and 18 nodules were 1 cm or less. Thirty-seven nodules were pathologically proven to be HCCs after resection. They included one well-differentiated (w), seven well- to moderately differentiated (wm), 22 moderately differentiated (m), five moderately to poorly differentiated (mp), and two poorly differentiated (p) HCCs. Histological examination of non-cancerous liver parenchyma in the 28 patients who underwent hepatic resection showed 10 cases of cirrhosis and 18 cases of chronic hepatitis or fibrosis, as classified by the Japanese Liver Cancer Study Group. The remaining 13 nodules were clinically judged to be HCCs by a study coordinator described later because they showed nodular enhancement on the first phase of CTHA, “corona enhancement” on the second phase of CTHA, and a perfusion defect on CTAP and also were detected on intraoperative ultrasound at the corresponding site. The nodules did not show strong hyperintensity on T2-weighted fast spin-echo (FSE). Follow-up CT or MRI at least six months after surgery ascertained the absence of HCC nodules in the remaining liver.

MRI and Combined CTHA and CTAP

MRI was performed on a whole-body 1.5 Tesla scanner (Intera Achieva Nova Dual; Philips Medical Systems, Best, Netherlands) equipped with a four-element sensitivity encoding (SENSE) body coil. Imaging included an axial dual-echo T2-weighted FSE, axial dual-echo T1-weighted fast field echo (FFE), axial dual-echo T2*-weighted FFE, and axial DWI. All sequences covered the whole liver. Detailed imaging parameters were as follows: a) T2-weighted FSE, respiratory trigger, 1313/80 and 160 (repetition time [TR] msec/echo time [TE] msec), echo train length = 18, matrix = 256 × 143, field of view (FOV) = 36 cm × 28.7 cm, SENSE factor = 1.3, spectral pre-saturation inversion recovery, section thickness = 8 mm, intersection gap = 2 mm, signal average = 1, 20 sections acquired, and acquisition time = approximately 4–5 minutes; b) T1-weighted FFE, breath-hold, TR/TE = 165 msec/2.3 msec and 4.6 msec, flip angle = 75°, matrix = 256 × 143, FOV = 36 cm × 28.7 cm, SENSE factor = 1.4, section thickness = 8 mm, intersection gap = 2 mm, signals average = 2, 20 sections acquired, and acquisition time = 18 seconds; c) T2*-weighted FFE, breath-hold, TR/TE = 165 msec/9.2 msec and 18.4 msec, flip angle = 75° matrix = 256 × 143, FOV = 36 cm × 28.7 cm, SENSE factor = 1.4, section thickness = 8 mm, intersection gap = 2 mm, signals average = 2, 20 sections acquired, and acquisition time = 18 seconds; d) DWI, respiratory trigger, single shot echo-planar imaging, TR/TE = 2386 msec/72 msec, matrix = 128 × 81, FOV = 36 cm × 32.6 cm, section thickness = 8 mm, intersection gap = 2 mm, half scan factor = 0.698, signal average = 1, spectral pre-saturation inversion recovery, SENSE factor = 2, b-factors = 0 and 1000, diffusion gradients applied in 3 axes, 20 sections acquired, and acquisition time = two to three minutes. Data acquisition was twice performed before and 10 minutes after intravenous injection of 0.016 mmol Fe/kg of ferucarbotran (Resovist; Nihon Schering or Bayer, Osaka, Japan).

