Use of Imaging in Radiotherapy for Head and Neck Cancer

Abstract

Wider availability of and technical improvements in anatomic and biological imaging have facilitated the implementation of high-precision three-dimensional conformal and intensity-modulated radiotherapy (RT) in head and neck cancer. The integration of recent advances in functional and molecular imaging have already improved staging, RT delivery, response prediction, and follow-up. Rational clinical use of all modalities should be encouraged, especially in the setting of imaging-intensive investigational RT protocols such as adaptive therapy. Expanded development of imaging markers which can predict radioresistance or outcome could further customize treatment. Continued successful use of innovative imaging in routine clinical practice will ultimately depend on well-designed clinical studies with adequate follow-up.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

Head and neck cancers (HNC) include malignancies originating in the oropharynx, oral cavity, hypopharynx, and larynx. In the western world, more than 90% of these are squamous cell carcinomas (Hustinx and Lucignani 2010). Nasopharyngeal malignancy is also a subtype of HNC, although with a somewhat different epidemiology and natural history. Most early-stage head and neck tumours can be cured by surgery or radiotherapy (RT), which have similar locoregional control rates. The selection of treatment modality in early disease depends on patient condition and preference, organ function, and cosmesis. The majority of patients with HNC, however, have locoregionally advanced disease at diagnosis, and are treated with either curative-intent RT plus or minus chemotherapy, or radical surgery followed by post-operative (chemo)radiotherapy (Hustinx and Lucignani 2010). HNC preferentially spreads to cervical lymph nodes (LN) which, if involved, are an adverse prognostic factor. For patients whose disease is too advanced to be addressed radically, palliative RT may be administered for symptom control. RT is therefore part of the treatment approach in the majority of HNC patients at some point during the course of their disease.

The clinical challenge in radical RT is to attain the highest probability of cure with the least toxicity to surrounding normal structures. It is assumed that improvements in locoregional control can be obtained by increasing the dose of radiation delivered to the gross tumour as well as areas at risk for microscopic involvement. Although radiation damages both normal and neoplastic cells, most normal tissues have a small advantage in their ability to recover from RT injury. In curative treatment, this difference is exploited by using many small dose fractions delivered daily over a period of several weeks to obtain a therapeutic advantage. The total dose that can be delivered is limited by the damage caused to the surrounding normal tissues and consequent risk of complications, which also increases with increased RT dose. Therefore, the more normal tissue that can be spared, the greater potential exists for escalating the tumour dose. Recent advances maximise the ability to accomplish this, including the integration of more powerful computers, hardware developments in delivery machines, and especially advances in imaging.

Wider application of and/or technical improvements in ultrasound (US), computed tomography (CT), magnetic resonance imaging (MRI), and positron emission scanning (PET) have made it possible to design high-precision three-dimensional (3D) conformal and intensity-modulated radiotherapy (IMRT). In HNC for example, the obvious advantage of sparing structures such as the salivary glands from damage has pushed IMRT to the forefront as the new standard of care faster than in other cancer sites. However, success requires exact radiological identification of both the tumour and normal structures. Functional imaging is an area of active investigation and metabolic, physiologic, genotypic, or phenotypic data may contribute to our ability to delivery RT safely and effectively in the near future.

This chapter will focus on recent advances in anatomical and functional imaging as they apply to the field of Radiation Oncology. After a general overview of RT in HNC, the advantages and limitations of various imaging techniques used in RT will be described. This will be followed by a review of the current and potential applications of these technologies in Radiation Oncology, although a comprehensive description of the technical details of these techniques is beyond the scope of this chapter.

2 Radiotherapy for Head and Neck Cancer: General Principles

In the infancy of Radiation Oncology, surface anatomy was used to delineate the site to be treated (Table 1). This necessarily resulted in treatment fields encompassing large amounts of normal tissue, and dose calculation could be performed at only a few selected points. In the 1950s, with the advent of the fluoroscopic simulator, internal bony landmarks were used to guide field determination. Two-dimensional (2D) treatment planning was performed in a single plane with target doses displayed as isodose lines. Typically, patients with HNC planned using this method were treated with two lateral fields encompassing the primary and nodal areas, with a separate low neck field for the supraclavicular fossae (Fig. 1). This field arrangement required the portals to be reduced in size at some point to limit the dose to critical normal structures such as the spinal cord. Further dose was then delivered through smaller ‘boost’ fields to the grossly involved areas.

Table 1 Progress in radiotherapy planning

Fig. 1

Standard 2D radiation portals: 2 lateral opposed fields (a) and a lower anterior field (b). Abbreviation: GTV= gross tumour volume

However, it was extremely difficult with 2D treatment to deliver a tumour ablative dose while limiting damage to an organ at risk (OAR) that may be just a few millimetres away. The toxicity from this type of head and neck RT was therefore among the worst seen in Radiation Oncology. In the 2D field setup, both parotid glands commonly received full dose radiation. The most common acute side effects consisted of mucositis resulting in dysphagia and odynophagia, salivary changes, and dermatitis. Late toxicities, usually permanent, included xerostomia, fibrosis, myelitis, and osteonecrosis. Acute and late toxicity was one of the limitations in the 2D era resulting in an inability to increase the total dose above approximately 74 Gy; others included irradiation of extensive areas of mucosa and the critical issue of matching of sequential fields.

Over the past 20–30 years, these issues led to the development of CT-based RT, during which time CT simulation has become routine. Tumours and OARs are outlined on serial axial CT slices by the radiation oncologist and then reconstructed in three-dimensions. Computer-based treatment planning systems show a ‘beam’s eye view’ projection of the patient’s anatomy as seen from the radiation source, demonstrating which structures are present in the beam’s path. Based on this geometry, beam and table angles and field sizes can be optimised for each field. Following international guidelines for dose computation, a treatment plan is developed using electron densities and tissue depths from the planning CT. Isodoses are volumetrically overlaid on the CT slice dataset forming a 3D model of the patient (Fig. 2). Dose–volume histograms, graphical representations of the dose delivered to a certain structure by volume, are also displayed. The beams in this 3D conformal radiation therapy (3DCRT) technique are more precisely targeted than in the 2D era, but within each, there is uniform dose intensity.

Fig. 2

Comparison between field arrangement and isodoses obtained with 2D, 3D, and IMRT planning showing sparing of contralateral parotid gland from high-dose radiation. a 2D plan isodoses, b 3D plan isodoses, c 5 field IMRT isodoses, d 3D-CRT setup with 4 fields, e 5 field IMRT field arrangement

To improve the therapeutic index further, IMRT was introduced in the 1990s. As its name implies, IMRT allows the modulation of intensity within each radiation beam, so that each may contain one or more areas of high intensity, and other areas of lower intensity. By modulating the intensity within each field as well as the number of fields, radiation dose can be elegantly sculpted around the tumour (Fig. 2). The head and neck is an ideal site for this highly conformal RT due to the complex geometry of this area, the extremely small distances between tumour and critical structures, and the significant impact of radiation-induced toxicity on quality of life (QoL). Advantages of IMRT include: the ability to deliver treatment to an irregularly-shaped or concave volume; administration of different doses to different areas simultaneously [the simultaneous integrated boost (SIB) technique]; eliminating the need for field adjustments and matching; and greater sparing of normal structures such as the spinal cord or salivary glands (Bhide et al. 2010). IMRT and 3DCRT have been prospectively compared in three randomised HNC trials, two of which are published (Kam et al. 2007; Pow et al. 2006). To date, it appears that local control and disease-specific survival are equivalent, with the suggestion of improved salivary function and QoL in the IMRT patients.

