Introduction

Lung cancer is the most frequent cancer worldwide and the most common cause of cancer death [1]. Brain metastases occur in 10–20% of patients with localized lung cancer [2] and severely impact the quality of life and the overall survival (OS) [3]. The therapeutic strategy for brain metastasis is mainly based on local treatments such as surgery, radiosurgery or whole-brain radiotherapy [4] and depends on the number, localization and size of brain metastases, control of extra-central nervous system disease and the general condition of patients. Systemic therapy alone may be considered in asymptomatic patients [5].

Current guidelines for brain metastases assessment are based on size variation on conventional magnetic resonance imaging (MRI), but radiological enlargement during the early post-treatment period does not necessarily represent tumour progression, especially in patients treated by radiosurgery or targeted therapy [6]. Given the potential benefits of second-line therapies [5], it is crucial to determine predictors of response during the early follow-up. Previous studies have demonstrated the increasing abilities of advanced MRI techniques to predict the response in patients with brain metastases, including the apparent diffusion coefficient of diffusion-weighted imaging [7], the relative cerebral blood volume using dynamic susceptibility-contrast MRI perfusion [8] and the metabolic changes using MR spectroscopy [9]. However, the relevance of these techniques remains widely debated for patient management in routine clinical practice.

Dynamic contrast-enhanced (DCE) MRI is a minimally invasive method (requiring intravenous injection of gadolinium) which enables the assessment of the blood-brain barrier (BBB) integrity and neoangiogenesis within brain tumours [10]. When a two-compartment model is used [11], DCE MRI provides different pharmacokinetic parameters: (i) the volume transfer constant (K trans), reflecting the BBB permeability; (ii) the extravascular extracellular space per unit volume of tissue (v e); (iii) the plasmatic space per unit volume of tissue (v p); (iv) the rate constant (k ep), which is a K trans/v e ratio.

The diagnostic and prognostic value of DCE MRI has been well demonstrated in gliomas [12, 13] and extracranial neoplasms [14, 15] but it has been scarcely evaluated in the post-treatment assessment of brain metastases despite promising results [16, 17].

We hypothesized that increased BBB permeability and neoangiogenesis in brain metastases were associated with a better response to antineoplastic therapy and that the early variation of the DCE MRI parameters was predictive of the objective response.

The purpose of our study was to determine the diagnostic accuracy of pharmacokinetic parameters measured by DCE MRI to predict the response of brain metastases to antineoplastic therapy in patients with lung cancer.

Materials and methods

Patients

This was a monocentric prospective observational longitudinal diagnostic study without randomization (NCT02250755), approved by the institutional review board (no. 2012-A011164-39). Written informed consent was obtained from all patients.

We prospectively screened all adult patients with histologically proven lung cancer, and newly diagnosed brain metastases on systematic computed tomography scan treated at our institution between April 2012 and December 2014. Exclusion criteria were (a) neurosurgical treatment of brain metastases; (b) prior brain metastases; (c) Karnofsky performance status (KPS) <60; (d) contraindication to MRI and (e) inability to consent. As a result, 44 patients were included (33 men, 11 women) (Fig. 1).

Fig. 1
figure 1

Flow chart of the study. CT computed tomography, DCE dynamic contrast-enhanced, KPS Karnofsky performance status, MRI magnetic resonance imaging

Full size image

Pathological analysis of primary lung cancer, age, KPS, presence of extracranial metastases and oncologic treatment (systemic therapy, whole-brain radiotherapy, radiosurgery) were prospectively recorded at the diagnosis and during a follow-up period of 6 months following the treatment onset. Treatment strategy was decided by multidisciplinary agreement according to current guidelines [4].

MRI protocol

MRI studies were performed at baseline (within 1 month before treatment), early (7–10 weeks) post-treatment period and midterm (5–7 months) post-treatment period on a 3-T MRI scanner (Philips Achieva, Philips Medical Systems, Best, the Netherlands). Conventional sequences, including a post-contrast 3D gradient-echo T1-weighted sequence, were performed in all MRI examinations (Supplemental Table 1). A DCE MR sequence was added to baseline and early MRI. The dynamic acquisition (TR/TE = 14/3 ms, flip angle = 15°, voxel = 1.25 × 1.25 × 3 mm, 22 axial sections, 65 dynamics, temporal resolution = 6 s, duration = 6 min 46 s) was performed after two fast-gradient echo T1-weighted sequences with different flip angles (α = 5° and 20°). Gadolinium (gadoteric acid, DOTAREM®, 0.1 mmol/kg, relaxivity = 3.3 mM−1 s−1 at 3 T, injection rate 5–7 mL/s) was administrated intravenously 48 s after the beginning of the acquisition.

