Prognostic role of diffusion weighted and dynamic contrast-enhanced MRI in loco-regionally advanced head and neck cancer treated with concomitant chemoradiotherapy

The study was approved by the National Medical Ethics Committee of the Republic of Slovenia (No. 22k/03/13) and written informed consent was obtained from all patients. Twenty patients with locally and/or regionally advanced (stages III-IVB) and histologically proven p16/HPV-negative SCC of the oro- or hypopharynx were enrolled in the study. p16/HPV status was assessed as described elsewhere.16 All patients were treated with cisplatin-based cCRT as recommended by the institutional multidisciplinary tumour board.

Patients were irradiated with a 6-MV linear accelerator photon beam using a concomitant boost intensity-modulated radiation technique. The dose to the primary tumour and enlarged lymph nodes was 70 Gy (gross tumour volume with a margin to compensate microscopic tumour extensions), whereas elective doses of 63 Gy (intermediate-risk volume - around larger nodes and non-palpable but radiologically suspicious nodes) and 56 Gy (low-risk volume) were applied for neck regions of probable microscopic disease in daily fractions of 2.0, 1.8 and 1.6 Gy, respectively, over 7 weeks. During radiotherapy, cisplatin in a dose of 40 mg/m2/week was concurrently administered. No antiangiogenic therapy was added to the protocol.

During therapy, toxicity was monitored on a weekly basis. After treatment completion, patients were examined for toxicity and tumour response every 2 to 3 months in the first and second year and every six months thereafter.

All patients underwent three MR examinations with DWI and DCE-MRI: (1) 0–7 days before the treatment; (2) after 1 week (i.e., after the fifth fraction of radiotherapy, i.e. 10 Gy) and (3) 2.5–3 months after the completion of cCRT to assess radiological response to treatment.

MR imaging was performed on 3T MAGNETOM Trio, A Tim System (Siemens Medical Systems®, Erlangen, Germany) with a neck array coil. The diagnostic imaging protocol included axial T2-weighted sequences with short tau inversion recovery (STIR) from the base of the scull to aortic arch (TR/TE 5010/71 ms, TI 170 ms, flip angle (FA) 70°, receiver bandwidth 287 Hz/pixel, matrix size 256 x 256, slice thickness 3 mm, gap 0.3 mm and field of view (FOV) 18 x 18 cm). Axial slices that covered the entire primary tumour were selected for DWI and DCE-MR imaging.

DWI images were acquired in axial plane using pulsed spin-echo echo-planar image sequence (TR/TE 3600/86 ms, receiver bandwidth 1302 Hz/pixel, matrix size 180 x 192, slice thickness 5 mm, gap 1.5 mm, FOV 230 cm2 and total acquisition time 5.04 min). Five different b-values (b = 0, 100, 200, 500 and 1000 s/mm2) were used with all diffusion-sensitizing gradients applied in three orthogonal directions to obtain trace-weighted images.

DCE-MRI was performed using 3D fast low angle shot (FLASH) sequence optimised for spatial and temporal resolution (TR/TE 5/1.16 ms, FA 15°, receiver bandwidth 490 Hz/pixel, matrix size 220 cm2, slice thickness 4 mm, temporal resolution 4 s, total acquisition time 5 min. T1 mapping was used to convert signal intensities into gadolinium concentration. The T1 map was calculated from pre-contrast multiple flip angle images (6°, 10° in 15°). A ĸ-space weighted image contrast algorithm was used to generate images with full spatial resolution of 128 x 128. Baseline images were acquired for 28 s. While the imaging continued, the gadobutrol (Gadovist®, Bayer HealthCare Pharmaceuticals) was administered intravenously in a dose of 0.1 mmol/kg body weight with a flow-rate of 3.5 ml/s followed by a 20 ml of saline flush with power injector.

Finally, post-contrast axial T1 weighted volumetric interpolated breath-hold examination (VIBE) sequences were acquired from skull base to aortic arch (TR/TE 3,26/1,26 ms, voxel size 1,1 x 0,9 x 1,5 mm with 4 mm averages, receiver bandwidth 640 Hz/pixel, matrix size 218 x 288 and FOV 250 cm2).

Post-processing of all the images was performed at a workstation running commercially available software Olea Sphere^® 3.0 MR Head & Neck expanded applications (Olea Medical^®, La Ciotat, France). Before data analysis, motion correction algorithm was applied that allowed for pairwise in-plane (acquisition plane) rigid co-registration of all raw perfusion images of a given slice location with well-chosen reference image over time.

