The application of functional imaging techniques to personalise chemoradiotherapy in upper gastrointestinal malignancies.

Functional imaging gives information about physiological heterogeneity in tumours. The utility of functional imaging tests in providing predictive and prognostic information after chemoradiotherapy for both oesophageal cancer and pancreatic cancer will be reviewed. The benefit of incorporating functional imaging into radiotherapy planning is also evaluated. In cancers of the upper gastrointestinal tract, the vast majority of functional imaging studies have used (18)F-fluorodeoxyglucose positron emission tomography (FDG-PET). Few studies in locally advanced pancreatic cancer have investigated the utility of functional imaging in risk-stratifying patients or aiding target volume definition. Certain themes from the oesophageal data emerge, including the need for a multiparametric assessment of functional images and the added value of response assessment rather than relying on single time point measures. The sensitivity and specificity of FDG-PET to predict treatment response and survival are not currently high enough to inform treatment decisions. This suggests that a multimodal, multiparametric approach may be required. FDG-PET improves target volume definition in oesophageal cancer by improving the accuracy of tumour length definition and by improving the nodal staging of patients. The ideal functional imaging test would accurately identify patients who are unlikely to achieve a pathological complete response after chemoradiotherapy and would aid the delineation of a biological target volume that could be used for treatment intensification. The current limitations of published studies prevent integrating imaging-derived parameters into decision making on an individual patient basis. These limitations should inform future trial design in oesophageal and pancreatic cancers.

Statement of Search Strategies Used and Sources of Information

Four separate searches were completed on Ovid MEDLINE® in-process and other non-indexed citations and Ovid MEDLINE® 1994 to present. The searches targeted literature on: (i) oesophageal cancer, chemoradiotherapy (CRT) and functional imaging; (ii) pancreatic cancer, CRT and functional imaging; (iii) oesophageal cancer, functional imaging and target volume definition; (iv) pancreatic cancer, functional imaging and target volume definition. All English language abstracts were reviewed and unrelated articles were excluded. Trials of neoadjuvant chemotherapy alone or mixed cohorts of chemotherapy and CRT were excluded if separate analyses of these treatment modalities were not described. Studies were grouped into those that carried out functional imaging before CRT, before and during CRT, pre- and post-CRT and post-CRT only.

Introduction

The utility of functional imaging to predict chemoradiotherapy (CRT) treatment response and prognosis or to define target volumes for radiotherapy for upper gastrointestinal tumours remains uncertain. Functional imaging can provide information about the heterogeneity of physiological properties within tumours. Correlating functional imaging-derived parameters with treatment response and long-term treatment outcome may offer a means of risk-stratifying patients and ultimately guide treatment decisions. Certain physiological parameters are associated with resistance to radiotherapy.

Physiological processes that can be assessed with imaging techniques include glucose metabolism, cell proliferation, hypoxia, perfusion and water diffusion. 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET), which reflects glucose uptake and retention, is by far the most commonly used functional imaging test.

Both neoadjuvant CRT and definitive CRT are treatment options in oesophageal cancer. Definitive CRT has a 2 year local failure rate of around 50% [1-3] and most local failures occur within the gross tumour volume (GTV) [4]. A pathological complete response (pCR) is seen in 30% of cases after CRT [5-7]. If rates of pCR could be improved by image-guided treatment intensification, CRT followed by selective salvage oesophagectomy may become the preferred treatment. The early identification of non-responders would also define a group of patients who should proceed to early surgery.

Locally advanced pancreatic cancer (LAPC) has a poor prognosis, with a median survival ranging from 5 to 19 months [8]. The LAP07 trial has recently reported that CRT after induction chemotherapy confers no survival advantage compared with continuing with chemotherapy alone (overall survival 15.2 and 16.4 months, respectively) [9]. The failure of CRT to improve treatment outcome is, perhaps, a little surprising, given that for 25–29% of patients with LAPC, the first site of disease progression is at the site of the original tumour [10,11]. Escalating the radiotherapy dose to the pancreas seems attractive, but is limited by normal tissue toxicity, particularly in the duodenum [12]. If a method of identifying patients who have a high risk of local failure could be identified, a dose-escalation regimen that allows a higher rate of treatment-associated toxicity may be seen as worthwhile. After neoadjuvant CRT, those with <10% of viable tumour cells have a median overall survival of 39 months compared with only 15 months in those who have >10% of viable tumour cells remaining [13].

Accurate GTV definition is essential in radiotherapy planning to reduce geographical misses and limit the involvement of normal tissues in the treatment volume. Incorporating functional imaging into GTV delineation is attractive for a number of reasons – not least to reduce intra- and interobserver variability. It may allow an automation of the target delineation process and identify areas that may benefit from radiotherapy dose boosting. Computed tomography is usually used in target volume delineation for radiotherapy planning. Computed tomography has its limitations – most notably in defining mediastinal lymph node involvement in oesophageal cancer, which is improved with FDG-PET.

