Diffusion-weighted magnetic resonance imaging (DWMRI) of head and neck squamous cell carcinoma: could it be an imaging biomarker for prediction of response to chemoradiation therapy

Chemoradiation therapy (CRT) has become a primary definitive treatment modality for head and neck squamous cell carcinoma (HNSCC); however, not all patients respond completely to treatment. Ability to identify those patients, who would not achieve complete response, before or early during the course of CRT will allow treatment modifications to improve outcome and overall survival. The aim of this prospective study was to assess the usefulness of diffusion-weighted imaging (DWI) in prediction of early therapeutic response of HNSCC after CRT. Local control was achieved in 22 patients out of 46 patients with pathologically proven HNSCC treated by chemoradiation therapy and local failure was detected in 24 patients out of 46 patients. Pretreatment mean apparent diffusion coefficient (ADCpre) was significantly higher in local failure group (1.1 ± 0.2 × 10−3 mm2/s) than local control group (0.89 ± 0.1 × 10−3 mm2/s). An optimal cut-off value of more than 0.94 × 10−3 mm2/s was predictive of local failure with sensitivity 83.33%, specificity 59.9%, PPV 69%, NPV 76.5%. Early intra-treatment percentage change of ADC (ΔADC) was significantly lower in local failure group (21.8% ± 21.3) than in local control group (45.2% ± 27.8). An optimal cut-off value of ≤ 33% was predictive of local failure after CRT with sensitivity of 71.34%, specificity of 60%, PPV of 62.5%, and NPV of 69.2%. Diffusion-weighted MRI could be a potential predictive biomarker for therapeutic response of HNSCC to CRT. Primary tumors with higher pretreatment mean ADC, and a smaller early intratreatment percentage increase of mean ADC would be more likely to fail treatment.


Background
Squamous cell carcinoma accounts for 95% of all head and neck cancers. The choice of treatment modality depends on how to provide cure with maximum possible preservation of organ function [1].
During recent years, chemoradiation therapy (CRT) has become a primary definitive treatment modality for head and neck cancer, allowing for preservation of organ function (swallowing and speech) with survival rates equal to surgery; however, about 30% of patients do not respond to CRT leading to locoregional failure [2,3].
The ability to identify patients who would not respond to CRT allows for modification of treatment strategies such as adding a boost of radiation therapy at the end of initial treatment, additional targeted therapies, or even shifting treatment strategy from CRT to surgery and subsequently improvement of cure and survival rates, also this will avoid the unnecessary toxic side effects of CRT, but this requires early diagnostic biomarkers that can predict later tumor response and identify patients who would benefit from these modifications [4][5][6].
Predictive imaging biomarkers before or early after initiation of radiation therapy of HNC are of great interest in recent researches, including PET-CT, perfusion-CT, and functional MRI studies either perfusion or diffusion imaging studies [7][8][9].
MRI is an attractive tool because it does not involve ionizing radiation, better soft tissue evaluation, and is the imaging method of choice in head and neck squamous cell carcinoma for initial anatomical staging and follow-up [10]. Diffusion-weighted magnetic resonance imaging (DWI) is a quantitative relatively rapid MRI sequence that can be easily incorporated into routine patient examination without need for exogenous contrast agent. It has the ability to act as a potential marker for tissue cellularity which affects tumor response to CRT [11,12].
DWI is a noninvasive imaging technique that measures the relative mobility of water molecules in different tissue microstructures which depends on cell size, density, integrity of cellular membrane and tissue vascularity. In the diffusion sequence, two large gradient pulses are used, equal in strength (which is determined by the b value) but in opposite directions. The first gradient pulse induces a phase shift of water molecules, and the second pulse for rephrasing; however, the moving water molecules will lead to incomplete rephrasing and signal loss [11,13]. The amount of signal loss on diffusion images with increasing b values can be quantified using the apparent diffusion coefficient (ADC) which shows an inverse correlation with tissue cellularity. Hypercellular tumors are associated with less signal loss and low ADC values while hypocellular tumors or tumors containing multiple necrotic foci are associated with more signal loss and high ADC [2,8,14]. DWI has another great advantage that it is sensitive to microscopic pathologic changes before they became visible on conventional MRI sequences [8,15].
