Blood oxygen level‐dependent liver MRI: Can It predict microvascular invasion in HCC?

To assess Blood Oxygen Level‐Dependent (BOLD) Magnetic Resonance Imaging (MRI) for noninvasive preoperative prediction of Microvascular Invasion (MVI) in Hepatocellular Carcinoma (HCC).

LIVER RESECTION AND transplantation are the only curative treatment options currently available for hepatocellular carcinoma (HCC) (1)(2)(3). However, long-term survival is still poor as a result of a high rate of recurrence among other drawbacks, such as lack of reliable prognostic factors, phenotypic diversity of the disease, and the lack of effective systemic treatment. Given the limited organ availability as well as morbidity and or mortality risks associated with surgical options, it is critical to select patients who are likely to have longterm curative outcome without recurrence. Among many outcome prediction parameters such as tumor grade, differentiation, size, multiplicity, and vascular invasion (macro and microscopic), microvascular invasion (MVI) has been correlated to be one of the most significant independent risk factors affecting recurrencefree survival following curative resection and or liver transplantation (4). Preoperative prediction of microvascular invasion could allow appropriate patient selection for liver transplantation and predicting prognosis.
Although the combined use of imaging modalities, including MRI, computed tomography (CT), and ultrasonography, can detect tumor invasion of the major branches of the portal and hepatic veins in 81-95% of cases at the time of diagnosis, imaging studies currently do not have the ability to detect microvascular invasion (5)(6)(7)(8)(9)(10)(11)(12)(13)(14). The diagnosis of MVI cannot be reliably achieved by a biopsy due to sampling errors (15) as only a small region of the tumor is being evaluated in a biopsy as compared to surgical pathology. Intratumoral hypoxia has been shown to enhance proliferation, angiogenesis, metastasis, chemoresistance, and radioresistance of HCC resulting in overall increased tumor invasiveness (16)(17)(18). Blood oxygen level dependent (BOLD) MRI is a noninvasive diagnostic method capable of assessing tumor oxygenation and indirectly hypoxia, by detecting signal changes secondary to changes in blood flow and oxygenation. The purpose of this study was to correlate prospectively the ability of BOLD MRI to predict MVI preoperatively compared with histopathology in patients with HCC who were undergoing liver resection.

Patients
This was a prospective single center HIPAA compliant, institutional review board approved study with enrolled patients giving approval to participate by means of written informed consent. Over a 2-year period from September 2009 to 2011, 28 patients with suspected HCC who were to undergo liver resection (partial hepatectomy) where accrued to this study and underwent routine staging liver MRI along with BOLD MR before surgery with an average time interval between MRI and surgery of 25 days (range, 1 to 138 days). Exclusion criteria included specifically lack of histopathology; prior tumor treatment such as transarterial chemoembolization, radiofrequency ablation, or chemoradiation; and general contraindications to MRI and or gadolinium-based contrast agents. Two patients were excluded as their surgery was canceled. Thus, our final study cohort included 26 patients (21 men and 5 women age range, 34-77 years with mean age of 61 years) who underwent liver MRI and subsequent surgical resection and histopathology. Serum alpha fetoprotein (AFP) levels were in the range of 2-15408 mg/L with mean of 1101 mg/L. Chronic viral hepatitis B and or C was present in 65% of patients.

MR Imaging
MR imaging was performed on all subjects on a 1.5 Tesla (T) MR system (Siemens Avanto, Siemens Healthcare, Erlangen, Germany). Torso phased-array coils were used. All subjects underwent a routine Liver MR protocol ( Table 1) that included coronal single-shot T2-weighted HASTE, fat suppressed motion corrected axial T2 fast spin echo (BLADE), fat suppressed axial single-shot T2 HASTE (long TE), axial gradient-refocused echo T1-weighted in-phase and out of-phase, axial diffusion-weighted imaging (DWI), and three-dimensional (3D) T1-weighted imaging (T1WI) before and after IV injection of gadolinium contrast material. Bold MR imaging was an addition to this routine liver protocol before gadolinium contrast injection. The BOLD MR imaging comprised of a 12 multiecho gradient refocused echo (GRE) T2* imaging pulse sequence with TE ranging from 1 to 41 ms. BOLD MR imaging was performed before and after inhalation of oxygen. Oxygen (100%, 5L/min) was administered through a nasal mask for 5 min before a repeat BOLD acquisition was obtained following oxygenation. The post oxygenation BOLD imaging was followed by multiphasic gadolinium enhanced T1WI.