The utility of functional imaging tests in providing predictive and prognostic information after CRT for both oesophageal cancer and pancreatic cancer will be reviewed. A separate review of the benefit of incorporating functional imaging into radiotherapy treatment planning will be included. The limitations of the evidence will be discussed and recommendations as to how the integration of functional imaging into the risk stratification of patients with locally advanced oesophageal and pancreatic cancers will be made.

Methods

Four separate searches were completed on Ovid MEDLINE® in-process and other non-indexed citations and Ovid MEDLINE® 1994 to present. The searches targeted literature on: (i) oesophageal cancer, CRT and functional imaging; (ii) pancreatic cancer, CRT and functional imaging; (iii) oesophageal cancer, functional imaging and target volume definition; (iv) pancreatic cancer, functional imaging and target volume definition.

All English language abstracts were reviewed and unrelated articles were excluded. Trials of neoadjuvant chemotherapy alone or mixed cohorts of chemotherapy and CRT were excluded if separate analyses of these treatment modalities were not described. Studies were grouped into those that carried out functional imaging before CRT, before and during CRT, pre- and post-CRT and post-CRT only.

Results

The database search to identify studies concerned with treatment response prediction in oesophageal cancer returned 181 results and three additional studies were identified from the references of these studies. Of these, 141 were excluded after full-text review, leaving 43 studies for review. Eighty-one studies concerning target volume definition in oesophageal cancer were identified by the database search. After full-text review, only 13 were included. The numbers in pancreatic cancer were lower – the database search identified 66 studies concerning functional imaging as a means of predicting CRT response, only six of which were eligible after full-text review. Only one study using functional imaging to guide target volume definition in pancreatic cancer was identified by this search strategy. Apart from one series that used diffusion-weighted magnetic resonance imaging (MRI) [14] and another that used a putative hypoxia PET tracer (18F-fluoroerythronitromidazole) [15], all series used FDG-PET as the imaging modality of choice. Although other functional imaging modalities, such as dynamic contrast enhanced MRI, have been shown to be feasible in cancers of the upper gastrointestinal tract [16], they have not been used in response prediction or target volume definition studies.

Tables 1–4 summarise the data that showed a positive correlation with treatment outcome or prognosis. Many studies that carried out imaging at more than one time point commented upon the usefulness of the imaging at each time point. A clear trend immediately becomes apparent; imaging before CRT, when analysed independently, offers little to no predictive or prognostic information [26,33,38,41]. Recent studies that have gleaned as much information as is possible from pre-CRT FDG-PET by carrying out a textural analysis have improved upon this to a degree: one series reported an area under the received operator characteristic (ROC) curve of 0.85 when a technique that calculates the variability in the size and the intensity of homogenous uptake areas within the tumour was used [17].

It can be seen that studies that have an area under the curve (AUC) on ROC analysis greater than 0.9 (and therefore offering relatively robust predictive power) have used a multi-parametric assessment. Combining functional parameters with anatomical-derived indices improved the predictive function of the tests [28,29]. Tests that applied parameter thresholds based upon ROC curves seem to have done so to optimise the sensitivity of the test. The appropriateness of this approach needs external validation.

Predictive and Prognostic Utility of Functional Imaging in Oesophageal Cancer

A variety of ways of defining ‘response’ have been used, with only five studies using pCR after CRT as the end point to be predicted [33,22,24,25,34]. The most commonly used FDG-PET-derived parameter used in response prediction is the maximum standardised uptake value (SUVmax) - either as an absolute value at specific time points or as a relative change between two scan dates. A number of studies have shown the failure of SUVmax to predict treatment outcome or survival [36,38,40]. Parameters that try to include more information from across a region of interest, such as the SUVmean, SUVpeak (the average of SUVs clustered around the SUVmax) or FDG uptake heterogeneity or skewness, have been shown to offer more predictive information [17,33,36]. Other methods include adding a volumetric measure to the SUV, such as metabolic tumour volume and total lesion glycolysis. In oesophageal squamous cell carcinoma (SCC), the ROC curve AUC improved from 0.71 to 0.93 when a response in functional tumour volume after CRT was used rather than SUV assessment alone [31].

SUVmax measured on baseline imaging, when used as an independent factor, has failed to show any predictive utility [26,33,38,41]. This is also true when baseline imaging parameters from studies that used dual time point assessments were analysed independently of the later imaging, particularly if SUVmax was used [5,23,49]. These data have not been included in Table 1. Of the 20 studies listed in Table 1 that carried out FDG-PET at two time points, only one reported an association between baseline SUVmax and treatment outcome – the ROC curve AUC was 0.555 [21]. Tixier et al. [17] were able to predict the radiological response using only baseline FDG-PET with a sensitivity of 92% only when textural features such as local homogeneity, entropy and size zone were calculated. An initial SUVmax greater than the median value of 12.8 was associated with a poorer overall survival (17.1 versus 33.4 months; P = 0.002) in a large retrospective analysis of baseline FDG-PET in patients treated with CRT as definitive treatment [39]. An apparent diffusion coefficient, a parameter derived from diffusion-weighted MRI, less than the mean was associated with a radiological response in one series [14].