Due to the highly conformal nature of the dose distribution and steep dose gradients, accurate localisation of areas to be treated is necessary to avoid geographic miss (Troost et al. 2010a). It is therefore essential that radiation oncologists have an excellent knowledge about the anatomy of the head and neck, as well as the use, advantages and disadvantages, and interpretation of different imaging modalities.

3 Overview of Imaging Used in Radiotherapy

Imaging modalities used commonly in RT delivery can be divided into two main categories: anatomical, which provide structural or morphological information, and functional, which elucidate biological or molecular features. General tumour discrimination and spatial resolution are summarised in Table 2, and use, advantages, and limitations from a Radiation Oncology perspective are outlined below. Inter- and intra-observer variation must be taken into account in assessing comparative accuracy, but this may be offset to some degree by using complementary information from multiple imaging modalities.

Table 2 Characteristics of imaging modalities used in radiotherapy

3.1 CT

3.1.1 Use

CT simulation scans are essential in RT as computer-based treatment planning algorithms require CT electron density data for dose calculation. PET or MRI alone cannot be used but require co-registration or ‘fusion’ with CT (Dandekar et al. 2010). CT simulation scans are performed with the patient immobilized in the treatment position, on a flat table top to simulate the RT delivery machine (linear accelerator) patient couch. Radio-opaque markers on skin reference marks are used in patient set-up at the time of treatment. CT slices are typically 2–10 mm thick depending on the site to be imaged, and the scanner is ideally networked to treatment planning software to enable direct data transfer.

3.1.2 Advantages

• good contrast between air, fat, and bone

• widespread availability

• relatively inexpensive

• good geometric accuracy

• visualisation of anatomy for comparison with diagnostic imaging

• spiral and multi-slice CT with reduced scanning times

• contrast agents improve tumour discrimination

• provides electron density information for tissue inhomogeneity corrections for accurate dose modelling

• multiplanar reconstruction

3.1.3 Disadvantages

• CT scans represent anatomic information regardless of biological changes making differentiation between malignancy and benign processes that cause morphologic changes difficult (Dandekar et al. 2010)

• differentiation difficulty between tumour edges and adjacent normal soft tissue

• artifacts caused by contrast agents and high-density structures such as tooth fillings can obscure anatomy and pathology in proximity

• high-density structures also hamper the conversion of tissue signal to electron density

• radiation dose

3.2 Magnetic Resonance Imaging

3.2.1 Use

MRI provides multi-planar anatomical information with high-contrast resolution. CT/MRI fusion is standard for HNC RT planning, especially for nasopharyngeal, oral tongue, and paranasal sinus carcinomas (Fig. 3). MRI-visible markers can be used to indicate skin reference marks.

Fig. 3

Example of CT/MRI fusion image. A 48-year-old male presented with recurrence of a squamous cell carcinoma of the facial skin with perineural spread along the infraorbital nerve into the fossa pterygopalatina, and extension to the lateral left cavernous sinus (arrow)

3.2.2 Advantages

• no patient radiation dose

• excellent delineation of soft tissue, intracranial infiltration and perineural spread

• flexibility in imaging sequences and planes allowing optimisation according to tissue of interest

• functional and molecular imaging capabilities such as diffusion-weighted (DW) MRI

3.2.3 Limitations

• does not provide inhomogeneity (electron density) information

• patient positioning may be a problem as scanner aperture is usually small

• long scan times

• geometrical distortion created at the edges of the field must be corrected

• susceptibility artifacts at interfaces between bone and air (Daisne et al. 2003)

• cannot be performed simultaneously with diagnostic CT or CT simulation and fusion can be challenging if slight differences in patient position are present

• use of immobilisation masks (required for RT positioning) during MRI can preclude the use of a dedicated head and neck coil

3.3 PET

3.3.1 Use

Technological advances in PET scanning have increased the use of this modality for RT. The hybrid PET/CT, introduced in 2001, is the technique of choice for imaging HNC for the purposes of treatment planning (Hustinx and Lucignani 2010). As with MRI, PET–CT should be acquired in the treatment position on a flat table top with appropriate RT immobilisation. The presence of RT department staff can help ensure accuracy, consistency, and patient comfort in positioning (Dandekar et al. 2010). Registration of the CT component with the CT simulation scan improves image fusion accuracy (Dandekar et al. 2010).

3.3.2 Advantages

• PET can define a metabolically active biological target volume (BTV) (Dandekar et al. 2010)

• detection of distant metastases or second primary tumours

• hardware fusion of hybrid PET–CT scanners provide better localisation than stand-alone PET scanners (Dandekar et al. 2010)

• new tracers under development

• during follow-up, differentiation of fibrotic scar from residual viable tumour after RT

3.3.3 Limitations

• false positive readings due to physiological tracer uptake in brain, salivary glands, laryngeal muscles, skeletal muscle, heart, and brown fat

• false positives can also be caused by infection, inflammation, or reactive LN

• uptake of ¹⁸F-FDG is influenced by tumour blood flow, blood glucose level, glucose consumption, and activity of glucose transporters and hexokinase (Dandekar et al. 2010)

• relatively low specificity

• poor spatial resolution

• radiation dose

• lack of a standardised method of signal segmentation

• does not provide accurate information on external and internal body contours (Daisne et al. 2003)

4 Applications of Imaging Data in Radiation Oncology

Imaging of tumours and surrounding normal tissues is used for several purposes:

• a. diagnosis and staging

• b. RT planning

• c. treatment verification

• d. response prediction

• e. follow-up

4.1 Diagnosis and Staging

Conventional staging includes clinical examination, US, examination under anaesthesia, endoscopy, CT, and MRI to determine size and location of the primary tumour, extent of LN involvement, and to exclude distant metastases and concurrent second primaries.

4.1.1 Primary Tumour and Cervical Lymph Nodes

US can be used as a diagnostic tool for unexplained neck masses. In 42 patients who ultimately had a confirmed tissue diagnosis of HNC, the sensitivity and specificity of US was 96.8 and 93.3% respectively (Hwang et al. 2009).

¹⁸F-FDG PET might be superior to CT or MRI at detecting involved cervical LN (Laudenbacher et al. 1995; Adams et al. 1998). The average sensitivity and specificity of PET is 84 and 96%, respectively, compared to 69 and 68% for CT or MRI. However, other studies have suggested that ¹⁸F-FDG PET offers little additional information (Stoeckli et al. 2002; Benchaou et al. 1996). Addition of ¹⁸F-FDG PET/CT to CT and MRI did increase the specificity (86 vs. 60%) and sensitivity (89 vs. 62%) for the detection of pathological retropharyngeal LN (Chu 2009). PET is limited in its ability to detect small lesions. Brink et al. (2002) showed a sensitivity of 83% for nodes larger than 1 cm³, but for smaller LN the sensitivity dropped to 71%.

A recent meta-analysis included 1,236 analysable patients from 32 retrospective and prospective studies investigating ¹⁸F-FDG PET scans used for neck staging (Kyzas et al. 2008). The authors excluded from consideration studies with a risk of verification bias, those with planned neck dissection, those without histologic correlation, and those including patients being investigated for recurrence. PET interpretation was both qualitative and quantitative and in only five studies was it performed in a blind fashion. All included studies which used single-modality PET scanners rather than integrated CT/PET. The overall sensitivity and specificity (79 and 86%, respectively) was slightly but not statistically higher than conventional imaging. In patients with clinically negative necks (N = 311), sensitivity of PET was 50%, and not statistically better than conventional imaging. The method of PET interpretation did not impact diagnostic accuracy (Kyzas et al. 2008).