DCE MRI analysis

DCE MRI analysis was performed by a neuroradiologist (15 years of experience) blinded to clinical and volumetric data, using the PRIDE software (Philips). K trans, k ep, v e and v p maps were generated automatically according to the bicompartmental modified Tofts model [11]. Preprocessing steps consisted of rigid realignment for motion correction and exclusion of the voxels located outside the skull. A T1 mapping was computed from three fast-gradient echo T1-weighted acquisitions with different flip angles (α = 5°, 15°, 20°). An individual arterial input function was determined: several regions of interest (ROI) were manually placed in the internal carotid arteries, middle cerebral arteries and superior sagittal sinus to obtain the best enhancement curve. For DCE MRI measurements, mean K trans, k ep, v e and v p were measured in an ROI manually placed on the enhancing part of each lesion on the dynamic acquisition, at the level of the maximum section area to minimize partial volume effect, and copied to the K trans, k ep, v e and v p maps (Fig. 2). Necrotic and cystic areas were carefully excluded. The minimum area of the ROI was 20 mm2. At early follow-up, ∆K trans, ∆k ep, ∆v e and ∆v p were calculated according to the following formula: ∆K trans = (early K trans – baseline K trans)/baseline K trans. Changes in DCE MRI parameters had no impact on the subsequent treatment strategy.

Fig. 2
figure 2

Illustrative case. Patient with a left precentral metastasis on T1-weighted images at baseline (a), early follow-up (b) and midterm follow-up (c) and parametric map of v e at baseline (d) and midterm follow-up (e). A volumetric regression of the lesion was observed on the conventional T1-weighted sequence with gadolinium injection between baseline (a) and early follow-up (b). On the contrary, DCE MRI parametric map demonstrated an increase of the v e value between baseline (d) and early follow-up (e), which predicted the volumetric progression of the lesion at midterm follow-up (c). The green dotted line represents the region of interest used for DCE MRI measurement at baseline (a, d) and early follow-up (b, e), corresponding to the enhancing part of the lesion at the level of maximum section area with careful exclusion of necrotic components

Full size image

To assess the reliability of the method, mean K trans, k ep, v e and v p were blindly measured on baseline MRI by two independent observers (with 6 and 15 years of experience).

Volumetric analysis

A neuroradiologist (25 years of experience), blinded to DCE MRI and clinical data, measured the volume of each metastasis, using a semi-automated segmentation method, on the post-contrast 3D T1-weighted sequence. The midterm response, determined according to the volume variation of each brain metastasis from baseline to midterm using a volumetric adaptation of the RECIST 1.1 criteria, was considered as the reference standard: a decrease of 65% or more was considered a positive response, an increase of more than 73% was considered as progression and a variation between −65% and +73% was considered as stable disease [18].

Statistical analysis

The interobserver reliability was calculated with Fleiss intraclass correlation coefficient (ICC) using a two-way mixed model and interpreted as follows: poor agreement, <0.45; fair to good agreement, between 0.45 and 0.75; and excellent agreement, >0.75 [19].

The primary endpoint was the diagnostic accuracy of DCE MRI parameters in predicting the midterm response, in the terms of metastases assessment. The association between DCE MRI parameters and the midterm response was explored using mixed logistic regression models to take into account the within-subject correlation (DCE MRI parameters as fixed effects, patients as random effect). Receiver-operating characteristics (ROC) curves were computed using the predicted probabilities from the fitted mixed logistic model to estimate the area under the curve (AUC). Optimal thresholds were calculated by maximizing the sensitivity and specificity using the fixed effects. We compared the diagnostic accuracy of conventional MRI with and without the addition of the best parameter using a logistic model. The correlation between ∆K trans, ∆k ep, ∆v e, ∆v p and the volumetric variation of each brain metastasis at midterm follow-up was analysed using a Spearman’s rank correlation coefficient (r).

