Functional MRI for Treatment Evaluation in Patients with Head and Neck Squamous Cell Carcinoma: A Review of the Literature from a Radiologist Perspective

Purpose of review To show the role of functional MRI in patients treated for head and neck squamous cell carcinoma. Recent findings MRI is commonly used for treatment evaluation in patients with head and neck tumors. However, anatomical MRI has its limits in differentiating between post-treatment effects and tumor recurrence. Recent studies showed promising results of functional MRI for response evaluation. Summary This review analyzes possibilities and limitations of functional MRI sequences separately to obtain insight in the post-therapy setting. Diffusion, perfusion and spectroscopy show promise, especially when utilized complimentary to each other. These functional MRI sequences aid in the early detection which might improve survival by increasing effectiveness of salvage therapy. Future multicenter longitudinal prospective studies are needed to provide standardized guidelines for the use of functional MRI in daily clinical practice.


Introduction
Head and neck cancer affects 550,000 new cases and 380,000 deaths worldwide annually [1][2][3]. Head and neck squamous cell carcinomas comprise over 90% of the head and neck carcinomas [4]. Patients frequently present with a locally advanced stage for which the current therapy is multimodal including surgery, radiation therapy and/or chemotherapy [5][6][7][8]. Many patients demonstrate unfavorable treatment response, with locoregional recurrence seen in about 30-60% [7]. This is in about 2/3 due to primary tumor recurrence, 1/3 due to regional nodal metastasis and in 1/3 due to both primary tumor recurrence as well as regional nodal metastasis [9].
Conventional anatomical MRI techniques are commonly used for treatment evaluation, but are often not able to have reliable assess treatment response. Surgery as well as chemoradiotherapy induce false positives on imaging as a This article is part of the Topical collection on ENT Imaging. result of inducing benign changes involving architectural distortion, fibrosis and/or necrosis [10 •• , 11, 12, 13 • , 14 • ]. These benign treatment-induced changes should be differentiated from residual and/or recurrent tumor on imaging to prevent unjustified alteration in treatment plan, e.g. salvage therapy or (dis)continuation of therapy. Early detection of local recurrence could lead to timely salvage therapy which can lead to an increase in overall survival [15][16][17].
Post-treatment surveillance can consist of ultrasound, PET-CT [18][19][20][21] and MRI [10 •• , 11, 12, 13  This review will analyze the functional MR imaging sequences with regards to their possibilities and limitations in head and neck squamous cell carcinoma. Clinical implications, applicability and possibilities of these sequences for treatment evaluation will be addressed.
However, anatomical MRI techniques are often unable to accurately identify treatment response showing a pooled sensitivity and specificity for local treatment response evaluation in HNSCC of 84 and 82%, respectively [22 •• ]. This is due to benign treatment effects such as inflammation, fibrosis and necrosis as a result of surgery and chemoradiotherapy. These post-therapy changes show overlapping signal characteristics with tumor. Most problematic for the primary site is that inflammation and tumor both show high T2 signal and enhancement after contrast injection. Lymph node assessment is most hindered by reactive lymph node that can be slightly enlarged similar to nodal metastasis. Furthermore, normal sized nodes can still contain tumor. See Table 1 for a detailed description of the signal intensities on anatomical MRI post-treatment.
Higher diagnostic accuracy than 84% sensitivity and 82% specificity post-therapeutically is needed to differentiate treatment effects from true malignancy for the local tumor site and the regional lymph nodes to reliable either initiate new therapy, adjust the current therapy or discontinue unjustified therapy.