Primary Analysis
The primary analysis comprised of quantitative estimations of R2*. All the MRI scans including the BOLD data were annonymized and saved to a password protected encrypted hard drive for analysis purposes. R2* Analysis was obtained using commercial software (Image J, NIH) ( Fig. 1). Multiple ROIs were placed on HCC, liver, and muscle on multiple slices to obtain T2* and thereafter R2* (1/T2* in 1/s) for HCC, liver, and muscle. Areas of obvious susceptibility artifacts were excluded from ROI measurement. The ROI tracing was performed manually taking care to be within confines of tumor at all times. Multiple R2* indices such as ratios and or differences in R2* between tumor and liver before and after oxygenation were calculated to seek discriminative thresholds for prediction of MVI.

Secondary Analysis
The secondary imaging analysis included multiparametric clinical, morphological, and DWI assessment. Detailed morphological observations were recorded performed by two readers in consensus using the departmental PACS viewer for review regarding the tumor size, shape location, signal intensity on T1-and T2-weighted images, multiphasic enhancement patterns and contrast washout, presence or absence of tumoral capsule, satellite nodules of tumor, and intra and extrahepatic metastatic disease. DWI analysis was done by a single observer calculating the ADC values from the ADC maps generated during the DWI imaging acquisition.
Regions of interest (ROIs) were drawn on the ADC maps in areas of interest encompassing the tumor as well as background liver on a MRI postprocessing workstation (Siemens Healthcare, Erlangen, Germany).

Pathological Assessment
Tumor characteristics were evaluated by review of the pathological specimens. Tumor size was measured as the largest diameter of the major tumor in centimeters. Macrovascular invasion is defined as gross vascular invasion into major portal vessels or hepatic veins. Microvascular invasion was determined on pathologic analysis as microscopic vascular invasion of small vessels within the peritumoral parenchyma of the liver (Figs. 2, 3). The predominant histopathologic grade of differentiation of the tumors was assessed according to Edmondson-Steiner criteria (G1, well differentiated; G2, moderately differentiated; G3, poorly differentiated; G4, undifferentiated). Immunohistochemistry using the monoclonal antibody QBEnd10 (anti-CD34) was performed in cases microvascular invasion is suspected but not unequivocal to help to confirm or exclude microvascular invasion. For the purpose of this study, sections from the blocks that had already been used for routine histological examination were stained by the streptavidin-biotin complex immunohistochemical technique. The slides were counterstained with hematoxylin, and blindly and randomly examined by two pathologists who had already evaluated hematoxylin and eosin (HE) sections. The presence of neoplastic cells inside the lumina of the vessels whose endothelium had been immunohistochemically stained at the periphery of the HCC nodules was carefully searched for the presence of attached tumor embolus.

Data and Statistical Analysis
Binary logistic regression was used to assess the ability of R2 parameters and quantitative morphological parameters to predict MVI. Stepwise variable selection in the context of binary logistic regression was also performed to evaluate the combination of MRI and morphological parameters in predicting MVI. Furthermore, BOLD MRI parameters (R2* values) and quantitative morphological outcomes (AFP value, length dimension, volume, and ADC values) were compared by means of the Wilcoxon rank-sum test (also known as the Mann-Whitney U test). Fisher's exact test was used to compare between patients with and without MVI in regards to the qualitative morphological parameters. The analysis was done using the statistical software IBM SPSS Version 20.

MRI Findings and Analysis
All 26 HCC were identified and imaging analysis was performed both quantitatively as well as qualitatively for primary and secondary analysis as described in the methods section. In the primary analysis, the mean baseline (preoxygenation) R2* values (1/s) in tumors with and without MVI were 35 6 12 (24-74) and 38 612 (25-58), respectively. The mean post oxygenation R2* values (1/s) in tumors with and without MVI were 36 6 10 (24-72) and 42 6 17 (24-83), respectively. The mean difference (delta) in R2* values (1/s) before and after oxygenation with and without MVI were À1 (À15 to þ7) and À4 (À50 to þ26). Various other indices to reflect R2* ratios between the HCC and liver before and after oxygenation as well ratios of difference in R2* values between HCC and liver were calculated to seek any discriminative threshold of correlation with presence or absence of MVI on histopathology. Table 2 displays descriptive information about the MRI parameters of interest in each group and the P value from the Wilcoxon rank-  Thus, overall, no significant differences were found between those who had MVI and those who did not with regard to the BOLD MRI parameters (R2*).

