Dynamic Contrast-Enhanced MRI–Derived Intracellular Water Lifetime (τi): A Prognostic Marker for Patients with Head and Neck Squamous Cell Carcinomas

The authors evaluated 60 patients with dynamic contrast-enhanced MR imaging before treatment. Median, mean intracellular water molecule lifetime, and volume transfer constant values from metastatic nodes were computed from each patient. Kaplan-Meier analyses were performed to associate mean intracellular water molecule lifetime and volume transfer constant and their combination with overall survival and beyond. Patients with high mean intracellular water molecule lifetime had overall survival significantly prolonged by 5 years compared with those with low mean intracellular water molecule lifetime. Patients with high mean intracellular water molecule lifetime had significantly longer overall survival at long-term duration than those with low mean intracellular water molecule lifetime. Volume transfer constant was a significant predictor for only the 5-year follow-up period. They conclude that a combined analysis of mean intracellular water molecule lifetime and volume transfer constant provided the best model to predict overall survival in patients with squamous cell carcinomas of the head and neck. BACKGROUND AND PURPOSE: Shutter-speed model analysis of dynamic contrast-enhanced MR imaging allows estimation of mean intracellular water molecule lifetime (a measure of cellular energy metabolism) and volume transfer constant (a measure of hemodynamics). The purpose of this study was to investigate the prognostic utility of pretreatment mean intracellular water molecule lifetime and volume transfer constant in predicting overall survival in patients with squamous cell carcinomas of the head and neck and to stratify p16-positive patients based upon survival outcome. MATERIALS AND METHODS: A cohort of 60 patients underwent dynamic contrast-enhanced MR imaging before treatment. Median, mean intracellular water molecule lifetime and volume transfer constant values from metastatic nodes were computed from each patient. Kaplan-Meier analyses were performed to associate mean intracellular water molecule lifetime and volume transfer constant and their combination with overall survival for the first 2 years, 5 years, and beyond (median duration, >7 years). RESULTS: By the last date of observation, 18 patients had died, and median follow-up for surviving patients (n = 42) was 8.32 years. Patients with high mean intracellular water molecule lifetime (4 deaths) had significantly (P = .01) prolonged overall survival by 5 years compared with those with low mean intracellular water molecule lifetime (13 deaths). Similarly, patients with high mean intracellular water molecule lifetime (4 deaths) had significantly (P = .006) longer overall survival at long-term duration than those with low mean intracellular water molecule lifetime (14 deaths). However, volume transfer constant was a significant predictor for only the 5-year follow-up period. There was some evidence (P < .10) to suggest that mean intracellular water molecule lifetime and volume transfer constant were associated with overall survival for the first 2 years. Patients with high mean intracellular water molecule lifetime and high volume transfer constant were associated with significantly (P < .01) longer overall survival compared with other groups for all follow-up periods. In addition, p16-positive patients with high mean intracellular water molecule lifetime and high volume transfer constant demonstrated a trend toward the longest overall survival. CONCLUSIONS: A combined analysis of mean intracellular water molecule lifetime and volume transfer constant provided the best model to predict overall survival in patients with squamous cell carcinomas of the head and neck.

cular compartment to tumor interstitium. The potential of pretreatment K trans in predicting short-term response 4-7 as well as overall survival (OS) in patients with HNSCC has been reported. 8,9 Patients treated with chemoradiation therapy (CRT) and with high baseline K trans from metastatic nodes were associated with an improved prognosis compared with patients with low baseline K trans . 8,9 In addition to K trans , shutter-speed model analysis of DCE-MRI derives a novel imaging biomarker known as mean intracellular water lifetime ( i ), [10][11][12] which has been suggested to be a metabolic marker. 12 The unique strength of i lies in the fact that it is less sensitive to arterial input function scaling variations than K trans , 13 indicating that i is a more reproducible and reliable marker. The parameter i has been used to characterize breast, 14 prostate, 15,16 esophageal, 17 and hepatocellular cancer. 18 A recent study 19 also demonstrated the prognostic utility of i in predicting survival in patients with hepatocellular carcinomas.
Although K trans reflects tumor perfusion and vascular permeability, 3 i provides unique information related to tumor cell characteristics such as cell size, cell membrane permeability, and cellular metabolic activity. 12 Given that K trans and i provide complementary information about the tumor microenvironment, we believe that a combined analysis may be more useful than individual parameters in predicting prognosis in patients with HNSCC.
Thus, the purpose of the present study was to assess the prognostic value of pretreatment i and K trans in predicting OS in patients with HNSCC. In addition, in a subset of patients, we explored the prognostic potential of i and K trans in p16-associated HNSCC given the fundamental differences in tumor biology and prognosis of these patients. [20][21][22]