Technical Background of Functional MRI Techniques
Diffusion Weighted Imaging DWI measures cellular density and cytoarchitecture using the measurement of water diffusivity. Random diffusion results from the Brownian motion of water molecules. Motion of water molecules is hindered, restricted, by interactions with other molecules and cellular barriers such as fibers, cell membranes and macromolecules. Diffusion abnormalities of water molecules thus reflect changes of tissue organization at a cellular level affecting the MR signal of a DWI sequence as can be seen in a number of processes including malignancy [ DWI sequences are based on a T2-weighted sequence. At least two b values are needed to analyze motion of water. DWI is done at different b values (in s/mm 2 ), which represent the duration between the gradient pulses used. Simplified, it is the time that water is allowed to diffuse before the distance is measured. Most commonly, a b0 and b800 or b1000 value are used for head and neck imaging. Diffusion is quantified using ADC in mm 2 /s. Having measured at least two different b values (e.g. b0 and b800), the logarithm of relative signal intensity of a tissue is plotted on the y axis against the b values on the x axis. The slope of the line fitted through the plots describes the ADC. This mono-exponential fitting represents a rough approximation of ADC and is most often used in clinical routine. This parameter is independent of the magnetic field strength. Lower values indicate more restricted diffusion. However, mono-exponential fitted ADC values cannot separate the pure molecular diffusion from the motion of water molecules in the capillary network [32]. Low b values are most influenced by the capillary component which influences the ADC values. Multi-exponential models using several b values are more suitable for accurate quantification of diffusion without perfusion contamination [30, 32, 33 • ].
Acquiring multiple b values yields techniques such as intravoxel incoherent motion (IVIM) and diffusion kurtosis imaging (DKI). IVIM imaging can distinguish between pure molecular diffusion and motion of water molecules in the capillary network through a single DWI acquisition Another, multiple b value method, DKI, represents the extent to which the diffusion pattern of the water molecules deviates from a perfect Gaussian curve that is assumed calculating standard ADC values. Table 2 includes the most commonly used parameters for the different diffusion techniques.

MR Perfusion
Perfusion is defined as the steady-state delivery of blood to tissue. Several perfusion techniques are available; dynamic contrast-enhanced (DCE) perfusion, dynamic susceptibility contrast (DSC) perfusion and arterial spin labeling (ASL) all yielding different parameters (see Table 2). DCE perfusion is most commonly used for the head and neck area. DCE is based on the T1 relaxivity effects of contrast agents. DCE perfusion has been reported as a technique which is able to characterize perfusion and vascularization of tissues [ DSC perfusion exploits the susceptibility-induced signal loss after administration of contrast on T2-weighted sequences, most commonly a quick T2* gradient echo sequence. It is based on inhomogeneity of the magnetic field during the passage of a short bolus of contrast through a capillary bed [27]. As result on the T2* sequence, blood products, calcifications and aerated structures result in artificial signal loss. Mean transit time, blood flow and blood volume can be calculated. However, in the head and neck area a multitude of artifacts are present (e.g. voluntary/involuntary motion, breathing, air-to-tissue surface artifacts) [10 •• , 11, 12, 13 • , 14 • ], affecting the reliability of the results acquired with DSC.
ASL is a perfusion technique without injection of contrast. Arterial blood is magnetized below the volume of  interest. After a certain period, the magnetized blood flows into the volume of interest and its derived signal is measured. Blood flow can be calculated, which could reflect neovascularity and angiogenic activity of malignancy [32]. ASL also uses T1 relaxation, but is challenging as timing of the signal read-out should be precise. Acquiring the volume of interest too late, and the magnetized arterial blood has already passed. However, ASL is feasible in head and neck cancer using an Locker-Locker sequence [41] or a pseudo-continuous sequence [42].

MR Spectroscopy
MRS is a technique that detects the presence of specific metabolites. Different metabolites have small differences in their intrinsic vibration frequency and thereby result in small differences in signal of 1 H protons. Spectroscopy is ADC;; and b800-1000 :: thus well-suited to detect changes in the components of tissue due to tumor after suppression of the abundant water signal [43]. Single voxel and multivoxel techniques are able to characterize tissue including the measurements of lactate, N-acetylaspartate, creatine and choline. Spectroscopy should be regarded as complimentary to the already acknowledged functional MRI techniques in assessing HNSCC.

Response Evaluation During Therapy Diffusion Weighted Imaging
A rise in ADC is seen after the treatment in HNSCC (Fig. 1)  . Three other studies found thresholds of \ 14-24% to be predictive for regional failure in using clinical outcome data with at least 2 years follow-up [26,45,46]. However, it must be noted that imaging is generally not performed within the first couple of weeks in standard clinical practice. It is of great importance to interpret ADC analysis in conjunction with anatomical imaging. Areas of necrosis may take longer to resolve than solid areas. In the interim, the necrosis may become organized and show a fall in ADC value [24 •• ]. Therefore, it is critical to identify sites of necrosis that need to be excluded from ADC analysis [24 •• ]. Furthermore, the development of mature scar tissue may also decrease the ADC value [27]. The same holds for compact fibrosis which can demonstrate lowered ADC values and low to intermediate T2 signal.

