The prediction of the treatment response of cervical nodes using intravoxel incoherent motion diffusion-weighted imaging

European Journal of Radiology 92 (2017) 93–102

Contents lists available at ScienceDirect

European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad

Research papers

The prediction of the treatment response of cervical nodes using intravoxel incoherent motion diffusion-weighted imaging

MARK

⁎

Simona Marzia, , Francesca Piludub, Giuseppe Sanguinetic, Laura Maruccic, Alessia Farnetic, Irene Terrenatod, Raul Pellinie, Maria Benevolof, Renato Covellof, Antonello Vidirib a

Medical Physics Laboratory, Regina Elena National Cancer Institute, Via Elio Chianesi 53, 00144 Rome, Italy Radiology and Diagnostic Imaging Department, Regina Elena National Cancer Institute, Via Elio Chianesi 53, 00144 Rome, Italy c Department of Radiotherapy, Regina Elena National Cancer Institute, Via Elio Chianesi 53, 00144 Rome, Italy d Biostatistics-Scientific Direction, Regina Elena National Cancer Institute, Via Elio Chianesi 53, 00144 Rome, Italy e Department of Otolaryngology & Head and Neck Surgery, Regina Elena National Cancer Institute, Via Elio Chianesi 53, 00144 Rome, Italy f Department of Pathology, Regina Elena National Cancer Institute, Via Elio Chianesi 53, 00144 Rome, Italy b

A R T I C L E I N F O

A B S T R A C T

Keywords: Diffusion-weighted imaging Head and neck cancer Intravoxel incoherent motion imaging Chemo-radiotherapy

Purpose: To investigate the predictive role of Intravoxel Incoherent Motion Diffusion-Weighted Imaging (IVIMDWI) parameters on cervical nodal response to chemo-radiotherapy (CRT) of head and neck squamous cell carcinoma (HNSCC). Materials and methods: Patients with pathologically confirmed HNSCC were included in the present prospective study, having at least one positive cervical lymph node (LN). They received concomitant CRT and underwent three serial IVIM-DWI investigations: before, at mid-treatment and after treatment completion. Tissue diffusion coefficient D, perfusion-related diffusion coefficient D* and perfusion fraction f were calculated by a biexponential fit. The two-sided Mann–Whitney rank test was used to compare the imaging parameters of patients with regional failure (RF) and regional control (RC). A p value lower than 0.05 was considered to be statistically significant. Results: Thirty-four patients were accrued. Twenty-four out of 34 LN (70.6%) showed persistent RC after a median follow-up time of 27.6 months (range: 12.0–50.2 months), while ten cases of RF (29.4%) were confirmed with a median time of 6.8 months (range: 1.5–19.5 months). Patients with RC showed significantly lower pretreatment D values compared to the RF patients (p = 0.038). At mid-treatment, the patients with RF showed significantly higher D values (p = 0.025), and exhibited larger percent reductions in f and the product D* × f from the baseline (p = 0.008 and < 0.001, respectively). No additional information was provided by the examination at the end of treatment. Conclusion: Pre-treatment and mid-treatment IVIM-DWI showed potential for prediction of treatment response of cervical LN in HNSCC patients.

1. Introduction Head and Neck Squamous Cell Carcinoma (HNSCC) is the fifth most prevalent type of cancer worldwide. It accounts for approximately 3–4% of all malignancies [1]. Primary chemo-radiotherapy (CRT) is the treatment of choice for locally advanced head and neck squamous cell carcinomas (HNSCC) to attempt organ and function preservation [2]. However, some patients will not benefit from CRT, and must rely on the surgical clearance of the residual/recurrent cancer. The identification of patients who were responsive to treatment, based on pre-treatment or intra-treatment information, may be advantageous to post-treatment

⁎

Corresponding author at: Viale Inigo Campioni 63, 00144 Rome, Italy. E-mail address: [email protected] (S. Marzi).

http://dx.doi.org/10.1016/j.ejrad.2017.05.002 Received 2 January 2017; Received in revised form 4 April 2017; Accepted 1 May 2017 0720-048X/ © 2017 Elsevier B.V. All rights reserved.

evaluation as unnecessary treatment is avoided for potentially nonresponsive patients. It is recognized that morphologic imaging alone is limited, as the treatment response is not only based on the changes in tumor size but also on the initial biological characteristics of the tumor, such as its oxygenation status, proliferative capacity, and perfusion characteristics [3–5]. Thus, the use of functional and metabolic imaging modalities [3] for the identification of novel biomarkers may help to elucidate the pathogenesis of CRT-resistant tumors, and aid clinicians in optimizing and tailoring individual treatment options for patients. The diffusion-weighted imaging (DWI) is a promising, functional

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

Considering the clinical relevance of the human papilloma virus (HPV) to the radiation sensitivity of oropharyngeal cancers [18], we determined the HPV status on this subset of patients.

