Evaluation of response in patients with hepatocellular carcinoma treated with intratumoral dendritic cell vaccination using intravoxel incoherent motion (IVIM) MRI and histogram analysis

Introduction

Hepatocellular carcinoma (HCC) is a major global health problem and the second leading cause of death by cancer worldwide (1). A majority of patients have advanced disease at diagnosis without options for curable treatment with surgical resection, transplantation, or locoregional therapy. In recent years, several clinical trials examining the potential of immunotherapy for the treatment of HCC have provided encouraging results in terms of both safety and efficacy (2). A problem when evaluating immunotherapy is that tumors respond differently to immunotherapies compared to treatment with chemotherapeutic drugs. Immunotherapy can have antitumor effects without significantly reducing tumor size, raising questions about the validity of the conventional RECIST 1.1. response criteria in such trials (3). Instead, creative use of new functional imaging methods, such as diffusion-weighted magnetic resonance imaging (DW-MRI) can offer opportunities for faster, cheaper, and more efficient preclinical and clinical testing of new immunotherapeutics in HCC.

Intravoxel incoherent motion (IVIM) MRI, using multiple b-values and a bi-exponential model for parameter estimation, is able to simultaneously quantify perfusion and diffusion in tumors without the use of contrast media. IVIM gives rise to three quantitative parameters: the true diffusion coefficient D; the pseudo diffusion coefficient D*; and the perfusion fraction f (4). IVIM imaging has clinical potential for diagnosing cancer (5), predicting prognoses (6), and therapeutic evaluation (7).

A well-known problem with quantitative MRI of tumors is that averaging the signal intensity variations within a region of interest (ROI) does not take the heterogeneity of malignant tumors into account. Histogram analysis is a mathematical method that aims to alleviate this by quantifying the heterogeneity of ROI data by measures such as standard deviation, skewness, kurtosis, and percentile of parameters (8). Relative to analysis of only ROI averages, histogram analysis may uncover biomarkers that can provide additional, clinically relevant information (9).

Dendritic cell (DC) vaccines are responsible for T-cell stimulation and antitumor immune response enhancement (2). In a phase I study, the dendritic cell vaccine ilixadencel (consisting of allogeneic, off-the-shelf, proinflammatory monocyte-derived dendritic cells) was administered to 17 patients with advanced HCC tumors (six in combinations with sorafenib). Ilixadencel was injected by ultrasound guidance in the same HCC lesion on three different occasions. Desirable immunologic response was shown in 11 of 15 (73%) evaluable patients in the form of increased frequency of tumor-specific CD8 + T cells in peripheral blood (10). DW-MRI suitable for IVIM evaluation was performed at baseline and before the second and third treatments at weeks 2–3 and at weeks 5–8, respectively, and at the follow-up three months after the first injection.

The aim of the present study was to evaluate the potential of DW-MRI using IVIM in the evaluation of patients with HCC treated with intratumoral injections with ilixadencel. We hypothesized that IVIM may provide metrics on tissue perfusion and cellularity that can be used to assess HCC response to tumor vaccination with ilixadencel, particularly when using histogram analysis.

Material and Methods

Patients and treatments

The trial (NCT01974661) was conducted at Sahlgrenska University Hospital, Gothenburg, Sweden. Patients eligible for the study were aged at least 18 years and had at least one measurable HCC lesion on computed tomography (CT) or MRI, i.e. measuring at least 20 mm at its longest unidimensional diameter. The HCC tumors were diagnosed by histopathology or according to the non-invasive European Association for the Study of the Liver (EASL) criteria (11) and the patients were not accessible for curative treatment or transarterial chemoembolization (TACE). In total, 17 patients were included in the study and patient characteristics are shown in Table 1.

OPEN IN VIEWER

Table 1.

Patient characteristics.

Characteristics	Values
Age (years)	67 (57–76)
Sex (M:F)	15:2
Underlying liver disease/cirrhosis	11/17
Hepatitis C	8/17
Size of index tumor (cm)	2.9 (2.1–16.4)
Bystander HCC lesions	11/17
AFP (µg/L)	619 (2–548,000)
Albumin (g/L)	35 (29–40)
CRP at inclusion (mg/L)	26 (2–140)
Time interval to 1st MR (days)	19 (13–43)
Time interval to 2nd MR (days)	45 (35–55)
Time interval to follow-up MR (days)	81 (76–90)

Values are given as n or median (range).

AFP, alpha fetoprotein; CRP, C-reactive protein; HCC, hepatocellular carcinoma; MR, magnetic resonance.

