Preoperative Diffusion-Weighted Imaging of Single Brain Metastases Correlates with Patient Survival Times

Background MRI-based diffusion-weighted imaging (DWI) visualizes the local differences in water diffusion in vivo. The prognostic value of DWI signal intensities on the source images and apparent diffusion coefficient (ADC) maps respectively has not yet been studied in brain metastases (BM). Methods We included into this retrospective analysis all patients operated for single BM at our institution between 2002 and 2010, in whom presurgical DWI and BM tissue samples were available. We recorded relevant clinical data, assessed DWI signal intensity and apparent diffusion coefficient (ADC) values and performed histopathological analysis of BM tissues. Statistical analyses including uni- and multivariate survival analyses were performed. Results 65 patients (34 female, 31 male) with a median overall survival time (OS) of 15 months (range 0–99 months) were available for this study. 19 (29.2%) patients presented with hyper-, 3 (4.6%) with iso-, and 43 (66.2%) with hypointense DWI. ADCmean values could be determined in 32 (49.2%) patients, ranged from 456.4 to 1691.8*10−6 mm2/s (median 969.5) and showed a highly significant correlation with DWI signal intensity. DWI hyperintensity correlated significantly with high amount of interstitial reticulin deposition. In univariate analysis, patients with hyperintense DWI (5 months) and low ADCmean values (7 months) had significantly worse OS than patients with iso/hypointense DWI (16 months) and high ADCmean values (30 months), respectively. In multivariate survival analysis, high ADCmean values retained independent statistical significance. Conclusions Preoperative DWI findings strongly and independently correlate with OS in patients operated for single BM and are related to interstitial fibrosis. Inclusion of DWI parameters into established risk stratification scores for BM patients should be considered.


Introduction
Metastases to the brain are a frequent complication of cancer and are associated with high morbidity and mortality. Primary tumor types vary in their propensity to form brain metastases (BM) with lung cancer, breast cancer and melanoma showing the highest incidences of central nervous system (CNS) involvement. [1,2] Treatment so far relies mainly on surgery and radiotherapy, although some targeted drugs have shown clinically meaningful activity in distinct molecular tumor subtypes and are beginning to enter clinical practice. [3,4,5].
The prognosis of BM patients is poor with median overall survival times of only few months. Several risk stratification scores have been developed such as the recursive portioning analysis (RPA), the graded prognostic assessment (GPA) and the diagnosis specific graded prognostic assessment (DS-GPA). [6,7,8] These scores are based on parameters with established prognostic impact including the Karnofsky performance status (KPS), patient age, status of the primary tumor, presence of extracranial metastases and the number of BM. [6,7,8] Median overall survival (OS) from diagnosis of BM varies extensively from 3 months in the least favourable group, up to 25.3 months in the most favourable groups which includes also long term survivors. [8,9] Neuroradiological variables, with the exception of the number of BM, are not considered for prognostic risk stratification so far.
Magnetic Resonance Imaging (MRI) using pre-and postcontrast T1-weighted imaging, T2-weighted imaging and fluid attenuated inversion recovery (FLAIR) is the modality of choice for radiological evaluation of brain tumors. [10] Increasingly, additional advanced radiological techniques like magnetic resonance spectroscopy (MRS), perfusion MRI, or diffusion-weighted imaging (DWI) are used to characterize brain lesions in order to provide further clinically relevant information. [11,12] DWI is an MRI method based on the visualisation of the mobility of water molecules in the extracellular space. A low diffusion capacity due to a restricted mobility of the water molecules in the extracellular space results in a hyperintense signal in DWI and low apparent diffusion coefficient (ADC) values. In contrast, a high diffusion capacity due to an increased mobility of water molecules results in a hypo-or isointense DWI signals and high ADC values. [12] DWI parameters have been shown to correlate with various histopathological characteristics such as tumor type, tumor grade, Ki67 tumor cell proliferation index, cellularity, or amount of interstitial fibrosis and survival prognosis in several intra-and extracranial tumor types. [13,14,15,16,17,18,19,20,21]. However, the prognostic value of DWI and its correlation with histomorphological findings in patients with BM has not been systematically studied so far.
In the present study, we investigated the prognostic impact of DWI signal intensity and performed a correlative analysis with tissue-based parameters in a homogenous cohort of patients with single BM and surgery as first line treatment for BM.