imaging modality for patients with HNSCC. It provides an early assessment of the tumor response to CRT [4], and allows for the quick assessment of the lesion without injecting contrast medium. According to several studies, DWI provides quantitative information on the apparent diffusion of water molecules in biological tissue, based on the measurement of the signal attenuation coefficient, or the apparent diffusion coefficient (ADC) [6]. ADC quantification indirectly evaluates tumor cellularity; whereby, a higher resistance towards water diffusivity is typically observed in cancers as a consequence of the increased cell density of the tumoral tissue [6]. The ADC parameter can be analytically derived following postprocessing of the diffusion-weighted images. Nonetheless, experimental and clinical studies have suggested that tissue perfusion may influence the value of ADC; this observation further indicates that a simple monoexponential fit may be inadequate for a correct description of the signal attenuation curve in response to increasing b values [7]. A specialized imaging technique, known as intravoxel incoherent motion (IVIM) imaging, was proposed to account for the effects of microcapillary perfusion on DWI measurements [8]. It acquires multiple diffusion b values, and provides the quantitative parameters that separately reflect tissue diffusivity and tissue microcapillary perfusion. The use of IVIMDWI allows for the derivation of three additional parameters to ADC: f, the perfusion fraction; D*, the perfusion-related diffusion coefficient; and D, the diffusion coefficient related to tissue molecular diffusion. The potential utility of IVIM imaging in patients with HNSCC is due to the influence of the tumor vasculature and oxygenation status on radioresistance. In several studies using perfusion computed tomography or dynamic-contrast enhanced magnetic resonance imaging (DCEMRI), the decreased perfusion levels were associated with a higher failure rate, which was attributed to a reduced sensitivity to radiationinduced free radical damage [9–11]. Although there has been growing interest in the use of IVIM-DWI in HNSCC [12,13], only a few studies investigated the potential utility of IVIM-DWI to predict the treatment response to CRT [14,15]. The primary aim of this study is to evaluate the predictive role of IVIM-DWI parameters, evaluated prior to radiation therapy (RT), at mid-RT, and at the end of RT, in the assessment of the cervical lymph nodes’ long-term response to CRT.

2.2. Follow-up criteria Follow-up consisted of clinical assessments and imaging examinations. The first follow-up MRI scan was performed 8 weeks after the end of RT; afterwards, MRI was performed every 6 months for the first two years, and then once a year. FDG-PET-CT was performed 12 weeks after the end of RT, and then once a year. At the time of the analysis, patients were categorized into two groups: those with regional control (RC), and those with regional failure (RF). RF was defined as the presence of residual disease following CRT or the recurrence of nodal disease (after initial clearance) during the follow-up. All nodal failures were pathologically confirmed. 2.3. MR imaging protocol MRI was performed on a 1.5-T system (Optima MR 450w, GE Health-care, Milwaukee, WI, USA) with dedicated 16-channel receiveonly radiofrequency coils: a head coil, a surface neck coil, and a spine coil. The patients underwent three serial MRI examinations: before RT, half-way through the course of RT (at the 16th or 17th fraction of RT), and at the end of RT (on the same day as the last dose fraction). The MRI examinations included fast spin-eco (FSE), T2-weighted images, and DWI. FSE T2-weighted images on the coronal plane were first obtained, followed by axial FSE T2-weighted images (acquisition matrix 256 × 256, field of view 26–28 cm, TR/TE = 2260 ms/119; slice thickness 4 mm, spacing between slices 5 mm), acquired from the level of the skull base to the thoracic inlet. DWI were obtained via single-shot spin-echo, and echo-planar imaging (acquisition matrix, 128 × 128; field of view, 26–28 cm; TR/TE 4500 ms/77 ms; slice thickness 4 mm; spacing between slices 5 mm, bandwidth 1953 Hz/pixel). The diffusionsensitizing gradient duration and diffusion time were 19 and 29 ms, respectively. Nine different b values (b = 0, 25, 50, 75, 100, 150, 300, 500 and 800 s/mm2) were used, with the diffusion-sensitizing gradients applied in three orthogonal directions to obtain trace-weighted images. The reduced signal-to-noise ratio (SNR) with the largest b values were accounted for by choosing three signal averages for b values ranging from 0 to 300 s/mm2, four for b values of 500 s/mm2, and five for b values of 800 s/mm2. A scan time reduction factor of two was used, with a resulting scan duration of 6 min and 13 s. The imaging protocol used prior to RT included additional sequences for more complete tumor characterization such as pre-contrast T1-weighted images, and the multi-phase post-contrast T1-weighted series.

2. Methods and materials 2.1. Patient population & treatment A single-institution prospective trial, aimed at evaluating the role of IVIM-DWI in predicting treatment response of HNSSC to CRT, was approved by our institutional ethics committee (RS 266/12). Eligible patients fulfilled all the following inclusion criteria: a) histologicallyconfirmed diagnosis of HNSCC; b) absence of distant metastases (M0); c) treatment with definitive concomitant chemotherapy and intensitymodulated radiation therapy (IMRT); d) age > 18 years old; e) informed, written consent. Exclusion criteria included general contraindications for MRI. In the present investigation, we focused our analyses on the incidence of metastatic lymph node (LN). Patients who were included in the present study were required to have at least one cervical lymph node metastasis based on volumetric and/or morphologic criteria (i.e. alterations of the internal architecture, presence of extracapsular spread) as defined by Rao et al. [16]. Patients were staged according to the tumor, node, metastasis (TNM) staging system of the American Academy of Otolaryngology, Head and Neck Surgery [17]. All patients received IMRT and concomitant chemotherapy (cisplatin 100 mg/m2 for three cycles every 21 days). A seven-field, simultaneous, integrated boost technique was used to deliver 70 Gy, in 33 fractions, to the site of the macroscopic disease (primary tumor and affected LN), 60 Gy to the regions at high risk of developing microscopic disease, and 54 Gy to the regions at low risk of developing microscopic disease.