Combined CTHA and CTAP scanning were performed as part of the preoperative angiographic examination using an Aquilion (multidetector row CT; 16 rows), Toshiba, Japan. During a single breath-hold period a craniocaudal scan was performed with the following parameters: 120 kVp, 300 mA, collimation = 1 mm, reconstructed to a slice thickness = 3 mm, table speed = 15 mm/rotation and pitch of 15. For CTAP, the catheter was placed in the superior mesenteric artery or, if the replaced right hepatic artery branched from it, in the portion distal to the branching site. Following a transarterial infusion of prostaglandin E1 (10 mg), the data acquisition was initiated 30 seconds after the initiation of a transcatheter arterial injection of 100 mL of iopamidol solution (150 mgI/mL) (Iopamiron 150; Nihon Schering or Bayer; Osaka, Japan) at a rate of 2.5 mL/seconds using an automated power injector. For CTHA, the catheter was placed in the proper, right, or left hepatic arteries. The data acquisition was performed twice during single-breath-hold after initiation of injection of iopamidol solution. The injection rate (1.0–2.5 mL/second) was determined as the maximal injection rate that could be used without leading to backward flow upon the hepatic angiography. The delay time of the first phase (seven to eight seconds) was determined based on when a hepatic tumor was most strongly enhanced on the hepatic angiography. The second phase (20–22 seconds) was initiated approximately four seconds after scanning was completed for the first phase. The contrast medium (20–46 mL) was injected until completion of scanning of the first phase.

Intraoperative ultrasound was performed during the surgical exploration with a 5-MHz curved array probe. Liver lesions detected on ultrasound were recorded by the leading surgeons and were treated with RFA, microwave coagulation therapy or ethanol injection. These nodules were correlated with the findings of combined CTHA and CTAP scanned preoperatively with respect to segmental location.

Image Analysis

Imaging analysis focused on the presence of HCC. Two combined MR sequence sets were prepared for comparison: an SPIO-enhanced MRI set (T2-weighted FSE and T1-/T2*-weighted FFE scanned before and after administration of ferucarbotran) and an SPIO-enhanced MRI + DWI set (SPIO-enhanced MRI with DWI scanned before and after administration of ferucarbotran). Apparent-diffusion coefficient map was not given for evaluation. For the receiver operating characteristic (ROC) analysis, one radiologist, who had knowledge of the clinical and histological findings, served as a study coordinator. This person initially reviewed all the MR, combined CTHA and CTAP, and intraoperative ultrasound images. He correlated the radiological findings, operation records, and histology, the latter being based on official pathology reports. The location (one of the eight liver segments according to Couinaud classification [13]), size and image number of HCC on MR images were recorded for each patient. Two MR sequence sets were interpreted independently by three abdominal radiologists with 14, 10, and six years of experience, respectively. Readers were blinded to patient identity, lesion number, and the results of other imaging studies, although a history of treatment such as hepatectomy and RFA was provided prior to interpretation. First, the study coordinator presented only a SPIO-enhanced MRI set to each reader separately and in random order in terms of patients. The readers were asked to grade the likelihood of presence of HCC in each liver segment on a four-point confidence scale: 1 = definitely absent, 2 = probably absent, 3 = probably present, and 4 = definitely present. If an HCC was thought to extend across two liver segments, the readers recorded a more predominant segment that the tumor occupied. Immediately after evaluation of the SPIO-enhanced MRI set, DWI images scanned before and after administration of ferucarbotran were provided for interpretation. The presence of HCC was reassessed using the SPIO-enhanced MRI + DWI. The readers judged an HCC to be present on the SPIO-enhanced MRI set when a focal and nodular area with definite decreased uptake of SPIO was identified on T1-, T2*-, or T2-weighted imaging, and when it did not show strong hyperintensity on T2-weighted FSE before administration of ferucarbotran, which could be suggestive of a cyst or a hemangioma. On DWI, malignant hepatic lesions have been reported to show mild to moderate hyperintensity at b= 0 second/mm2 and to remain hyperintense compared with liver parenchyma at high b values (6). Therefore, the considerations for identification of HCC on the SPIO-enhanced MRI + DWI set were as follows: a) When a questionable lesion for HCC is present on SPIO-enhanced MRI set, the readers judge it as positive if it showed hyperintensity on DWI at b = 1000 seconds/mm2, b) When a possible hyperintense lesion is present on DWI at b = 1000 seconds/mm2, the readers judge it as positive if it showed decreased uptake of SPIO after reassessing SPIO-enhanced MRI set, and c) the readers may consider as positive when the presence of HCC is strongly suspected on either the SPIO-enhanced MRI set or DWI.