The aim of this prospective study was to evaluate the potential usefulness of pretreatment and early intratreatment DWI in prediction of therapeutic response of HNSCC to CRT.

Methods
The local ethics committee approved the study protocol and written informed consent was obtained from all participants. Seventy-four patients with pathologically proven, previously untreated head and neck squamous cell carcinoma, planned to be treated with CRT, presented to the Outpatient Clinic of the Clinical Oncology Department from March 2017 to October 2019 and referred to the Radiodiagnosis Department for imaging were enrolled in the study. Twenty-eight patients were excluded due to starting induction chemotherapy before MRI (n = 7), change of management plan from CRT to surgery or palliative treatment (n = 4), technical difficulties leading to diffusion image distortion (n = 3), patient's death due to unrelated causes and final response status could not be determined (n = 4), claustrophobia (n = 3), withdrawal from the study (n = 1), and too small lesions to be detected on MRI (n = 6). Eventually, 46 patients were eligible for the study. Patient's characteristics are demonstrated at Table 1. All tumors were staged according to the American Joint Committee on cancer TNM staging system, 8th edition, 2017.
All patients were subjected to complete history taking, clinical and endoscopic assessment as well as histopathological diagnosis of SCC. Pretreatment diagnostic studies were done including contrast-enhanced CT and MRI examinations with diffusion-weighted imaging.
All patients had been treated with 3D conformal radiation therapy (The radiation therapy regimen included a total fractionated dosage of 70 Gy that was given in 35 daily fractions at a daily dose of 200 cGy per fraction.). Thirty-nine patients had received concurrent chemotherapy (cisplatin 40 mg/m 2 that was given weekly along with the radiotherapy for 7 weeks) and 7 patients had been treated by radiotherapy alone. Thirty-two patients had received induction chemotherapy prior to definitive chemoradiation therapy; the induction chemotherapy regimen included 3-4 cycles of cisplatin (25 mg/m 2 /day for 3-4 days) and 5-fluorouracil (1000 mg/m 2 /day for 4 days), the cycle was repeated every 3 weeks.
MRI was repeated without contrast administration 2-3 weeks after start of CRT in 31 patients, two of them were excluded as the tumor became too small to be measured.
The endpoint of the study was assessment of therapeutic response 4-12 weeks after the end of CRT, (mean = 7 weeks), by clinical, endoscopic assessment, and MRI examination. The response was determined according to RECI ST 1.1 criteria and patients were classified into two groups: Local control (n = 22, 47.8% of patients) was determined as complete disappearance of the tumor (CR), that is further confirmed by histopathological examination (in three patients only) (biopsy was done due to presence of suspicious lesions on endoscopic assessment) as well as further follow up by clinical and cross-sectional imaging assessment for a period of 4-19 months (mean = 9.5 months) that revealed no development of any suspicious lesions as well as stationary or decreasing existing posttherapeutic lesions. Local failure (patients who did not achieve complete response) (n = 24, 52.2% of patients) was determined as presence of persistent or progressive tumor (stable or progressive disease, SD or PD), residual tumor (partial response, PR), or death of the patient during or after treatment due to the disease. The presence of residual tumor was confirmed by histopathological examination (in three patients only), PET-CT studies (in another three patients, with findings matching with residual tumor), and/or further follow-up imaging studies that revealed progression in size of the previously detected lesions.
The MR studies were performed on 3-T MR system (Ingenia, Philips, Healthcare, Best, the Netherlands) using 16-channel head-neck surface coil. All sequences extended from the skull base to the thoracic inlet, covering the primary lesion and all nodal stations. The imaging protocol included: , all are multishot TSE sequences. T1-weighted sequence was repeated after intravenous GAD administration in the axial, coronal, and sagittal planes. Diffusion-weighted imaging: in our institution, using single-shot spin-echo echo planar imaging (SS-EPI) in diffusion imaging of the neck on 3-T MR system was associated with significant susceptibility artifacts and geometric distortion, so we preferred to use non-EPI TSE-based DWI that showed much less susceptibility artifacts. It was done in the axial plane (TR/TE 3000/ 72 ms, FOV 230 × 230 × 200, FA 90, slice thickness 3 mm, intersection gap 1 mm, reconstruction matrix 240, parallel imaging SENSE, fat suppression SPIR) with 2 b values of 0 and 1000 mm 2 /s.