2.4. Tumor delineation Three contiguous sections of LN, each covering the largest crosssectional area of the lesion, were identified on DWI with b = 800 s/ mm2, by two expert HN radiologists (A.V. and F.P.) in consensus, with more than 15 and 6 years of experience, respectively. Morphological T2-weighted images and/or post-contrast T1weighted images were used as a guide for tumor delineation. Arterial or venous structures, and bony components were excluded from the volume of interest (VOI). In patients with several LN metastases, the largest LN was initially selected from the DWI sequence. To improve consistency in the identification of the chosen lymph node across the three serial MRI scans, contouring was performed at the end of treatment when all the scans were available for each patient. For cases involving regional recurrence in an LN different from that of the largest one, the exact site of the nodal recurrence was identified by the HN radiologists on each MRI scan, and used for the analyses. 94

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

Fig. 1. Metastatic lymph node in a 77-year-old man affected by a nasopharynx carcinoma: the three largest sections of the lymph node on the pre-treatment diffusion-weighted images obtained with b of 800 s/mm2, with the user-defined lesion contours in dashed line (a). The axial T2-weighted images used as a guide for lesion delineation (b).

between the measured data and the theoretical values predicted from the model. In order to yield clinically meaningful results, the optimal solutions for S0, D* and f were constrained between 0 and 2000, 0 and 50 × 10−3 mm2/s, and 0 and 50%, respectively. To avoid the risk of a single process being trapped in suboptimal solutions, 100 simultaneous optimization processes were launched for each fit, starting from randomly chosen initial values of S0, D* and f between the bounds defined above. The RSS value of each optimization process was recorded and the minimum RSS value was used to assess the best solution for S0, D* and f. The product of D* by f was consequently derived, considering that a linear relationship has been supposed between this product and blood flow in the brain [22]. Finally, the conventional ADC was calculated using data at b-values of 0, 300, 500, and 800 s/mm2. The Levenberg-Marquardt algorithm was used to perform the mono-exponential fits of both D and ADC. Dedicated scripts, developed in Matlab code (Release 7.10.0, The Mathworks Inc., Natick, MA), were used to perform the calculations. An example of the signal decay derived from the median value inside the user-defined sections of a lymph node is illustrated in Fig. 2, with a schematic representation which describes the two steps analysis used to determine D first and secondly S0, D* and f.

The entire volume of each LN, as manually drawn on the appropriate slice, was also quantified on morphological images. The 3D Slicer Software (Version 4.1.1) [19] was used for visualizing the image sets, and for the lesion segmentation. An example of lymph node delineation is illustrated in Fig. 1 for a 77-year-old man affected by a nasopharyngeal carcinoma. 2.5. Calculation of IVIM-diffusion parameters and ADC To reduce the instability of the SNR, the signal from the voxels within the VOI was integrated to obtain a total signal intensity distribution, and the median signal intensity for each b value was derived. The signal variation with increasing b values was modeled using the following bi-exponential function [8]:

Sb = S0 ·[(1−f)·e−b·D + f·e−b·(D+D*) ]

(1)

where Sb is the signal intensity with diffusion weighting b, S0 is the signal intensity for a b value of 0 s/mm2, f is the fractional volume of capillary blood, D is the diffusion coefficient (in mm2/s), and D*is the perfusion-related diffusion coefficient (in mm2/s).In order to avoid over fitting of the bi-exponential model, the four parameters, S0, D*, D and f, were not simultaneously estimated and an analysis method in two steps was implemented [20]. At first, D was calculated using only data at 300, 500, and 800 s/mm2 from the following equation: Sb = S0 · e−b·D, assuming that the contribution of D* to the signal decay can be neglected for b values greater than 150–200 s/mm2 [8]. In a second step, fixing D at the value estimated above, the parameters, S0, D* and f, were determined from Eq. (1) using a nonlinear, constrained minimization algorithm [21], minimizing the residual sum of squares (RSS)

2.6. Extracapsular spread The presence of extracapsular spread (poorly defined margins, presence of capsular contour irregularity and infiltration of adjacent fat or muscle planes on T2-weighetd images [23]) was investigated for each lymph node at baseline. The information was treated as binary 95

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

Table 2 Summary statistics of ADC and IVIM-DWI parameters for patients with regional control (RC) and regional failure (RF), at different time points.

D

Status

Pre-RT

Mid-RT

End-RT

Change at Mid-RT

Change(%) at Mid-RT

RC

0.83 (0.32) 1.05 (0.63) 0.038

1.09 (0.45) 1.28 (0.86) 0.025

1.40 (0.33) 1.35 (0.31) 0.526

0.24(0.45)

23.5 (68.9)

0.45 (0.77)

41.4 (82.8)

12.6 (41.1) 14.2 (25.0) 0.774

18.0 (28.1) 10.2 (9.7) 0.127

10.0 (16.4) 19.8 (35.2) 0.270

8.6 (8.0) 12.9 (14.3) 0.205

12.5 (7.4) 8.6 (10.4) 0.252

14.1 (15.4) 12.9 (11.0) 0.404

111 (155) 156 (222) 0.381

177 (248) 74 (105) 0.037

118 (207) 118 (451) 0.815

0.99 (0.38) 1.29 (0.75) 0.041

1.41 (0.55) 1.50 (0.87) 0.286

1.57 (0.50) 1.53 (0.65) 0.713

RF p-value D*

RC RF p-value

f

Fig. 2. Example of diffusion-weighted signal attenuation curve inside a lymph node and schematic representation of the two steps analysis used to determine D first and secondly D* and f (fitting curve in dashed line).

RC RF p-value

Table 1 Patient characteristics.