Patients
This retrospective analysis of pre-existing imaging and clinical data was institutional review board-approved and was compliant with the Health Insurance Portability and Accountability Act. On the basis of previous CT/MR imaging reports, all patients were assessed for the presence of at least 1 metastatic cervical lymph node measuring Ͼ1 cm 3 and biopsy-proved histopathologic diagnosis of HNSCC. The exclusion criteria included prior CRT or a history of cancer other than HNSCC. A total of 72 patients with newly diagnosed HNSCC met the inclusion criteria and were recruited between January 2005 and August 2009. TNM staging was used to determine the disease status. Each patient received appropriate therapy to deliver the maximum clinical benefit, which included upfront neck dissection (n ϭ 3), concurrent CRT (n ϭ 46), or induction chemotherapy followed by CRT (n ϭ 23). Three patients who underwent upfront neck dissection and 9 patients who had either corrupted MR imaging data or insufficient clinical data were excluded from the data analysis. Therefore, OS analyses were performed on the remaining 60 patients (mean age Ϯ SD, 62.34 Ϯ 9.18 years; 49 men, 11 women). Tumor location and staging from these patients at the initial presentation are summarized in the Table.

Data Acquisition
All patients underwent MR imaging before surgery and CRT on a 1.5T Sonata scanner (Siemens, Erlangen, Germany; n ϭ 37) or on a 3T Magnetom Trio scanner (Siemens; n ϭ 35). Structural imaging included axial T2-weighted and T1-weighted images with standard parameters. Inversion-recovery-prepared T1-weighted images were acquired by using TIs of 60, 200, 400, 800, and 1600 ms before the acquisition of DCE-MRI data for T1 quantification.
As described previously, 5 DCE-MRI was performed by using a rapid 3D-spoiled gradient-echo sequence modified to acquire 8 angle-interleaved subaperture images from the full-echo radial data. 23 Imaging parameters were: TR, 5.0 ms; TE, 4.2 ms; 256 readout points per view; 256 views (32 views per subaperture, 8 subapertures); field of view, 260 ϫ 260 mm 2 ; number of sections, 8; section thickness, 5 mm. Fat saturation was applied once every 8 excitations. Spatial saturation was applied once every 32 excitations to minimize the flow effect while minimizing acquisition time. This scheme resulted in a temporal resolution of 2.5 seconds for each subaperture image with full spatial resolution of 256 ϫ 256 by using a dynamic k-space-weighted image reconstruction contrast algorithm.

Image Processing
All images (T2, T1, postcontrast T1-weighted, and DCE-MRI) were coregistered by using a 2-step nonrigid image registration technique. 5 The ROIs were drawn on the solid portion of the largest nodal mass by using anatomic images. Care was taken to avoid necrotic/cystic or hemorrhagic parts as well as surrounding blood vessels. Pharmacokinetic analysis of DCE-MRI data was performed for each voxel in the selected ROIs by using a shutterspeed model. 5 formed semiautomatically from an ROI on one of the carotid arteries located near the metastatic lymph node. 5,10 Median pretreatment i and K trans were computed by using only the central 4 sections to avoid erroneous results from wraparound artifacts in the edge sections.

Clinical Follow-Up and Data Analysis
The clinical follow-up period was measured from the end date of CRT to the date of death for deceased patients or to the date of last observation for surviving patients. Median pretreatment i and K trans values were 0.125 seconds and 0.409 minutes Ϫ1 , respectively, and were used as thresholds to divide patients into 2 groups (at or above and below the threshold value). Patients in the high i group had a mean Ϯ SD i of 0.276 Ϯ 0.086 seconds and a median of 0.269 seconds. Patients in the low i group had a mean Ϯ SD i of 0.070 Ϯ 0.028 seconds and a median of 0.071 seconds. Similarly, patients in the high K trans group had a mean Ϯ SD K trans of 0.90 Ϯ 0.54 minutes Ϫ1 with a median of 0.88 minutes Ϫ1 , whereas patients in the low K trans group had a mean Ϯ SD K trans of 0.188 Ϯ 0.108 minutes Ϫ1 and a median of 0.196 minutes Ϫ1 . Using the 2-year (short-term), 5-year (intermediate-term), and all available follow-up (long-term [median, 7.83 years; range, 0.07-10.7 years]) as clinical end points, OS was analyzed. In addition to using i and K trans as independent predictors, combinations of these parameters (high i /high K trans , high i /low K trans , low i /high K trans , and low i /low K trans ) were used. Kaplan-Meier survival curves were plotted and compared by using log-rank tests. Using a Cox regression model, hazard ratios (HRs) of deaths and associated 95% CI were estimated for i and K trans first separately, and then for the different combinations of these parameters (high i /high K trans , high i /low K trans , low i /high K trans , and low i /low K trans ). In addition, a Wald test was performed to evaluate the joint effect of these combinations. A P value Ͻ.05 was considered significant. All data analyses were performed by using SPSS for Windows version 18.0 (IBM, Armonk, New York).