DISCUSSION
Treatment decisions for HCC currently based on tumor number, size, and liver function continue to be hampered by poor long-term survival and high recurrence rates (19)(20)(21). The combination of tumor recurrence with vascular invasion limits additional attempts at various therapies, such as repeat hepatic resections, percutaneous ethanol injection, microwave coagulation therapy, and radiofrequency ablation, thereby contributing to poor survival (22)(23)(24). Tumor invasion into microscopic branches of the portal and/ or hepatic veins (MVI) is associated with increased risk of early tumor recurrences following surgical  treatment of HCC and shorter survival (25)(26)(27)(28)(29)(30)(31)(32). Accurate prediction of MVI would impact treatment decisions based on anticipated tumor biology, therapeutic response and clinical outcomes, but because current imaging techniques as well as preoperative biopsy are unable to accurately predict MVI in HCC, this important prognostic marker remains unconsidered while making treatment decisions for HCC. Lack of sufficient oxygenation, hypoxia, is a common tumor microenvironmental characteristic caused by the imbalance between oxygen supply by abnormal tumor vasculature and demand by rapidly proliferating tumor cells. Evidence suggests that the heterodimeric transcription factor hypoxia inducible factor 1 a (HIF-1a) controls the expression of a variety of genes, which play crucial roles in the acute and chronic adaptation of tumor cells to oxygen deficiency, including enhanced erythropoiesis and upregulated glycolysis, promotion of cell survival, inhibition of apoptosis, inhibition of cell differentiation, and increased angiogenesis. Hypoxic stress accelerates the invasion of hepatoma by up-regulating ETS-1 and the matrix metalloproteinases family by the HIF-1a-independent pathway. These adaptive changes of gene expression in neoplastic cells results in tumor invasion, metastases, and chemoradiation resistance (33)(34)(35)(36)(37). Specifically related to the liver, fibrogenesis associated with cirrhosis reduces hepatic blood flow leading to hypoxia. High proliferation of tumor cells also induces local hypoxia within HCC also stimulating angiogenesis to support the tumor growth by inducing the expression of angiogenic factors (16,17). To summarize, hypoxia enhances proliferation, angiogenesis, metastasis, and chemo-and radioresistance of HCC resulting in increased tumor invasiveness (MVI), which is essentially the primary step for future development of macrovascular invasion and or metastases in HCC. Thus, the level of intratumoral oxygenation, precisely hypoxia, could be a contributing factor to development of MVI.
A noninvasive technique that could estimate tumor oxygenation would find broad applications in prediction of MVI in HCC. The blood oxygen level-dependent (BOLD) MRI imaging technique has the unique capability to study tumor pathophysiology noninvasively. By accentuating the susceptibility effect of deoxyhemoglobin (dHb) in the blood with gradient-echo techniques, image contrast reflects the blood oxygen level. Changes in R2 (1/T2) reflect the capillary microvasculature whereas R2* (1/T2*) is sensitive to both microand macrovasculature (38). Some studies have shown a linear relationship between R2* and content of deoxyhemoglobin while others have reported a quadratic function between blood R2* and oxygen saturation, suggesting that the technique should be more sensitive in regions with low oxygen saturation, e.g., in tumors (39,40). Tumor basal R2* may potentially be considered as an intrinsic marker of pO2, because it is related to the oxygenation state of hemoglobin and to the arterial blood pO2, which is in equilibrium with tissue p02. MRI using R2* quantification (BOLD MRI) has been reported to be a promising tool for noninvasive imaging of prostate cancer hypoxia (41,42).
In this study, we correlated the presence of MVI on histology by MRI estimation of BOLD effect (R2*) in HCC. We performed tumoral and background liver R2* estimations at baseline as well as after oxygen inhalation. Oxygen inhalation was performed to extract physiological changes induced in tumoral oxygen content which might allude to better determination of intratumoral hypoxia and thereby MVI. Our data analysis of absolute estimations of baseline R2* and postoxygenation R2* did not reveal a statistically significant threshold for prediction of MVI. We also assessed the difference in R2* before and after oxygenation, which also failed to provide a positive result statistically. Calculations of the ratios of R2* values between HCC and liver parenchyma both pre and post oxygenation were also unable to deliver a quantitative threshold to predict MVI. Finally, ratio of the delta R2* values between HCC and liver before and after