Quantitative Perfusion-Weighted Magnetic Resonance Imaging in Uveal Melanoma

Abstract

Purpose: Perfusion-weighted imaging (PWI; magnetic resonance imaging [MRI]) has been shown to provide valuable biological tumor information in uveal melanoma (UM). Clinically used semiquantitative methods do not account for tumor pigmentation and eye movement. We hypothesize that a quantitative PWI method that incorporates these, provides a more accurate description of tumor perfusion than the current clinical method. The aim of this study was to test this in patients with UM before and after radiotherapy.

Methods: Perfusion-weighted 3T MRIs were retrospectively analyzed in 47 patients with UM before and after radiotherapy. Tofts pharmacokinetic modeling was performed to determine vascular permeability (K^trans), extracellular extravascular space (v_e), and reflux rate (k_ep). These were compared with semiquantitative clinical parameters including peak intensity and outflow percentage.

Results: The effect of tumor pigmentation on peak intensity and outflow percentage was statistically significant (P < 0.01) and relative peak intensity was significantly different between melanotic and amelanotic tumors (1.5 vs. 1.9, P < 0.01). Before radiotherapy, median tumor K^trans was 0.63 min⁻¹ (range = 0.06–1.42 min⁻¹), median v_e was 0.23 (range = 0.09–0.63), and median k_ep was 2.3 min⁻¹ (range = 0.6–5.0 min⁻¹). After radiotherapy, 85% showed a decrease in K^trans and k_ep (P < 0.01). Changes in tumor pigmentation before and after radiotherapy were small and not significant (median increase in T1 of 33 ms, P = 0.55).

Conclusions: Quantitative PWI parameters decreased significantly after radiotherapy and can therefore can serve as an early biomarker for treatment response assessment. However, due to the nonsignificant changes in tumor pigmentation before and after radiotherapy, the current semiquantitative method appears to be sufficiently sensitive for detection of changes in tumor perfusion.

Uveal melanoma (UM) is the most frequently occurring primary intraocular malignancy in adults.¹ In the clinical care for patients with UM, magnetic resonance imaging (MRI) is increasingly being used.^2,3 In addition to three-dimensional high-resolution anatomical images, dedicated functional sequences can provide tumor biomarkers for differential diagnosis and response assessment.^4–6 In this context, perfusion-weighted MRI (PWI) is a technique that can obtain information on the tumor perfusion by administering an intravenous contrast agent while continuously imaging the eye, to visualize the arrival, uptake, and outflow of the contrast agent (Fig. 1). 

Figure 1.

 

Overview of methods. Additional diagnostic sequences for this patient can be found in Supplementary Figure S1. (A) Tumors were delineated on the contrast-enhanced 3D T1-weighted image (3DT1gd). (B) All time points of the dynamic scan, the flip angle series and the B1-map were registered to contrast-enhanced 3D T1-weighted scan. In this figure, a mean time intensity curve is shown before and after registration, with the expected and actual tumor position at t = 200 seconds shown. (C) A T1-map was made for each tumor using the flip angle series and B1-map. (D) Voxel-wise pharmacokinetic (PK) modeling was performed using the T1-map, B1-map, and time-concentration curve. PK modeling was performed up to t = 180 seconds.

In clinical practice, a semiquantitative PWI analysis is generally performed,⁷ in which a representative region of interest (ROI) is drawn in one or several of the slices. By showing the average signal intensities of the ROI on each of the time points, a time intensity curve (TIC) is obtained. From this TIC, tumor parameters such as shape of the curve (e.g. a wash-out curve), relative peak intensity, and outflow percentage can be derived. These semiquantitative parameters have shown to be valuable for differential diagnosis,^7,8 prognosis,⁹ and early response assessment¹⁰ of UM. 