Data analysis
Analysis of MR examinations was performed manually by two independent radiologists with 25 and 8 years of experience in head and neck radiology. The primary tumor was identified on the conventional TSE MR sequences, then visual analysis of the DW images and ADC map was done in conjunction with the coregistered anatomical images (DW images could not be analyzed separately due to their low spatial resolution). The volume of the primary tumor was measured in axial fat suppressed T2 images or post-GAD fat suppressed T1 weighted images in which the tumor was well delineated, by placing freehand regions of interest (ROIs) around the margins of the tumor in all imaging slices including necrotic or cystic areas and excluding areas filled with air or subjected to image distortion. The tumor volume was measured in the pretreatment (TV pre ) as well as midtreatment (TV mid ) MRI studies, then the percentage change of tumor volume (ΔTV) was computed: ΔTV = (TV pre − TV mid )/TV pre × 100 Mean ADC of the whole tumor was measured on the ADC map in the same way by placing freehand regions of interest (ROIs) around the margins of the tumor in all imaging slices including necrotic or cystic areas, then exported as xml file to be retrieved by SPSS program for histogram analysis and calculation of ADC skewness and kurtosis in the whole tumor that were considered as indicators for tumor heterogeneity. ADC skewness represents the skew of the shape of ADC distribution curve within the tumor, it is positive when the long tail of the curve is to the right and negative when the long tail of the curve if to the left. ADC kurtosis represents the sharpness of the peak of ADC distribution curve, broader peak denoted lower kurtosis, and more heterogenous ADC distribution while sharper peak denotes higher kurtosis and less heterogenous ADC distribution.
On midtreatment studies, mean ADC (ADC mid ) was measured in the same way, then percentage change of ADC (ΔADC) was computed: ΔADC = (ADC mid − ADC pre )/ ADC pre × 100 Horos dicom viewer was used for image interpretation, data analysis, and measurements.

Statistical analysis
Kolmogorov-Smirnov tests were used to assess the normality of distribution of our data. Because the MRI data was not normally distributed, Mann-Whitney test was performed to look for differences in primary tumor volume and ADC measurements between local control and local failure groups. P value of < 0.05 was considered significant. Receiver-operating characteristics (ROC) analysis with the area under the curve (AUC) was employed to assess the discriminatory capability of pretreatment mean ADC (ADC pre ) and early intratreatment percentage change of mean ADC (ΔADC) with calculation of the optimal cutoff value, sensitivity, specificity, PPV, and NPV for both. Univariate analysis of other clinical variables that may affect the therapeutic response were done including age, TNM staging classification and histological grade of differentiation using chi-square test. All data analyses were performed using a statistical tool, SPSS (Statistical Package for the Social Sciences, Version 23; IBM, Armonk, NY).

Results
The distribution of pretreatment mean ADC values (ADC pre ) and their midtreatment percentage change (ΔADC) among local control and local failure patient groups is demonstrated in the first figure (Fig. 1). Pretreatment mean ADC was significantly higher in local failure than local control patient groups (P = 0.001). The pretreatment tumor volume (TV pre ) was higher in local failure group than local control group but it was not statistically significant (P > 0.05) ( Table 2). The pretreatment mean ADC of the primary tumors (ADC pre ) increased significantly 2-3 weeks after the start of CRT in the whole group of patients (P < 0.001) as well as in both local control (P = 0.001) and local failure (P = 0.002) patient groups (Table 3). Percentage increase in early intratreatment mean ADC (ΔADC) was significantly higher in local control than local failure patient groups (P = 0.008). The absolute mid treatment tumor volume (TV mid ) was significantly higher in local failure patient group (P = 0.007), while the percentage change of tumor volume (ΔTV) was higher in local control group, yet without statistical significance (P = 0.1) ( Table 2). Univariate analysis for other clinical variables showed no significant difference between both groups (P > 0.05).  On histogram analysis of pretreatment ADC of the whole tumors, no statistically significant difference was found in ADC skewness or kurtosis (P > 0.05) between both groups (Table 2), while tumors with local failure were likely to have higher percentage of tumor voxels with high ADC values than tumors with local control as shown in Fig. 6 illustrating histogram of ADC distribution through the primary tumors of the representative cases.