For imaging of the primary tumour, as reviewed by Troost et al. (2010a) PET has an overall sensitivity of 93–100%, specificity of 90–100%, and accuracy of 94–98%. For patients with histologically-proven squamous cell carcinoma in a neck LN of unknown origin despite conventional workup, PET can detect the primary in up to 25% of cases (Troost et al. 2010a).

4.1.2 Pathoradiologic Correlation Studies

Eighteen patients with stage II–IVB or recurrent HNC underwent ¹⁸F-FDG PET and contrast-enhanced CT followed by surgical resection within 1 month (Burri et al. 2008). PET scans were reviewed in a blinded fashion and no baseline MRIs were performed. Three-dimensional pathologic data were available for 12 tumours. Mean tumour standardised uptake value (SUV) was 14.2 (range 2.6–43.3). PET had a sensitivity of 94% for primary tumours (compared to 82% for CT) but correctly identified five tumours missed by CT. PET sensitivity was 90% for neck disease compared to 67% for CT; specificities were identical at 78%. Although multiple values were investigated, a SUVmax threshold of 40% or greater showed the best compromise between accuracy and risk of underestimating disease extent, and also correlated best with pathologic volumes (Burri et al. 2008).

Three protocols for staging of HNC were compared in 44 patients who underwent surgical resection of the primary and neck dissection within 6 weeks of imaging (Rodrigues et al. 2009). The imaging protocols were contrast-enhanced CT, whole body ¹⁸F-FDG PET/CT, and dedicated head and neck high-resolution contrast-enhanced ¹⁸F-FDG PET/CT. There was no statistical difference between the two PET protocols, but both were significantly better than CT in evaluation of the primary tumour. The dedicated head and neck PET outperformed whole body PET in detection of cervical LN metastases, particularly those <15 mm (Rodrigues et al. 2009).

In a landmark study, Daisne et al. (2004) compared co-registered CT, MR, and ¹⁸F-FDG PET scans in 29 patients with HNC, of whom nine (all T4 larynx) underwent radical surgery. The surgical specimen was also coregistered with the images in an internally validated technique accurate to within 2.1 mm. Gross tumour volumes (GTVs) were retrospectively delineated on CT and MRI by one investigator and reviewed by another. For PET scans, GTVs were delineated automatically using a signal-to-noise ratio segmentation algorithm. Gross tumour infiltration in the surgical specimen was determined by a pathologist without knowledge of imaging results. The authors found that the average GTV_PET was significantly smaller than GTV_CT or GTV_MRI; however, GTV_PET were not totally encompassed by that of the other two modalities. CT and MRI volumes were not significantly different in size, but did not completely overlap. In comparison with the reference surgical specimen, all imaging modalities overestimated actual tumour size with GTV_PET closest to reality. CT volumes were up to 107% larger, and PET overestimated by up to 46%. Areas visualised best by CT were the peritumoural soft tissues as well as tumour extension in the region of the pharyngeal wall. MRI best depicted infiltration of muscle, cartilage, extralaryngeal tissues and the parapharyngeal space. All three imaging modalities failed to identify a small fraction of macroscopic tumour, approximately 10%, mainly superficial mucosal or extralaryngeal extension (Daisne et al. 2004).

Alternative PET tracers investigated for HNC staging include 0-2-fluro-(18F)-ethyl-l-tyrosine (FET). In one such study, Balogova et al. prospectively compared the diagnostic performance of ¹⁸F-FDG versus ¹⁸FET-PET/CT. ¹⁸FET-PET had significantly improved specificity (100 vs. 63%) but lower sensitivity (64 vs. 95%). The authors concluded that based on the sensitivity, ¹⁸FET was not a suitable tracer for HNC staging (Balogova et al. 2008).

Diffusion-weighted MRI detects differences in tissue microenvironment due to random displacement of water molecules. This movement, occuring between pairs of opposing magnetic field gradients, is detectable as signal loss proportional to the amount of movement and the strength of the gradient. These differences are quantified as apparent diffusion coefficients (ADC) which are inversely correlated with tissue cellularity. From 2004 to 2006, 33 patients scheduled for surgical treatment of biopsy-proven HNC underwent 1.5 T turbo-spin echo (TSE) MRI and DW-MRI the day before surgery (Vandecaveye et al. 2009). DW-MRI images were read a mean of 18 months after surgery in a blinded fashion. The specimen was progressively dissected then matched to the TSE MRI. Nodal stage was determined by consensus between a pathologist, radiologist, and radiation oncologist. Six hundred and fifty LNs were identified in the surgical specimen, 301 identified by DW-MRI and 76 were metastatic. The majority of nodal metastases showed no distinct radiologic morphologic abnormalities. TSE MRI had a sensitivity of 46%, specificity of 96%, and an accuracy of 83% for nodal disease, with DW-MRI sensitivity 84%, specificity 94%, and accuracy 91%. Neither of the MRI protocol could detect LN metastases smaller than 4 mm. The ADC was significantly lower for metastatic than for benign LN likely due to the hypercellularity, cellular polymorphism and increased mitoses of malignancy (Vandecaveye et al. 2009). It must be kept in mind that a false decrease in ADC may be due to nodal reactive changes (germinal centres and fibrotic stroma) acting as microstructural barriers. These changes may be most apparent in small LN close to metastatic ones due to a tumour-induced immunologic response (Vandecaveye et al. 2009). The high negative predictive value of DW-MRI for lymphatic metastatic disease may help decision-making in the setting of planned neck dissection or determination of RT treatment volume.

4.1.3 Distant Metastases and Second Primaries

¹⁸F-FDG PET may detect previously unrecognized distant metastases or second primary malignancies (Hustinx and Lucignani 2010). Senft et al. (2008) investigated the added value of ¹⁸F-FDG PET in screening HNC patients at high risk for distant metastases (multiple or bilateral LN, low-neck LN, LN >6 cm, tumour recurrence, or second primary tumours). Of 92 evaluable patients, the mean age was 59 and more than 50% had either oropharynx or oral cavity primaries. All underwent chest CT and PET and the gold standard was clinical status after 12 months’ follow-up. Each imaging modality was interpreted without knowledge of the other. Thirty-one percent of patients with a negative CT chest at baseline developed distant metastases or a second primary by 12 months. Fifteen percent of patients with negative screening PET developed distant metastases and 2% a second primary within 12 months. Compared to CT, PET had a higher sensitivity with a slightly higher negative predictive value and accuracy, while the combination of both had the highest sensitivity (63%). The authors’ concluded that screening with both modalities should be performed (Senft et al. 2008).

Haerle et al. (2010) retrospectively reviewed ¹⁸F-FDG PET/CT results in 311 patients of mixed HNC subsites undergoing staging panendoscopy between 2002 and 2007. Ninety percent had stages III or IV disease and underwent PET primarily to screen for distant metastases. The gold standard was histologic information (when available) or clinical status after 6 months’ follow-up. Panendoscopy detected second primaries in 4.5% compared with 6.1% detected by PET within the coverage area of endoscopy. All lesions detected by endoscopy were also detected by PET, which found five additional malignancies. Sensitivity was 100% for PET/CT versus 74% for panendoscopy; specificity was 96.5 and 99.7%, respectively. The authors felt that PET was not cost-effective in early-stage disease, as panendoscopy was sufficient due to its high accuracy. In patients with advanced HNC, ¹⁸F-FDG PET is recommended to routinely exclude distant metastases. But with advanced disease and a negative PET, endoscopy could be restricted to the area of the primary tumour (Haerle et al. 2010).