Table 1

Predictive utility of functional imaging in oesophageal cancer

Reference	n	% preoperative	% adenocarcinoma	Tumour radiation dose and chemotherapy agents	Response assessment	Imaging modality	Imaging parameter	Sensitivity, specificityPPV, NPV	AUC	Comments
Imaging at baseline only
[17]	41	0	24	60 Gy (median dose) in 1.8 Gy fractions	RECIST: CR versus non-CR	FDG-PET	SUVmax ≤6	46, 91–, –	0.7
				Carboplatin or cisplatin/5-FU			SUVmean	62, 81–, –
							SUVpeak	62, 81–, –
							Local homogeneity	92, 56–, –
							Local entropy	92, 69–, –
							Size zone	92, 69–, –	0.85
							Intensity variability	85, 75–, –
[14]	80	14	0	40 Gy in 20 fractionsCisplatin/5-FU	RECIST: NR versus PR and CR	DW-MRI	ADC < mean (1.1 × 103)			86% versus 25% ‘response’ (P < 0.01)
Imaging before and during CRT
[18]	38	100	100	40 Gy in 20 fractions5-FU	Response = <10% tumour cells	FDG-PET	SUVmax reduction ≥ 30% after 2 weeks CRTSUVmax reduction ≥52% post-CRTBaseline SUVmax >3.8	93, 8893, 8889, 57–, –95, 50–, –		27/38 had repeat imaging after 20 Gy (2 weeks)All 38 had pre- and post-CRT imaging
[19]	37	100	100	40 Gy in 15 fractionsCisplatin/5-FU	Mandard TRG	FDG-PET	SUVmax reduction ≥ 26.4%	63, 7263, 72	0.674	Imaging before and in 2nd week of CRT
[20]	100	100	82	41.4 Gy in 23 fractionsCarboplatin/paclitaxel	Mandard TRG	FDG-PET	0% SUVmax reduction	91, 5076, 75	0.71	AUC for all patients 0.71, adenocarcinoma 0.71, squamous cell carcinoma 0.35.Imaging at baseline and after 2 weeks CRT.
[21]	48	0	0	50 Gy in 25 fractionsPlatinum/5-FU	Clinical CR at 3 months	FDG-PET	Baseline SUVmaxBaseline metabolic TV (physician defined) >18.6 cm3	50, 87–, –	0.5550.701	Imaging at baseline and 21 days into CRTRelative change in parameters offers no information
Imaging before and after CRT
[22]	36	100		40 Gy in 20 fractions5-FU	pCR	FDG-PET	mCR	67, 050, –
[23]	83	100	88	50.4 Gy in 28 fractionsVarious regimens	Residual disease	FDG-PET	Post-CRT PET:‘abnormal’SUVmax >2SUVmax >4	85, 29–, –76, 19–, –26, 95–, –
[24]	32	100	100	45.6 Gy in twice daily 1.2 Gy fractions or 46 Gy in 23 fractions over 4 weeksCisplatin with 5-FU or capecitabine	pCR	FDG-PET	mCRmCR in those with pre-CRT SUVmax >4.0	27, 9575, 7133, 100100, 65
[7]	41	100	81 (in cohort of 64 patients)	Median dose 50.4 Gy all delivered in 1.8–2.0 Gy fractions (22 patients received hyperfractionated)Various regimens	pCR or microscopic residual disease	FDG-PET	Post-CRT SUVmax <4	61, 6083, 33		Only 43/64 patients received CRT and oesophagectomy.
[25]	62	100	0	45.6 Gy in twice daily 1.2 Gy fractions or 46 Gy in 23 fractionsCisplatin/5-FU or cisplatin/capecitabine	pCR	FDG-PET	mCR	51, 6779, 64		RR 16.5
[26]	25	100	88	50.4 Gy in 28 fractionsVarious regimens	Mandard TRG	FDG-PET	Functional TV <29 cm3	71, 78–, –	0.80
							Post-SUVmax <4.35	100, 69–, –	0.82
							Post-SUVmean <3.55	89, 79–, –	0.85
							SUVmean reduction >32.3%	75, 63–, –	0.64
[27]	20	100	100	35 Gy in 15 fractionsCisplatin/5-FU	Mandard TRG	FDG-PET	SUVmax reduction >50%	67, 71–, –
[28]	51	100	100	50.4 Gy fractionation NRCisplatin/5-FU	Response = <10% viable tumour cells	FDG-PET	SUVmax reduction >43%PET/CT TV reduction >63%TLG reduction >78%	86, 6664, 8791, 9086, 9391, 9391, 93	0.8430.918	TV volume calculated by PET TL × CT diameter
[29]	47	100	76	50.4 Gy fractionation NRCisplatin/5-FU	Response = <10% viable tumour cells	FDG-PET	Functional tumour length reduction >33%SUVmean reduction >43%	91, 86–, –92, 61–, –	0.9190.833
[30]	86	100	62	50.4 Gy fractionation NRCisplatin/5-FU(for ‘most patients’)	CR = < 1% viable tumour	FDG-PET	SUVmax reduction ≥ 64%	64, 81–, –	0.75
[31]	49	100	0	50.4 Gy, fractionation NRCisplatin/5-FU	Response = <10% viable tumour cells	FDG-PET	“Diameter-SUV index” reduction >55%	91, 93–, –	0.931
							SUVmax reduction > 42%	82, 70–, –	0.713
[32]	55	100	44	36 Gy in 20 fractionsCisplatin/5-FU	Response = <10% viable tumour cells	FDG-PET	SUVmax reduction of: 22% for AC	50, 90–, –	0.667
							70% for squamous cell carcinoma	42, 100–, –	0.698
[33]	37	57	73	50.4 Gy in 28 fractionsVarious regimens	pCR	FDG-PET	Post-CRT MTV2.5			Post- MTV2.5 correlates with pCR (P = 0.01). Volume threshold not recorded.
[34]	60	100	0	40–45 Gy in 1.8 Gy fractionsCisplatin/paclitaxel	pCR	FDG-PET	Reduction in functional tumour length >33%Reduction SUVmax >75%	81, 8175, 8488, 8778, 87
[35]	46		48	45 Gy in 25 fractionsCisplatin/5-FU	Visual (major or non-major response)	FDG-PET	Pre-SUVmaxPost-SUVmaxΔSUVmax		0.5730.4670.589
[36]	20		85	50.4 Gy in 28 fractionsCisplatin/5-FU	Visual	FDG-PET	SUVmax declineSUVmax pre/postInertiaCorrelationCluster predominance		0.760.7/0.610.850.80.78