ROC analysis for ADC pre determined an optimal cutoff value of > 0.94 × 10 −3 mm 2 /s to predict local failure after CRT with sensitivity of 83.33% specificity of 59.09%, positive predictive value of 69%, and negative predictive value of 76.5%. ROC analysis for ΔADC determined an optimal cut-off value of ≤ 33% to predict local failure after CRT with sensitivity of 71.34%, specificity of 60%, positive predictive value of 62.5%, and negative predictive value of 69.2% (Fig. 7) (Table 4).

Discussion
This prospective study assessed the correlation between ADC measurements of primary HNSCC before and early during the course of CRT and therapeutic response after completion of treatment. Patients were classified according to their response to treatment into two groups, local control group (who achieved complete response after CRT), and local failure group (in whom residual tumor was detected or the patients had died due to the disease during or after treatment). Pretreatment mean ADC was significantly higher in local failure group (1.1 ± 0.2 × 10 −3 mm 2 /s) than local control group (0.89 ± 0.1 × 10 −3 mm 2 /s). An optimal cut-off value of more than 0.94 × 10 −3 mm 2 /s was predictive of local failure. High ADC value of the tumor reflects its low cellularity, high stromal content, micronecrosis, and mostly negative HPV status, all contributes to tumor resistance to CRT and treatment failure [16,17]. Mean ADC increased significantly 2-3 weeks after start of CRT in both local failure and local control patient groups, this was attributed to cellular death and damage of cell membrane barriers which facilitates diffusion of water molecules [18][19][20]. Significant smaller percentage increase of mean ADC was found in local failure group (21.8% ± 21.3) than in local control group (45.2% ± 27.8). A threshold of less than 33% increase of mean ADC was significant predictor of local failure.
These results were in agreement with other two studies, one study assessed the role of DWI in monitoring therapy in HNSCC and found that pretreatment ADC was significantly lower in complete responders with cutoff value of ≤ 0.98 × 10 −3 mm 2 /s was identified as significant predictor of local control [21]. The second study was performed on metastatic HNSCC nodes; they found that complete responders had lower pretreatment ADC (1.04 ± 0.19 × 10 −3 mm 2 /s) than partial/non-responders (1.35 ± 0.3 × 10 −3 mm 2 /s) and identified optimal cut-off value of 1.11 × 10 −3 mm 2 /s for differentiating complete and partial responders [22].
Other studies assessing the efficacy of pretreatment ADC measurements in predicting therapeutic response found that complete responders exhibited lower pretreatment ADC than non-responders but they were not statistically significant [3,5,6,15].
Other several studies were also compatible with this study as regards the early intratreatment percentage change of mean ADC being significantly higher in local control patient group [5,15,23,24]. King et al. [24] found that the intratreatment ADC fall early or late during treatment had 100% positive predictive value of locoregional failure; however, ADC rise was found in both groups, but they did not compare between percentage rise in both groups.
The other three studies were done by King et al. [5] Matoba et al. [15], and Vandecaveye et al. [23] who assessed the correlation between locoregional outcome and ADC values as well as tumor volume measurements in primary tumors and metastatic nodes before and 3 weeks after start of treatment, they found that the intratreatment percentage change of ADC was significant predictor for locoregional control with optimal thresholds of ≥ 15.5%, 24%, and 25% to predict locoregional control after CRT respectively. A meta-analysis was performed including nine studies assessing the role of DWI (before and early during treatment) in predicting therapeutic response of HNSCC to CRT, they found that DWI had good predictive accuracy (AUC = 0.84) for locoregional failure with the optimum cut-off values for pretreatment ADC ranging from 0.86 to 1.11 × 10 −3 mm 2 /s and from 14 to 52.7% for early intratreatment percentage change in ADC [25].