Quantitative DCE-MRI parameters were volume transfer constant (K_trans), extracellular extravascular volume fraction (v_e) and plasma volume fraction (V_p). The pharmacokinetic modelling was done on a pixel-by-pixel basis using the extended Tofts model - a two compartment model, which is suitable for any freely diffusible tracer. The equation modelling the contrast agent’s concentration was the following17:

where C (t) is the tissue contrast agent concentration time course, V_p is plasma volume fraction, c_p is plasma contrast agent concentration, K_trans is the volume transfer constant and K_ep is the transfer function from extracellular extravascular space to the plasma space. The Arterial Input Function (AIF) for the pharmacokinetic analysis was derived from automatic selection of perfusion weighted image pixels selected in suitable arteries using a dedicated algorithm. The following multi-parametric maps with K_trans, V_e and V_p were then obtained automatically.

Apparent diffusion coefficient (ADC) maps were computed by a single exponential fit using the DW signal intensity-b value curves.

ADC and DCE-MRI derived parameters as well as tumour volumes were measured at baseline and after 10 Gy. A radiologist (three years of experience in head and neck imaging) placed three regions of interest (ROIs) and analysed the multi-parametric ADC and DCE maps independently, with the T1 post-contrast images serving as the reference images. The ROIs were drawn manually on all imaging sections encircling solid-appearing portions of primary tumours and metastatic lymph node or nodal masses. Special attention was paid not to include necrotic and cystic areas (hyper-intense areas on STIR images), large feeding vessels and surrounding normal tissue.

Tumour volumes were calculated by manually encircling tumour borders on each T1 post-contrast sequence on all axial slices. Soft wear algorithm then calculated the volume by using the equation:

Mean value of the selected ROIs from primary tumour was calculated for each parameter at each time point. Relative changes of the ADC and DCE-MRI parametric values as well as tumour volumes were calculated by dividing the mathematical difference of the corresponding parametric and tumour volume values after 10Gy and at pre-treatment by the pre-treatment parameter value for individual patient using the formula:

where P represents any given parameter (ADC, K_trans, V_e, V_p or tumour volume), P_10Gy represents absolute value of the parameter after 10Gy, and P₀ absolute value of the pre-treatment parameter.

All data analysis and graphs were performed using statistical software SPSS 19.0 (IBM SPSS Statistics for Windows, Version 19.0. Armonk, NY: IBM Corp.). Shapiro-Wilk test and Levene test were used for normality testing and homogeneity-of-variance testing, respectively. The continuous variables are presented by median value and range. The relative changes in parameters between pretreatment values and at 10 Gy are expressed in percentage (%). OS (event: death from any cause) and DFS (event: persistent tumour after cCRT, loco/regional recurrence, systemic dissemination) was calculated from the first day of therapy using Kaplan-Meier curves. For assessment of prognostic value of MRI-derived parameters, patients were divided into two groups, according to treatment effect (DFS: without vs. with persistent tumour after cCRT or local/regional recurrence or systemic metastasis) or survival status at the end of observation period (OS: alive vs. death).

Non-parametric tests (Wilcoxon Signed Rank test, Mann-Whitney U test) were employed for comparative analysis of MRI-derived parameters under study. DFS and OS were plotted using the Kaplan-Meier method and significant predictors for survival outcomes were identified by univariate Cox regression analysis. All variables that reached the level of statistical significance in univariate analyses were entered into the multivariate Cox regression model, and stepwise forward selection was used to identify the independent predictors of DFS and OS. Receiver operating characteristic (ROC) curve analysis was applied to determine the discriminatory power of the ADC and DCE-MRI parameters as well as tumour volume and clinical parameters for predicting survival rates. Area under curve (AUC) was computed and the optimal cut-off values were calculated by maximizing sensitivity and specificity. Two-tailed p values less than 0.05 were considered statistically significant.

All of the 20 consecutive patients (19 men, 1 woman; median age 58 years, range 46–67 years) with loco-regionally advanced HNSCC (oropharynx 13, hypopharynx 7) were free of systemic disease at presentation. The baseline clinical characteristics of the patients are listed in Table 1.

The median follow-up times of the entire study cohort and of the surviving patients were 27.2 months (range, 2.6–51.4 months) and 30.2 months (range, 18.2–51.4 months), respectively. In all patients, radiological complete tumour response to cCRT was determined 2.5–3 months post-therapy. Disease reappearance was diagnosed in 5 patients: 3 (15%) patients had recurrence at primary site, and 2 patients (10%) developed distant metastases (lungs). At the time of analysis, five patients (25%) were dead. Locoregional disease was the cause of death in three patients (with survival rates of 5.2, 10.6, and 22.9 months post-therapy), distant metastasis in one patient (14.1 months) and acute lower respiratory tract infection in the remaining patient (2.6 months, considered as treatment failure). A mean DFS was 38.9 months (2.6–51.4 months, 95% Confidence Interval [CI] 30.7–47.1). A mean OS time was 40.6 months (2.6–51.4 months, 95% CI 32.4–48.8). The 1- and 2- year DFS survival rates were 88.8% and 88.2% and for OS they were 84.4% and 73.8%.