4.2 Radiotherapy Planning: Anatomic Information

In treatment planning, it is essential to accurately differentiate tumour from normal adjacent tissue. Evidence is growing that complementary morphological information from imaging such as MRI can help decrease variability in tumour delineation.

4.2.1 CT

CT is routinely used for delineation of tumour volumes and is considered to be the gold standard at present. CT-based delineation of metastatic LN is usually less prone to error than that of the primary disease due to differentiation from surrounding fatty tissue (Troost et al. 2010a). However, both inter- and intra-observer variability in volume delineation exists. Hermans et al. investigated this variability in CT studies of 13 laryngeal tumours reviewed by five different observers, each of who repeated the delineation exercise four times. Both inter-, and to a lesser extent, intraobserver variability had a statistically significant effect on resulting volumes. The most experienced participant obtained the most stable mean tumour volume (Hermans et al. 1998). The authors concluded that variability can be reduced by having a single experienced reader to perform all delineations. In practice, however, each physician typically performs the RT planning for their own patients, so several groups have published guidelines to facilitate reproducible delineation of primary tumours and LN regions (Gregoire et al. 2003; Chao et al. 2002; Eisbruch et al. 2002).

4.2.2 MRI

Because of the superior soft-tissue contrast of MRI, it is the preferred imaging modality to demonstrate oral cavity, nasopharyngeal and oropharynx malignancies, for example (Troost et al. 2010a). Rash et al. studied the potential impact of the combined use of CT and MRI on volume delineation in advanced HNC. Four participants outlined the GTV in six patients with cancers involving the base of skull on CT, as well as axial, coronal, and sagittal MRI. GTV_MRI were on average 30% smaller, with less interobserver variation than CT-derived GTVs (Rash et al. 1997). Enami et al. examined MRI and CT in eight nasopharyngeal patients. Compared with CT, MRI-based targets were 74% larger and more irregularly shaped. On average, composite CT + MRI GTVs were 10% larger than GTVs drawn from MRI alone. The authors concluded that fusion of MRI and CT images would significantly reduce the possibility of geographic miss in this tumour site (Enami et al. 2003).

4.3 Radiotherapy Planning: Biological Information

4.3.1 ¹⁸F-FDG PET

Besides the location, size and extent of the tumour, knowledge about biological features is useful in the management of HNC. Potential roles of ¹⁸F-FDG PET include: reduction of inter-observer variability in tumour delineation, reduction of the size of the GTV, identification of tumour extension missed by CT or MRI, decision-making in the case of marginally enlarged LN, the identification of parts of the tumour potentially benefiting from additional radiation dose, and adaptation of the treatment plan during RT (Troost et al. 2010a). Biological or molecular techniques may allow improved detection of micrometastases either directly, or as a result of better assessment of the characteristics of the primary (Mahfouz et al. 2010).

The definition of “metabolic” or “biological” tumour volume remains a challenging one, however, and there is no consensus regarding the optimal method for segmenting a metabolically active lesion (Hustinx and Lucignani 2010). In order to use ¹⁸F-FDG PET to optimally delineate volumes, the way the PET images are interpreted must be known. The best method of PET interpretation for RT applications is unknown at this time and options include:

• visual (qualitative)—generally leads to larger volumes, is operator-dependent, and susceptible to window–level settings

• SUV-based methods—percent maximum peak SUV, fixed arbitrary SUV, threshold SUV, fixed percent SUV relative to the maximum activity in the tumour (usually 40–50%)

• background cutoff (defined with respect to the background FDG signal)—can be automated, and may be affected less by heterogeneity of lesion tracer uptake

• automated or semiautomated methods—provide consistent target delineation with a significant saving in contouring time

The impact of PET imaging on target delineation in RT for HNC has recently been investigated. Syed et al. (2005) examined the impact of ¹⁸F-FDG PET/CT in 24 patients and concluded that PET/CT significantly increases interobserver agreement in disease localisation. Scarfone evaluated the influence of ¹⁸F-FDG PET on tumour volumes of six patients which were delineated on CT and then modified based on PET data. The resulting GTV_PET was larger than the GTV_CT by an average of 15%. The final GTV_PET LN volume was on average 17% larger (Scarfone et al. 2004). Koshy et al. also used PET in 36 patients with HNC as part of their RT planning. In 14%, PET/CT fusion altered RT volumes, and dose was altered in four patients (Koshy et al. 2005). PET–CT detected 39 positive nodes in contrast to 28 detected by clinical examination, CT, or MRI in a study by Nishioka et al. In four patients, the nodal stage increased, which impacted on target delineation. Parotid-sparing became possible in 71% of patients whose neck was determined to be tumour-free on PET, and, except for one patient, no recurrences were seen 18 months after treatment of the GTV_PET (Nishioka et al. 2002).

Schinagl compared co-registered ¹⁸F-FDG PET with CT for assessment of LN in 78 patients. Of 108 nodes classified on CT as enlarged (shortest axial diameter of >10 mm), 75% were also identified as enlarged by PET using visual interpretation, and between 43 and 59% by other PET segmentation methods. Of 100 LN classified as marginally enlarged (7–10 mm), only a minority were visualised by PET. Volume and shape of the resulting GTVs were influenced heavily by the method of segmentation. All automated segmentation methods yielded significantly smaller GTVs than those based on clinical information and CT; visual interpretation yielded volumes closer to that of GTV_CT. Depending on the segmentation tool used, more than 20% of the GTV_PET was located outside the GTV_CT. However without a histologic gold standard, it is possible that PET uptake in this area could be a result of peritumoural inflammation (Schinagl 2009).

Studies addressing integration of PET into IMRT treatment planning, recently reviewed by Troost et al. (2010a), are encouraging, although tend to be retrospective, small and heterogeneous, with short follow-up and historical controls. Paulino et al. compared the GTV identified on CT to that obtained from ¹⁸F-FDG PET in 40 patients with HNC treated with IMRT. The GTV_PET was smaller, the same size, or larger than the GTV_CT in 75, 8, and 18% of cases, respectively. In approximately 25% of patients, tumour would have been underdosed if PET had not been performed. The authors recommended use of both CT and PET in determining the GTV for IMRT in head and neck malignancy (Paulino et al. 2005).

Yu et al. used a co-registered multimodality pattern analysis segmentation system (COMPASS) to automatically delineate radiation targets in HNC using ¹⁸F-FDG-PET and CT. This innovative system extracts textural features such as coarseness and contrast from PET and CT voxels, each of which is labelled “normal” or “abnormal”. The COMPASS system was applied to the images of ten patients and results compared with those of three PET threshold-based methods (SUV of 2.5, 50% maximal intensity and signal/background ratio). The gold standard was considered to be the manual contouring of three radiation oncologists. The COMPASS tumour delineations were more similar to those of the physicians than those based on other methods, especially in regions adjacent to tissues with high PET uptake. Validated automated segmentation methods may in future decrease inter-observer variability and uncertainty in target delineation (Yu et al. 2009).