MR Perfusion
Vascular HNSCCs are thought to have better treatment response compared to less vascular HNSCCs because of better delivery of chemotherapeutic agents and greater radiosensitivity [24 •• ]. On the other hand, vascular tumors may have a poorer outcome because they are thought they have greater metastatic potential [24 •• ]. Reports suggest that a fall in blood volume is associated with poor overall survival. On the other hand, an increased area under the curve is associated with local control [39]. The early rise in volume transfer (Ktrans) is speculated to result from damaged blood vessels causing them to temporarily become leakier, which potentially could increase the delivery of chemotherapeutic agents into the tumor.
Also plasma flow has shown to react in patients undergoing induction chemotherapy for the regional tumor [47]. The median baseline tumor plasma flow was 53 ml/100 ml/ min in 25 responders and 24 ml/100 ml/min in 12 nonresponders. In lymph nodes, differences were not significantly different between non-responders and responders [47]. After appropriate validation, this method may be potentially used to guide treatment modification in patients.

MR Spectroscopy
To the best of our knowledge, only one in vitro study of tumor specimens by has shown significantly elevated pretreatment choline-to-creatine ratios in a poor response group, but these findings could not be confirmed in an in vivo human study using choline/creatine ratios as well choline/water ratios [48].

Imaging Primary Tumor Site Post-therapy Diffusion Weighted Imaging
Anatomical MRI is mandatory for an accurate delineation of anatomical details (see Table 1). However, anatomical MRI is hindered by interpretation difficulties in the detection of local primary tumor recurrence [10 •• , 11, 12, 13 • , 14 • ]. A diffusion-derived b 800 or b1000 map provides high lesion-to-background contrast, outperforming conventional T2-weighted sequences in this aspect. The accompanying ADC indicates whether the high signal on the b value map is indeed due to tumor recurrence if low signal is seen on the ADC map. If the high signal on the b value map is accompanied by high signal on the ADC map it is not due to tumor and represents T2-shine-through, or increased diffusivity (see also Table 2 for interpretation of functional MRI). Fibrosis also lacks diffusion restriction (Fig. 2). A large meta-analysis showed a higher diagnostic accuracy for ADC compared to anatomical MRI. Anatomical MRI yielded a pooled sensitivity and specificity of 84 and 82%, respectively. ADC showed a pooled sensitivity and specificity of 89 and 86%, respectively [22 •• ]. More recent studies demonstrate a similar diagnostic accuracy for ADC values [46]. Even higher b values up to b2000 do not increase the diagnostic accuracy [44,49]. Using both a b1000 and b2000 and ADC ratio (= ADC 2000 /ADC 1000 9 100%) can be calculated. The ADC ratio might increase the diagnostic accuracy although results are variable with a sensitivity and specificity of 63 and 84%, respectively, for one study [44]. This is a small study with 32 patients, thus should be further studied in a large population.
Diffusion restriction results from high cellularity as in tumor, but can be also induced due to inflammation and abscesses. Moreover, restricted diffusivity can be seen in normal structures (e.g. Waldeyer's ring or normal lymph nodes) because these structures have an inherent high cellularity [

MR Perfusion
A cross-sectional study demonstrated significant differences between DCE perfusion parameters comparing the blood volume of scar tissue and tumor recurrence in HNSCC [50]. Its potential use in treatment follow-up was also shown in a small retrospective study [51]. Although DSC is not the most used perfusion method in the head and neck area, a higher wash-in on DSC has been related with tumor recurrence instead of treatment changes in a prospective study [33 • ]. However, diagnostic accuracy studies to differentiate treatment changes from tumor recurrence or residual with DCE or DSC perfusion are lacking. Although, visual assessment is possible (see also Table 2 for interpretation of functional MRI), further quantification is currently hindered by standardization of scan parameters and thresholds. In our experience, the area under the curve (AUC) summing the enhancement in a certain voxel, delineates abnormalities most easily with high values for tumor. Relative enhancement provides more insight in the magnitude of enhancement compared with the pre-contrast values. Region of interest analyses could demonstrate relative enhancements curves with the internal carotid artery as reference. A rapid wash-in comparable with the carotid artery followed by a wash out or plateau phase is indicative of tumor (Fig. 3), while slowly progressive enhancements indicate benign treatment changes (Fig. 4).