In terms of patient response, OS and progression-free survival (PFS) at 6 months were determined using the Kaplan–Meier method. We studied the association between clinical prognostic factors (age < 60 years, KPS ≥ 90, number of brain metastasis [solitary versus more than one], presence of extracranial metastases) [20] and OS or PFS using log-rank tests. The association between baseline DCE MRI parameters and OS or PFS was explored using univariate Cox’s regression. Hazard ratio and its 95% confidence interval (CI) were derived from Cox’s regression models as effect size measures. We assessed the log-linearity assumption for each parameter using Martingale residual plots and the proportional hazards assumption by plotting the Schoenfeld residuals against the rank of survival time [21]. For patients with multiple brain metastases, K trans, k ep, v e and v p were defined as the maximum value among the patient’s metastases.

Data were analysed using the SAS software, release 9.4 (SAS Institute, Cary, NC). A p value less than 0.05 indicated a statistically significant difference.

Results

Population

Baseline characteristics of the 44 patients (with 123 brain metastases) are presented in Tables 1 and 2. Median age was 58 years. The most frequent histopathology was adenocarcinoma (68.2% of the patients), followed by small cell carcinoma (25% of the patients). Brain metastases were diagnosed after a median time of 9.1 months [interquartile range (IQR) = 3–14.2] from lung cancer diagnosis. At the time of brain metastases diagnosis, 36 (81.8%) patients had concurrent extracranial metastases. Brain metastases were symptomatic in 6 (13.6%) patients. The median KPS was 90. Solitary brain metastases were found in 17 (38.6%) patients, 2 or 3 brain metastases in 14 (31.8%) and more than 3 brain metastases in 11 (25%). Median pretreatment volume of the brain metastases was 400 mm3 [IQR = 110–1240]. At baseline, median K trans, k ep, v e and v p were 0.04 min−1 [IQR = 0.02–0.06], 0.23 min−1 [IQR = 0.16–0.31], 0.18 [IQR = 0.13–0.25] and 0.04 [IQR = 0.03–0.05], respectively. The measurement reliability evaluated by ICC demonstrated an excellent agreement for K trans, k ep, v e and v p (Table 3).

Table 1 Clinical, histopathological and radiological characteristics of the patients at baseline (n = 44 patients)
Full size table
Table 2 Treatment strategy (n = 44 patients)
Full size table
Table 3 ICC for the DCE MRI pharmacokinetic parameters measured by two observers (n = 123 metastases)
Full size table

Follow-up

Median time from baseline MRI to treatment onset was 19 days [IQR = 7–27]. The early MRI was performed at a median time of 8 weeks [IQR = 7–10] after the treatment onset and the midterm MRI at a median time of 6 months [IQR = 5–7]. One patient had a modification of treatment between early and midterm follow-up, consisting of a modification of the systemic therapy regimen.

Clinical outcome

Twenty four patients were alive at midterm follow-up, with an OS rate of 54.5% [95% CI = 38.8, 67.8]; of these, 18 had stable disease or better, giving a PFS rate of 40.9% [95% CI = 26.5, 54.8] (Fig. 3). Causes of death were neurological in 5 (25%) patients, extracranial in 11 (55%) [thoracic progression in 6, pneumonia in 3, liver failure in 1, renal failure in 1] and unknown in 4 (20%). OS at midterm follow-up was significantly better in patients less than 60 years of age (p = 0.04) and with KPS ≥ 90 (p = 0.002) (Fig. 4). There was a trend towards worse OS with multiple brain metastases (p = 0.08) and the presence of extracranial metastasis (p = 0.06); however, this was not statistically significant. Similar results were observed for PFS (data not shown). Conversely, baseline DCE MRI parameters were not significantly associated with the OS or PFS (Table 4).