4.3.2 Dose Painting

Most treatment failures occur in the high-dose RT volume, suggesting that the dose delivered is insufficient to ablate malignant cells. Dose escalation to the entire target volume is not possible, however, due to normal tissue toxicity. The term ‘dose painting’ implies non-uniformity within a target region, to direct dose to subregions of tumour in order to improve local control. IMRT has the ability to deliver non-uniform dose distributions, but the question is, which areas should be targeted? Focusing a high-dose boost to a BTV, the hypermetabolic area on a PET scan could improve response rates (Ling et al. 2000). Likewise, eliminating PET-negative regions from prophylactic radiation volumes could reduce the dose to OARs. Therefore, imaging methods are being developed to provide biological information regarding tumour hypoxia, proliferation, apoptosis, angiogenesis and receptor status, and all aspects of the tumour microenvironment relevant to radiation resistance (Chapman et al. 2003; Bussink et al. 2010).

4.3.3 Hypoxia

Hypoxic cells are 2.5–3 times more resistant to radiation than well-oxygenated cells and hypoxia is therefore a major determinant of tumour response to RT. A variety of methods have been used to measure oxygenation of tumours including needle electrodes (Nordsmark and Overgaard 2004). This technique has several drawbacks including invasiveness, lack of discrimination between necrotic and viable hypoxic tissue, sampling error, and requirement of accessibility. Non-invasive techniques to identify hypoxia have recently sparked interest in terms of suggesting areas which could benefit from dose escalation (Troost et al. 2010a).

Tumour perfusion rate was investigated by Hermans et al. with dynamic CT in 105 patients treated with RT. Intravenous contrast was rapidly injected while dynamic axial data acquisition was performed at the level of the largest tumour dimension. Perfusion rate was calculated using the time–density curve of the tumour and the maximal arterial density. When patients were stratified according to median perfusion value, those with lower perfusion rates had a significantly higher local failure rate (Hermans et al. 2003).

⁶²Cu-diacetylbis-N-4-methyl-thiosemicaarbazone (Cu-ATSM) has been investigated as a PET marker for hypoxia because of its rapid uptake, activity ratio, rapid blood clearance, and washout from normally oxygenated cells. Chao et al. examined the feasibility of dose escalation to areas identified as hypoxic by Cu-ATSM PET in an IMRT planning study. Based on co-registration of Cu-ATSM PET to CT images, 80 Gy total dose in 35 fractions could be safely delivered to the hypoxic target volume, and 70 Gy/35 to the remainder (Chao et al. 2001).

¹⁸F fluorinated misonidazole (¹⁸FMISO) is reduced and incorporated into cells under hypoxic conditions and has also been investigated as a PET tracer (Koh et al. 1992; Chapman et al. 1983). Jansen explored the microenvironment of neck LN metastases in HNC patients, examining the relationship between tumour perfusion measured by dynamic contrast-enhanced MRI (DCE-MRI) and hypoxia measured by ¹⁸FMISO-PET. Matched regions of interest from both modalities were analysed in 13 newly diagnosed patients. The authors found a strong negative correlation between perfusion parameters and ¹⁸FMISO SUV, supporting the hypothesis that metastatic neck LN are poorly perfused and therefore likely relatively hypoxic (Jansen et al. 2010).

Eschmann et al. (2005) compared ¹⁸FMISO-PET uptake with clinical outcome after treatment. Twenty-six patients with HNC were scanned before RT, and ¹⁸FMISO uptake was quantified using SUV and tumour-to-background ratios. Patients with local recurrence could be separated from disease-free patients by the SUV 4 h after injection, at which timepoint all recurrences had SUV >2. All patients with a tumour-to-muscle ratio >1.6 experienced tumour recurrence. The authors also reported temporally inconsistent ¹⁸FMISO distribution in that the location of regions which met the criteria for hypoxic changed with time.

The potential of ¹⁸F-FDG-PET, ¹⁸FMISO-PET, DW-MRI and dynamic contrast-enhanced (DCE)-MRI to provide a potential BTV for dose painting was evaluated by Dirix and colleagues. After intravenous injection of a paramagnetic contrast agent, DCE-MRI shows tissue signal intensity increase on T1-weighted scans (Dirix et al. 2009). Changes in signal intensity on DCE-MRI are related to tumour permeability, perfusion and interstitial pressures, which may influence treatment response. Fifteen patients with locally advanced HNC participated in a study of sequential CT/PET with ¹⁸F-FDG PET, ¹⁸FMISO PET and 1.5 T MRI performed before, during, and after radical RT (Dirix et al. 2009). GTVs delineated by the treating radiation oncologist were retrospectively retrieved while the GTV_PET was automatically segmented based on source-to-background ratio. The GTV_MRI was manually delineated by a radiation oncologist and radiologist in consensus. There was an excellent correlation between CT- and MRI-based volumes, and all volumes showed significant shrinkage over the course of RT, by approximately 50% (Dirix et al. 2009). Both the GTV_PET and GTV_DW-MRI were significantly smaller than the GTV_CT. Over a median follow-up of 30.7 months, disease recurred in seven patients; all recurrences were located within the area of overlap of the GTV_CT, GTV_MRI, and GTV_PET. There was little residual hypoxia on follow-up ¹⁸FMISO PET scan during the fourth week of treatment. DW-MRI showed residual disease in three patients and all developed locoregional recurrence. At 8 weeks after the end of treatment, ¹⁸F-FDG PET suggested residual disease in two patients, both of whom ultimately recurred. DFS correlated negatively with baseline maximum tissue-to-blood ¹⁸FMISO ratio (T/Bmax), size of the hypoxic volume at baseline, and with T/Bmax on the ¹⁸FMISO scan during treatment. Three of the locoregional recurrences were outside the hypoxic volume defined by baseline ¹⁸FMISO PET. Compared with lesions that remained controlled, those which recurred had significantly lower ADC on DW-MRI during and after RT, and a significantly higher initial slope on baseline DCE-MRI. The authors caution that interpretation of DW- and DCE-MRI scans in this area is not straightforward, and suggest the use of quantitative or semiquantitative measurements for volume definition assistance (Dirix et al. 2009).

Newer generations of nitroimidazoles have also been studied, including ¹⁸F-labelled fluoroerythronitroimidazole (FETNIM), fluoroetanidazole (FETA), and 2-(2-nitro-1[H]-imidazol-1-yl)-N-(2,2,3,3,3-pentafluoropropyl)-acetamide (EF5). FETNIM and FETA are both stable in the context of non-oxygen-dependent metabolism and have higher tumour-to-background contrast secondary to increased hydrophilicity (Lehtio et al. 2001; Rasey et al. 1999; Yang et al. 1995). EF5 has been used for immunohistochemical detection of hypoxia, with a future possibility of correlating this with PET results (Ziemer et al. 2003). FETNIM has been used clinically to determine the oxygenation status of patients undergoing RT.

Temporal and geographic stability is a significant concern for hypoxia-based dose escalation (Troost et al. 2010b). Acutely this is related to microscopic changes in oxygenation caused by intermittent opening and closing of vessels (Chapman et al. 1983) (Fig. 4). Reoxygenation of regions which are hypoxic at baseline will likely occur over the course of curative-intent RT. Also unknown is what RT dose is required to eliminate hypoxic subpopulations (Troost et al. 2010b). More work is required to elucidate the evolution in intratumoural hypoxia before single hypoxia-tracer PET images can be used as a basis for dose escalation (Dirix et al. 2009).