For other tumor sites, the more quantitative pharmacokinetic (PK) modeling has shown to provide a more accurate description of the tumor's perfusion characteristics, as the exchange of contrast agent from the blood to the tissue’s extracellular extravascular space is quantified.^11–15 The Tofts model is one of the most commonly used methods to perform PK modeling of MRI perfusion data.^16–19 In this model, the transfer rate of gadolinium from blood plasma to (tumor) tissue is modeled by K^trans, which depends on the vascular permeability, whereas v_e describes the fraction of extracellular extravascular space. For each voxel, these quantities can be obtained. From these quantitative parameters, k_ep, the reflux rate at which the gadolinium is transferred back into the vasculature, can be derived. 

Recently, Jaarsma et al.¹⁸ adapted this quantitative PWI analysis method for intra-ocular malignancies, showing that PK modeling can potentially provide more accurate biomarkers of the lesion's vasculature as it corrects for both eye motion and the tumor pigmentation. In particular the confounding effect of varying melanin content of UM, could overshadow the differences in tumor vasculature as it has a strong effect on the observed signal enhancement due to its effect on longitudinal relaxation time (T1).^7,18 

We hypothesize that for UM, the quantitative PWI method, which corrects for tumor pigmentation, is more accurate and will provide significantly different results than the current clinical method. The aim of this study was to test this hypothesis in a cohort of patients with UM before and after treatment. 

Methods

Patient Population

All patients with UM who received an MRI between May 2019 and March 2021 as part of clinical care or in the context of a scientific study were evaluated. Only patients with a tumor prominence larger than 2.5 mm (excluding sclera) on MRI were included, as the proposed PWI method is less reliable for thinner tumors due to partial volume effects.¹⁸ Data were analyzed retrospectively after approval of the local ethics committee, in accordance with the Declaration of Helsinki. For all patients, pretreatment MRIs were available, and post-treatment scans, acquired approximately 1 to 3 months after treatment, were available for 26 of 47 patients. Tumors with a prominence below 7 mm were generally treated with ¹⁰⁶Ru brachytherapy (apex dose 130 GyE²⁰), whereas larger tumors and juxtapapillary tumors were treated with proton beam therapy (tumor dose 60 GyRBE²¹) or enucleation. 

Image Acquisition

Patients were scanned on a 3 tesla MRI scanner (Ingenia Elition, Philips Healthcare, The Netherlands) with a 4.7 cm surface receive coil (Philips Healthcare) according to the protocol by Ferreira et al.^4,7 Scans used in this study included fat-suppressed contrast-enhanced 3D T1-weighted images, a variable flip angle series used for T1-mapping, a DREAM series for B1⁺-mapping,²² and a dynamic contrast-enhanced time series with a temporal resolution of 1.9 seconds per dynamic. The dynamic time series was acquired with a spatial resolution of 1.25 × 1.25 × 1.5 mm³ using the TWIST method.^18,23 The dynamic time series was acquired using a bolus of 0.1 mmol/kg gadoterate meglumine (DOTAREM; Guerbet, Roissy CdG Cedex, France), administered 6 seconds after the start of the scan, using a power injector with an injection rate of 2 mL/second.¹⁸ More scan parameters can be found in the Table. 

Table.

 

Scan Parameters

 

Quantitative Analysis

Tumors were semi-automatically delineated on the contrast-enhanced 3D T1-weighted images in MeVisLab (MeVis Medical Solutions, Bremen, Germany) using a subdivision surface fit²⁴ and manual correction by an ophthalmic MRI expert with 10 years of experience based on the clinical evaluation of the images by a neuroradiologist with at least 20 years of experience. Tumor prominence was computed automatically based on the 3D tumor contours, according to a previously published method.²⁵ 

The quantitative eye-specific PWI analysis was performed in Matlab 2019b (MathWorks, Natick, MA, USA) based on the method described by Jaarsma et al.¹⁸ First, the variable flip angle series, B1⁺-mapping scans and all separate time points of the dynamic contrast enhanced (DCE)-MRI were rigidly registered to the contrast-enhanced 3D T1-weighted scan using Elastix version 4.9.0²⁶ in 2 steps: first unmasked, then using an eye mask to correct for eye motion. Then, a T1-map was calculated using the variable flip angle series and the B1⁺-map.¹⁸ The tumor contours were eroded by two reconstruction voxels (total 0.6 mm) to eliminate any effects of partial voluming and sub-voxel eye motion after registration. Subsequently, voxel-wise PK modeling was performed according to the formulas of Tofts¹⁹ for the first 90 time points (172 seconds), resulting in a K^trans, v_e, and k_ep value per voxel for all tumors, with k_ep defined as the ratio between K^trans and v_e. An overview of the methods is shown in Figure 1. 