Fig. 3
figure 3

Kaplan–Meier overall (a) and progression-free (b) survival curves of the 44 patients at midterm follow-up. Death occurred in 20 patients during the follow-up, progression in 26 patients

Full size image
Fig. 4
figure 4

Kaplan–Meier survival curves of the 44 patients for 4 clinical prognostic factors demonstrated a better survival in patients < 60 years (a), with KPS ≥ 90 (b), with a solitary brain metastasis (c) and without extracranial metastasis (d)

Full size image
Table 4 Risk of death and progression at midterm follow-up by pretreatment values of pharmacokinetic parameters (n = 44 patients)
Full size table

Radiological outcome

Regarding to volumetric analyses at midterm follow-up, the objective response was evaluated for 61 brain metastases in 24 surviving patients. Positive response on MRI was observed in 46 brain metastases and non-response in the 15 remaining metastases (progression in 7 and stable disease in 8). When the volumetric adaptation of the RECIST 1.1 criteria was used, conventional MRI at early follow-up had a sensitivity of 0.76 [95% CI = 0.61, 0.87] and a specificity of 0.73 [95% CI = 0.51, 0.95] in predicting a positive response at midterm follow-up.

Regarding DCE MRI analyses, the baseline values of DCE MRI pharmacokinetic parameters and their early variation, according to the objective response at midterm follow-up, are presented in Table 5. Baseline K trans, k ep, v e and v p were not associated with the objective response at midterm follow-up (Table 6). However, early ∆K trans, ∆v e and ∆v p were significantly associated with the objective response at midterm follow-up (p = 0.02, p = 0.001 and p = 0.02, respectively) (Table 6). The best parameter to predict a positive response was ∆v e with an AUC of 0.93 [95% CI = 0.87, 0.99] (Figs. 2 and 5). The optimal threshold for ∆v e was < −12.5% with a sensitivity of 0.87 [95% CI = 0.74, 0.95] and a specificity of 0.80 [95% CI = 0.52, 0.96]. The addition of Δv e to conventional MRI in the prediction model significantly improved the diagnostic accuracy: AUC = 0.82 [95% CI = 0.70, 0.95] for conventional MRI vs AUC = 0.94 [95% CI = 0.88, 1.00] for conventional MRI + Δve, p = 0.03 (Fig. 6).

Table 5 Baseline values of DCE MRI pharmacokinetic parameters and their variation at early follow-up according to the objective response at midterm follow-up (n = 61 metastases)
Full size table
Table 6 Association between DCE MRI pharmacokinetic parameters and positive response at midterm follow-up (n = 61 metastases)
Full size table
Fig. 5
figure 5

ROC curves analysis for DCE MRI parameters. ∆v e provided the best predictive value for a positive response at midterm follow-up with an area under the receiver operating curve of 0.93 [95% CI = 0.87, 0.99]. AUC area under the receiver operating characteristics curve

Full size image
Fig. 6
figure 6

ROC curves analysis for conventional MRI with and without the addition of ∆v e. The addition of Δv e to conventional MRI in the prediction model significantly improved the diagnostic accuracy: AUC = 0.82 [95% CI = 0.70, 0.95] for model 1 (conventional MRI) vs AUC = 0.94 [95% CI = 0.88, 1.00] for model 2 (conventional MRI + Δv e), p = 0.03. AUC area under the receiver operating characteristics curve

Full size image

In each treatment subgroup, we found a strong correlation between early ∆v e and the volumetric variation of each brain metastasis at midterm follow-up (radiosurgery, r = 0.843, p = 0.002; radiotherapy, r = 0.589, p < 0.001; systemic therapy, r = 0.813, p < 0.001).

In the subgroup of patients with multiple brain metastases, the risk of global non-response at midterm follow-up was higher for patients having at least one metastasis with ∆v e > −12.5% (3/6) in comparison with patients without metastasis with ∆v e > −12.5% (1/5).

Discussion

In this study conducted in patients with brain metastases from lung cancer, treated either by radiosurgery, whole-brain radiotherapy or systemic therapy alone, we found that the early variation of v e (7–10 weeks post-treatment) was the best predictor of response at midterm follow-up. We also found a moderate but significant association between ∆K trans and the objective response.

To our knowledge, the prognostic value of the early variation of v e in brain metastases has not been previously reported. In brain neoplasms, neoangiogenesis generates leaky vessels, with altered endothelial tight junctions, leading to plasma extravasation into the extravascular extracellular space [22]. DCE MRI enables a quantitative assessment of the extravascular extracellular space expansion within each voxel through the pharmacokinetic parameter v e, which is elevated in brain metastases [23], glioblastomas and various extracranial tumours [24, 25]. The antiangiogenic effect of radiotherapy and systemic therapy is responsible for a reduction of vascular permeability and plasma extravasation in brain tumours. Accordingly, glioblastomas treated by antiangiogenic therapies [26, 27] and/or radiotherapy [28] exhibit a 20–30% post-treatment reduction of v e. Histopathological studies in mice confirmed decreased water content in glioblastomas treated by antiangiogenic therapy and demonstrated that it was associated with better survival [29].