Statistical Analysis

Initially, interobserver agreement for the two combined MR sequence sets was assessed to establish the reliability of image interpretation in this study. Interobserver agreement between pairs of readers was evaluated by using weighted κ statistics, with a κ value of less than 0.20 indicating poor agreement, 0.20–0.39 fair agreement, 0.40–0.59 moderate agreement, 0.60–0.79 substantial agreement, and 0.80 and higher excellent agreement.

For each MR sequence set, a binominal ROC curve was fitted to each reader's confidence ratings by using a maximum-likelihood estimation (14). The diagnostic accuracy of each MR sequence set for each reader was estimated by calculating the area under the ROC curve, Az, with the aid of the ROCKIT software package (University of Chicago, IL, USA). An average Az value for each MR sequence set for all three readers was compared statistically by using a paired t-test.

An HCC was considered to be present when the readers assigned a grade of 3 or 4, or to be absent when the readers assigned a grade of 1 or 2. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of each MR sequence set on a segment-by-segment basis were calculated. The sensitivity on the SPIO-enhanced MRI + DWI set was also compared with that on the SPIO-enhanced MRI set when the tumor diameter was limited to >1 cm, or ≤1 cm. The statistical significance of differences in sensitivity and specificity was evaluated using McNemar's test (15).

Finally, we examined whether the sensitivity for each MRI sequence set was correlated with histological differentiation of HCC. The difference in histological differentiation between detectable and undetectable nodules, which were limited to 37 histopathologically proven HCCs, in each sequence set was statistically compared using Fisher's exact test.

For all tests, a P value of ≤0.05 indicated a statistically significant difference.

RESULTS

According to the review of the study coordinator, five segments in three patients had been completely resected due to a previous surgery. Therefore, the number of liver segments evaluated was in total 235. The number of positive segments for HCC was 50, while that of negative segments was 185. There was no segment where more than two HCC nodules were present.

Kappa Analysis

The weighted κ values indicating agreement among the three readers are summarized in Table 1. There was excellent or substantial agreement between each combination of two readers.

Table 1. Interobserver Agreement for Detection of HCC for Each MR Sequence Set*
Reader (pair)SPIO-enhanced MRI setSPIO-enhanced MR I+ DWI set
  • *

    The data are κ values.

1 vs. 20.8250.810
2 vs. 30.7620.806
1 vs. 30.7180.768

ROC Analysis

The Az values for the ROC curves for each reader are summarized in Table 2. The trends for the ROC curves for the two sequence sets were similar across the three readers (Fig. 1a–c). The average Az value of the three readers for SPIO-enhanced MRI+DWI set was significantly higher than that of the SPIO-enhanced MRI set (P = 0.025).

Table 2. Az Values from ROC Analysis for Detection of HCC for Each MR Sequence Set*
ReaderSPIO-enhanced MRI setSPIO-enhanced MRI + DWI set
  • *

    The data from readers 1, 2, and 3 are Az values. Values for “Average” are mean ± standard deviation.

  • a

    The average Az value for the SPIO-enhanced MRI + DWI set is significantly higher than that of the SPIO-enhanced MRI set (P < 0.05).

10.7750.846
20.8050.839
30.8820.923
Average0.820 ± 0.055a0.870 ± 0.046
Figure 1.

ROC curves for detection of HCC for the two MRI sequence sets for readers 1 (a), 2 (b), and 3 (c). The curves display the performance of each reader for the SPIO-MRI set (square) and the SPIO-MRI+DWI set (diamond). TPF and FPF represent the true-positive and false-positive fractions, respectively.