Clinical Parameters

Relative peak intensity and outflow percentage, the most commonly used clinical PWI parameters,⁷ were determined. Relative peak intensity was defined as the ratio between signal intensity of the highest peak within the first 90 seconds and mean signal intensity of the second through fourth time point. Outflow percentage was defined as the difference between the intensity 120 seconds after the peak and the peak intensity, divided by the peak intensity. Based on outflow percentage, tumors were assigned a TIC type, with outflow percentages of −5% or lower corresponding to wash-out curves, between −5% and 5% to plateau curves, and higher than 5% corresponding to progressive curves, respectively. 

Statistics

Medians and 25th to 75th percentiles were determined for B1⁺, T1, K^trans, v_e, and k_ep, relative peak intensity and outflow percentage before and after treatment. The difference in K^trans and relative peak intensity between melanotic and amelanotic tumors was determined using an unpaired t-test or Mann-Whitney test. Tumors were considered amelanotic if they had a median T1 larger than 1000 ms, based on a previous study in which all amelanotic tumors had a T1 above 1000 ms.¹⁸ Additionally, the significance of the difference in K^trans, v_e, and k_ep between the two time points was determined using a paired t-test or Wilcoxon signed rank test. As tumor pigmentation affects longitudinal relaxation time (T1) and therefore the amount of apparent enhancement,^7,18 tumor T1 was compared before and after treatment in the same way. 

To test if the quantitative PWI method was significantly different from the current clinical method the following comparisons were performed: first, multiple linear regression was performed with K^trans and T1 for relative peak intensity, as T1 has been proposed to be a confounding factor in the analysis of tumor vascularity. Similarly, the multiple linear regression was performed with k_ep and T1 for outflow percentage. Furthermore, to test if the quantitative PWI method was different from the current clinical method for the follow-up after treatment, Pearson's or Kendall's correlation coefficient was determined for the change in K^trans and change in relative peak intensity, and for the change in k_ep and the change in outflow percentage. 

Additionally, because tumor prominence is an important prognostic factor,²⁷ Pearson's or Kendall's correlation coefficient was determined for tumor prominence and PWI characteristics, such as peak intensity and outflow percentage, for tumor prominence and median tumor T1, and for tumor prominence and PWI parameters such as K^trans, v_e, and k_ep. 

For all statistical analyses, P values lower than 0.05 were considered statistically significant and the choice between parametric and nonparametric testing depended on the result of the Shapiro-Wilks test for normality. 

Results

Forty-seven (47) patients were included, of whom 26 underwent an MRI-scan both before and after treatment. Median tumor prominence excluding sclera was 6.6 mm (range = 2.7–13.8 mm). Fifty-three percent (53%) of all patients were treated with proton beam therapy, whereas 30% received ¹⁰⁶Ru brachytherapy and 17% of patients underwent enucleation. Median tumor B1⁺ was 91% (range = 78–103%) and median tumor T1 was 920 ms (range = 460–1500 ms). For the subgroup of 26 patients who underwent scans before and after treatment, the majority underwent proton beam therapy (24/26, 92%), whereas 2 patients (8%) received brachytherapy. Median relative peak intensity and outflow percentage before treatment were 1.63 (range = 1.19–2.66) and −8% (range = −24% to 3%), respectively. Relative peak intensity was significantly higher for patients with a higher T1: median relative peak intensities for melanotic and amelanotic tumors were 1.5 (range = 1.2–1.9) and 1.9 (range = 1.3–2.6), respectively (P < 0.01). Before treatment, 68% of tumors had a wash-out type curve and the remaining 32% had a plateau type curve. 