D* × f

RC RF

Characteristics Age (years) Median (range) Gender (M/F) Primary tumour site, n(%) Oropharynx Nasopharynx Hypopharynx and Larynx No evidence

RC

RF

Total sample

p-value p-value

0.170* 55 (28–82)

59.5 (49–77)

ADC

54.5 (28–79)

RF

0.169** 20/4

10/0

30/4

p-value 0.370**

9 (37.5) 11 (45.8) 3 (12.5)

5 (50.0) 2 (20.0) 3 (30.0)

14 (41.2) 13 (38.2) 6 (17.7)

1 (4.2)

0

1 (2.9)

Extracapsular spread, n(%) Yes No

18 (75.0) 6 (25.0)

6 (60.0) 4 (40.0)

24 (70.6) 10 (29.4)

T stage, n(%) T0 and T1 T2 T3 T4

6 (25.0) 11 (45.8) 3 (12.5) 4 (16.7)

2 2 3 3

(20.0) (20.0) (30.0) (30.0)

8 (23.5) 13 (38.2) 6 (17.7) 7 (20.6)

N stage, n(%) N1 N2 N3

6 (25.0) 15 (62.5) 3 (12.5)

0 8 (80.0) 2 (20.0)

6 (17.7) 23 (67.6) 5 (14.7)

RC

0.387

0.760

0.0(22.2)

−0.1(253)

−3.1(26.6)

−29.7 (114.7)

0.263

0.134

3.5 (9.5)

53.8 (100.0)

−6.2 (12)

−35.7 (42.8)

0.015

0.008

97.0 (300)

82.5 (307)

−114.2(227.7)

−69 (127)

0.003

< 0.001

0.38 (0.41)

32.7 (53.8)

0.15(0.60)

9.3 (79.3)

0.286

0.186

Abbreviations: perfusion-related diffusion coefficient, D*(10−3 mm2/s); tissue diffusion coefficient, D(10−3 mm2/s); perfusion fraction, f(%), product of D* by f, D* × f (10−3 mm2/s·%); apparent diffusion coefficient, ADC(10−3 mm2/s). Data are expressed as median values and interquartile range. P values refer to the Mann-Whitney test.

0.430***

to search the optimal threshold for the most significant parameters, based on the Youden’s index (which maximizes both sensitivity and specificity). Sensitivity, specificity, positive predicted value (PPV) and negative predicted value (NPV) were thus calculated. A p value lower than 0.05 was considered to be statistically significant. Statistical analyses were carried out using SPSS software (SPSS version 21, SPSS Inc., Chicago, IL, USA).

0.370**

0.213**

3. Results 3.1. Patient population and follow-up data

* Mann-Whitney non parametric test. ** Chi-Square non parametric test. *** Fisher Exact test. Abbreviations: Regional control, RC; regional failure, RF.

The sample size used in the present study was determined based on the literature data available at the beginning of the study, in particular it was based on the difference in the ADC values between complete responders and partial responders reported by Kim et al. [25]. Considering a drop-out rate equal to 10%, a total of 40 patients was found to be appropriate to achieve a statistical power of 80%, with a significance level of 0.05. After enrolment, 6 patients were excluded from the study: 2 patients withdrew their consent and 4 decided to be treated elsewhere for logistical reasons. Therefore, a total of 34 patients were included in the study from January 2010 to December 2013. Patient and tumor characteristics are shown in Table 1. Patients with RC and RF did not show statistically significant differences in terms of baseline characteristics (Table 1). All patients suffered from squamous cell carcinomas of the upper aerodigestive tract of variable grades of differentiation with the exception of three patients affected by an undifferentiated carcinoma of the nasopharynx. Four patients refused the mid treatment MRI examination. Moreover, there were missing mid-treatment data, due to susceptibility artifacts, for two patients. A complete disappearance of nodal diseases following CRT was evident in eleven patients, and consequently, image

variable (0, no or 1, yes). 2.7. Statistical analyses Statistically significant differences in variables between patients with RC and RF were calculated using the non-parametric Chi-square, or Fisher exact test, when appropriate, for categorical variables while the non-parametric Mann–Whitney test was applied for continuous variables. The Wilcoxon test for paired samples was used to assess whether changes in the same variable between two different time points were significant. Box-and-whisker plots were used to display statistical summaries of the parameters. On each box, the central mark is the median. The edges of the box are the 25th and 75th percentiles, and the horizontal lines extend from the minimum to the maximum value, outliers being plotted with square markers. Receiver operative characteristics (ROC) curves were used in order 96

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

Fig. 3. Box-and-whisker plots of diffusion coefficient D (a), perfusion fraction f (b), the product of perfusion-related diffusion coefficient D* by f (c) and apparent diffusion coefficient ADC (d), for patients with regional control (RC) and regional failure (RF), at baseline, at mid-RT and at end-RT. Table 3 Area under the receiver operating characteristic curve (AUC) and 95% CIs for the most significant imaging parameters as predictors for regional failure. The optimal cut-off and the corresponding Sensitivity, Specificity, Positive Predictive Value (PPV) and Negative Predictive Value (NPV) are reported. Parameter Before RT

AUC

95% CI

Cut-off

Sens(%)

Spec(%)

PPV(%)

NPV(%)

D ADC

73.3 72.9

55.4–86.9 55.0–86.6

> 0.97 > 1.20

70.0 70.0

75.0 79.2

53.8 58.3

85.7 86.4

Parameter at Mid-RT

AUC

95% CI

Cut-off

Sens(%)

Spec(%)

PPV(%)

NPV(%)

D D* D* × f

77.5 68.8 75.6

57.9–91.0 48.6–84.8 55.8–89.7

> 1.12 ≤15.6 ≤92.9

100 87.5 75.0

60 50 80.0

50.0 42.1 61.0

100.0 90.6 88.5

Change at Mid-RT(%)

AUC

95% CI

Cut-off

Sens(%)

Spec(%)

PPV(%)

NPV(%)

Δf Δ(D* × f) ΔADC

82.9 88.1 66.2

63.5–94.5 70.3–97.1 45.0–82

≤−19.1% ≤−48.3% ≤15.5%

87.5 87.5 75.0

89.5 90.0 70.0

77.6 78.5 50.0

94.5 94.5 87.5

Abbreviations as in Table 2.