oxygenation as well as percentage change in R2* values in HCC before and after oxygenation also did not result in a significant result. Our negative results, although disappointing, may be relevant to other investigators in this area. The use of BOLD contrast in tumors is a relatively new area of research and brings with its challenges of understanding and interpretation. As with any technique, BOLD MRI has both advantages and disadvantages. One advantage of BOLD MRI is that it is noninvasive BOLD MRI also has high spatial resolution, allowing it to address the issue of the spatial heterogeneity of the tumor response. To distinguish the contribution from the inflow and blood oxygenation to the BOLD signals, multiple gradient-echo imaging sequences is used instead of using conventional gradient-echo techniques. Carbogen induced changes in R2* or basal R2*, which reflect vascular development, may also be monitored with BOLD MRI to predict radiotherapy sensitivity. As for disadvantages, BOLD MRI is unfortunately an indirect method for monitoring tumor pO2. This is the result of the extreme sensitivity of changes in R2 * to the basal state of tumor oxygenation and blood volume fraction. The intra-and intertumoral distribution of these parameters may be greatly heterogeneous, making it very difficult to compare estimated pO2 changes between two regions or individuals. Even more problematic is the fact that the change in R2* is not always indicative of the change in pO2 only. Concomitant changes in blood volume, blood pH, and metabolic status can lead to smaller-than-expected or even negative changes in R2*. In our secondary analysis, we also performed multiparametric morphological and DWI analysis to assess for any significant factors that could correlate with presence of MVI. However, neither of these parameters yield a statistically encouraging result. Although we did observe an association of MVI with the presence of an incomplete tumoral capsule (Figs. 4, 5) the correlation was not statistically relevant. This secondary result is, however, in keeping with prior and more recent studies (14) that have attempted to correlate tumoral morphologic characteristics with prediction of MVI and returned negative results.
There are several limitations to our study that we would like to acknowledge. First, is the small sample size of only 26 subjects. However, this was an exploratory pilot study and herein we report our preliminary findings. Second, a technical limitation may be due to employment of a 1.5T MR system. Possibly higher field strengths of 3T may widen the quantitative R2* spectrum and perhaps have better statistical significance. The choice of a 1.5T system was based on well established MR image quality results for body imaging and well established values of T2 and T2* for normal liver tissue. A further limitation may be usage of nasal oxygen as methods to induce changes in tumoral oxygenation. Perhaps this does not have the desired impact, and there may be a need to explore stronger stimuli such as carbogen or CO2 breathing. Furthermore, they may be limitations to detecting or demonstrating hypoxia in HCC on the basis of fast blood flow in these mostly hypervascular tumors that may not allow enough time for hypoxia to be reflected in the BOLD signal. Limitations relating to data analysis include inclusion of only up to a maximum of five sections for BOLD evaluation. Because many tumors were larger than 5 cm, we may not be sampling the entire tumor for analysis and tumoral heterogeneity may be responsible for inaccurate oxygenation status estimation. There is a technical limitation to the number of slices that can be obtained in a single breathhold given the nature of multi-echo gradient echo acquisition. The move to free breathing or triggered acquisitions instead of breathheld acquisitions would result in temporal and spatial discrepancies and could only worsen the results. Finally, a significant limitation could revolve around the simplistic assumption that BOLD contrast necessarily reflects hypoxia that necessarily correlates with MVI. The lack of correlation between BOLD MR and MVI might be accounted for by the complexity of hypoxia. Factors such as blood volume will affect BOLD contrast and may interfere with the depiction of hypoxia. Hypoxia is also not a static entity, but can be rather than dynamic with perhaps acute and chronic hypoxic states affecting BOLD contrast differently impacting final results.
In conclusion, BOLD MR imaging does not appear to be promising as an accurate method for preoperative prediction of MVI in HCC under the circumstances wherein it was tested by us. Morphological imaging is already known to be unable to determine MVI as well. The prediction of MVI by existing functional and or morphological imaging methods remains elusive.