Another systematic review was done including 11 studies evaluating the value of DWI in prediction of response of HNSCC to CRT [26]. Five studies found significantly higher pre-treatment ADC values in patients with locoregional failure than in those with locoregional control with cut-off values ranging from 0.86 to 1.2 × 10 −3 mm 2 /s [21,[27][28][29][30]. The other 6 studies did not show significant difference in ADC between the two groups [5,15,[31][32][33][34].
Tumor heterogeneity may influence its response to treatment. Tumors with multiple foci of necrosis may exhibit more resistance to treatment. These  necrotic foci would measure high ADC values. ADC histogram analysis could be a tool for assessment of tumor heterogeneity in terms of ADC skewness and kurtosis. In this study, no significant difference in pretreatment ADC skewness and kurtosis was found between local control and local failure patient groups. This was compatible with two studies performed by King et al. [5] and Martens et al. [3] who also found no significant difference in pretreatment ADC skewness and kurtosis between local failure and local control groups; however, in the former, they found significant difference in intratreatment ADC skewness and kurtosis between both groups. This study revealed that tumors with local control has higher percentage of tumor voxels with low ADC values, this was compatible with study performed by Srinivasan et al. [35] who assessed the value of pretreatment mean ADC and ADC histogram analysis in predicting therapeutic response of HNSCC in correlation with outcome 2 years after treatment, they found that tumors with lower pretreatment mean ADC and more than 45% of volume below ADC threshold of 1.15 × 10 −3 mm 2 /s may have better outcome.
In this study, pretreatment tumor volume was higher in local failure group yet without statistical significance as with tumor volume changes 2-3 weeks after start of treatment while midtreatment tumor volume was significantly higher in local failure group. These results were compatible with study done by King et al. [5] who found that the intratreatment tumor volume was significantly higher in local failure and was the only predictor that remains significant after multivariate regression analysis.
Diffusion-weighted imaging of the neck is challenging because the neck is a narrow region with complex anatomical structures, multiple tissue-fat, and tissue-air interfaces specifically in our studied areas including all mucosal surfaces, also it is subjected to motion artifacts caused by respiratory movements and unavoidable swallowing movements, as well as metallic artifacts caused by dental prothesis [7]. Also, DWI is highly susceptible to artifacts than other MR imaging sequences, so in this study, non-EPI TSE-based DWI was preferred rather than usual single-shot spin-echo echo-planner DWI used in routine neck imaging to avoid these susceptibility artifacts. TSE-DWI is well established in imaging of cholesteatoma to avoid artifacts of bone-air interface at the skull base [36]. Another study assessed its value in nasopharyngeal imaging differentiating recurrent tumor from post-radiation fibrosis [37]. To the best of our knowledge, limited studies are available in the literature about application of non-EPI TSE-DWI in routine neck imaging. In this study, TSE-DWI was promising giving better imaging quality with less susceptibility artifacts; however, further studies are required to compare its results with single-shot spin-echo echoplanner DWI and to be sure that they give similar ADC measurements.
This study had several limitations such as small number of patients, including SCCs of different sites and stages as well as different treatment strategies, also several factors that are known to affect the biological behavior of SCC such as smoking and HPV status were not assessed in this study. This study was performed on the primary tumors; however, nodal assessment is also important as the nodal status greatly affects the prognosis of HNSCC. The tumor status at 4-12 weeks after the end of treatment was the end point of this study, further studies with long term follow up is required to assess the efficacy of DWI in predicting overall and disease-free survival. Finally, these results still need to be validated in a larger multicenter trial with optimization of sequence parameters, methods of data analysis, and exact timing of assessment to allow for its clinical use.

Conclusions
Diffusion-weighted MRI could be a potential predictive biomarker for therapeutic response of HNSCC to CRT. Primary tumors with higher pretreatment mean ADC, and a smaller early intratreatment percentage increase of mean ADC would be more likely to fail treatment.