Fig. 4

Temporal change in hypoxia. A 57-year-old patient presented with stage T3N2b squamous cell carcinoma of the oropharynx. Baseline ¹⁸F-FDG PET (a) shows uptake at the primary tumour (black arrow) and regional lymph node (grey arrow). ¹⁸FMISO PET acquired 1 day later (b1–2) indicates hypoxia in both the primary tumour (b-1 black arrow) and lymph node (b-2 grey arrow). Repeat ¹⁸FMISO PET after 4 weeks of radiotherapy (c) shows decreased ¹⁸FMISO uptake

4.3.4 Proliferation

Uncontrolled proliferation is one of the hallmarks of malignancy. Tumour subvolumes with evidence of high rates of proliferation may also benefit from dose escalation. The degree of metabolic activity detected by ¹⁸F-FDG may be a surrogate for tumour cell density as slowly proliferating tumours generally show less uptake than rapidly proliferating ones. ¹⁸F-FDG is prone to false positives, however, for the reasons outlined in Sect. 3. Therefore, more specific tracers have been investigated.

Radiolabelled nucleosides can be used to quantify DNA synthesis and radiolabelled amino acids can be used as markers for protein synthesis. In terms of nucleoside tracers, the most studied is ¹⁸F-labelled fluorothymidine (FLT), a pyrimidine analogue (Dandekar et al. 2010). FLT has high-specific activity and good tumour-to-background ratio (Shields et al. 1998). Cobben et al. studied ¹⁸FLT PET for visualisation of laryngeal cancer in 21 patients in comparison with ¹⁸F-FDG PET. While an equal number of tumours were detected with each modality, higher absolute uptake of ¹⁸F-FDG than ¹⁸FLT was reported (Cobben et al. 2004). Because less tracer is taken up compared to glucose uptake, sensitivity of FLT imaging may be relatively low (Buck et al. 2003). Troost et al. (2010b) attempted to use ¹⁸FLT PET for adaptive RT in oropharyngeal patients, but concluded that PET segmentation methods could not be directly extrapolated to ¹⁸FLT as the resulting volumes were unsatisfactory. ¹⁸FLT PET may be useful in future to direct RT to primary tumour sites as it is not influenced by peritumoural inflammation. However, it is unlikely that ¹⁸FLT neck imaging will be practical as false positives are caused by uptake in germinal centres of metastatic LN.

The most studied amino acid tracers are ¹¹C-methionine (MET) and ¹¹C-tyrosine, which are surrogate markers of high rates of protein metabolism (Dandekar et al. 2010). Amino acid tracers may better discriminate between tumour and inflammation because inflammatory cells have lower protein than glucose metabolism (Kubota et al. 1995). Geets et al. studied the role of MET PET for delineation of tumour volume in pharyngo-laryngeal squamous cell carcinomas. Twenty-three patients were imaged with CT, ¹⁸F-FDG PET, and MET PET before treatment (RT or surgery). MET PET volumes did not differ from CT volumes, while the ¹⁸F-FDG volumes were significantly smaller. The authors concluded that MET PET does not provide additional value information over CT, probably because of the confounding effect of high-MET uptake by the normal mucosa and salivary glands (Geets et al. 2004).

Another indicator of proliferation is an increased cellular choline level. ¹H MR-spectroscopy can be used to estimate choline levels in human tumours. This procedure has successfully localised active disease in other cancer sites but data in HNC have not yet been reported.

4.3.5 Apoptosis

Apoptosis can be an indicator of intrinsic radiosensitivity because it is a major pathway of cell death after ionising irradiation. Annexin V is a protein which binds to membrane-bound phosphatidyl serine and is exposed on the surface of cells undergoing apoptosis. ^99mTc-radiolabeled annexin V is therefore being tested to quantify apoptosis (Green and Steinmetz 2002). Van de Wiele et al. (2003) performed quantitative tumour apoptosis imaging using ^99mTc-radiolabelled annexin V and SPECT in 20 patients planned for surgical resection. Quantitative annexin V tumour uptake values correlated well with the number of apoptotic cells on assays for apoptosis-induced DNA fragmentation, provided samples had minimal necrosis. The authors concluded that this method can be used to monitor treatment response, if treatment does not alter the diffusion-related uptake of peptides. ^99mTc-radiolabelled annexin V scintigraphy has been used to monitor radiation-induced apoptotic cell death in other cancer sites, but has not yet been used in the setting of HNC.

4.3.6 Receptor Status

Proliferation of many tumours is regulated by factors that bind to membrane or intracellular receptors to activate signal transduction pathways. Imaging of receptor status therefore might provide information that can be used to guide treatment and prognosis. For example, epidermal growth factor receptor (EGFR) is a transmembrane glycoprotein involved in activating several pathways associated with proliferation, migration, stromal invasion, angiogenesis, and resistance to cell death-inducing signals (Dancey 2004). EGFR is overexpressed in 80–100% of HNC (Grandis et al. 1996), and this overexpression is associated with both increased metastatic potential and poor prognosis (Ang et al. 2002). Contrast agents to visualise EGFR are under development, such as radiolabelled cetuximab.

In summary, functional imaging to guide RT planning is promising, but still needs to be validated through clinical studies before it can be incorporated in daily practice. Outcome data on this approach are pending at this time and more work remains before most of the tracers discussed can be used for patient treatment decisions.

4.4 Treatment Verification

Treatment verification refers to the use of imaging to ensure that RT fields are appropriately localised throughout the entire treatment period, taking into consideration factors such as patient immobilisation and positioning. Commonly in head and neck RT, masks or shells are custom-made for the patient’s anatomy and are then fixed to the table to prevent movement during treatment (intrafraction) and on a daily basis (interfraction). This ensures the ability to re-establish the patient’s position each day, which can be affected by several factors: changes in tumour size and shape; organ movement such as respiration or swallowing; anatomic variation; and patient set-up errors.

Image-guided radiotherapy (IGRT), defined as the use of on-board imaging to improve patient set-up accuracy (Castadot et al. 2010), allows verification of the correct position of the RT field prior to treatment initiation. There are several ways to accomplish this. Traditionally, a ‘port film’ used the portion of the accelerator beam exiting from the patient to expose commercially available radiographic film. This has less diagnostic quality but it is sufficient for comparison to planning images for gross errors. The next generation is called electronic portal imaging (EPI), which uses fluorescent screens, 2D ion chambers, or matrix flat panel imagers and are again compared with a reference image. These 2D online electronic images have made it possible to correct patient positioning in real time, just before radiation is delivered (Fig. 5). On-board cone beam CT has recently been introduced, which allows 3D verification of fields in relation to other structures, such as OARs. To create these images, an extra tube is mounted on the linear accelerator producing a divergent X-ray beam, which is captured using a portal detector opposed to the CT tube. Both make a 360° rotation around the patient. One study by Han et al. (2008) evaluating the impact of daily helical megavoltage CT-guided set-up correction reported that the median dose to critical normal structures (spinal cord and parotid glands) increased significantly when daily corrections were not applied. Techniques of adaptive RT designed to further customise treatment by minimising both systematic and random positioning errors are well reviewed by Castodot et al. (2010).