Primary tumor recurrences were not reported in patients with RC, while three of the ten patients with RF showed evidence of local disease. In three patients, regional recurrence did not occur in the largest LN (via cross-sectional measurement); these LN were subsequently used for the analyses. In the RC group, there were 24, 20 and 13 patients at pre-RT, midRT, and end-RT, respectively. There were 10, 8 and 9 patients for the RF group at pre-RT, mid-RT, and end-RT, respectively.

analysis was not possible. The median time interval between initial staging and start of CRT was 7 days (CI95%:1.6–19.1 days). All selected lymph nodes were larger than 1 cm on axial images and all received a mean dose of 70 Gy. Twenty-four out of 34 LN (70.6%) showed durable RC after a median follow-up time of 27.6 months (range: 12.0-50.2 months), while ten cases of RF (29.4%) were confirmed during a complementary neck dissection following CRT (N = 6), or during the follow-up (N = 4), with a median time of 6.8 months (range: 1.5–19.5 months). 97

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

Fig. 4. A case showing regional failure in a 61-year-old male patient affected by a hypopharynx carcinoma with bilateral neck nodes. Pretreatment scans (a), mid-treatment scans (b), endtreatment scans (c), and scans acquired 8 weeks after the end of RT (d) are shown. T2-weighted images and diffusion-weighted images that were obtained with a b value of 800 s/mm2, indicate a large, initial necrotic lymph node in the right cervical region (indicated by the arrow), and evidence of residual disease at the end of the treatment, and during the first followup. The lymph node had a high initial D value (1.39 ·10−3 mm2/s), with moderate variation evident at mid-RT (1.27 ·10−3 mm2/s), as depicted by the comparison between the signal attenuation curves from the baseline to mid-RT (Fig. 5a). The f changed from 11.7% to 4.3%, with a percent variation of −63% from the baseline. The D* × f also decreased from 50.5 (10−3 mm2/s· %) to 15.2 (% · 10−3 mm2/s).

0.016) compared to initial values. None of the changes of any parameter from mid-RT to end-RT was significant for either LN group. The predictive power and optimal threshold for the most significant parameters, such as those predicting for RF, are reported in Table 3. At baseline, ADC and D values had the greatest predictive power, with an area under the ROC curve (AUC) of 72.9% and 73.3%, respectively; the optimal cut-off value were > 1.20 × 10−3 mm2/s for ADC and > 0.97 × 10−3 mm2/s for D. At mid-RT, D and D* × f had the highest predictive power, with an AUC of 77.5% and 75.6%, respectively; the optimal cut-off values were > 1.12 × 10−3 mm2/s for D and ≤ 92.9 × 10−3 mm2/s·% for D* × f. At baseline, the NPV were 86.4% and 85.7 for ADC and D, respectively, and 100% and 88.5% for D and D* × f, respectively, at mid-treatment. The corresponding PPV for ADC, D at baseline, and for D and D* × f at mid-treatment, were 58.3%, 53.8%, 50% and 61%, respectively. During treatment the percent changes of f and D* × f were the most predictive variables, with an AUC of 82.9% and 88.1%, respectively. The optimal cut-off values were ≤−19.1% for f and ≤−48.3% for D* × f (Table 3). Two representative cases of patients with RF and RC are illustrated in Figs. 4 and 5, respectively, with the corresponding plots of the signal

3.2. Analysis of IVIM-DWI parameters and ADC Summary statistics of the diffusion parameters and their variations during treatment by nodal status, are shown in Table 2 and illustrated in Fig. 3. On average, patients with RC showed significantly lower pretreatment D values (0.83 × 10−3 mm2/s versus 1.05 × 10−3 mm2/s, p = 0.038) compared to RF patients, as well as lower ADC values (0.99 × 10−3 mm2/s versus 1.29 × 10−3 mm2/s, p = 0.041). At midRT, the patients that developed RF showed significantly higher D values (1.28 × 10−3 mm2/s versus 1.09 × 10−3 mm2/s, p = 0.025), and exhibited significant reductions in f and D* × f from the baseline, both in absolute units and percentages (Table 2). At the end of RT, no significant differences were found for both ADC and IVIM-DWI parameters by disease status. In patients with RC, D significantly increased from baseline to mid treatment (p = 0.027), while the same parameter showed only a trend in the RF group (p = 0.078). Similarly, compared to baseline, f and D* × f values were significantly larger at mid-RT (p = 0.027 and 0.05, respectively) for RC patients while the opposite was observed for RF patients: both f and D* × f were lower at mid-RT (p = 0.078 and 98

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

Fig. 5. A case showing regional control in a 70-year-old man affected by nasopharynx carcinoma. Pretreatment scans (a), mid-treatment scans (b), end-treatment scans (c), and long-term follow-up scans acquired 35 months after the end of RT (d) are shown. T2-weighted images and diffusion-weighted images that were obtained with a b value of 800 s/mm2, depict a solidshaped node in the left region (indicated by the arrow), which rapidly decreased in size during treatment; evidence of residual disease was not observed on the long-term follow-up scans. The lymph node had very low initial D values (0.66 ·10−3 mm2/s), which rapidly increased during treatment (1.07 ·10−3 mm2/s), as depicted by the comparison between the signal attenuation curves from the baseline to mid-RT (Fig. 5b). The f changed from 6.8% to 9.8%, with a percent variation of +44.1% from the baseline. The D* × f also increased from 50.1 (10−3 mm2/s· %) to 87.2 (10−3 mm2/s· %).

4. Discussion

decay, as a function of the b value, illustrated in Fig. 6.