ADC, apparent diffusion coefficient ; AUC, area under the curve; CR, complete response; CT, computed tomography; CRT, chemoradiotherapy; DW-MRI, diffusion-weighted magnetic resonance imaging; FDG-PET, 18F-fluorodeoxyglucose positron emission tomography; mCR, metabolic complete response; MTV, metabolic tumour volume; NPV, negative predictive value; NR, not recorded; pCR, pathological complete response; PPV, positive predictive value; PR, partial response; RR, relative risk; TLG, total lesion glycolysis; TRG, tumour regression grade; TL, tumour length; TV, tumour volume; SUV, standardised uptake value; ; 5-FU, 5-fluorouracil;

A metabolic complete response seems to be associated with a good prognosis. When post-CRT FDG-PET is used in patients managed by definitive CRT, metabolic complete response (defined as a SUVmax < 3) is associated with improved overall survival and rates of local recurrence equal to that of patients receiving trimodality therapy [43].

The relative reduction in SUVmax may offer more predictive information than absolute values [20,34], although this is not always the case, particularly for those with adenocarcinoma [27]. A variety of imaging response thresholds have been used, for example a reduction in SUVmax ranging from 26.4 to 30% when FDG-PET was repeated during CRT or from 32.3 to 75% when repeated post-CRT (see Table 1). However, the sensitivity and specificity of these tests remain poor.

Tables 1 and 2 demonstrate that in most oesophageal carcinoma studies, a mixture of adenocarcinoma and SCC has been included. When adenocarcinoma-only patients were included, the predictive power of a reduction in SUVmax from baseline compared with the second week of CRT no longer provided any prognostic information [19]. This was also observed in a series where a relative reduction in SUVmax from baseline to post-CRT FDG-PETs showed a significant correlation between a pathological response for oesophageal SCC but not adenocarcinoma [32]. Only one study offered different thresholds for adenocarcinoma and SCC; 22 and 70% reduction in SUVmax, respectively [32]. This improved the specificity of the test to 90% for adenocarcinoma and 100% for SCC.

Hypoxia is a well-known cause of chemoradioresistance. The SUVmax and SUVmean values on 18F-fluoroerythronitromidazole (FENTIM) PET (a putative hypoxia tracer) showed good test–retest repeatability, but were not associated with a pathological response or survival [15]. Although tumour hypoxia is not just the result of inadequate perfusion, a decrease in blood flow on perfusion computed tomography correlated with tumour size reduction after CRT. Although a low tumour blood flow was also associated with a shorter median survival, this cohort had mixed treatment modalities and as only 12 patients had CRT, it is difficult to extrapolate these data to the CRT group [50].

Functional Imaging as a Means of Target Volume Definition in Oesophageal Cancer

Incorporating FDG-PET into radiotherapy planning improves the accuracy of target volume definition and reduces geographical misses. The degree of agreement between tumour volumes delineated by different methods is assessed using a conformality index. A conformality index of 1.0 indicates total agreement, whereas 0 indicates that the two volumes are not spatially related at all. Computed tomography-defined GTVs excluded >5% of the FDG-avid disease in 61% of patients [51] in one series. In this series, the conformality index of the GTVs derived from computed tomography and computed tomography co-registered with FDG-PET was 0.68. In another series of 16 patients, the mean conformality index of a computed tomography-derived and FDG-PET/computed tomography-derived GTV was 0.46 (range 0.13–0.80) [52].