Fig. 5

Comparison between digitally reconstructed radiograph derived from CT simulation images (a), and a port film, obtained during irradiation (b),which are compared to verify patient positioning and size and shape of the treatment portal

4.4.1 Adaptive Radiotherapy: 4D

As alluded to above, it is common for tumours and to some extent OARs to change over the course of curative RT. Tumour and nodal volumes shrink by up to 3% per day, changing size, shape, and position, sometimes asymmetrically. External face and neck contour modifications may be seen as patients lose weight and muscle mass. This further alters the anatomy and geometry of the disease in relation to critical normal structures. Parotid glands not only decrease in volume but also shift medially into the high-dose region with time (Castodot et al. 2010; Barker et al. 2004). Resolving post-operative changes or even disease progression during treatment can also be seen. Spatiotemporal instability of the target and normal structures and/or geometric uncertainty of patient positioning are critical in IMRT because of the sharp dose gradients involved (Castadot et al. 2010). The result of these changes is that location and dose delivered may differ significantly from what was planned.

Four-dimensional (4D) RT is the next generation in conformal treatment, which attempts to account for these changes over time. Latest generation linear accelerators are now equipped to visualise soft-tissue structures with CT while fluoroscopy tracks the breathing cycle in real time. Changes are mapped over the period of RT and treatment volumes are changed accordingly in “adaptive” or response-based therapy. Significant alterations may even involve revising the entire radiation plan (Dandekar et al. 2010).

Megavoltage cone beam CT (MVCT) allows the reconstruction of the actual delivered dose based on a patient’s specific anatomy in real time (Pouliot et al. 2003; Ghilezan et al. 2004). Acquiring a low-dose MVCT image on the treatment machine immediately prior to therapy allows online verification of patient position. The same imaging device also records the portal dose during treatment. By converting this portal dose to primary fluence at the plane of the detector and backprojecting through the MVCT model of the patient, the 3D dose delivered at the time of treatment can be obtained (Welsh et al. 2002). This may allow modulation of the next day’s RT dose based on the actual dose received to that point, as opposed to delivering the previously planned daily dose (Welsh et al. 2002). Optimisation in this manner could compensate for dosimetric alterations caused by anatomical or positional changes during treatment (Castadot et al. 2010).

Ten patients with stage III/IV pharyngolaryngeal tumours receiving SIB chemoradiotherapy were imaged using contrast-enhanced CT and ¹⁸FDG-PET at baseline and then weekly during weeks 2–5 to investigate whether imaging reassessment influenced dose delivery (Geets et al. 2007). Eight patients also underwent MRI. Volumes were manually delineated by one investigator. PET-based volumes were delineated using an in-house validated gradient-based segmentation method. The mean primary tumour GTV significantly decreased over the course of treatment (P < 0.001). Functional imaging primary tumour GTVs were always smaller than those based on anatomical imaging. There was no clinically relevant difference between CT- and MRI-based volumes. By completion of 45 Gy, mean primary tumours had decreased by 54–70% compared to baseline, with all imaging modalities suggesting similar rates of reduction. PET-based adaptive IMRT produced significantly tighter dose distributions (i.e. 15–40% reduction of the volume of tissue receiving >60 Gy), but this did not significantly reduce OAR doses. One challenge with use of PET during RT, especially after the fourth week, is that images become more difficult to interpret due to inflammation of the tumour and normal tissue (Geets et al. 2007). While a potential advantage of adaptive RT is the compensation for underdosage of target volumes or overdosage of OARs, because of extra staff workload and cost, the optimal implementation strategy remains to be defined (Castadot et al. 2010).

4.5 Response Prediction Using Biological Imaging

Recognition of nonresponders or early progressors after RT is essential as these patients might benefit from intensified treatment or early surgical salvage (Farrag et al. 2010). The time window for successful salvage surgery is unfortunately limited because operability is determined not only by the likelihood of obtaining oncologic control but also by surgical reconstructive options available (Vandecaveye et al. 2007).

4.5.1 Prediction During Therapy

Farrag et al. investigated whether ¹⁸F-FDG PET during treatment can predict outcome for HNC patients treated with radical RT ± chemotherapy. Over three years, 43 consecutive patients were evaluated by history, examination, lab tests, panendoscopy, CT, and MRI. All patients also underwent PET at baseline and at the end of the fourth week of therapy (approx 47 Gy). Scans were analysed by two physicians with access to complete clinical information. Sixty-three percent of patients had either hypopharynx or oropharynx cancer, 37% received cisplatin-based chemotherapy, and 44% had T3 or T4 disease. RT was given via a SIB technique to a total dose of 66–70.5 Gy to the primary disease and involved LN. Median follow-up was 12.7 months. Two-year overall and disease-free survival was 66 and 52%, respectively. Median SUVmax decreased from baseline to follow-up PET. Both low SUVmax at baseline and on follow-up PET were significantly correlated with overall survival. There was a non-significant trend for patients with a complete metabolic response at follow-up PET to have better outcomes (Farrag et al. 2010).

Baseline and mid-treatment ¹⁸FMISO PET/CT scanning has also been prospectively investigated (Lee et al. 2009). Twenty HNC patients, 90% with locally advanced oropharyngeal primaries, underwent concurrent chemotherapy and IMRT. All had pre-treatment ¹⁸F-FDG and ¹⁸FMISO-PET/CT scans, with a mid-treatment ¹⁸FMISO PET scan 4 weeks after the start of treatment. However, the authors found that ¹⁸FMISO findings on the mid-treatment scan did not correlate with patient outcome (Lee et al. 2009).

Hentschel et al. used ¹⁸FDG-PET to adapt RT based on the evolution of tumour metabolic activity. While the median SUVmax decreased, in this study the median sizes of GTV_PET and metabolic volumes actually increased during RT, likely as a result of therapy-induced inflammation. The authors concluded that treatment volumes cannot be reduced through PET-based adaptive replanning in this manner, probably due to the use of this particular type of algorithm (Hentschel et al. 2009).

4.5.2 Prediction After Therapy

After completion of RT, ¹⁸F-FDG PET may assist in detection of residual viable tumour. This may be particularly useful in patients with persistently enlarged LN, in whom neck dissection may be considered (Hustinx and Lucignani 2010). However, scan interpretation after therapy is difficult because of the altered glucose consumption, necrosis, and infiltration of inflammatory cells within irradiated tissue (Farrag et al. 2010). The appropriate time to image patients post-therapy is not yet clear. Some authors have suggested that an interval of up to 12 weeks may be required before cell debris is sufficiently cleared (Bussink et al. 2010). However imaging done too late may miss the window of opportunity for salvage (Murphy et al. 2010). The reader is referred to “Positron Emission Tomography in Head and Neck Cancer” for more detail.

A retrospective study of DW-MRI in response prediction included 38 patients with pathologically confirmed HNC treated to at least 60 Gy (Hatakenaka et al. 2010). Most received concurrent chemotherapy. Salvage treatment (neck dissection and/or intra-arterial chemotherapy) was ultimately delivered to eight patients. Pretreatment DW-MRI was performed a median of 7 days prior to initial treatment. The primary lesion region of interest was determined by consensus between two radiologists and one radiation technologist without knowledge of locoregional disease status. Tumour volume, T stage, treatment, and ADC correlated significantly with local disease status. ADC had a sensitivity of 77.8%, specificity of 100%, PPV of 100%, and NPV of 80% for the prediction of local failure in stage T3/T4 disease. Low ADC values pretreatment was correlated with high local control. Kim et al. (2009) reported the usefulness of pretreatment ADC for predicting treatment response in 33 cases. They calculated the ADC values of the neck nodes, but not from the primary lesions. Pretreatment ADC value of complete responders was significantly lower than that from partial responders. The authors also performed a repeat MRI after 1 week of RT. The patients with a complete response showed a significantly higher increase in ADC than the partial responders by the first week of chemoradiotherapy. These data suggest that ADC can be used as a marker for prediction and early detection of response to therapy. In general, tumours showing a high ADC increase have a better treatment response than those with little or no ADC increase. The authors speculated that the apoptosis and necrosis of responding tumours present fewer microstructural barriers, thereby increasing ADC.