Before treatment, median tumor K^trans, v_e, and k_ep were 0.63 min⁻¹ (range = 0.06–1.42 min⁻¹), 0.23 (range = 0.09–0.63), and 2.3 min⁻¹ (range = 0.6–5.0 min⁻¹), respectively (Fig. 2A). Time-intensity curves of two representative patients, one with a highly enhancing wash-out curve and one with a plateau curve showing less enhancement, are shown in Figure 2B. Results for all patients can be found in Supplementary Table S1. The eye-specific quantitative PWI analysis used in this paper corrects for T1, resulting in a smaller, although still significant, difference in K^trans between melanotic and amelanotic tumors (median = 0.47 min⁻¹ vs. 0.67 min⁻¹, P = 0.03; Fig. 3). Amelanotic tumors had a smaller tumor prominence than melanotic tumors (5.9 mm vs. 7.7 mm, P = 0.03). Correlations between prominence and PWI parameters before treatment were weak and not significant (Supplementary Fig. S2). 

Figure 2.

 

(A) Median with IQR of K^trans and k_ep for each tumor before treatment. (B) Examples of time-concentration curves for patients with low and high K^trans and k_ep values. Mean time-concentration curves of the entire tumor are shown.

Quantitative PWI Measures Compared to Semiquantitative Clinical Parameters

Generally, tumors with high K^trans values (Fig. 2B, green curve) had a higher peak intensity than tumors with lower K^trans values (see Fig. 2B, blue curve), which was also reflected in regression coefficient of 0.84 (Fig. 4A, both K^trans and T1 P < 0.01). Due to the significant effect of tumor pigmentation, different patients with similar K^trans values showed larger differences in relative peak intensity (see Fig. 4A). Furthermore, both k_ep and T1 were significant factors for the outflow percentage (P < 0.001), with a regression coefficient of 0.80 (P < 0.001). 

Figure 3.

 

Relative peak intensity (semiquantitative clinical parameter) and K^trans (quantitative PWI analysis) for melanotic and amelanotic tumors. Large differences exist between relative peak intensity (A) due to the confounding effect of the difference in native T1, which is corrected in the calculation of K^trans (B), although differences between melanotic and amelanotic tumors remain. Tumors were considered amelanotic if they had a median tumor T1 higher than 1000 ms.

Figure 4.

 

Comparison between quantitative and semi-quantitative PWI parameters. RPI, relative peak intensity; OP, outflow percentage. For the regression analysis, T1 was measured in seconds. (A) The regression coefficient among K^trans, T1, and relative peak intensity was significant. The importance of correcting for T1 can be observed here: patients with similar K^trans values have varying peak intensities. (B) The regression coefficients among k_ep, T1, and outflow percentage was significant. Values with similar k_ep values correspond to varying outflow percentages.

Follow-Up

The median time between treatment and the post-treatment MRI scan was 69 days (range = 26–112 days). Eighty-five percent of the patients showed a decrease in K^trans and k_ep after treatment, with the median K^trans for all patients decreasing from 0.62 min⁻¹ before treatment to 0.41 min⁻¹ after treatment (P < 0.01) and median k_ep from 2.5 before treatment to 1.8 after treatment (P < 0.01; Fig. 5). No significant difference was observed for v_e, with median values of 0.23 and 0.24 before and after treatment, respectively (P = 0.95). Although K^trans and k_ep decreased for the majority of patients, two patients showed an increase in K^trans (Fig. 5C), and one patient showed an increase in k_ep (Figs. 5C, 5D). For these patients, the TIC remained the wash-out type and for the patients with increasing K^trans, the peak intensity was higher after treatment than before treatment. Change in K^trans correlated significantly with change in relative peak intensity (Pearson's r = 0.72, P < 0.01; Supplementary Fig. S3a) and change in k_ep correlated significantly with change in outflow percentage (Pearson's r = −0.83, P < 0.01; see Supplementary Fig. S3b). Differences in T1 between the time points were small and not significant (median increase of 33 ms, P = 0.55; Supplementary Fig. S4). 

Figure 5.