Stratifying HPV/p16-Positive and p16-Negative Patients
Human papillomavirus (HPV) status was determined from tissue specimens by immunohistochemical evaluation of p16 expression by using a commercially available monoclonal antibody. Tissue samples were available from only 32 patients, and these were divided into 2 groups: positive (n ϭ 21) or negative (n ϭ 11) for p16 expression. 24 There were no significant differences (P Ͼ .05) in age, treatment regimen, i , and K trans between patients who had and those who did not have tissue specimens for p16 expression. Using p16 as an independent variable, OS analyses were performed from these 32 patients. To further stratify p16-positive and p16-negative patients, separate OS analyses were performed by using i and K trans as independent variables and by using combinations of these parameters.

RESULTS
Representative anatomic images and i and K trans maps from a patient who was alive by the last date of observation with a follow-up duration of 8.19 years and from a patient who died 2.12 years after the end of CRT are shown in Figs 1 and 2, respectively.
In the first 2-year follow-up period, 13 of 60 patients died of the disease. In 5 years, the number of deceased patients was 17, whereas a total of 18 patients died by the last date of observation. The median follow-up for surviving patients (n ϭ 42) was 8.32 years (range, 5.42-10.7 years).

Prognostic Utility of i
In the first 2 years, a trend toward longer OS was noted for patients with higher i (4 deaths) compared with those with lower i (9 deaths; log-rank P ϭ .09). The probability for survival in patients with high i was 86.7% (95% CI, 68.3%-94.8%), whereas it was 70% (95% CI, 60.3%-83.1%) for patients with low i .
Interestingly, by 5 years, significantly longer OS was observed for patients with higher i (4 deaths) compared with those with lower i (13 deaths; log-rank P ϭ .01). Similarly, patients with higher i (4 deaths) had significantly prolonged OS compared with those with lower i (14 deaths; log-rank P ϭ .006; Fig 3) when long-term follow-up duration was considered (median duration, Ͼ7 years). At 5 years, survival probability for patients with high i was 86.7% (95% CI, 68.3%-94.8%), and for patients with low i , it was 56.8% (95% CI, 37.4%-72.1%).
Cox regression analysis demonstrated a nonsignificant difference (P ϭ .13) in OS between the 2 groups for the 2-year follow-up period, with an HR of 2.43 (95% CI, 0.75-7.92). However, a P value of .02 with HR of 3.71 (95% CI, 1.20 -11.39) for 5 years and a P value of .01 with HR of 4.24 (95% CI, 1.38 -12.97) were observed in predicting long-term OS.
By 5 years, significantly prolonged OS for patients with higher K trans (5 deaths) was noted compared with those with lower K trans (12 deaths; log-rank P ϭ .03). Survival probability for patients with high K trans was 83.9% (95% CI, 65.5%-93.0%) and for patients with low K trans was 58.6% (95% CI, 38.8%-74.0%). However, only a trend toward prolonged OS was observed for patients with higher K trans (6 deaths) compared with those with lower K trans (12 deaths; log-rank P ϭ .06) when long-term OS was evaluated (Fig 4).

