Monitoring tumour response during chemo-radiotherapy: a parametric method using FDG-PET/CT images in patients with oesophageal cancer

Positron emission tomography with fluoro-2-deoxy-d-glucose (FDG-PET/CT) is commonly used in the initial staging and post-treatment follow-up of many cancer patients [1, 2]. Since metabolic changes under treatment are likely to precede anatomic alterations [3‐5], FDG-PET/CT is actively investigated as a way to assess tumour response to chemotherapy (CT) and radiotherapy (RT) in lymphomas [6, 7], non-small cell lung [8‐10], head and neck [11], and colorectal and breast cancers [12].

Oesophageal cancer is the third most frequent gastro-intestinal cancer, with a poor prognosis and high mortality rates (5-year survival rates around 4% to 10% [13]). In clinical practice, FDG-PET has a well-established role in the diagnosis and staging of oesophageal cancer [14]. Comparisons of FDG uptake before and after treatment reported a better outcome in patients with complete metabolic response [15, 16]. Definitive chemo-radiotherapy (CRT) has become the first-line therapeutic standard in patients with locally advanced tumours [17, 18]. At the time of diagnosis, less than 50% of these patients have potentially operable tumours [19]. The early identification (e.g. around 25 to 30 Gy) of the tumours that do not respond to CRT would suggest to go for surgery and avoid full-dose CRT toxicity. Due to the heterogeneity of the published series [20], FDG-PET/CT cannot be yet recommended in routine practice to guide the initial treatment of patients with oesophageal cancer [21], particularly in those with squamous cell carcinoma [22].

The guidelines for tumour response assessment [23] include several indices quantifying FDG uptake (standard uptake value (SUV)max, SUVmean, SUVpeak and total lesion glycolysis (TLG)) or metabolic volume (tumour longitudinal length (TL) and tumour volume (TV)) [24, 25]. These indices are calculated on a regional basis, i.e. represent index values measured over the whole tumour. Complex changes in tumour uptake/volume, namely heterogeneity in tumour response, can therefore be overlooked. An automatic method using parametric imaging (PI) has been proposed to quantify FDG uptake variations in metastatic colorectal cancers [26]. The main interest of this approach is its ability to describe heterogeneous tumour response at the voxel level.

In a previous analysis of 57 patients with oesophageal squamous cell carcinoma (SCC) and recruited in a prospective study [27], we showed that the parameters derived from baseline FDG-PET were good predictors of outcome after CRT: larger tumour volume and higher SUVmax/TLG were associated to poor outcome at 3 months. Higher SUVmax values were also predictors of poor outcome at 1 year. FDG-PET performed during CRT at day 21 appeared to be of lower clinical relevance. We present a reanalysis of the same series, where the PI method was used to investigate the predictive value of the metabolic variations observed between FDG-PET/CT performed before and during curative-intent CRT. Our goal was to demonstrate that the PI method was applicable to oesophageal SCC treated with concomitant radiotherapy and chemotherapy (and not limited to colorectal cancers after chemotherapy as initially described) and that it would provide additional information to conventional clinical and FDG-PET/CT data.

Fifty-seven patients were prospectively included. Patients 8, 16, 25, 31 and 40 were secondarily excluded because of disease progression at the time of diagnosis. Patient 14 died during RCT. In comparison with the initial study [27], and for technical reasons related to the impossibility to use CT images in the co-registration method in the software (non joined slices), the PET₁ of patients 42 and 45 and the PET₂ of patients 6, 27 and 50 were not available. As a result, 46 patients were fully evaluable.

Table 1 shows the characteristics of the 46 evaluable patients. Five patients (11%) had cancer extending to more than one anatomical third of the oesophagus. Twenty-nine (63%) and 22 (48%) patients were alive without disease at 3 months and at 1 year, respectively.At least one green cluster was observed in all the 46 evaluable patients, demonstrating some SUV decrease between the two PET examinations. Eight patients had two green clusters, none had three green clusters, one had four green clusters, and two had five green clusters. Fifteen patients had at least one red cluster, nine had two red clusters, two had three red clusters, one had four red clusters, and one had five red clusters. The presence of both green and red clusters in the same patient illustrates the spatially heterogeneous evolution of lesions' uptake. An example is shown in Figure 1.