 

Both K^trans and k_ep decrease after treatment. (A) Violin plots of K^trans and k_ep values before and after treatment, (B) examples of time-concentration curves with corresponding PWI parameters for one patient, (C) changes in K^trans and k_ep after treatment, (D) patient with increase in K^trans and k_ep after treatment.

Discussion

In this paper, an eye-specific quantitative PWI analysis, which incorporates the confounding effect of tumor pigmentation, was compared to current clinical method for patients with UM before and after treatment. In addition, this is the first description of PK parameters of a large cohort of patients with UM. 

Overall, the values of the quantitative PWI parameters found in this cohort correspond well to an earlier study with the same method, but a different cohort,¹⁸ as well as to another study from a different center.²⁸ The study of Kamrava et al.,²⁹ however, reported a lower transfer constant from blood to tissue (K^trans) which could be caused by partial voluming, due to their relatively poor slice thickness of 3 mm, but could also be the result of a different contrast agent. These differences accentuate that the reference values proposed in this paper are specific to the used acquisition and analysis methods. Furthermore, the clinically used, semiquantitative, parameters (relative peak intensity, outflow percentage, and TIC type) observed in this cohort were comparable to prior UM studies,^7,8,30 with wash-out curves being the most prevalent and plateau curves occurring in approximately one third of the patients. No progressive curves were observed before treatment, a finding consistent with the literature.² 

Previous research has shown that semiquantitative PWI parameters can serve as an early biomarker for treatment response.¹⁰ Here, we show that these early changes are also reflected in the tumor's PK parameters as both K^trans and k_ep decreased significantly after treatment. A decrease in K^trans after treatment corresponds to decreased tumor perfusion, which is also reflected by the lower relative peak intensities after treatment. Similarly, the decrease in k^ep corresponds to less reflux back into the blood vessels, which is qualitatively reflected in lower washout after the initial enhanced after contrast administration. 

A similar post-treatment decrease of quantitative PWI parameters has been reported in other tumor sites, such as breast, head and neck, and rectal cancer.^12,13,15,31 However, for patients with small changes in PWI parameters, it is difficult to determine whether this is truly due to the absence of biological changes, as no research has been performed on the reproducibility of this eye-specific quantitative PWI analysis. 

Comparison With Semiquantitative Clinical Evaluation and Added Value

Although several differences exist between the proposed quantitative PWI method and the clinically used semiquantitative method, significant regression coefficients were observed between K^trans and peak intensity, and k_ep and outflow percentage. The main difference between the two approaches is that in the quantitative method the actual gadolinium concentration is calculated, whereas the semiquantitative method evaluates relative signal intensities, a metric which is partly determined by the tumor T1. The implications of this confounding effect are shown, as patients with different peak intensities can have the same K^trans. Incorporating these differences in tumor T1 results in a more accurate comparison of lesions, which is especially relevant in UM, as this T1 differs between melanotic and amelanotic UM. Interestingly, with the quantitative method still a difference in K^trans between melanotic and amelanotic tumors was observed, which might be related to the difference in tumor size or other differences between the two groups, as pigmentation appears to be related to genetic status and prognosis.^1,32 

For treatment response assessment, incorporating this confounding effect of pigmentation appears to be less relevant, as minimal changes in tumor T1 before and after treatment were observed, suggesting that the semiquantitative method is sufficient to assess whether changes associated with treatment response are present. 

Although the correlation between the quantitative PWI method and current clinical method was significant, relative peak intensity differed significantly between the melanotic and amelanotic subgroups. Therefore, the more advanced quantitative method may be valuable for differential diagnosis of intraocular lesions, as, conversely to UM, these are generally non-pigmented (except melanocytoma). For orbital lesions, the PK parameters have already been described by Xu et al.³³ and Ro et al.,³⁴ who found larger K^trans and k_ep values for malignant orbital masses compared to benign lesions. However, studies describing intraocular masses in terms of quantitative PWI parameters are sparse.^35,36 In clinical practice, we therefore advise to use separate reference values for pigmented and unpigmented UM when comparing the TIC-related metrics, such as peak enhancement, between lesions. 