The FDG-PET-derived GTVs tended to be smaller than those outlined on computed tomography alone in some series [51-54] and significantly larger in others [55]. When a visual assessment of FDG-PET images fused with the planning computed tomography was integrated into treatment planning, the GTV was decreased by >25% in 12% of patients and increased by >25% in 6% of patients [56]. The series by Schreurs et al. [54] showed that 28% of patients had a >2 cm craniocaudal, anteroposterior or lateral mismatch between GTVs derived from computed tomography and FDG-PET-derived GTV.

FDG-PET improves both intra- and interobserver variability in GTV definition for tumours of the gastro-oesophageal junction. The mean interobserver standard deviation of tumour length decreased from 10 mm to 8 mm (P = 0.02) with the addition of FDG-PET/computed tomography. This was also true for intraobserver agreement with the mean standard deviation in tumour length reducing from 5.3 mm to 1.8 mm (P = 0.001), with corresponding improvement in conformity index – 0.73 for PET/computed tomography versus 0.69 for computed tomography (P = 0.05) [57]. This improvement in interobserver variability was not replicated in another study, despite the incidence of geographical miss of FDG-avid disease being reduced [54].

The most obvious way a GTV can be altered by the inclusion of PET images is through the inclusion of previously unrecognised involved lymph nodes [55] and a greater accuracy in defining tumour length. An absolute SUV threshold of 2.36 has been shown to have a sensitivity and specificity of 76.2% and 96.0%, respectively, in predicting positive nodal involvement [58]. Tumour length defined by an FDG SUV of 2.5 had a better correlation with tumour length defined by endoscopic ultrasound (EUS) than computed tomography-defined tumour length. EUS does, however, seem to be a more robust method of identifying pathological lymph nodes than FDG-PET [53] and remains the gold standard.

The timing of FDG-PET is important. In a series that repeated FDG-PET just before radiotherapy treatment planning, rather than relying on the diagnostic imaging, new FDG-avid lymph nodes were identified in 18% of patients and 13% had new metastatic disease [59]. The median time between imaging time points was 22 days in this series.

The best way of segmenting FDG-PET imaging to aid, or even semi-automate, GTV delineation remains unclear. Measurement of the tumour at surgical resection has allowed correlation of a variety of SUV thresholds on FDG-PET with actual measured tumour length [60]. An SUV that was 40% of the maximum for the tumour grossly underestimated the tumour length seen after resection. Another series has suggested that the SUV threshold for target volume definition to define tumour length needs to be decided on an individual patient basis [61]. In this analysis, it was found that the optimal SUV threshold used to define the tumour varied with tumour length and SUVmax. For example, long tumours or those with a low SUVmax required a higher percentage threshold to make the resultant tumour length correlate with that seen at pathology. This led the authors to conclude that an absolute SUV of 2.5 might be the best compromise if FDG-PET alone was to be used to define the cranial and caudal limit of the tumour. This suggestion was supported by another series where an FDG SUV of 2.5 correlated very well with tumour length measured on computed tomography. Using a personalised SUV threshold based on SUVmax led to a poorer correlation [62]. Again, although an absolute SUV correlated well with tumour length, the conformality index of the resultant volume remained poor (0.57).

The only study to attempt to validate a PET tracer that is associated with hypoxia to derive target volumes was unsuccessful. In a study of 10 patients, the correlation coefficient of the hypoxic volumes derived on two separate FETNIM-PET studies was only 0.21 [15].

Predictive and Prognostic Utility of Functional Imaging in Pancreatic Cancer

All studies to assess the predictive and prognostic utility of functional imaging in pancreatic cancer used FDG-PET. Only one study that showed a correlation with FDG-PET parameters with treatment response was found. Higher baseline SUVmax was associated with a histopathological response – the predictive function of FDG-PET in this series was increased by combining the baseline SUVmax with the relative SUV response after CRT [44]. It should be noted that this series defined histopathological response as <50% viable tumour cells seen in a resection specimen. This is perhaps inevitable given that only 2% achieved a pCR.

Four studies looked for correlation between FDG-PET and patient prognosis. An association with low baseline SUVmax and larger SUVmax reduction after CRT was observed (see Table 4).