Ethics Statement
The study was approved by the local ethics committee of the Medical University of Vienna, Austria. No written consent was given by the patients for their information to be stored in the database and used for research, because this study was performed in a retrospective manner in line with local regulations. The institutional ethics committee waived the need for written informed consent from the participants for this project (Ethics committee protocol number 641/2011).

Patients
We identified all patients with radiologically proven single BM who underwent surgery as a first-line-therapy for a single BM between April 2002 and December 2010 and whose presurgical MRI work-up included DWI. Availability of at least one tissue block for research purposes with viable BM tissue and full information on the clinical course including date of diagnosis, administered therapies, Karnofsky performance score, GPA and date of death or date of last follow-up investigation were mandatory for inclusion. All clinical parameters were retrieved by chart review and from the database of National Cancer Registry of Austria and the Austrian Brain Tumor Registry. [22].

Imaging Analysis
All imaging analyses were performed by one investigator (TS) blinded to all clinical and histological data. In conventional MRI (contrast-enhanced and native T1-weighted images, FLAIR and T2-weighted images, as available) the maximum diameter and localization of the single BM was determined. In DWI, the BM was semiquantitatively judged to be either hypointense, isointense, or hyperintense in comparison to normal non-pathological brain tissue. In BM which showed heterogeneous signal behaviour, diffusion intensity was graduat-ed upon the predominant (.70% of the metastasis) signal behaviour. In the cases with available ADC maps, ADC values were derived as described previously in up to 5 non-overlapping areas of interest (each at least 50 mm2) in solid, non-necrotic, non-macrohemorrhagic areas of the BM. In each case, the mean ADC value (ADCmean) was calculated from the ADC values of all areas of interest. [13].

Tissue Based Analysis
Tissue based analysis was performed blinded to clinical and radiological data. Histological confirmation of BM was evaluated on routinely performed hematoxylin and eosin (H&E) stained tissue sections by a specialist in neuropathology. Cellularity was evaluated semiquantatively on an H&E section as low, moderate and high. Gomori silver impregnation stain for reticulin was performed according to laboratory standard. The amount of extracellular reticulin fibers was semiquantitatively grouped as follows: prominent interstitial fibrosis (more than 25% of the tumor tissue displaying a dense interstitial meshwork of reticulin fibres); little interstitial fibrosis (less than 25% of the tumor tissue displaying a dense interstitial meshwork of reticulin fibres). Ki67 (antibody MIB1, Dako, Glostrup, Denmark) immunostaining and analysis was performed as previously published. [23] Ki67 proliferation index was obtained by counting 500 cells and giving the percentage of positive cells (0-100%). [23] Differentiation of the tumor tissue was divided into well, moderately and poorly differentiated based on the tumor organization, the cell polymorphism and the mitotic activity.

Statistical Analysis
For correlation of parameters the Spearmans correlation coefficient, Chi square test or the Mann-Whitney U test were used as appropriate. Overall survival (OS) was defined as time from first diagnosis of BM until death or last day of follow up. For all tests, a two-sided p-value of ,0.05 was considered as statistically significant. For univariate survival analysis Kaplan-Meier curves and the log-rank test were used. Variables that showed statistically significant prognostic value at univariate survival analysis were entered in multivariate survival analysis using the Cox regression model.
The statistical software package SPSS version 19 (SPSS Inc, Chicago, IL, USA) was used for all calculations.

Survival Analyses
Median OS from first diagnosis of BM to death was 15 months (range 0-99 months) in the entire population.
In univariate analysis, patients with hypo/isointense DWI signal intensity showed a significantly longer survival with a median OS of 16 months (95% CI: 10.79-21.25) than patients with hyperintense DWI signal intensity with a median OS of 5 months (95% CI: 0-12.47; p = 0.029; log rank test; figure 2). Patients with high ADCmean values showed a significantly longer survival with a median OS of 30 months (95% CI: 13.97-46.03) than patients with low ADCmean values with a median OS of 7 months (95% CI: 1.51-12.49; p = 0.02; log rank test; figure 2). Furthermore, primary tumor type, Karnofsky performance score ,70, lack of adjuvant WBRT after neurosurgery and high GPA were significantly associated with unfavourable OS in univariate analysis (p,0.05). Adjuvant chemotherapy after surgery for BM had no statistically significant impact on OS (p.0.05).
All factors with statistically significant impact on OS in the univariate analysis were included in multivariate analysis (AD-Cmean, primary tumor type, KPS and adjuvant WBRT (yes/no)). Primary tumor type as well as KPS and adjuvant WBRT did not remain statistically significant in multivariate analysis. Only high ADCmean values remained as statistically significant independent