For all the included patients, the pre-treatment values of ADC and DCE-MRI derived parameters K_trans, V_e and V_p were 0.81 (0.62–1.07) x 10^-3 mm2/s, 0.48 (0.13–0.79) min^-1, 0.32 (0.15–0.64) and 0.17 (0.06–0.43), respectively. A statistically significant change from pre-treatment to 10 Gy was seen in all parameters: an increase of 22.5% (1.7 to 52.4%) in ADC (p < 0.001) and of 70% (73.2 to 357.14%) in V_p (p = 0.015) and a reduction of 50.3% (-93.4 to 179.2%) in K_trans (p = 0.003) and of 35.2% (-88.9 to 126.8%) in V_e (p = 0.024). Pre-treatment tumour volume was 10.5 ml (2.6–54.9 ml). However, the change after 10 Gy from pre-treatment was not statistically different (-15.0%; -76.3 to 254.6%; p = 0.823) (Table 2).

The values of ADC and DCE-MRI derived parameters before treatment and the relative change of parameters after 10 Gy from baseline were compared between patients alive at the end of follow-up and those who died (Table 2). Pre-treatment ADC (0.76 x 10^-3 mm2/s; 0.62–0.91 x 10^-3 mm2/s) was significantly lower (p = 0.042) in patients which were still alive at the end of follow-up versus the patients who died (0.96 x 10^-3 mm2/s; 0.76–1.07 x 10^-3 mm2/s). Alive patients also showed a significantly higher (p = 0.002) pre-treatment DCE-MRI

derived parameter K_trans (0.57 min^-1; 0.23–0.79 min^{- 1}) compared with patients that died during follow-up (0.22 min^-1; 0.13–0.35 min^-1). In addition, alive patients demonstrated a reduction in K_rans after 10 Gy from baseline (-60.0%; -93.0 to -43.2%) whereas the patients who died showed no change or an increase in K_rans (4.2%; -37.1 to 179.2%) (p < 0.001). There was no difference in the pre-treatment values of other perfusion parameters and tumour volumes (p > 0.05) nor in the relative changes from 10 Gy to baseline of ADC, V _e, V_p and tumour volume (p > 0.05). Illustrative examples of DW and DCE MRI images of alive and deceased patient at the end of follow up are shown in Figure 1 and 2, respectively.

Figure 1

Figure 2

On contrary, no significant difference was observed when values of all the functional parameters and tumour volumes where compared between patients that showed disease progression during follow-up and patients without disease progression (p > 0.05). However, in patients with noted disease progression there was a trend of larger pretreatment tumour volume (14.0 ml; 8.9–55.0 ml) compared to the group with disease controlled (6.6 ml; 2.6–50.8 ml) (p = 0.054).

When assessing potential prognostic factors for DFS, none of them reached the level of statistical significance on univariate analysis (Table 3). Pretreatment parameters that significantly influenced OS on univariate analysis were ADC (p = 0.012), K_trans (p = 0.026), and tumour volume (p = 0.048) (higher K_trans and lower ADC and tumour volume were associated with longer survival). In addition, higher reduction in K_trans (p = 0.014) from baseline to 10 Gy was also identified as a significant prognosticator for OS but not also any of the clinical parameters (age, tumour location, TNM stage). Multivariate Cox regression analysis identified only higher pre-treatment K_trans (p = 0.026; 95% CI: 0.000–0.132) as an independent predictor of OS.

When ROC curve analysis was performed, pretreatment K_trans. provided an excellent diagnostic accuracy with an area under curve (AUC) of 0.95 (p = 0.003; 95% CI: 0.85–1.00) and a sensitivity and specificity of 93.3% and 80.0%, respectively (Figure 1). The cut-off value for longer survival was set at ≥ 0.29 min^-1. In addition, a reduction in v_e after 10 Gy (survivors: median -40.0%, range -88.9 % to 65.4%; deceased patients: median -14.7%, range -40.0 to 126.8%) yielded a good diagnostic accuracy with AUC 0.80 (p = 0.055; 95% CI: 0.569–1.000), sensitivity 73.3% and specificity 80.0%. Pre-treatment v_e showed only poor diagnostic usefulness accuracy (AUC 0.63), whereas other parameters failed to show any diagnostic accuracy (AUC < 0.6). When patients were grouped according to the K_trans cutoff value of 0.29 min^-1, the log rank test confirmed a statistically significant difference in the length of OS between the two groups with mean survival of 48.9 months (95% CI: 44.2–53.6) in patients with pre-treatment K_trans ≥ 0.29 min^-1 and 12.8 months (95%CI: 4.1–21.4) in patients with pre-treatment K_trans < 0.29 min^-1 (p < 0.001) (Figure 3).

Figure 3

Figure 4