Table 4

Prognostic utility of functional imaging in pancreatic cancer

Reference	n	Total tumour radiation dose and chemotherapy agents	Survival end point	Imaging modality	Imaging parameter	Sensitivity, specificity PPV, NPV %	AUC	Comments
[45]	NR	50.4 Gy in 28 fractions or 50 Gy in 40 fractionsGemcitabine	OS	FDG-PET	Pre-SUVmax <7.0			Improved median OS (magnitude NR) (P < 0.05)
[46]	15	NR	TTP	FDG-PET	SUVmax reduction >50%			Mean TTP 399 versus 233 days (P < 0.05)
[47]	32	50.4 Gy in 28 fractions5-FU	OS	FDG-PET	SUVmax reduction >63.7%			Median OS 17.0 versus 9.8 months (P = 0.009)Median LRPFS 12.3 versus 6.9 months (P = 0.02)
[48]	30	50.4 Gy in 28 fractions5-FU	OS	FDG-PET	FDG-PET CT derived GTV <91.1 cm3	79.6, 91.7– , –	0.777	Median OS 14.1 versus 9.5 months (P = 0.005)

PPV, positive predictive value; NPV, negative predictive value; AUC, area under the curve; 5-FU, 5-fluorouracil; OS, overall survival; FDG-PET, 18F-fluorodeoxyglucose positron emission tomography; SUV, standardised uptake value; CT, computed tomography; GTV, gross tumour volume.

TTP.

LRPFS.

Functional Imaging as a Mean of Target Volume Definition in Pancreatic Cancer

Only one series has looked at the effect of functional imaging on GTV definition in pancreatic cancer. In a cohort of patients with LAPC, a computed tomography-defined GTV was used as the reference volume for comparison with a GTV that was delineated after fusion of the FDG-PET with the planning computed tomography. An SUV ≥42% of SUVmax was used when viewing the FDG-PET. The PET-derived GTV was larger by 29.7%, due to extension of primary tumours and additional nodes. Figures 1 and 2 show the potential benefit of including FDG-PET in radiotherapy planning for LAPC. Figure 3 illustrates one of the potential limitations of FDG-PET in LAPC, namely the failure of FDG uptake to differentiate between tumour glucose metabolism and uptake in inflammatory cells. No published data on the correlation of functional imaging with histopathology have been reported.

Fig 3

Axial images from 18F-fluorodeoxyglucose positron emission tomography (FDG-PET) in a patient with locally advanced pancreatic cancer. Although FDG avidity can be used to inform target volume definition, the process cannot be fully automated. FDG-uptake can be seen to correspond with a mass within the pancreatic head (A). This area of avidity runs the length of the stent within the common bile duct, including areas beyond that of the tumour mass (B). FDG avidity is also associated with inflammatory cell glucose metabolism.

Discussion

Although the aim of this review was to assess the predictive and prognostic utility and the additional benefit of functional imaging on target volume definition in oesophageal and pancreatic cancer, the lack of heterogeneity in the functional imaging modalities used and the large degree of variation in technique and reporting make comparison of the results challenging.

There is a strong suggestion that a single parameter (e.g. SUVmax) derived from a pretreatment imaging study is unlikely to offer a predictor of pathological response that is robust enough to drive treatment decision-making. An obvious limitation of the studies that used FDG-PET is the heavy reliance on the SUVmax within the tumour – a single-pixel measure that is subject to considerable noise effect [63]. This is in keeping with the observation that oesophageal tumours were more likely to respond to CRT if the number of pixels with a high SUV value was small [36], suggesting that noise effect could artificially elevate the SUVmax. There is an emerging trend to utilise much more of the information that is included in the scan rather than a single point value. Approaches including total glycolytic volume [63], texture features (descriptive measures of tracer uptake heterogeneity) [17,36] and even the simple method of combining tumour diameter with SUVmax to produce a ‘diameter-SUV index’ [31] may offer better predictive and prognostic utility. Using a support vector model that incorporated a number of features, the treatment outcome for all 20 patients treated with CRT for oesophageal cancer could be accurately predicted when all features, including FDG-PET textural analysis, were taken into account [64].

The timing of the FDG-PET response assessment is probably crucial. Failure of post-CRT functional imaging to accurately predict the pathological response may be due to post-CRT oesophagitis [25] or because cell ‘stunning’ effects, which have nothing to do with tumour cell viability [18], confound the picture. Some experts therefore advocate re-imaging earlier in the course of treatment as the onset of treatment-associated oesophagitis is around 2 weeks [65,66] and as the reduction in FDG uptake at this time point may be more representative of cell death rather than stunning. Moreover, deferring reassessment until after CRT has been completed does not give the opportunity of therapeutic intervention, such as radiotherapy dose escalation, in those who are failing to have an optimal response.

It is clear that if FDG-PET is to be used for radiotherapy planning, it should be carried out as close to the planning computed tomography scan as possible [59]. Disease progression from the time of diagnostic scanning to treatment planning could lead to the failure of inclusion of the positive lymph nodes in the treatment volume or progressing with a radical treatment plan in the presence of metastatic disease. The inclusion of hybrid PET/computed tomography scanning into routine planning computed tomography is worthy of consideration.