In the present study, we investigated whether conventional ADC and IVIM parameters predicted the tumor control of cervical lymph nodes in HNSCC patients undergoing CRT. Preliminary results from this investigation have been presented to the 2016 Annual Meeting of American Society for Radiation Oncology [25]. As only three patients developed a primary tumor recurrence, we focused our analyses on metastatic lymph nodes. Moreover, primary tumors are generally located in areas prone to motion and susceptibility artifacts, which make the derivation of IVIM parameters and the quantitative analysis of DWI significantly more challenging [24]. We found lower pretreatment D and ADC values in patients with RC compared to patients with RF. Consistently to published literature, mostly limited to ADC evaluation only [24,26–30], we found lower pretreatment ADC values in patients with RC than those with RF. In the latter group, a high ADC value particularly in LN may reflect the presence of necrotic areas with lower tumor oxygenation, which would negatively impact the tumor response probability. The changes in ADC from baseline to mid-RT did not show significant differences between patients with RC and RF, even though patients with RF tended to be associated with smaller ADC changes compared to patients with RC, according to previous studies

3.3. HPV status in oropharyngeal SCCs The HPV status was determined in 12 out of 14 patients with oropharyngeal SCC; as the clinical relevance of the HPV tumor status became apparent only during the course of the present investigation, it was not available in two patients. As shown in Table 4, HPV status was distributed similarly between patients with RC and RF.

3.4. Analysis of LN volume over time There were no significant differences observed between the patients with RC and RF for the LN volumes at baseline, and at mid-treatment. Additionally, the differences in the percent changes between RC and RF patients from baseline to mid-RT were not significant (Table 5). It should be noted that in the RC group some patients had persistently enlarged lymph nodes after CRT; in these cases the residual lymph node at the end of CRT showed a very low T2-weighted signal which is indicative of fibrosis. The absence of viable tumor cells in these lymph nodes was confirmed by the subsequent follow-up examinations, with a minimum follow-up time of 12.0 months. 99

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

Table 5 Summary statistics of lymph node volume size for patients with patient with regional control (RC) and regional failure (RF), at different time points.

Vol (cm3)

Status

Pre-RT

Mid-RT

End-RT

Change(%) at Mid-RT

RC RF p-value

6.7 (7.7) 11.2 (15.5) 0.227

3.7 (10.2) 7.3 (10.7) 0.281

1.6 (4.0) 2.2 (5.5) 0.728

−34.0 (49.7) −31.3 (39.2) 1.000

Data are expressed as median values and interquartile range. P values refer to the MannWhitney test.

[24,26,29]. We found that D progressively increased throughout the treatment, as the number of tumor cells decreased. At mid-treatment, D became significantly different between RC and RF patients, whereas a negligible difference was evident in ADC values between the two patient groups. Therefore, the perfusion-free diffusion coefficient, D, seems more sensitive to the variations in the cellular microstructure, and may better reflect the radiation effects associated with cell damage, compared to ADC values, which are susceptible to the effects of perfusion and diffusion [7,8]. While, the D value at baseline and the D value at mid-RT had high predictive values, with fair AUC values, and remarkably high sensitivity and NPV scores. Concerning perfusion-related parameters, the baseline values of f, D* and D* × f were lower in patients with RC, with a significant increase in f observed at mid-RT. Hauser et al. [14] also reported higher initial f values in patients with RF in a small (N = 15) cohort of patients, suggesting that a larger vascular compartment in LN may reflect a higher tumor aggressiveness, as documented by histopathological and sonographic studies [31,32]. A possible explanation for the increase in f observed at mid-RT may be the faster re-oxygenation rate of initially hypoxic cells in patients with RC that ultimately results in a substantial raise of vascularization. These observations are also consistent with the findings of a similar study of Ding et al. [15] in patients with HPV-associated or oropharyngeal carcinoma. However, as Ding et al. discontinued treatment response evaluation at a short-term follow-up (2–3 months), a direct comparison with our results is not straightforward. Therefore, to our knowledge, the predictive role of IVIM on response of HNSCC to CRT, both at baseline and during treatment, based on a long-term follow-up has never been investigated and reported so far. A progressive decrease in tumor blood flow during RT in patients with RF was reported in a study of Truong et al. [11] using perfusion CT. In the same patient subgroup, we found an initial decrease in perfusion-related parameters, followed by an increase during the second half of treatment. Therefore, vascular changes in patients with RF during a course of CRT may be less consistent and predictable than initially observed [11]. The observed fluctuations may reflect competing effects, a reduction in vascular density due to the progressive vascular damage and an increase in vessel permeability due to vascular edema, with an unpredictable final result on tumor vascularization. The D* × f value at mid-RT and the changes in f and D* × f from baseline to mid-RT had the highest predictive values, improving both specificity and PPV values, compared to D. These parameters are promising for the evaluation of nodal response following CRT and thus deserve further investigation. It should be noted that f, D* and D* × f showed a statistical dispersion larger than that of ADC and D (Table 2 and Fig. 1), confirming that the perfusion-related parameters are more sensitive and susceptible to poor SNR levels [7,33]. Therefore, in order to obtain reliable D* and f values, we decided to perform our IVIM analyses on VOI-based measurements, as DWI is prone to have low SNR within single voxels. Our findings indicate that the diffusion parameters are superior to volumetric data in predicting treatment response. Some investigators reported a correlation between baseline nodal volume and response

Fig. 6. Plots of the signal attenuation, as a function of b, at baseline and at mid-RT, in the representative patient with local regional failure illustrated in Fig. 4. A slight difference between the rates at which the signal decreases is evident, from baseline to mid-RT (a). Plots of the signal attenuation, as a function of b, in the representative patient with local regional control illustrated in Fig. 5, showing a marked change between the signal attenuation curves from baseline to mid-RT (b). The plot at mid-RT shows a large increase in tissue water diffusivity. This is indicative of a lower cell density owing to the effects of progressive radiation damage.

