Radiologic Monitoring of Hepatocellular Carcinoma Tumor Viability after Transhepatic Arterial Chemoembolization: Estimating the Accuracy of Contrast-enhanced Cross-sectional Imaging with Histopathologic Correlation

Materials and Methods 

An institutional review board exemption was obtained from the sponsoring institution (Columbia University Medical Center, New York, New York) for this retrospective study. From February 2002 to October 2005, a total of 55 patients who had undergone chemoembolization for HCC underwent successful liver transplantation at our institution. Of these 55 patients, 31 had a follow-up cross-sectional imaging study before transplantation that was available for review. We undertook a retrospective review of the imaging studies in this subgroup of 31 patients and correlated the presence or absence of tumor viability on imaging, as well as the presence of satellite lesions, with histopathologic analysis performed at the time of transplantation. Demographics of the patient population are summarized in Table 1.

Table 1. Study Demographics (N = 31)

	Characteristic	Value
	Age (y)
	Mean	60
	Range	51–74
	Sex (M/F)	28/3
	Imaging modality
	CT	20(64.5)
	MR	10(32.3)
	Both	1(3.2)
	Imaging–pathology interval
	Mean (d)	32
	Median (d)	27
	Imaging–pathology interval per patient (d)
	<60	26
	61–90	4
	90–120	1

Note.—Values in parentheses are percentages.

Chemoembolization was performed in the usual standard fashion. In brief, all patients were discussed at our weekly multidisciplinary liver conference. Chemoembolization was performed at a segmental or a subsegmental level with a combination of 100 mg of cisplatin (Bristol-Myers Squibb, Princeton, New Jersey), 50 mg of doxorubicin hydrochloride (Adriamycin; Pharmacia/Upjohn, Kalamazoo, Michigan), and 10 mg mitomycin-C (Bedford Laboratories, Bedford, Ohio) reconstituted with 8.5 mL of nonionic iodinated contrast medium and 1.5 mL of sterile water and further emulsified in a 1:1 ratio with Ethiodol (Savage Laboratories, Melville, New York). Particulate embolic material (100–300-μm; Embosphere particles; Biosphere Medical, Rockland, Massachusetts) was mixed into the emulsification based on the vascularity and size of the tumor and administered concurrently with the emulsion. Substasis or “pruning” was considered as the endpoint for all chemoembolization procedures. All patients were admitted for observation and discharged within 24–48 hours after the procedure. Because all patients were on the transplant list, close clinical follow-up was maintained by the transplantation team through a combination of biochemical markers, Model for End-stage Liver Disease score, and cross-sectional imaging.

Surveillance imaging after chemoembolization consisted of a CT or MR imaging study. Quadruple-phase contrast-enhanced helical CT of the liver was performed on a 16-slice CT scanner (Siemens, Erlangen, Germany) with optimal bolus timing to clearly delineate unenhanced, arterial, portal, and equilibrium phases. Images were obtained during a single breath-hold helical acquisition of 25–30 seconds in a craniocaudal direction, and were reconstructed every 5 mm to provide contiguous sections. After the acquisition of the unenhanced images, the images of the hepatic arterial, portal venous, and equilibrium phases were obtained with delays of 30, 70, and 180 seconds, respectively, after injection of 100–120 mL of nonionic iodinated contrast material through the antecubital vein at a rate of 3 mL/sec.

MR imaging of the liver was performed on a 1.5-T HD Excite 12.0 scanner (GE Healthcare, Berkshire, United Kingdom) at 1.5 T with use of the body coil for transmission and an eight-channel body-phased array coil for signal reception. The liver acquisition with volume acceleration (LAVA) pulse sequence employed three-dimensional spoiled gradient-echo imaging with an inversion recovery fat suppression pulse applied to each slice loop. Two-fold parallel imaging was used to obtain a 384 × 256 matrix with approximately 40 4.4-mm-thick slices, which were twofold zero-filled down to 2.2-mm slice spacing during reconstruction. This LAVA sequence was obtained in the axial plane 30 seconds, 1 minute, and 3 minutes after dynamic gadolinium (30 mL) injection at a rate of 2–3 mL/sec. One final 5-minute LAVA sequence was then obtained in the coronal plane.