The method of using functional imaging to derive GTVs needs to be standardised. Some trials have used ‘side by side’/sequential viewing of images [55,57], whereas others have either relied upon image registration software [51,54] or using hybrid PET/computed tomography scanners [52]. The conformality index is often used to describe the reliability of a method of target volume definition compared with a current standard. This could be potentially problematic in upper gastrointestinal malignancies, particularly pancreatic cancer, where in the presence of a large, physiologically quiescent, stromal component to the tumour, the conformality index will always be low despite the functional imaging test identifying a region of interest that may contain all viable tumour cells.

The investigation of PET tracers that give information on a variety of specific physiological processes, such as hypoxia, may be beneficial. The use of FDG-PET has a good scientific basis, in addition to being a pragmatic choice because of wide availability and relatively low costs. In a preclinical model, FDG-avid tumours required an increase in radiation dose to improved local control rates, whereas tumours with low FDG-avidity did not benefit from an increased radiation dose [67], suggesting that FDG-PET may be an appropriate means of defining an area that would benefit from dose boosting. Targeting hypoxic areas within the GTV is attractive given that hypoxia leads to chemoradioresistance. Local treatment failure is an important consideration, both for oesophageal and pancreatic tumours, so dose-escalating a hypoxic subvolume is ideologically appealing. Other hypoxic tracers, such as 18F-misonidazole or 64Cu-ATSM (diacetyl-bis (N4-methylthiosemicarbazone)) should be investigated, as they may increase confidence in PET-derived hypoxic volumes. In preclinical oesophageal cancer models, FLT uptake has been shown to be a rapidly responding marker of response to CRT [68]. Using FLT as a PET tracer seems to be attractive, as cellular retention of FLT relies upon phosphorylation by tyrosine kinase 1, which is only expressed in late G1 and S phase. Targeting only proliferating cells may be beneficial and may improve some of the poor predictive and prognostic utility associated with FDG uptake.

A greater understanding of the need for four-dimensional PET scanning is required. There can be considerable movement of oesophageal tumours throughout the respiratory cycle, particularly in the craniocaudal direction in lower thoracic tumours [69]. Data acquisition in a static PET scan is a slow process (over minutes). Uptake detection is therefore averaged throughout the time of acquisition and across the whole respiratory cycle. This will be of particular relevance if individualised thresholds are used. Four-dimensional PET imaging may allow a greater confidence in individual voxel SUV values and in boundaries of transition between tracer uptake thresholds. Pancreatic tumours move throughout the breathing cycle. No studies to date have investigated four-dimensional PET in pancreatic cancer. The role of PET/MRI also remains uncertain.

Conclusion

Further studies are required to increase the confidence in the predictive and prognostic power of functional imaging in upper gastrointestinal malignancies. A multimodality, multiparametric assessment of the tumours at more than one time point to increase the likelihood of finding predictive indices should be systematically explored. Attempts should be made to reduce the time from imaging to the start of CRT. The imaging modality used should give information about a physiological process that is associated with treatment resistance. With increased confidence in this imaging modality, the functional imaging could then be used for biological target volume definition. Delivering a higher radiotherapy dose to areas of the tumour that are less likely to respond, or integrating physiological modulating agents into the CRT regimen, may increase the likelihood of pCR without increasing treatment toxicity. This approach would, however, requires a robust means of risk stratification that can be carried out early in the treatment schedule.