Furthermore, although on a group level a correlation between the clinical and quantitative methods was observed, on a patient level the lack of motion correction in the clinical method might be a limitation. Due to the registration, the proposed quantitative method might be more robust to slow gaze drifts, which can be missed during the clinical evaluation.¹⁰ However, as this eye-specific registration is not available in clinically used software packages and is therefore time intensive, the current clinical practice of mitigating movement artifacts by the manual selection of a suitable neighboring time point is likely sufficient for most cases. 

Similarly, the fully quantitative method is a voxel-wise analysis, in contrast to the clinical method, where an averaged representative 2D ROI is evaluated. This voxel-wise analysis allows for a more detailed evaluation of the mass, which might be valuable for heterogeneous tumors.^7,37,38 Clinically, it is proposed to obtain several ROIs to correctly evaluate these masses,^7,10 but this is only feasible to a certain extent. Although the proposed quantitative PWI method provides quantifiable insights into tumor heterogeneity, the clinical value of this heterogeneity is currently unknown. 

MRI is gaining a more prominent place in the clinical care of patients with UM: it is, for example, now routinely used in the proton therapy planning,^39–41 as it provides a geometrically accurate three dimensional visualization of the tumor and surrounding structures.⁴² With these developments, various magnetic resonance biomarkers have become accessible in routine clinical care. Diffusion weighted imaging, for example, can aid in the differential diagnosis^2,7 and reportedly holds promise in prognosis determination.⁴³ Similar applications, including treatment response assessment have been reported of PWI.^2,7,10,29 In this context, the quantitative analysis as proposed by Tofts et al.¹⁹ has been validated in phantom studies^44–46 and was adapted for eye-specific applications by Jaarsma et al.,¹⁸ correcting for confounding factors such as tumor pigmentation and eye movement. However, this method requires several calibration scans and post-processing steps. As the semiquantitative analysis is more accessible, it is used clinically, and, in this paper, we show that these semiquantitative parameters strongly correlate to the validated Tofts method and are therefore sufficiently accurate for early treatment response assessment.¹⁰ Furthermore, reference values are given for melanotic and amelanotic UM, simplifying the use of the semiquantitative method in the differential diagnosis of amelanotic intraocular masses. Future research will have to show whether semiquantitative PWI is of added value for estimation of prognosis,²⁹ as the differences in perfusion between several risk profiles might be smaller. 

Limitations

The proposed method corrects for tumor pigmentation and eye motion. However, due to the limited field of view, no personal arterial input function (AIF) could be determined. Although population-based AIFs are widely used,^47,48 especially K^trans estimations may be less accurate on the patient level, especially for subjects with an impaired cardiac output. Here, an overestimation of the AIF would result in an underestimation of K^trans.⁴⁹ However, in this context, it is important to realize that patients with UM are not predisposed to cardiovascular abnormalities. Furthermore, adding a personal AIF to this method would require enlarging the field of view, resulting in a decrease in either spatial or temporal resolution, which might be unfavorable, especially in small UM. 

Conclusions

In this paper, we show that a quantitative PWI method that corrects for tumor pigmentation can serve as an early biomarker for treatment response assessment, as both K^trans and k_ep decreased significantly after radiotherapy. However, due to the nonsignificant changes in tumor pigmentation before and after treatment, the current clinical semiquantitative method appears to be sufficient as a biomarker for changes after ocular radiotherapy. For differential diagnosis, the quantitative PWI method may be preferred, although studies on quantitative PWI for intraocular masses are sparse. In clinical practice, we therefore advise to use separate reference values for pigmented and unpigmented intra-ocular masses when interpreting semiquantitative parameters such as peak enhancement. 

Acknowledgments

Funded by Varian, a Siemens Healthineers company. LK, MJC and JWB furthermore received research support from Philips. 

Disclosure: L. Klaassen, Varian (F), Philips (F); M.G. Jaarsma-Coes, Philips (F); M. Marinkovic, None; G.P.M. Luyten, None; C.R.N. Rasch, Varian (F); T.A. Ferreira, None; J.-W.M. Beenakker, Varian (F), Philips (F)