Sensitivity, Specificity, PPV, and NPV

Table 3 shows the sensitivity, specificity, PPV, and NPV of the two sequence sets for detecting HCC. Although the SPIO-enhanced MRI+DWI set tended to be more sensitive than the SPIO-enhanced MRI set across all readers (Fig. 2), no statistical significance in difference was obtained (P = 0.15). There was no significant difference in specificity between the two sequence sets (P = 0.58). The PPV of the SPIO-enhanced MRI set was higher than that of the SPIO-enhanced MRI + DWI set for reader 1, while the PPV of the SPIO-enhanced MRI set was smaller than that of the SPIO-enhanced MRI + DWI set for reader 3. However, the NPV of the SPIO-enhanced MRI + DWI set was higher than that of the SPIO-enhanced MRI set for all three readers. There was no significant difference in sensitivity between the two sequence sets even if the tumor diameter was limited to >1 cm or ≤1 cm (P = 0.25 or 0.51) (Table 4). The sensitivity of the two sequence sets for detecting HCC with a diameter of 1cm or less was very low. There was a significant difference in tumor differentiation between detectable and undetectable HCC nodules in both the SPIO-enhanced MRI set and the SPIO-enhanced MRI + DWI set (P = 0.001, 0.0002, respectively) (Table 5). Only 50% (12/24) and 45.8% (11/24) of the well-differentiated and well to moderately differentiated HCCs were detected in the SPIO-enhanced MRI and SPIO-enhanced MRI + DWI sets, respectively. Retrospective analysis of the false positive lesions revealed that readers 1, 2 and 3 misdiagnosed hepatic vessels (six) or fibrosis (three) as HCCs for one, two and eight segments on the SPIO-enhanced MRI set, respectively. Especially, seven of the eight segments that reader 3 misdiagnosed were judged to be negative in the SPIO-enhanced MRI + DWI set (Fig. 3). On the other hand, four lesions misdiagnosed by at least one of the three readers on DWI were attributed to cirrhosis nodule or heterogeneous liver fibrosis (Fig. 4). Any reader did not detect three of the four lesions on the SPIO-enhanced MRI set. Two false-positive lesions detected by reader 2 on DWI were tiny lymph nodes located at the porta hepatis, which were not detected on the SPIO-enhanced MRI set. Readers 1 and 2 misdiagnosed a hemangioma as positive on the SPIO-enhanced MRI + DWI set.

Table 3. Sensitivity, Specificity, PPV and NPV of Each MRI Sequence Set for Detection of HCC*
ReaderSPIO-enhanced MRI setSPIO-enhanced MRI + DWI set
  • *

    Data are sensitivities, specificities, PPVs and NPVs expressed as a percentage. The numbers in parentheses for sensitivity and specificity are the total number of segments assigned a confidence level of 3 or 4 for sensitivity and 1 or 2 for specificity by all 3 readers, respectively.

Sensitivity (N = 50)  
 160.0 (30)64.0 (32)
 266.0 (33)72.0 (36)
 372.0 (36)74.0 (37)
 Average66.0 (99)70.0 (105)
Specificity (N = 185)  
 199.5 (184)98.5 (182)
 298.9 (183)98.9 (183)
 395.7 (177)98.4 (182)
 Average98.0 (544)98.6 (547)
PPV  
 196.891.4
 294.294.7
 381.892.5
 Total90.092.9
NPV  
 190.291.0
 291.592.9
 392.693.3
 Total91.492.4
Figure 2.

A 71-year-old female with a history of partial hepatectomy of segment VIII for HCC. The patient had two recurrent HCCs in segments III and VI. Axial MR images obtained with (a) and (c) T2*-weighted FFE (TE = 18.4) after administration of SPIO, and (b) and (d) DWI with b factor of 1000 before administration of SPIO. (a) A discrete hyperintense nodule suggestive of recurrence is identified in segment VI (arrow). (b) A hyperintese focus is seen at the corresponding site (arrow). (c) A vague, hyperintense focus is suspected at the surface of segment III (arrow). The readers 1 and 2 selected a confidence level of 1 and the reader 3 a confidence level of 2 in the SPIO-enhanced MRI + DWI set. (d) A discrete hyperintense nodule is seen at the corresponding site of segment III (arrow). The reader 1 selected a confidence level of 4, while the readers 2 and 3 a confidence level of 3 in the SPIO-enhanced MRI + DWI set. In this case, the SPIO-enhanced MRI + DWI set is more sensitive than the SPIO-enhanced MRI set across all readers.

Figure 3.