DW-MRI has also been used prospectively to assess salivary gland function before and at a mean of 9 months after parotid-sparing RT in eight patients (Dirix et al. 2008). RT was delivered to a dose of 60–72 Gy. The mean total dose to the spared parotid gland was 20 Gy, below the 25 Gy required for functional sparing. There were no dose constraints for the other parotid or either submandibular gland; typically these glands received >40 Gy. Before RT, a biphasic response to stimulation was confirmed in both parotids of all patients, identical to the pattern seen in healthy volunteers (Thoeny et al. 2005). In unspared glands, ADC was significantly higher after RT, possibly due to fibrosis, necrosis, or both. In the spared parotid, ADC value after RT was not significantly different compared to before RT. A comparable response to stimulation as before RT was observed in the spared but not the unspared parotid (Dirix et al. 2008). This suggestion of preserved function in the spared gland was confirmed by low patient-rated xerostomia scores and salivary gland scintigraphy results. The authors hypothesise that based on this data, DW-MRI allows non-invasive assessment of salivary gland functional changes with sufficiently high spatial resolution that it could be used to improve models of dose-response relationships (Dirix et al. 2008).

4.6 Follow-Up

Long-term follow-up after radical RT for HNC aims to evaluate treatment response, manage toxicity, exclude persistent disease, detect locoregional recurrence, and monitor for distant metastases or second primary tumours (Manikantan et al. 2009). Follow-up also provides an opportunity for psychological support, symptom management, and rehabilitation of voice and swallowing. The lack of consensus in the literature on the optimum frequency of follow-up imaging reflects significant variation in clinical practice. Published surveillance protocols often lack a clear evidence base and can be resource-intensive (Manikantan et al. 2009).

Manikantan et al. reviewed the English literature (1980–2009) on follow-up with any imaging modality to compile recommendations for effective surveillance. Publications included chest X-ray, chest CT, ¹⁸F-FDG PET, endoscopy, and tumour markers. There are no randomised controlled trials on the subject although expert guidelines have been published by collaborative groups. Manikantan summarised 55 studies, producing a flow chart for suggested timelines of surveillance. The first visit should take place 4–8 weeks after treatment completion and include history and physical exam. Endoscopy is indicated in symptomatic patients, and a new baseline CT or MRI should be obtained 3–6 months after RT. Thyroid function studies should be performed after any RT to the lower neck. PET should only be done if there is a discrepancy between physical findings and other imaging. Untreated clinically negative necks should be followed with examination and US-guided FNA when necessary. Chest CT should be performed in symptomatic patients and those with suspected lung metastases or second primaries. The authors recommend more intense surveillance in the first 3 years and for patients in whom a salvage option is available. There are no data proving greater efficiency for higher cost strategies (Manikantan et al. 2009). A second literature survey on ¹⁸FDG PET in HNC indicates that when compared with CT, PET has a higher sensitivity (93 vs. 54%) and specificity (83 vs. 74%) for diagnosing recurrences, and higher sensitivity (84 vs. 60%) and specificity (95 vs. 39%) for monitoring effects of therapy (Gambhir et al. 2001).

However, analysis of follow-up imaging is not always straightforward. Tissue changes after RT may be striking as they evolve over time (Glastonbury et al. 2010). The complexities of image interpretation after RT were pictorially reviewed by Glastonbury et al. Radiation-induced changes may mask or mimic residual disease. Acute RT-related changes include mucositis, interstitial oedema, ischaemia, and inflammation. Late side effects are associated with fibrosis and atrophy and can appear years after treatment. Severe but uncommon late effects include osteoradionecrosis and aerodigestive tract stenosis. Deep ulcers, new enhancing masses, lymphadenopathy, or bone or cartilage destruction must be considered recurrence until proven otherwise (Glastonbury et al. 2010). Dependence on size criteria limits the use of anatomical imaging in the detection of residual subcentimeter nodal disease and in definitive exclusion of tumour in persistently enlarged LN (Vandecaveye et al. 2007).

Emerging reports suggest that DW-MRI may help to distinguish persistent tumour from post-RT changes. RT-associated tissue changes are expected to show lower cellularity (and therefore higher ADC), compared to tumour which is highly cellular (with lower ADC) (Vandecaveye et al. 2007). Twenty-six patients with suspected tumour recurrence after chemoradiotherapy planned for salvage surgery were prospectively imaged. DW-MRI and TSE sequences were performed the day before surgery in all patients; most also underwent contrast-enhanced CT and some had panendoscopy, biopsy ,and/or ¹⁸F-FDG PET. Pathology was considered the gold standard (surgical specimen in 23, biopsy in 3). The ADC was measured in each region with suspected recurrence, in normal-appearing LN and normal-appearing but irradiated head and neck tissue. Fifteen of nineteen suspected lesions were positive for malignancy and all surrounding tissues showed RT-induced tissue changes or necrosis. ADC values were indeed significantly lower for tumour than for tissue with post-RT changes only. When compared with ¹⁸F-FDG PET (N = 17), DW-MRI correctly excluded tumour in three patients with hypermetabolic areas on PET, and correctly identified subcentimeter contralateral LN metastases without PET uptake in two. DW-MRI identified clinically relevant neck nodal metastases as small as 6 mm. Compared with CT, TSE, MRI and PETand , DW-MRI showed a low false positive rate and correctly charactersed non-tumoural persistent ulcers, persistently enlarged LN and inflammatory lesions. The authors concluded that DW-MRI is optimally interpreted in conjunction with anatomic imaging (Vandecaveye et al. 2007).

5 Conclusions and Future Challenges

No single imaging intervention is perfectly sensitive, specific, inexpensive, safe, and convenient (Manikantan et al. 2009). Anatomic and functional imaging provide complementary information which has already improved staging, RT delivery, response prediction, and follow-up. Knowledge about the capabilities and limitations of all imaging modalities is essential for their rational clinical use, especially in the setting of imaging-intensive investigational RT protocols such as adaptive therapy.

There remain hurdles to be overcome, including optimal timing of mid- and post-treatment imaging for each modality and accounting for spatiotemporal changes in tumour motion and hypoxia. Restrictions on PET use include the preferred method of segmentation and implementation of automated techniques. Many studies published to date have included small numbers of patients with a wide variety of disease sites, stages and treating institutions. This has the potential of confounding results due to differing image acquisition protocols, instrumentation and interpretation methods (Kyzas et al. 2008). Benefits of any imaging modality vary according to the local radiology experience available (Moeller et al. 2010). Interpretation of certain imaging modalities and especially imaging performed after high-dose RT may require special expertise.

Use of emerging technologies may address current challenges. For example, DW-MRI avoids the contrast required by dynamic MRI and CT that may have side effects, cost implications, and cannot be used in renal dysfunction (Hatakenaka et al. 2010). It may also be used as a non-invasive tool to select patients who could potentially benefit from treatment intensification such as dose painting (Dirix et al. 2009). Further development of imaging markers which can predict radioresistance or outcome could assist with prescription of RT based on tumour phenotype or genotype. Successful introduction of functional imaging into routine clinical practice will depend on continued clinical studies with sound methodology and adequate follow-up (Dirix et al. 2009).