In two previous studies evaluating DCE MRI in the assessment of brain metastases after radiotherapy, v e was not evaluated but the variation of K trans was reported as a good predictor of response. Jakubovic et al. [16], in 44 patients with 70 brain metastases treated by radiosurgery, found a lower K trans at 1 week in responding lesions. Similarly, Almeida-Freitas et al. reported an increase of K trans 4–8 weeks after radiosurgery associated with midterm progression in 26 patients with 34 brain metastases [17]. In the present study, we found a moderate association between ∆K trans and the objective response. Post-treatment vascular changes are highly time dependent and the timing of early MRI evaluation may account for this partial discrepancy. In the study of Jakubovic et al., K trans was significantly associated with tumour response when measured at 1 week after treatment, but the association was no longer significant at 1 month [16]. In our study, the early MRI evaluation was performed 7–10 weeks after treatment, which is in agreement with the routine follow-up of patients with brain metastases and therefore practical in a clinical setting. Nevertheless, the optimal timing of the early MRI evaluation remains to be determined.

As reported in earlier studies [16, 17], we did not find any association between pretreatment DCE MRI parameters and the objective response. BBB is considered to be the main limitation to the efficacy of chemotherapy on brain metastases [30, 31] and is the target of many research projects such as selective alteration of the BBB [32, 33] and development of drugs going through the BBB [34]. However, conversely to preclinical reports in animal models [35], baseline BBB permeability evaluated by K trans did not significantly differ between responding and non-responding lesions in the present study.

We showed that a cut-off value of ∆v e > −12.5% predicted the midterm response with an improved diagnostic accuracy in comparison to conventional MRI. According to current guidelines, conventional MRI with gadolinium injection is the standard reference for brain metastases assessment [6]. However, transient volume variations may occur in the early follow-up after radiotherapy [36] or targeted therapy [37] and are not predictive for long-term outcome. Thus, despite an increased scan time, an additional DCE MRI sequence allows a more comprehensive analysis of neoangiogenesis modifications. On the basis of our results and given the potential benefits of second-line treatments, metastases with a stable or increased v e (i.e. ∆v e > −12.5%) during the early post-treatment period would benefit from short-interval MRI follow-up to anticipate the diagnosis of tumour progression. In case of multiple brain metastases and heterogeneous response, the added value of DCE MRI remains undetermined but our preliminary results suggest that the risk of global non-response is higher in patients with at least one metastasis with ∆v e > −12.5%.

Our study had several limitations. First, the limited size of our cohort precluded accurate evaluation of DCE MRI in each treatment subgroup (radiosurgery vs whole-brain radiotherapy vs systemic therapy). The heterogeneity of treatment strategies is inherent to the progress of histopathological and genetic characterization of lung cancer which allows for individual treatment decision [5]. Nevertheless, the strong correlation between ∆v e and the volumetric variation at midterm follow-up in each treatment subgroup supported the prognostic value of ∆v e whatever the type of treatment. Second, we only included patients with brain metastases from lung cancer whereas previous studies included various primary tumours [16, 17]. This may explain in part the moderate association between ∆K trans and the objective response in this study, similarly to breast cancers in which the diagnostic accuracy of K trans in predicting complete response after neoadjuvant chemotherapy depends on immunochemistry subtypes [38]. The absence of association between survival and DCE MRI parameters may be explained by the high rate of extracranial causes of death in our study (75%), in line with others [39]. Finally, the low overall survival rate (54.5%) limited the follow-up efficiency. Surviving patients may have less active brain metastases causing a selection bias.

Conclusions

Our preliminary study showed the potential value of DCE MRI, and particularly the pharmacokinetic parameter v e, as a biomarker of the response of brain metastases from lung cancer to antineoplastic therapy. The early variation of v e may help the clinician to identify patients with high risk of progression, to individualize patient’s oncological follow-up and anticipate second-line treatments. Further studies including larger groups of patients are required to confirm the exact role of DCE MRI in the management of patients with brain metastases.