Table 4 HPV status and response to treatment for patient affected by oropharyngeal squamous cell carcinoma. HPV status

RC, n(%)

RF, n(%)

Total, n(%)

Negative Positive Total

4 (57.1) 3 (42.9) 7 (100)

4 (80) 1 (20) 5 (100)

8 (66.7) 4 (33.3) 12 (100)

Abbreviations: Regional control, RC; regional failure, RF.

100

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

rates [28,30], while others did not [24,26]. We found only a weak correlation between baseline nodal volume and response, reflecting the inconsistency of literature data. Concerning the HPV status in patients with oropharyngeal SCC, HPV-related tumors have a better prognosis and show higher response rates than HPV-negative ones [18,15,34], though this was not evident in our study presumably due to limited sample size. The role of MRI changes according to HPV status remains an open question and it deserves a study on a larger patient cohort. There were several limitations in the current study. Besides the small sample size, the tumor population was heterogeneous under several aspects including primary tumor sub-site. In a previous study we had found a correlation between the diffusion parameters and the anatomic site of the lesion [20]. Therefore, we are currently performing a prospective study on IVIM-DWI in oropharyngeal cancers only. This on-going study will enroll a larger patient population and includes also DCE-MRI at baseline to further validate IVIM parameters with conventional perfusion techniques. Finally, the present study cannot be rigorously and entirely considered a prospective one, as, in 3 patients, the individual lymph node was (re)selected based on treatment response and not on baseline volume size. While this may have introduced a bias, this gave us the opportunity to focus on the clinically meaningful part of the nodal disease.

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

5. Conclusions

[15]

IVIM-DWI at baseline and during treatment correlated significantly with the RC status of nodal metastases. The D values at baseline and at mid-RT had greater diagnostic accuracy than ADC values. The perfusion-related parameters f and D* × f showed potential to further improve the predictive power compared to D, suggesting that both vascularization and oxygenation correlate to HNSCC radioresistance.

[16]

[17]

[18]

Conflict of interest [19]

None of the authors have potential conflict of interest including any financial, personal or other relationships with other people or organizations that could inappropriately influence this work.

[20]

[21]

Funding

[22]

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

[23]

Acknowledgments

[24]

The authors are indebted to Michele Farella and to Pierfrancesco Rinaldi for their continued technical assistance.

[25]

[26]

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ejrad.2017.05.002.

[27]

References

[28]

[1] A. Jemal, R. Siegel, E. Ward, et al., Cancer statistics, 2008, CA. Cancer J. Clin. 58 (2) (2008) 71–96. [2] D.G. Pfister, S. Spencer, D.M. Brizel, et al., Head and neck cancers, Version 2.2014. Clinical practice guidelines in oncology, J. Natl. Compr. Canc. Netw. 12 (10) (2014) 1454–1487. [3] H. Quon, D.M. Brizel, Predictive and prognostic role of functional imaging of head and neck squamous cell carcinomas, Semin. Radiat. Oncol. 22 (2012) 220–232. [4] A.D. King, H.C. Thoeny, Functional MRI for the prediction of treatment response in head and neck squamous cell carcinoma: potential and limitations, Cancer Imaging 16 (1) (2016) 23 (016-0080-6). [5] C. Jentsch, B. Beuthien-Baumann, E.G. Troost, G. Shakirin, Validation of functional

[29]

[30]

[31]