Prognostic Utility of Combined Analysis Involving i and K trans
By 2 years, the longest OS was observed for patients with high i /high K trans (n ϭ 15, 2 deaths). On the other hand, patients with low i /low K trans (n ϭ 14, 7 deaths) had the shortest OS (log-rank P ϭ .02). In the group of patients (n ϭ 31) who had either high i /low K trans or low i /high K trans , 4 patients died. Moreover, patients with high i /low K trans had longer OS than patients with low i /high K trans . Similarly, at 5 years, patients with high i /high K trans (n ϭ 15, 2 deaths) had the longest OS, whereas patients with low i /low K trans (n ϭ 14, 10 deaths) had the shortest OS (log-rank P Ͻ .0001). Of the remaining 31 patients, 5 patients were deceased and patients with high i /low K trans had longer OS than patients with low i /high K trans . A similar pattern was observed for long-term analysis. Patients with high i /high K trans (n ϭ 15, 2 deaths) had the longest OS, and patients with low i /low K trans (n ϭ 14, 10 deaths) had the shortest OS (log-rank P Ͻ .0001). Of the remaining 31 patients, 6 died, and again, patients with high i /low K trans had longer OS than patients with low i /high K trans (Fig 5).   For the first 2-year follow-up period, multivariate Cox regression analyses revealed a P value of .048 with HRs of 0.21 (95% CI, 0.04 -1.00; P ϭ .05), 0.22 (95% CI, 0.05-1.07; P ϭ .06), and 0.20 (95% CI, 0.04 -0.95; P ϭ .04) for patients with high i /high K trans , high i /low K trans , and low i /high K trans , respectively, with respect to patients with low i /low K trans . Similarly, a P value of .003 with HRs of 0.13 (95% CI, 0.03-0.60; P ϭ .009), 0.14 (95% CI, 0.03-0.63; P ϭ .01), and 0.19 (95% CI, 0.05-0.69; P ϭ .01) for the first 5 years, and P value of .003 with HRs of 0.12 (95% CI, 0.03-0.58; P ϭ .007), 0.13 (95% CI, 0.03-0.61; P ϭ .01), and 0.26 (95% CI, 0.08 -0.82; P ϭ .02) for long-term OS were observed.

Stratifying HPV/p16-Positive and p16-Negative Patients
Of 32 patients with the availability of p16 status, 13 died by the date of last observation. Patients with p16-positive status had significantly longer OS than p16-negative patients (P ϭ .05; Fig 6A). Noticeably, no significant difference (P ϭ .17) in OS was observed between patients in whom p16-expression data were or were not available.
Patients with p16-positive status with higher i demonstrated nonsignificantly longer OS (P ϭ .20) than patients with lower i . Similarly, nonsignificant (P ϭ .61) differences were observed for K trans . Moreover, p16-positive patients with high i /high K trans had the longest OS, and patients with low i /low K trans had the shortest OS. However, these findings were also not significant (Fig 6B; P Ͼ .05). In addition, patients with p16-negative status and higher i demonstrated nonsignificantly (P ϭ .12) longer OS than those with lower i . We could not compare OS by using K trans because all patients in this subgroup had K trans higher than the median value.