Twenty patients (64.5%) had a CT examination and 10 patients (32.3%) had an MR imaging examination before tumor explantation; one patient (3.2%) had both (Table 1). The mean time interval from imaging to transplantation and subsequent histopathologic correlation was 32 days (median, 27 d). The time interval between imaging and pathologic examination was less than 60 days in 26 patients (83.9%), 61–90 days in four patients (12.9%), and 117 days in one patient (3.2%). Because the department had a large number of radiologists with varying experience, for the purpose of the present study, all images were reinterpreted by a senior radiologist (M.R.P.) with significant experience with body CT and MR imaging. All studies were blinded to this radiologist and were independent of end results and previous interpretations. Tumor viability was assessed based on residual nodular enhancement, increase in size of the tumor, and/or presence of new tumors. Presence of new segmental or main portal thrombus that enhanced was considered as gross vascular invasion. The imaging findings were then compared with the surgical and pathologic clinical reports detailing the anatomic and microscopic histopathology results obtained at the time of tumor resection, liver explantation, or transplantation.

Results 

Thirteen patients (41.9%) had viable tumor on imaging, compared with 22 patients (70.9%) with viable tumors on histopathologic examination; imaging findings were indeterminate in one patient. Overall sensitivity and specificity rates of imaging were 35% and 63.6%, respectively (Table 2). Of the 17 patients (54.8%) who had erroneous interpretation on imaging, 13 patients (41.9%) had viable tumors that were missed and four (12.9%) had findings suggestive of viable tumors on imaging that were found to be nonviable on pathologic examination (Fig 1). Among the 13 false-negative findings, eight patients (68.5%) had viable tumor that was only identifiable microscopically as small nests of viable tumor cells. However, five patients (38.5%) had macroscopic tumor identified grossly on explantation, including two patients with multifocal disease and one patient with adjacent organ and inferior vena cava invasion. Patients who had an interval from imaging to pathologic examination greater than 30 days were no more likely to have an incorrect diagnosis (41.7%) than patients with an interval of less than 30 days (47.4%; P > .05). Likewise, patients who had an interval from imaging to pathologic analysis greater than 60 days were not statistically more likely to have an incorrect diagnosis (60%) than those with an interval of less than 60 days (54%; P = .92; Table 3). Assuming no sampling error or error in pathologic classification, the overall accuracy rate for CT was 42.9% (nine of 21 patients), with 35.7% sensitivity and 57.1% specificity, and the overall accuracy rate for MR imaging was 54.5% (six of 11 patients), with 42.9% sensitivity and 75% specificity (Table 2). One patient underwent CT and MR imaging with a correct diagnosis on both. A greater majority of tumors missed (ie, false-negative findings) on MR imaging were likely to be microscopic disease (60% microscopic disease, 20% macroscopic disease) compared with CT (41.7% microscopic disease, 33.3% macroscopic disease).

Table 2. Accuracy, Sensitivity, Specificity, PPV, and NPV of CT vs MR Imaging

	Imaging Findings	HCC	No HCC	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	Overall Accuracy (%)
	All imaging			35	64	64	35	59
	Positive	7	4
	Negative	13	7
	Quadruple-phase CT			36	57	63	31	43
	Positive	5	3
	Negative	9	4
	LAVA MR			43	75	75	43	55
	Positive	3	1
	Negative	4	3

View Large Image Download to PowerPoint

Figure 1. Images from follow-up contrast-enhanced CT scans in two patients who have undergone chemoembolization illustrate false-negative and false-positive findings. (a) Noncontrast (top) and contrast-enhanced (middle) CT images demonstrate lack of contrast enhancement in the previously chemoembolized tumor in segment 2/3 (circled). These images suggest adequate treatment without any evidence of residual tumor. However, on explantation (bottom; hematoxylin and eosin stain, magnification ×4), viable tumor cells were visible, with highly pleomorphic nuclei and focal multinucleated tumor giant cells (arrows). (b) Noncontrast (top) and contrast-enhanced (middle) CT images in another patient who had undergone chemoembolization demonstrates an area of contrast enhancement (circles). However, on explantation (bottom; hematoxylin and eosin stain, magnification ×2), this proved to be a macroregenerative nodule with hepatocytes and a rim of fibrosis surrounding the nuclei (arrows).