101

imaging as a biomarker for radiation treatment response, Br. J. Radiol. 88 (1051) (2015) 20150014. A.R. Padhani, G. Liu, D.M. Koh, et al., Diffusion-weighted magnetic resonance imaging as a cancer biomarker: consensus and recommendations, Neoplasia 11 (2) (2009) 102–125. D.M. Koh, D.J. Collins, M.R. Orton, Intravoxel incoherent motion in body diffusionweighted MRI: reality and challenges, AJR Am. J. Roentgenol. 196 (6) (2011) 1351–1361. D. Le Bihan, E. Breton, D. Lallemand, M.L. Aubin, J. Vignaud, M. Laval-Jeantet, Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging, Radiology 168 (2) (1988) 497–505. R. Hermans, M. Meijerink, W. Van den Bogaert, A. Rijnders, C. Weltens, P. Lambin, Tumor perfusion rate determined noninvasively by dynamic computed tomography predicts outcome in head-and-neck cancer after radiotherapy, Int. J. Radiat. Oncol. Biol. Phys. 57 (5) (2003) 1351–1356. K. Newbold, I. Castellano, E. Charles-Edwards, et al., An exploratory study into the role of dynamic contrast-enhanced magnetic resonance imaging or perfusion computed tomography for detection of intratumoral hypoxia in head-and-neck cancer, Int. J. Radiat. Oncol. Biol. Phys. 74 (1) (2009) 29–37. M.T. Truong, N. Saito, A. Ozonoff, et al., Prediction of locoregional control in head and neck squamous cell carcinoma with serial CT perfusion during radiotherapy, AJNR Am. J. Neuroradiol. 32 (7) (2011) 1195–1201. M. Sumi, M. Van Cauteren, T. Sumi, M. Obara, Y. Ichikawa, T. Nakamura, Salivary gland tumors: use of intravoxel incoherent motion MR imaging for assessment of diffusion and perfusion for the differentiation of benign from malignant tumors, Radiology 263 (3) (2012) 770–777. J. Sakamoto, A. Imaizumi, Y. Sasaki, et al., Comparison of accuracy of intravoxel incoherent motion and apparent diffusion coefficient techniques for predicting malignancy of head and neck tumors using half-Fourier single-shot turbo spin-echo diffusion-weighted imaging, Magn. Reson. Imaging 32 (7) (2014) 860–866. T. Hauser, M. Essig, A. Jensen, et al., Prediction of treatment response in head and neck carcinomas using IVIM-DWI: Evaluation of lymph node metastasis, Eur. J. Radiol. 83 (5) (2014) 783–787. Y. Ding, J.D. Hazle, A.S. Mohamed, et al., Intravoxel incoherent motion imaging kinetics during chemoradiotherapy for human papillomavirus-associated squamous cell carcinoma of the oropharynx: preliminary results from a prospective pilot study, NMR Biomed. 28 (12) (2015) 1645–1654. N.G. Rao, G. Sanguineti, G. Chaljub, S.D. Newlands, S. Qiu, Do neck levels negative on initial CT need to be dissected after definitive radiation therapy with or without chemotherapy? Head Neck 30 (8) (2008) 1090–1098. K.T. Robbins, J.E. Medina, G.T. Wolfe, P.A. Levine, R.B. Sessions, C.W. Pruet, Standardizing neck dissection terminology: official report of the Academy's Committee for Head and Neck Surgery and Oncology, Arch. Otolaryngol. Head Neck Surg. 117 (6) (1991) 601–605. E. Benson, R. Li, D. Eisele, C. Fakhry, The clinical impact of HPV tumor status upon head and neck squamous cell carcinomas, Oral Oncol. 50 (6) (2014) 565–574. A. Fedorov, R. Beichel, J. Kalpathy-Cramer, et al., 3D Slicer as an image computing platform for the Quantitative Imaging Network, Magn. Reson. Imaging 30 (9) (2012) 1323–1341. S. Marzi, F. Piludu, A. Vidiri, Assessment of diffusion parameters by intravoxel incoherent motion MRI in head and neck squamous cell carcinoma, NMR Biomed. 26 (12) (2013) 1806–1814. R.H. Byrd, J.C. Gilbert, J. Nocedal, A trust region method based on interior point techniques for nonlinear programming, Math. Program. 89 (2000) 149–185. D. Le Bihan, R. Turner, The capillary network: a link between IVIM and classical perfusion, Magn. Reson. Med. 27 (1992) 171–178. J.K. Hoang, J. Vanka, B.J. Ludwig, C.M. Glastonbury, Evaluation of cervical lymph nodes in head and neck cancer with CT and MRI: tips, traps, and a systematic approach, AJR Am. J. Roentgenol. 200 (1) (2013) W17–25. S. Kim, L. Loevner, H. Quon, et al., Diffusion-weighted magnetic resonance imaging for predicting and detecting early response to chemoradiation therapy of squamous cell carcinomas of the head and neck, Clin. Cancer Res. 15 (3) (2009) 986–994. G. Sanguineti, S. Marzi, L. Marucci, A. Farneti, F. Piludu, A. Vidiri, Prediction of treatment response of cervical nodes using IVIM-DWI, Int. J. Radiat. Oncol. Biol. Phys. 96 (2) (2016) S69–S70. V. Vandecaveye, P. Dirix, F. De Keyzer, et al., Diffusion-weighted magnetic resonance imaging early after chemoradiotherapy to monitor treatment response in head-and-neck squamous cell carcinoma, Int. J. Radiat. Oncol. Biol. Phys. 82 (3) (2012) 1098–1107. D.P. Noij, P.J. Pouwels, R. Ljumanovic, et al., Predictive value of diffusion-weighted imaging without and with including contrast-enhanced magnetic resonance imaging in image analysis of head and neck squamous cellcarcinoma, Eur. J. Radiol. 83 (7) (2014) 1144–1151. M. Lambrecht, B. Van Calster, V. Vandecaveye, et al., Integrating pretreatment diffusion weighted MRI into a multivariable prognostic model for head and neck squamous cell carcinoma, Radiother. Oncol. 110 (3) (2014) 429–434. A.D. King, K.K. Chow, K.H. Yu, et al., Head and neck squamous cell carcinoma: diagnostic performance of diffusion-weighted MR imaging for the prediction of treatment response, Radiology 266 (2) (2013) 531–538. M. Matoba, H. Tuji, Y. Shimode, et al., Fractional change in apparent diffusion coefficient as an imaging biomarker for predicting treatment response in head and neck cancer treated with chemoradiotherapy, AJNR Am. J. Neuroradiol. 35 (2) (2014) 379–385. M.V. de Oliveira, E.P. Pereira Gomes, C.S. Pereira, et al., Prognostic value of microvessel density and p53 expression on the locoregional metastasis and survival of the patients with head and neck squamous cell carcinoma, Appl.

European Journal of Radiology 92 (2017) 93–102

S. Marzi et al.

[34] J.P. Driessen, A.J. van Bemmel, P.M. van Kempen, L.M. Janssen, C.H. Terhaard, F.A. Pameijer, S.M. Willems, I. Stegeman, W. Grolman, M.E. Philippens, Correlation of human papillomavirus status with apparent diffusion coefficient of diffusionweighted MRI in head and neck squamous cell carcinomas, Head Neck 38 (Suppl. 1) (2016) E613–8.

Immunohistochem. Mol. Morphol. 21 (5) (2013) 444–451. [32] A. Dietz, S. Delorme, V. Rudat, et al., Prognostic assessment of sonography and tumor volumetry in advanced cancer of the head and neck by use of doppler ultrasonography, Otolaryngol. Head Neck Surg. 122 (4) (2000) 596–601. [33] J. Pekar, C.T. Moonen, P.C. van Zijl, On the precision of diffusion/perfusion imaging by gradient sensitization, Magn. Reson. Med. 23 (1) (1992) 122–129.

102