DISCUSSION
Our observations suggest that patients with higher pretreatment i and K trans from metastatic nodes exhibit longer OS. These results are particularly encouraging given the adverse prognosis associated with nodal metastases and offer potential opportunities to develop alternative therapeutic approaches in the poor prognosis cohort.
Earlier studies 12,25,26 have reported that i is inversely correlated with cell membrane ion-pump activity, a measure of mitochondrial metabolism suggesting that i might be a sensitive indicator of cellular energy turnover.
Similarly, some studies have reported that increased metabolic activity is associated with lower i (median, 0.16 -1.03 seconds) in regions of prostrate 15,16 and esophageal cancer 17 compared with normal tissues. In yet another study, 18 lower i (0.11 Ϯ 0.02 seconds versus 0.29 Ϯ 0.53 seconds) was observed from normal liver parenchyma (a metabolically more active region) compared with hepatocellular carcinomas. The inverse correlation between i and tumor metabolism was corroborated when higher i was observed in complete responders compared with poor responders with breast carcinomas. 12,14 Taken together, these studies imply that malignant cells  . Patients with positive p16 expression and high i /high K trans (thick broken curve) had longer OS than p16-positive patients with low i /low K trans (thin broken curve). In addition, patients with high i /low K trans (gray solid curve) exhibited longer OS than patients with low i /high K trans (black solid curve) at all clinical end points. However, these differences were not significant (P Ͼ .05).
from tumors with high i have reduced metabolic energy available for their growth relative to malignant cells from tumors with low i . In the present study, patients with HNSCC with higher pretreatment i from the metastatic nodes were associated with prolonged OS measured at 3 different clinical end points, suggesting that malignant cells were associated with reduced metabolic energy. An inverse correlation of i with EF5, a marker of hypoxia, and direct correlation of elevated i regions with high blood flow have been reported. 27 An earlier study 28 reported that tumors with relatively higher blood flow and reduced hypoxia are associated with increased oxygenation, resulting in better access to chemotherapeutic drugs and radiosensitivity. Taken together, these studies and our observations suggest that tumors with higher i may harbor a favorable microenvironment for CRT. We postulate that the combined effect of lower hypoxia and reduced metabolic activity of the tumor cells might have contributed to higher pretreatment i that led to better treatment response and prolonged OS in patients with HNSCC. This notion is supported by a HR of 4.24 that was observed at the long-term follow-up period, implying a 4-fold higher risk of death in patients with low i than in patients with high i .
Pretreatment K trans (median value, 0.19 -0.64 minutes Ϫ1 ) has been reported to predict treatment response to CRT, 4-6 induction therapy, 7 and short-term survival in patients with HNSCC. 8,9 We also observed that higher pretreatment K trans was associated with better OS. Because K trans reflects a combination of tumor perfusion and microvascular permeability, our results and those of earlier published reports support the notion that tumor vascularity might be an important predictor of disease control in HNSCC. The fact that K trans predicted only 5-year OS while i was useful in predicting both 5-year and long-term OS suggests that i is a more robust prognostic imaging biomarker in evaluating HNSCC.
When i and K trans were combined, significantly longer OS was observed for patients with high i /high K trans for all clinical end points. We believe that synergistic interaction between these 2 parameters in predicting OS was greater than what would be expected from an individual parameter. The fact that patients with high i /high K trans demonstrated the longest OS agrees with the hypothesis that these patients may have had the most favorable conditions such as elevated tumor blood flow, lower hypoxia, and lower cellular metabolic energy for optimal therapeutic benefit of CRT. Another interesting finding was that patients with high i / low K trans exhibited longer OS than patients with low i /high K trans at all clinical end points. This indicates that i was a more dominant factor than K trans in the synergistic interaction leading to longer OS, further substantiating that i is a robust prognostic indicator in HNSCC.
We performed an exploratory analysis to investigate if i and K trans could predict OS in HPV/p16 patients because HPV/p16 is a key etiological factor in HNSCC, with distinct epidemiologic, clinical, and molecular characteristics. 21,22 Patients harboring the p16 gene are generally more sensitive to CRT and are associated with improved prognosis and prolonged survival. 20 The favorable outcome in p16-positive patients may be attributed to fewer and distinct somatic genetic alterations and positive immunologic response. 21,22 In agreement with prior studies, 20,21 significantly longer OS was observed for p16-positive patients compared with p16-negative patients in our study. A nonsignificant, but positive trend toward longer OS for p16-positive patients harboring high i and K trans suggests the prognostic utility of i and K trans may also be valid in stratifying p16-positive patients. We recognize that these results may be subject to selection bias inherent to the availability of tissue specimens. However, a nonsignificant difference in OS between patients who had and those who did not have tissues for p16 expression indicated that there was a minimal influence of sample size in the determination of OS.
While predicting short-term (6 months) treatment response to CRT, no significant differences in pretreatment i were observed between complete and partial responders. 5 However, in the present study, patients who had higher pretreatment i were associated with prolonged OS. The difference might be related to different follow-up duration, assessment of clinical outcome (local regional recurrence/evidence of residual disease on pathology versus OS), and different number of patients (n ϭ 33 versus n ϭ 60) enrolled between these 2 studies. We focused on assessing nodal metastases in predicting OS because lymphadenopathy in the neck is a well-known prognostic determinant in patients with HNSCC. 29 A DCE-MRI analysis of primary tumors would also be important; however, artifacts induced by physiologic motion precluded the analysis of primary tumors. We combined DCE-MRI data from MR systems with 2 different field strengths (1.5T and 3T). To account for this variability, blood T1 and contrast agent relaxivities corresponding to the field strengths were used for converting signal intensity curves to concentration curves and estimating pharmacokinetic parameters. Previously, 5 we reported similar trends for K trans in predicting treatment response in patients scanned at 2 different field strengths, suggesting that DCE-MRI-derived parameters are independent of magnetic field. Our supposition is corroborated by a study 30 in which no significant differences in signal enhancement ratio, time to peak enhancement, uptake, and washout rates were observed for breast lesions scanned with 1.5T and 3T MR systems.

CONCLUSIONS
Our data indicate that shutter-speed model analysis of DCE-MRI provides 2 important imaging biomarkers ( i and K trans ) presenting complementary physiologic information about the tumor microenvironment. Although i may be a potential independent prognostic imaging biomarker for predicting OS, combined analysis of i and K trans greatly improves the predictive power of these parameters in determining OS in HNSCC. Future prospective trials involving a larger patient population are required to confirm our findings.