Table 3. Time Interval between Imaging and Pathologic Examination

	Interval (d)	Correct Diagnosis	Incorrect Diagnosis
	<60	12	14
	>60	3	2

Time to pathologic examination was not correlated with a statistical difference in accuracy of detection: P = .92 on χ2 test with Yates correlation.

Discussion 

Assessment of tumor viability after chemoembolization is important for evaluation of tumor response, subsequent treatment planning, and evaluation for liver transplantation. Although histopathologic assessment of the treated tumor remains the most definitive method to determine viability, it is not feasible as a method for follow-up. Serum α-fetoprotein level remains a useful marker for disease recurrence, but as many as 59% of patients with HCC present with normal serum α-fetoprotein levels (7). Even if present, increasing α-fetoprotein levels could be caused by new intra- or extrahepatic metastatic disease, rather than local recurrence at the previously treated tumor site. Therefore, radiographic imaging remains the predominant methodology to assess tumor viability and disease progression after chemoembolization.

The optimal choice of imaging modality for follow-up after chemoembolization is still the subject of ongoing research. Multiple modes of ultrasonography (US) (8, 9, 10, 11), MR imaging (11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22), and CT (11, 15, 17, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39) have been used for follow-up, each with its own advantages and disadvantages. US is the least costly, does not subject the patient to ionizing radiation, and has been shown to be particularly useful for diagnosing concomitant portal venous thrombosis (40). Color and power Doppler US can demonstrate peritumoral vascularity in viable lesions and have shown high specificity (8, 9, 10, 11). However, besides being highly operator-dependent, US has limited utility in evaluating small lesions or lesions located deep within the liver or near the dome, and in patients with large abdominal girth (11, 41, 42).

CT remains the “workhorse” of imaging follow-up after chemoembolization because of its ease of use and availability. Extensive research exists on the diagnosis of HCC with contrast-enhanced multiphasic helical CT in chemoembolization-naive patients; however, there is a paucity of literature regarding the accuracy, sensitivity, and specificity of CT in evaluating response to chemoembolization and diagnosis of residual disease. One of the earliest reports (23), published before the use of iodized oil (ie, Ethiodol [Savage Labs] and Lipiodol [Laboratoire Andre Guerbet, Aulnay-sous-Bois, France]), compared contrast-enhanced CT imaging findings versus resected specimens in a study of 18 patients who had undergone chemoembolization for HCC. The necrotic areas observed on histologic examination corresponded closely to the low-density area seen on CT in 16 of the 18 cases (89%), with the remaining two patients having undetected residual tumor in the background of coagulative necrosis. However, with the addition of iodized oil during chemoembolization, interpretation of contrast-enhanced CT after chemoembolization is problematic because of beam-hardening artifacts from the iodized oil, which make it difficult to visualize marginal recurrences as well as residual viable tumor (33). Jang and colleagues (38) reviewed contrast-enhanced multiphasic CT images in 35 HCCs treated with chemoembolization that were subsequently resected. In their study (38), the overall accuracy rate was 78%, with a sensitivity of 72% and specificity of 91%. In the present study, we measured the accuracy of quadruple-phase CT at 42.9%, with 35.7% sensitivity and 57.1% specificity. The lower accuracy reported in the present study likely reflects the fact that the accuracy of CT is significantly diminished in the presence of small tumors.