The primary objective of the original study was to evaluate the safety of the dendritic cell (DC) vaccine ilixadencel administered intratumorally in patients with HCC. Results of this phase I trial have been reported previously (10). An optional secondary endpoint was to assess changes in tumor viability and capillary perfusion by DW-MRI and IVIM modeling.

Three intratumoral injections of ilixadencel were planned for each patient (the first on study day 1, the second during days 14–21, and the third 21–35 days after the second). Injections took place at the Department of Radiology and each injection was performed under ultrasound guidance. In cases of multiple tumors, the most accessible lesion was selected. Twelve patients were treated with a dose of 10 × 10⁶ viable DC cells and five patients received a dose of 20 × 10⁶ viable DC cells. The last six patients were treated with a combination of ilixadencel and approved doses of sorafenib. Tumor dimensions were assessed at baseline, at three and six months after the first injection, and every three months until disease progression or the end of the study. Objective response assessment was done using CT or MRI according to the HCC-specific, modified RECIST criteria (12).

The study was approved by the Ethics Review Board (Regionala Etikprövningsnämnden i Göteborg, approval number 365-13). The study was conducted in compliance with the principles of the World Medical Association Declaration of Helsinki. All patients gave oral and written informed consent before study participation in accordance with the Declaration of Helsinki before entering the trial.

Image processing and analysis

To correct for motion between diffusion-weighted images with different b-values, thus acquired at different breath-holds, images with b > 0 were registered to the image with b = 0. The image registration used a regularized method based on freeform deformation, which is able to compensate for the complex motion of the organs in the abdomen while being robust to overfitting (13). The target lesion was identified by an experienced abdominal radiologist (MA, with >25 years of experience in body MRI) based on imaging data and findings at the ultrasound-guided injections. Then the radiologist manually delineated a freehand ROI around the largest cross-sectional area of the tumor on the images with the highest b-value of each patient. IVIM parameters (D, D*, and f) were estimated voxel-by-voxel within the ROI using an in-house developed Bayesian method with lognormal prior distributions and mode as the measure of central tendency, as described previously (14). In addition, for reference, the apparent diffusion coefficient (ADC) was estimated based on b-values 0 and 400 s/mm². From each ROI, the 10th, 50th (median), and 90th percentiles of the respective IVIM parameter were calculated (Fig. 1). The signal-to-noise ratio (SNR) was calculated voxel-wise as the ratio between the signal in the images without diffusion weighting (b = 0) and the standard deviation of the residuals of the model fits. Mean SNR levels in tumors were 21 ± 6.9. All data were processed with in-house developed Matlab software R2014b (The MathWorks, Natick, MA, USA).

OPEN IN VIEWER

Fig. 1.

Example of histograms of IVIM parameters and ADC of a typical patient at baseline with 10th, 50th, and 90th percentiles indicated. The nth percentile is the point at which n/% of the pixel values that form the histogram are found to the left. ADC, apparent diffusion coefficient; IVIM, intravoxel incoherent motion.

Results

The median longest diameters (LD) of the treated tumors were 2.9 cm (range = 2.1–16.4 cm). The tumors did, in general, display a large degree of heterogeneity, thus motivating the use of descriptive statistical measures beyond average values (Fig. 2).

OPEN IN VIEWER

Fig. 2.

Example of IVIM parameter and ADC maps of a tumor (color) superimposed on an image without diffusion-weighting (grayscale). Same patient and examination as in Fig. 1. ADC, apparent diffusion coefficient; IVIM, intravoxel incoherent motion.

After the DW-MRI at baseline, the second DW-MRI was performed on average 19 days (range = 13–43 days) after baseline and the third on average 45 days (range = 35–55 days) after baseline. Follow-up MRI (DW-MRI in conjunction with contrast-enhanced, clinical MRI) was performed on average 81 days (range = 76–90 days) after the first injection. Following intratumoral vaccination with ilixadencel, the most pronounced trends for IVIM parameters and ADC were found for the perfusion-related IVIM parameters, especially for the lower percentiles (Fig. 3). Significantly lower values of the 10th percentile of D* were found at weeks 2–3 and after the two first vaccinations at weeks 5–8, relative to baseline (P < 0.05). A non-significant trend of lower median ROI f values compared to baseline could also be seen at weeks 2–3 (P = 0.07). All results from the Wilcoxon tests, including non-significant results, are found in Supplementary tables S1–S3.

OPEN IN VIEWER

Fig. 3.