A 55-year-old male with a history of RFA for HCC in segment VIII. The patient had a recurrent HCC only in segment V. Axial MR images obtained with (a) and (c) T2*-weighted FSE (TE = 9.2) after administration of SPIO, (b) DWI with b-factor of 1000 after administration of SPIO and (d) DWI with b-factor of 1000 before administration of SPIO. (a) A discrete hyperintense nodule suggestive of recurrence is identified in segment V (arrow). (b) A discrete hyperintense nodule is seen at the corresponding site (arrow). (c) A vague, hyperintense focus is suspected in segment VII (arrow). The readers 1, 2, and 3 selected confidence levels of 1, 2, and 3 in the SPIO-enhanced MRI set, respectively. An arrowhead represents a change after RFA. (d) No definite hyperintense nodule is identified at the corresponding site of segment VII. The readers 1 and 2 selected a confidence level of 1, while the reader 3 a confidence level of 2 in the SPIO-enhanced MRI + DWI set. In this case, the SPIO-enhanced MRI + DWI set was more specific than the SPIO-enhanced MRI set.

Figure 4.

A 67-year-old male with a solitary HCC in segment I. Axial MR images obtained with (a) and (c) T2*-weighted FFE (TE = 9.2) after administration of SPIO, (b) and (d) DWI with b factor of 1000 before administration of SPIO. (a) and (b) A discrete hyperintense nodule suggestive of HCC is identified in segment I (arrow). (c) No definite hyperintense nodule is identified in segment VI. All readers selected a confidence level of 1 in the SPIO-enhanced MRI set. (d) Although a tiny hyperintense focus is suspected in segment VI (arrow), this finding is not definite. However, the reader 3 selected a confidence level of 3 in the SPIO-MRI + DWI set. This is an example of a false positive lesion for DWI, probably due to cirrhosis nodule or heterogeneous liver fibrosis. No recurrence has been identified at the corresponding site for at least two years after surgery.

Table 4. Sensitivity of Each MRI Sequence Set for Detection of HCC When a Tumor Diameter Is Limited to >1 cm, or ≤1 cm*
ReaderSPIO-enhanced MRI setSPIO-enhanced MRI + DWI set
  • *

    Data are sensitivities. The numbers in parentheses are the total number of true positive segments assigned a confidence level of 3 or 4 by all 3 readers.

1 cm (N = 32)  
 181.3 (26)87.5 (28)
 287.5 (28)90.6 (29)
 393.8 (30)93.8 (30)
 Average87.5 (84)90.6 (87)
≤1 cm (N = 18)  
 122.2 (4)22.2 (4)
 227.8 (5)38.9 (7)
 333.8 (6)38.9 (7)
 Average27.8 (15)33.3 (18)
Table 5. Correlation of Sensitivity for the Detection of HCC with Histological Differentiation*
DifferentiationW (N = 1)Wm (N = 7)M (N = 22)Mp (N = 5)P (N =2)
  • *

    The numbers of detectable and undetectable nodules are the total number of true positive nodules assigned a confidence level of 3 or 4 by all 3 readers. There was significant difference in differentiation between detectable and undetectable nodules in both the SPIO-enhanced MRI set and the SPIO-enhanced MRI + DWI set using Fisher's exact test (P < 0.001).

  • W = well-differentiated HCCs, wm, well- to moderately differentiated HCCs, m = moderately differentiated HCCs, mp = moderately to poorly differentiated HCCs, p = poorly differentiated HCCs, n = number of pathologically proven HCCs that belong to each histological differentiation group.

SPIO-MRI     
 Sensitivity (%)0.057.177.3100.066.6
 Detectable nodules01251154
 Undetectable nodules391502
SPIO-MRI+DWI     
 Sensitivity (%)0.052.477.3100.0100.0
 Detectable nodules01151156
 Undetectable nodules3101500

DISCUSSION

In our study, the additional value of DWI was evaluated with respect to detection of HCC on SPIO-enhanced MRI. Detection and precise localization appear to be difficult on DWI alone because DWI provides poor anatomic description. Therefore, we decided to investigate whether DWI can improve the detection of HCC lesions on SPIO-enhanced MRI. As a result, the average Az value for SPIO-enhanced MRI + DWI set was significantly higher than that of the SPIO-enhanced MRI set. Considering that there was no statistically significant difference in sensitivity and specificity between the two sequence sets, DWI may play a primary role in increasing the confidence level of interpretation.