Discussion
In this study, we found a highly significant association of presurgical DWI parameters with OS times of patients operated for single BM. Both, DWI signal intensity as assessed semiquantitatively by visual impression and ADCmean values stratified patients into prognostic groups. The median OS of patients with tumors showing hyperintense DWI was 5 months compared to 16 months in patients with iso-or hypointense DWI signals (p = 0.029; log rank test). ADCmean values showed an even stronger separation of risk groups with patients with high ADCmean values showing more than 4-times longer median OS (30 months) than patients with low ADCmean values (7 months; p = 0.008; log rank test). The prognostic impact of ADC values was independent from known prognostic factors including GPA class, the primary tumor type and the KPS and also from postoperative therapy including adjuvant WBRT and chemotherapy in multivariate analysis (Hazard ratio 0.32, 95% CI 0.12-0.91; p = 0.03; Cox regression model).
While, to the best of our knowledge, the correlation of DWI parameters with patient outcomes has not been investigated in BM, some studies have postulated a prognostic value of DWI signal intensity in primary tumors. In colorectal cancer, low mean ADC values were shown to be associated with aggressive tumor behaviour, impaired response to therapy and decreased prognosis. [19] Similarly, a correlation of low ADC values and impaired OS and impaired progression free survival was postulated for several other extracranial cancers with high propensity to form BM like lung cancer or breast cancer. [21,24] For the instance of intracranial lesions, hyperintense DWI signal were shown to be associated with worse outcome in primary CNS lymphoma and pituitary macroadenoma. [16,25] In line with these findings we observed a statistically significant and independent adverse prognostic value of restricted diffusion in our homogenous cohort of patients with single BM. Therefore, DWI signal intensity could serve as a prognostic imaging biomarker for clinical decision making.
The DWI signal intensity was shown to correlate with histology and cellularity of various intra-and extracranial tumors, providing an indirect insight in a tumor's microarchitecture. [13,15,19,21,26] In high grade gliomas a hyperintense DWI signal with low ADCmean values resembles areas of high cellularity with high cytoplasm to nucleus ratio. [14,15] Similar, a correlation of hyperintense DWI signal intensity and poor tumor differentiation was shown for extracranial tumors like lung cancer, breast cancer or rectal cancer. [19,21,24] For the instance of BM, a low ADCmean value was shown to correlate with high tumor cellularity and poor tumor differentiation. [26,27] In our study, we could demonstrate a significant correlation of a prominent interstitial fibrosis with signs of restricted diffusion in DWI, which resembles the impaired mobility of water molecules in the intercellular space. The interstitial reticulin fiber network is part of the fibrotic collagen-rich tumor stroma and our data further emphasize the importance of the microenvironment in the pathobiology of BM. [28] In line with our results, several other tumor types with a restricted diffusion due to dense stromal matrix were shown to have an impaired survival prognosis. [29,30,31].
Our study has some limitations that need to be acknowledged. We performed a retrospective study in a single center and were therefore able to include only a limited number of cases, thus restricting the statistical power of our correlative analyses. On the other hand, our approach enabled us to analyse a well-defined patient cohort characterized by single brain metastasis homogenously treated by neurosurgical BM resection as the initial therapy. Another limitation related to the retrospective nature of our study is the fact that ADC maps could not be retrieved in all cases. In the absence of ADC maps, diffusion restriction as cause of DWI hyperintensity cannot be unequivocally differentiated from other phenomena such as T2-shine through. [32] Further, the accuracy of ADC values is potentially limted due to the usage of different MRI machines. [33] However, we found a strong correlation of ADC values and semiquantitaitve DWI signal intensity in the cohort of 32 patients of whom both parameters were available. Furthermore, in this cohort the prognostic impact of ADC values was even more pronounced than the semiquantitatively evaluated DWI signal intensity. Still, our findings need to be reproduced in independent data sets, preferably in prospective studies.
In conclusion, we could demonstrate the independent prognostic value of DWI findings in our large homogenous cohort of patients with a single BM and its correlation with tissue based characteristic, indicating the value of DWI signal intensity as an imaging biomarker. Future studies should prospectively evaluate the prognostic value and the inclusion in prognostic scores of DWI parameters.