Oi and colleagues (43) compared dynamic MR imaging versus early- and late-phase contrast-enhanced CT in 225 hepatic tumors smaller than 3 cm and reported overall detection rates of 62% and 52%, respectively. In addition, small vague areas of hypervascularity that represent arterial–portal shunts, commonly seen in patients with HCC, are a frequent cause of “pseudolesions” on CT, again limiting specificity (44). Other authors have qualitatively described CT after chemoembolization with histopathologic correlation (23, 25, 26, 27, 28, 31), but they have not reported diagnostic accuracy.

Although iodized oil (ie, Ethiodol or Lipiodol) can cause beam-hardening artifacts on CT, retention of the iodized oil itself has prognostic value, a factor that cannot be easily determined by US or MR imaging. Dense, uniform retention of Ethiodol in the tumor itself has been demonstrated as a reliable surrogate marker for complete or near-complete tumor necrosis (27, 28). Jinno and colleagues (26) compared oil retention after chemoembolization on CT with histopathologic findings of resected specimens in 32 patients with HCC. They classified all lesions into one of six CT patterns, ranging from “complete” to “defective” to “deficient” enhancement based on Lipiodol retention. A high degree of reliability was noted for complete Lipiodol enhancement and deficient Lipiodol enhancement (90%–100% necrosis and 0%–10% necrosis, respectively) (26). Other studies demonstrated similar results for tumors that demonstrated dense, uniform Ethiodol uptake (27, 28). However, the utility of moderate retention is limited because it can represent a wide range of necrosis from completely necrotic to predominantly viable (11, 27, 28). Newer imaging methodologies such as the measurement of attenuation values in the setting of quadruple-phase helical CT are likely to improve the accuracy of CT in differentiating viable tumor after chemoembolization therapy for HCC (39).

MR imaging has grown increasingly popular for follow-up after chemoembolization. It does not subject the patient to ionizing radiation and is not affected by artifacts produced by the use of Ethiodol. Early work from Yoshioko and colleagues (12) suggested that hypointensity in MR spin-echo T2-weighted sequences corresponded to coagulative necrosis, and could be used as a proxy for positive response to treatment. Unfortunately, a paradoxic T2-weighted hyperintensity often accompanies early stages of tumor necrosis, perhaps reflecting intratumoral hemorrhage. As T2-weighted hyperintensity could correspond to intratumoral hemorrhage, inflammation, liquefied necrosis, or even viable tumor, conventional T2-weighted imaging requires serial MR studies to demonstrate the evolution into T2-weighted hypointensity specific for coagulation necrosis.

Several authors, using various imaging sequences, have sought to evaluate the accuracy of MR follow-up after chemoembolization with histopathologic correlation (13, 16, 18, 19). Murakami et al (13) reported a sensitivity of 80% with the use of turbo-fast low-angle shot MR imaging with gadopentetate contrast medium in 10 patients after chemoembolization with Lipiodol. Viable tumor was shown to correlate with early hyperintense enhancement, whereas lack of early enhancement corresponded to necrotic regions of the tumor. Ito et al (16) reported a sensitivity of 92% with the use of multisection dynamic MR with gadopentetate contrast medium to assess tumor viability in 13 patients after chemoembolization with Lipiodol. Castrucci et al (18) compared gadolinium-enhanced spin-echo T1-weighted MR imaging with unenhanced T2-weighted MR imaging in a prospective study of 15 patients with hyperintense T2 lesions who underwent chemoembolization with Lipiodol followed by tumor resection 15 days later. Gadolinium-enhanced T1-weighted imaging had 96% accuracy, with 100% sensitivity and 78% specificity, whereas unenhanced T2-weighted imaging had a specificity of only 40%. Yan and colleagues (19) compared T1-weighted, T2-weighted, and fast multiplanar spoiled gradient-recalled MR imaging of 24 lesions in 22 patients with histopathologically examined lesions 4–9 weeks after chemoembolization. Fast multiplanar spoiled gradient-recalled imaging is a T1-weighted dual fast field echo technique that demonstrates a dynamic early-phase enhancement in viable tumors, followed by a decline in the late phase. The authors noted that conventional spin-echo T1-weighted and T2-weighted imaging characteristics after chemoembolization were highly variable, and that hyperintensity, isointensity, and hypointensity each could correspond to residual tumor or inflammation (19). Assuming all viable tumors were resected, fast multiplanar spoiled gradient-recalled MR imaging resulted in an overall accuracy rate of 96%, with 100% sensitivity and 83% specificity. In the present study, 11 patients were examined with use of gadolinium-enhanced T1 weighted LAVA imaging followed by histopathologic correlation at a mean of 32 days (Fig 2). LAVA is a newer three-dimensional dynamic spoiled gradient-echo T1-weighted sequence with fat suppression that is purported to result in higher spatial resolution, and therefore an increased sensitivity in the detection of small liver lesions (45). We found the accuracy of LAVA MR imaging to be 54.5%, with 42.9% sensitivity and 75% specificity.