MR Spectroscopy
MRS is not routinely used for the treatment evaluation of HNSCC. However, the presence of choline as indication of proliferation and cell membrane turnover yield high specificity of 100%, although false-negative are frequently present, resulting in a very low sensitivity of 44% [52].

Imaging Lymph Nodes Post-therapy Diffusion Weighted Imaging
Treatment evaluation of regional lymph node is less studied than the primary tumor site. A higher diagnostic accuracy for ADC over anatomical MRI is suggested [22 •• , 53-55]. Anatomical MRI sensitivity and specificity ranged between 67-90 and 33-97%, respectively [22 •• ]. For ADC, this was 78 and 88% in one study and 73 and 100% in another study [45,53]. However, the difference was statistically not significant. Benign lymph nodes demonstrate higher ADC values compared to malignant lymph nodes [54][55][56]. This is also demonstrated in lymph nodes between 5 and 10 mm [54][55][56]. However, mean ADC values for benign lymph nodes range from to 1.1 to 1.6 9 10 -3 mm 2 / s, while HNSCC metastatic nodes range between to 0.78 and 1.1 9 10 -3 mm 2 /s [24 •• , 25 •• ]. A threshold of 1.1, therefore, seems most appropriate to use, although overlap could result in false-positive and false-negative results. The diagnostic accuracy for post-treatment lymph nodes using the IVIM or DKI methods might be better using multiple b values. This remains speculative currently as diagnostic accuracy studies are lacking post-therapy. The values of the known decrease of kurtosis of lymph nodes during treatment [57,58] should be further established. The IVIMderived D values represent pure diffusion without perfusion components. Significantly higher D values are demonstrated in patients with regional failure in line with the ADC results [29 • , 33 • ]. However, another study showed no significant rise in D values but a higher initial f value (perfusion fraction) in locoregional failure compared to locoregional control [53].

Perfusion-Weighted Imaging
A few recent studies have demonstrated differences in perfusion parameters between benign lymph nodes and malignant lymph nodes [24 •• , 33 • , 41,42]. Perfusion of nodal metastasis might be increased (Fig. 5). Metastatic lymph nodes demonstrate higher blood flow and blood volume compared to benign lymph nodes on CT perfusion [41,42], which thus would be expected to be similar for DSC MRI perfusion. The capillary permeability (Ktrans) correlates with the hypoxia-induced transcription factor in the tissue, which is known to stimulate angiogenesis [59]. However, interpretation of MR perfusion in post-therapy lymph nodes is difficult and it remains to be elucidated whether differentiation of malignant and benign lymph nodes can be done reliable (Fig. 6).

MR Spectroscopy
Acquiring MRS in lymph nodes is currently not clinically applicable as the region of interest should be placed separately on each suspicious lymph node by a radiologist on site. If the technically challenges are overcome, the increased choline, decreased creatine and subsequently increased choline/creatine ratio of metastatic nodes need to be confirmed in larger studies [60][61][62].