IVIM parameter and ADC percentiles evolution over time. The graph shows group median for each parameter and percentile, with the upper and lower group quartiles shown as an error bar. The three percentiles analyzed in this study (10th, 50th, and 90th) are shown in green, yellow, and blue, respectively. The upper row of plots shows the absolute parameter values, while the lower row shows the parameter values relative to baseline. ADC, apparent diffusion coefficient; IVIM, intravoxel incoherent motion.

There were significant correlations between the 10th percentiles and median of D at baseline and change in tumor size after three months (r = 0.79, P < 0.01 and r = 0.72, P < 0.05, respectively) (Supplementary table S4, Fig. 4), i.e. lower values of D at baseline indicated less tumor growth.

OPEN IN VIEWER

Fig. 4.

IVIM parameter and ADC values at baseline vs. change in tumor size. Correlation between histogram metrics of D at baseline and relative change in tumor size from baseline to follow-up at three months. Blue round dots represent the 10th percentile, orange rhomboid dots the 50th percentile, and green squared dots the 90th percentile. ADC, apparent diffusion coefficient; IVIM, intravoxel incoherent motion.

OPEN IN VIEWER

Fig. 5.

IVIM parameter and ADC value change from baseline to before vaccination 2 vs. change in tumor size. Blue round dots represent the 10th percentile, orange rhomboid dots the 50th percentile, and green squared dots the 90th percentile. ADC, apparent diffusion coefficient; IVIM, intravoxel incoherent motion.

There was a near significant negative correlation between early change in the 10th percentiles of D and ADC at 2–3 weeks and change in tumor size at three months (r = −0.53, P = 0.14, and r = −0.65, P = 0.06, respectively) (Fig. 5). All results from the Spearman analysis between parameter values after the first vaccination and change in tumor size, including non-significant results, are found in Supplementary table S5.

No significant differences in baseline values or changes in median IVIM parameters or changes in tumor size were found between patients treated with single ilixadencel or patients treated with a combination of ilixadencel and sorafenib (P = 0.93).

Discussion

After intratumoral vaccination with the immune primer ilixadencel, we found a significant reduction in histogram-derived low percentiles of the perfusion-related IVIM parameter D*. There was also a near significant reduction in median f (perfusion fraction). The majority of HCC lesions are highly vascularized by arterioles only, reflecting neoangiogenesis in the developing phase of HCC. Arterial phase enhancement on dynamic contrast-enhanced imaging is the single most important imaging feature in patients with HCC. Measurement of changes in the viable arterial phase enhancing components of the tumors according to the mRECIST and LI-RADS criteria are the standard method for assessment of radiologic response in HCC (15,16). Although antiangiogenic effect has not been reported with ilixadencel, an objective response rate of 19% according to mRECIST has been reported with other types of immunotherapy (17). The causal relationship between the observed reduction in the perfusion-related IVIM parameters and any antiangiogenetic effect of ilixadencel is, of course, uncertain; there are circumstances that both speak for and against such a relationship.

The correlation between perfusion-related IVIM parameters and classical perfusion parameters obtained with dynamic contrast-enhanced imaging (DCE) is not straightforward. Some studies examining the association between these perfusion parameters in various tumors have found moderate correlations (18–20), while other studies have not been able to demonstrate such a connection (21,22). The mixed findings of correlations between IVIM and DCE can in part be explained by different physiological principles working at different temporal scales (22). D* is dependent on blood velocity as well as on mean capillary segment length while the degree of contrast enhancement on DCE is dependent not only on vessel density but also to a large extent on vascular permeability.

After radioembolization of HCC, Hectors et al. reported a reduction of the IVIM parameter D* as well as a reduction in tumor DCE MRI parameters six weeks after the treatment, reflecting reduction in arterial blood flow (23). For prediction of objective response, none of the assessed central tendency measures (mean and median) of DCE-MRI or IVIM parameters at baseline showed significant diagnostic performance, while several baseline heterogeneity parameters, such as kurtosis, skewness, and standard deviation, did.