Table 2

Prognostic utility of functional imaging in oesophageal cancer

Reference	n	% preoperative	% adenocarcinoma	Total tumour radiation dose and chemotherapy agents	Survival end point	Imaging modality	Imaging parameter	Sensitivity, specificity PPV, NPV %	AUC	Comments
Imaging at baseline only
[37]	47	100	87.2	45 Gy or 50.4 Gy, fractionation NRVarious regimens	Mean OS	FDG-PET	NPA >1			12.4 versus 19.6 monthsBaseline SUV measurements offer no prognostic information. NPA >1 = lymph node involvement
[38]	45	0	27	Mean 60 Gy in 1.8 Gy fractionsCisplatin/5-FU	OS	FDG-PET	Functional TV (cut-off value not recorded)TLG < 180 g			16 versus 5 months (P = 0.0005)21 versus 10 months (P = 0.01)
[39]	209	0	76	45 Gy in 25 fractions or 50.4 Gy in 28 fractions5-FU or capeitabine with taxane or platinum	OS	FDG-PET	SUV < median (12.7)			OS 33.4 versus 17.1 months (P = 0.002)
[14]	80	14	0	40 Gy in 20 fractionsCisplatin/5-FU	1 year survival	DW-MRI	ADC > 1.1 × 103			18 versus 42 months (P = 0.02)
Imaging before and during CRT
[18]	38	100	100	40 Gy in 20 fractions5-FU	OS	FDG-PET	SUVmax decrease ≥30% at 2 weeks			38 versus 18 months (P = 0.01)
[19]	37	100	100	40 Gy in 15 fractionsCisplatin/5-FU	OS	FDG-PET	SUVmax decrease ≥ 26.4%			Median OS NS
[40]	59	32	31	66 Gy in 33 fractionsCisplatin/5-FU	2 year OS	FDG-PET	Decrease in SUVmax>30%	83, 3457, 66		Imaging pre-CRT and after 20 Gy
							>50%	70, 5863, 65
							>70%	36, 8369, 56
[21]	48	0	0	50 Gy in 25 fractionsPlatinum/5-FU	1 year DFS	FDG-PET	Baseline SUVmax >11.9	64, 70–, –	0.670
							Baseline metabolic TV (physician defined) >14.0 cm3	60, 83–, –	0.706
Imaging before and after CRT
[22]	36	100		40 Gy in 20 fractions5-FU	OS	FDG-PET	Visual ‘major response’ on post-CRT imaging			16.3 versus 6.4 months (P = 0.005)
[23]	83	100	88	50.4 Gy in 28 fractionsVarious regimens	2 year OS	FDG-PET	Post- SUVmax <4			60% versus 34% (P = 0.01)
[25]	62	100	0	45.6 Gy in twice daily 1.2 Gy fractions or 46 Gy in 23 fractionsCisplatin/5-FU or cisplatin/capecitabine	DFS	FDG-PET	Decrease in SUVmax ≥80%mCR			31.4 versus 17.1 months (P = 0.025)Not reached versus 17.38 months (P = 0.006)For median OS, not reached versus 22.4 months (P = 0.033)
[26]	25	100	88	50.4 Gy in 28 fractionsVarious regimens	DFS	FDG-PET	Post-SUVmean <4.35			100% DFS during follow-up versus 53% recurrence by 9.5 months
[28]	51	100	100	50.4 Gy fractionation NRCisplatin/5-FU	DFSOS	FDG-PET	PET/CT TV reduction >63%			Mean DFS 29 versus 16 months (P < 00.001)Mean OS 34 versus 22 months (P < 0.001)
[29]	47	100	76	50.4 Gy fractionation NRCisplatin/5-FU	DFS	FDG-PET	Functional tumour length reduction >33%			Median DFS 33 versus 19 months (P < 0.001)
[31]	49	100	0	50.4 Gy, fractionation NRCisplatin/5-FU	DFS	FDG-PET	Reduction the ‘diameter-SUV index’ >55%			Mean DFS 32 versus 16 months (P < 0.001)
[32]	55	100	44	36 Gy in 20 fractionsCisplatin/5-FU	OS	FDG-PET	Baseline and relative reduction in SUVmax			No correlation
[33]	37	57	73	50.4 Gy in 28 fractionsVarious regimens	2 year OS	FDG-PET	Post- MTV2.5 ≤ 7.6 cm3TGA ≤ 26.9			84% versus 29% 2 year OS (P = 0.007)77% versus 37% 2 years OS (P = 0.04)
[41]	40	0	5	50-55 Gy Fractionation unclearVarious regimens	OS	FDG-PET	SUV ≥ 2.5 on post-CRT PET			Hazard ratio (death) 3.56 (95% confidence interval 1.04–12.15)
Post-CRT imaging only
[42]	53	0	45	50 Gy in 25 fractions or 35 Gy in 15 fractionsCisplatin/5-FU	2 year OS	FDG-PET	mCR (defined as uptake less than above/below initial tumour site or paramediastinal lung)			32% versus 78%Relative risk death increased 5.75 fold with failure to achieve mCR
[43]	105	46	75	50.4 Gy (fractionation NR)Platinum/5-FU in 90%	OS2 year OS	FDG-PET	SUVmax < 3			38 versus 11 months (P < 0.01)71% versus 11% (P < 0.01)

TV, tumour volume; TGA, total glycolytic activity); DFS, disease-free survival; OS, overall survival; AUC, area under the curve; NPV, negative predictive value; PPV, positive predictive value; CRT, chemoradiotherapy; 5-FU, 5-fluorouracil; TLG, total lesion glycolysis; SUV, standardised uptake value; mCR, metabolic complete response; FDG-PET, 18F-fluorodeoxyglucose positron emission tomography; CT, computed tomography; DW-MRI, diffusion-weighted magnetic resonance imaging; ADC, apparent diffusion coefficient

TGA.

NPA.

Table 3

Predictive utility of functional imaging in pancreatic cancer

Reference	n	% Preoperative	Total tumour radiation dose and chemotherapy agents	Response assessment (pathological response unless otherwise stated)	Imaging modality	Imaging parameter	Sensitivity, Specificity PPV, NPV %	AUC	Comments
Pre- and post-CRT imaging
[44]	40	100	50 Gy in 25 fractionsGemcitabine	Response = <50% viable tumour cells post-CRT	FDG-PET	Pre-SUVmax ≥4.7Decrease SUVmax >46% (median)Meets both cut-offs	67, 84– , –		71% ‘response’ versus 32% (P = 0.03)71% versus 26% (P = 0.01)

NPV, negative predictive value; PPV, positive predictive value; AUC, area under the curve; CRT, chemoradiotherapy; FDG-PET, 18F-fluorodeoxyglucose positron emission tomography; SUV, standardised uptake value.