Limitations
The limitations and potential pitfalls of the functional MRI sequences should be kept in mind during the interpretation. First, the lack of anatomical information at high b values in DWI is a drawback because of suppressed signal in most of the normal tissues. Therefore, DWI should not be interpreted alone, but in correlation with anatomical sequences. This is also true for perfusion and spectroscopy which means that all functional MRI sequences can never be used without the use of anatomical sequences. Moreover, all functional sequences are currently hindered by high variability of cut-offs and parameters used.
Second, it must be stressed that functional MRI remains technically challenging to perform due to artifacts (i.e. breathing, swallowing, involuntary motion and air-tissue interfaces) [ Diffusion-derived interpretation is mainly done using mean ADC values. Diffusion showed good reproducibility for baseline scans for the ADC value of the primary tumor and nodal metastasis [63]. The reproducibility of the ADC during treatment is also suggested to be good [64]. Mean values of the tumor or metastatic lymph node are not representative when they consist of both highly and poorly cellular (necrotic) portions. Mean ADC values should be measured in the areas with high cellularity only to overcome this limitation [24 •• , 25 •• ]. Even then, ADC interpretation remains challenging. A recent study suggested a reduced field of view (FOV) might increase accuracy [65]. Moreover, it has been suggested that multiple b values are more accurate as this method is able to distinguish the perfusion component resulting in a pure diffusion value. This perfusion might influence the ADC value, although  Studies with regards to MRS suggest a higher cholineto-creatine ratio in patients with poor prognosis, which corresponds with expected high rates of proliferation and membrane biosynthesis in aggressive tumors (increased rate of metabolism) [48]. However, MRS is not commonly used due to its technical challenges. The region of interest should be placed by a radiologist to ensure correct placement in the anatomically difficult head and neck area. Furthermore, motion artifact from the carotid artery, long scan durations and complex post-processing hinders clinical applicability [52,66].  6 Normal lymph node and nodal metastasis with diffusion and perfusion. A 66-year-old patient with a right sided pT1N1Mx floor of the mouth SCC demonstrated recurrent lymph nodes after postoperative radiation therapy. An enlarged metastasis lymph node was seen on the right side with diffusion restriction and increased relative enhancement and AUC (arrow). A contralateral lymph node was not enlarged and demonstrated slightly restricted diffusion as is also seen in normal lymphoid tissue. Perfusion showed a high AUC and relative enhancement with a rapid wash-in with plateau phase for both lymph nodes, although most pronounced in the metastatic lymph node. Interpretation of the perfusion of lymph nodes remains difficult and should be further investigated (Color figure online)

Future Developments and Challenges
Differentiation between malignancy and benign post-treatment effects such as fibrosis in HNSCC is of importance to guide clinical decisions. The head and neck is an area sensitive for artifacts and functional MR imaging requires advanced MRI post-processing software to evaluate HNSCC. Combined functional sequences are required to fully appreciate HNSCC post-therapy, in addition to the necessary anatomical sequences. This would result in long scan durations, but new developments could overcome time issues. A possible role of hybrid integrated PET/MR imaging might be demonstrated offering the potential to acquired anatomical and function data using different modalities. However, future research is needed to evaluate PET/MRI and its appropriate applications compared to existing techniques [67] and whether PET/MRI is of greater clinical value than PET/CT and retrospective image fusion techniques [68]. HNSCC is common and local residual and/or recurrence and nodal metastasis are seen in many patients. Diffusion is already frequently used. However, diffusion with multiple b values and perfusion required further confirmation of their added value in the post-therapy setting before wide-spread implementation. This is even more the case for spectroscopy. Future studies should focus on the added value of the different functional MRI sequences preferable by large prospective longitudinal multicenter studies comparing all sequences in the same population.
These studies are needed to assess the diagnostic accuracy of the functional MRI sequences separately and in combination. Another important aspect of these studies should be to define the optimum time for assessment of metabolic and physiological MRI parameters using functional techniques.
The functional parameters should be tested in relation to the histopathological changes in HNSCC, treatment effects and patient outcomes. These new trials must result in standardized cut-off values and ratios for the anatomical and functional MRI sequences to precisely define post-therapy changes from tumor progression. The use of standardized cut-off values might remain arbitrary because of the use of different MRI systems. Nevertheless, it would be a valuable guideline for the clinician in daily practice. Despite these possible limitations, implications into clinical practice would be an important step in making an accurate treatment decisions for HNSCC patients.

Conclusions
In summary, this review analyzed the role of specific functional MRI modalities in differentiating benign posttreatment effects from recurrence and/or residual malignancy and metastases in HNSCC.
Differentiation between malignant and benign posttreatment effects in HNSCC is of importance to guide clinical decisions. As anatomical MRI is not able to reliably differentiate post-therapy effect from tumor, functional techniques have been investigated and shown to be promising. This review showed that DWI can increase the diagnostic accuracy significantly for the primary tumor site and might also increase the diagnostic accuracy for the region lymph nodes after therapy. Diffusion is most easy to implement and is recommended to perform routinely in a clinical setting in HNSCC follow-up. Its use during treatment to predict outcome is interesting, but evidence is too low to implement.
Although perfusion parameters might be increased in tumor residual or recurrence and nodal metastasis, its diagnostic accuracy has yet to be established and is not routinely used clinically. DCE is least hindered by artifact and might be performed clinically if local experience is present.
Spectroscopy research is promising, but evidence is too sparse for clinical implementation in the near future. The role of hybrid PET/MR imaging is to be established.