Tumors are by definition heterogeneous, most evident on the microscopic scale (8). Due to tumor heterogeneity, histogram IVIM evaluation may provide a more complete physiological tumor assessment and an improved response prediction than central tendency (mean and median) parameters (8,9). Lower percentile values of ADC and D are regarded as a promising measurement method as they may sensitively detect focal regions with higher cellularity (24). We found that low 10th and 50th percentiles of D at baseline correlated with smaller size increase at three months (Fig. 4). This finding is somewhat counterintuitive as low ADC and D values are generally considered to be associated with low histological grade and poor prognosis. However, high ADC and D at baseline have been found to predict poor response to treatment with TACE in HCC and with chemotherapy in colorectal liver metastasis, although no such correlation has been demonstrated in systemic treatments of patients with HCC (25). A hypothetical explanation for our findings is that because high ADC and D values reflect tissues with greater elements of necrosis with subsequent increased acidity and hypoxia, factors known to be associated with poorer treatment response, a treatment effect could only be seen in cell-dense tissue with lower D values (26). We also found a near significant negative correlation between early change in ADC in the 10th percentile at 2–3 weeks and change in tumor size at three months (r = 0.65, P = 0058), which may reflect a link between early apoptosis and long-term outcome (27).

The reason why the 10th percentile of D* and not the 50th, 90th percentile of D* showed a significant decrease at weeks 2–3 and weeks 5–8 compared to baseline is not obvious but may be linked to areas with increased malignancy (9). D* is dependent not only on blood velocity but also on the mean capillary segment length, which is likely to be shorter in regions with tortuous, abnormally branching microvessels (28). This is a complicating factor when correlating estimations of microperfusion in HCC tumors with D* and DCE (22).

IVIM parameters, especially D*, in the liver and in HCC tumors have been reported to have poor-to-modest repeatability (29). To our knowledge, there has been no study on the reproducibility of measurements of histogram parameters in HCC tumors. To evaluate a treatment effect on a biomarker, at least two sets of images obtained at subsequent timepoints are required (24). Although acceptable test–retest repeatability is important, for this small and very sick patient group in our study, assessment of test–retest variability would not be meaningful or practically feasible.

However, use of carefully selected methods of data acquisition and analysis has been shown to improve repeatability (30). The particular means undertaken in this study were as follows: (1) to use a relatively large number of b-values, which has been shown to decrease IVIM parameter variability (31); (2) to acquire data during breath-hold and apply non-rigid motion correction to minimize the presence of artefacts and misalignment between images with different b-values (31,32); and (3) Bayesian model fitting, which has been shown to be superior to other methods for IVIM parameter estimation (33).

There are a number of pitfalls associated with IVIM analysis, of which most are associated with violations of IVIM model assumptions. The following are the most prominent: (1) the use of segmented model fitting with a too low b-value threshold, which leads to underestimation of the perfusion-related parameters and overestimation of the diffusion coefficient (34); (2) use of too high b-values, which may result in underestimation of the diffusion coefficient due to intravoxel diffusional variance, with errors also propagating into the perfusion-related parameters (28); and (3) signal voids at higher b-values due to gross motion resulting in the same errors as point 1 (35). It has even been suggested that the diffusion- and perfusion-related IVIM parameters cannot be estimated independently although no violations are made, meaning that, for example, artificial trends in perfusion parameters can be introduced by changes in diffusion (36). However, as the biexponential representation does not introduce a fundamental limitation on independence of the parameters, the explanation behind such effects remains unknown in the absence of model violations. Using advanced diffusion encoding, such as variable flow encoding, to enable independent measures of the different IVIM parameters can potentially solve this issue (36).

The present study has some limitations. First, the sample size was small, with some patients not completing all examinations. Second, in four analyzed patients sorafenib was added to the treatment. However, separate analysis of the cohorts treated with ilixadencel vaccination only and the patients treated with the combination therapy revealed no significant differences in resulting MRI parameters. Third, no test of intra-observer or interscan measurement reproducibility was made. Even if a biomarker has poor reproducibility, it can be useful if the scale of the parameter change in special situations exceeds the scale of the measurement error. Several studies have noted significant associations between perfusion-sensitive IVIM parameters and accepted flow measurements in various tumors, including HCC (18–20,23). Finally, no correction for multiple testing was made in order to reduce the risk of type II errors in this exploratory study.

In conclusion, using an optimized DWI acquisition protocol and histogram analysis, we found a reduction in perfusion-related IVIM parameters early after intratumoral injections of the DC-vaccine ilixadencel, indicative of a treatment-induced decrease in tumor perfusion. Reduced tumor perfusion is regarded as an imaging sign of tumor response in HCC. An association between the low percentiles of D at baseline and smaller tumor growth at three months was also found. The findings are suggestive of a treatment effect of intratumoral vaccinations with the immune primer ilixadencel, although the value of IVIM with histogram analysis needs to be validated by further prospective studies.

Supplemental Material