View Large Image Download to PowerPoint

Figure 2. Images from follow-up contrast-enhanced MR scans in two patients who had undergone chemoembolization illustrate true-negative and true-positive findings. (a) The image obtained before gadolinium enhancement (top) shows T1 bright signal in the lesion typical of hemorrhagic infarction after chemoembolization (circled). LAVA sequence after gadolinium injection (second from top) shows some bright signal in the lesion, which is shine-through from the bright T1 signal before gadolinium administration. Subtraction of the unenhanced image from the enhanced image (third from top) shows no enhancement within the lesion after chemoembolization, indicating no residual tumor. Pathologic examination (bottom; hematoxylin and eosin stain, magnification ×4) shows a necrotic nodule without any viable malignant cells (arrows). Embolic material is present in the vessels (ie, true-negative findings). (b) The image obtained before gadolinium enhancement (top) shows a subtle focus of T1 bright signal in segment 4 typical of hemorrhagic infarction (arrow). LAVA sequence after gadolinium injection (second from top) shows thick irregular enhancement anteriorly and laterally (arrows). Delayed-phase image after gadolinium enhancement (third from top) demonstrates that the arterial-enhancing portion of the tumor washes out (arrows). The combination of arterial enhancement and early washout is characteristic of tumor recurrence, which was confirmed on explantation. Hematoxylin and eosin–stained section on explantation (×10 objective, bottom) demonstrates viable tumor cells growing in a trabecular pattern with a mild degree of pleomorphism, representing a well differentiated tumor (arrows; ie, true-positive findings).

The present study has several limitations worth noting. First, the small sample size might not reflect the accuracy encountered in larger surveys of the reported imaging modalities. In addition, the small sample size in the present study required Yates modification of the χ2 analysis, and is of insufficient power to assess statistical significance between MR and CT groups. Second, the variable time from imaging to histopathologic analysis may have allowed for development of new tumors that were not present at the time of imaging. This would lead to an underestimation of the sensitivity of recurrence detection, and may explain the discrepancy in sensitivity as reported in earlier studies. Third, the presence of microscopic foci of viable cells may be below the current threshold of detection by current methodologies, and may not represent clinically relevant recurrence. In the future, prospective studies are needed that randomize patients scheduled for explantation to undergo imaging with different modalities immediately before explantation. In this way, the prospective imaging findings can be temporally correlated with the histopathologic analysis.

In conclusion, imaging represents the only currently feasible—albeit imperfect—means to evaluate accuracy of chemoembolization therapy. To determine the accuracy of each imaging modality, it is necessary to measure the sensitivity and specificity of different imaging modalities compared with direct histopathologic assessment of explanted tumors. Unfortunately, although many descriptive studies have been published on the imaging characteristics of lesions after chemoembolization, few studies attempt to quantify the accuracy for detection of viable tumor. The present study supports MR rather than quadruple-phase CT imaging for chemoembolization follow-up in terms of overall accuracy (54.5% vs 42.9%), sensitivity (42.9% vs 35.7%), and specificity (75% vs 57.1%) of detection of viable tumor. As the population of patients undergoing chemoembolization continues to increase, the need for a highly sensitive and specific follow-up imaging sequence or assay represents an opportunity for further innovation.