The previous two reports showed sensitivities of 82% and 94% for detecting liver metastatic lesions on DWI, respectively (5, 6). The slightly lower sensitivity in our study is probably related to two factors: the different lesion etiology (HCC vs. metastatic lesion) and the difference in the background liver. The former is attributed to the phenomenon that some HCCs, contrary to metastatic lesions, take up SPIO and show hypo- or isointensity relative to the surrounding liver parenchyma on SPIO-enhanced MRI (3). The latter can be explained by the fact that many patients with HCCs have chronic liver disease including cirrhosis on the basis. Taouli et al (16) have reported that the apparent diffusion coefficient decreases according to the degree of liver fibrosis and inflammatory activity. In other words, the liver parenchyma substantially comes to show heterogeneous hyperintensity on DWI using a high b factor as liver dysfunction progresses. It can get difficult to detect a small focal lesion in such a coarse background on DWI. Parikh et al (6) have shown high sensitivity of 80.5% for detecting HCC on DWI. The authors carried out their evaluation on DWI at b = 500 seconds/mm2, while we used a b factor of 1000 seconds/mm2 in this study because we considered that a hyperintense focus on DWI using a high b factor is more specific to malignant lesions (17). On DWI using a high b factor, the signal of hepatic cysts is isointense to liver parenchyma, and the signal intensity of hepatic hemangiomas is also remarkably attenuated (17). In contrast, DWI using a relatively low b factor can describe the weak abnormal hyperintensity shown by some tumors. The difference in b factor could affect the sensitivity for detecting HCC. However, the specificity on DWI using a low b factor would be decreased, although Parikh et al (6) did not evaluate for specificity in their study. In our study, no significant difference in specificity between the two sequence sets was observed. The result suggests that specificity for detecting HCC may not decline even if we refer to DWI using a high b factor.

One interesting result obtained in this study was that the PPV of the SPIO-enhanced MRI set was increased by adding DWI for reader 1 but decreased for reader 3. This phenomenon can be explained as a difference in the way to utilize DWI between the two readers. Reader 1 could detect more HCC lesions with DWI while he pointed out a false-positive lesion such as an atypical hemangioma with moderately high signal intensity on T2-weighted FSE by mistake. The signal of a hemangioma does not completely disappear even on DWI with a high b factor (17). A confidence level of 3 was selected for this lesion in the SPIO-enhanced MRI + DWI set by reader 2 as well. In contrast, reader 3 used DWI to exclude false-positive lesions detected in the SPIO-enhanced MRI set, most of which were attributed to the high signal intensity of vascular structures and fibrosis according to the reassessment of the study coordinator. As a result, the NPV of the SPIO-enhanced MRI + DWI set was higher than that of the SPIO-enhanced MRI set for all three readers. DWI can play a different role when it is added to SPIO-enhanced MRI although Xu et al (12) reported that combined use of DWI with dynamic MRI increased sensitivity for detection of small HCC lesions compared with dynamic MRI alone. A cirrhosis nodule was also one of the causes of “false-positive” lesions on DWI. It is defined as a focal area with active fibrosis or cirrhosis that can be present with infiltration of many inflammatory cells and atypical hyperplasia (12). In addition, misdiagnosed tiny lymph nodes as HCCs on DWI were located along the liver surface at the porta hepatis and mimicked intrahepatic lesions. This shortcoming derives from poor anatomic description of DWI. Appropriate utilization and sufficient recognition of pitfalls of DWI may therefore lead to further improvement of the detectability of HCC on SPIO-enhanced MRI.

Many of the “false-negative” lesions were attributed to their small size. Although the SPIO-enhanced MRI + DWI set tended to be more sensitive than the SPIO-enhanced MRI (P = 0.15), the sensitivity for detection of HCC with a diameter of ≤l cm was not significantly increased by adding DWI to SPIO-enhanced MRI. This result may be due to the small number of HCC lesions evaluated and the relatively large section thickness in our scanning protocol. Otherwise, the low spatial resolution of DWI could limit the identification of small lesions. The subcardiac signal loss caused by physiological movement is also a possible factor that could limit the performance for detecting HCC on DWI (5). However, there was no HCC lesion located at the dome of the left lobe in our study. Overall sensitivities of the lesions in the left and right lobes in the SPIO-enhanced MRI + DWI set was 64.3% and 75.8%, respectively (unpublished data). We think that the slight decrease in sensitivity for detecting the left lobe lesions can be explained with the difference in ratio of HCC lesions with a diameter of 1cm or less (left lobe versus right lobe; 39.2% vs. 31.8%). Therefore, this deficiency of DWI can be neglected in our study. Another factor of “false-negative” is considered to be histological grade of HCC. Imai et al (3) have demonstrated that some well-differentiated HCC show hypointense or isointense enhancement relative to the surrounding liver parenchyma on SPIO-enhanced MRI. This result indicates a greater or similar uptake of SPIO in well-differentiated HCC. Such tumors do not fulfill our diagnostic criteria of definite decreased uptake of SPIO on T1-, T2*- or T2-weighted imaging. In fact, only 50% (12/24) of the well-differentiated and well to moderately differentiated HCCs were judged to be positive in the SPIO-enhanced MRI set in this study. Another interest of ours was determining whether the identification of well-differentiated HCC can be improved by adding DWI to SPIO-enhanced MRI. However, our results suggest that DWI might not contribute to the detection of HCC with well-differentiated components. One of the reasons why a mass shows hyperintensity on DWI using a high b factor is restricted water diffusion due to the high cellular density, especially considered in cases of malignant lymphoma (18). The trabeculae of moderately differentiated HCC are thicker than those of well-differentiated HCC (19). In addition, the cellular density of poorly differentiated HCC is increased compared with that of moderately differentiated HCC (19). The histological morphology of HCC based on grade or differentiation may explain the result in which the detectability of HCC is increased as the histological grade or differentiation shifts toward malignancy.

Our study had several limitations. The first is that post-therapeutic changes were observed in three of the 235 segments evaluated. Although a history of treatment was provided to the readers prior to interpretation, this information may have made their decision for lesion detection more difficult. The second limitation is that DWI was scanned before and after administration of SPIO in our protocol. However, Naganawa et al (20) reported that the number of hepatic lesions detected on DWI increased after SPIO administration in the clinical study with a small number of patients. The excellent lesion-to-liver contrast of DWI after administration of SPIO may improve the detection of small lesions. Therefore, two DWI images scanned before and after administration of SPIO were provided to the readers. In this study a difference in detection of HCCs on unenhanced DWI and SPIO-enhanced DWI was not investigated. Further assessment is required to determine the more appropriate timing of scanning DWI for detection of HCC. Third, only DWI at b = 1000 seconds/mm2 was used for interpretation in the SPIO-enhanced MRI + DWI set. Scanning of DWI with different or multiple b factors could enhance performance for detecting HCC. Fourth, 8 mm slice thickness and 2 mm intersection gap were used for scanning of SPIO-enhanced MRI and DWI in our study. These parameters may partly account for relatively low sensitivity for detection of HCCs with a diameter of ≤1 cm. Fifth, the ideal gold standard used for lesion proof is the pathological examination of liver explants. The presence of false positive and negative segments for HCC cannot be excluded strictly in this study although follow-up CT or MRI after surgery showed the absence of HCC nodules in the remaining liver.

In conclusion, the SPIO-enhanced MRI + DWI set was found to outperform the SPIO-enhanced MRI set in depicting HCC. It appears to be of value to add DWI to the standard SPIO-enhanced MRI protocol for evaluation of HCC.

Acknowledgements

We thank Dr. Yoshihiko Maehara, Department of Surgery and Science, Kyushu University, for providing the clinical information for this manuscript. We also thank Dr. Masazumi Tsuneyoshi, Department of Anatomic Pathology, Kyushu University, for providing the pathological information for this manuscript.

Ancillary

Advertisement