The patients who died or those without complete remission at 3 months (Table 2) were more likely to have T4 cancer, tumour extending to more than one third of the oesophagus, nodal involvement, larger metabolic volumes on PET₁ and PET₂, and larger green (regression) or red (progression) volumes on FDG-PET/CT parametric analysis. At 1 year (Table 3), the patients who died or those without complete remission were younger, had larger tumours (more than one third of the oesophagus and larger metabolic volume on PET1) with more frequent nodal involvement and had larger red volumes (progression).

Multivariate analyses (Table 4) of outcome at 3 months showed that the risk of failure/death increased with younger age (p = 0.001), larger metabolic volume on PET₁ (p = 0.009) and larger green volume (p = 0.047). As for outcome at 1 year, the risk of failure/death increased with younger age (p = 0.006), nodal involvement (p = 0.08) and larger red volumes (p = 0.03).

In this prospective, multicentre study, an automated parametric method [26] was successfully used to investigate the prognostic value of variations in FDG uptake during curative-intent RCT in oesophageal squamous cell carcinoma. The co-registration of sequential PET examinations allows a voxel-based analysis, as a way to investigate spatial variability in tumour response to treatment. In multivariate analyses, a larger volume of red voxels, i.e. with increased uptake between baseline PET and PET performed at day 21, was significantly associated with a greater probability of treatment failure/death at 1 year.

Our results were obtained in a homogeneous population of patients with oesophageal squamous cell carcinoma. Oesophageal squamous cell carcinoma is very sensitive to CRT [30]. During CRT, SUVmax, SUVmean and TV decreased sharply (Table 2). The SUV value depends on several parameters: delay between injection and acquisition [31], noise level, spatial resolution in the reconstructed images, and region selected to estimate the SUV [32]. SUV measurements have been shown to vary between centres [33]; thus, multicentre studies require rigorous standardisation of the FDG-PET procedures [34]. In the present study, all of the paired FDG-PET/CTs for a given patient were performed in the same department, avoiding inter-centre variability. The cross-calibration of the participating centres was assessed in a previous clinical study [35]. The performance and quality control of the PET/CT devices were monitored by using a specific phantom developed by our group [29], and the post-injection delay was kept constant to facilitate inter-patient and inter-centre comparison. All quantitative analyses were performed in Rouen on the same workstation by one nuclear medicine physician. We relied on an experienced nuclear medicine physician to delineate the tumour metabolic volume (TV) since low FDG uptake on images acquired during treatment limits the use of automatic segmentation methods [8]. Acute inflammatory reactions, as reported in head and neck cancer [36], may also hamper the analyses of PET images during treatment. Oesophagitis was present in 9 out of 46 (19%) patients, and 3 of 46 had a nasogastric feeding tube.

In a previous analysis of the same group of patients [27], smaller TVs at baseline (PET₁) and at day 21 (PET₂) were associated to higher probabilities of response at 3 months, as already reported by other investigators [25, 37, 38]. However, we failed to demonstrate a prognostic value for Δ%SUVmax, Δ%SUVmean or Δ%TV, showing similar evolution of these quantitative parameters whatever the tumour response to treatment. These indices are calculated over the whole tumour and do not address spatial variations of response to treatment within the tumour. In the present report, we aimed at investigating intra-tumour heterogeneity in FDG uptake as an early measure of response to treatment, our results being in line with the encouraging ones published in oesophageal (mixing SCC and adenocarcinomas), cervical and head and neck cancers [39, 40]. In both analyses, the patients with CR (either at 3 months or at 1 year) were older than the ones who died or whose tumour relapsed. These differences were statistically significant at 3 months and 1 year in the present analysis, varying from our original report [27], where age was a predictor of outcome only at 1 year and not at 3 months, possibly as a consequence of the exclusion of two patients and/or the introduction of new variables (V_g and V_r) in the logistic model.

Our aim was to characterise the variations in FDG uptake at the voxel level. The parameters were extracted from an image set combining the FDG-PET/CTs performed before and during CRT. This procedure is similar to the use of ictal and interictal perfusion single-photon emission computed tomography (SPECT) in the SISCOM procedure [41]. Such an approach critically relies on registration accuracy. Our FDG-PET/CTs were co-registered over the thorax region using a rigid method, then visually inspected and, if necessary, re-registered with an affine registration method. The repeatability of the method has been previously validated, as well as the impact of misregistration on the generation of green and red clusters [26]. The classification of the voxels according to the differences in uptake at PET₁ and PET₂ and their position as regards the CT tumour volume were visually checked. Some voxels obviously belonging to the heart or related to oesophagitis had to be manually re-classified. Our previous analysis [27] failed to support FDG uptake or metabolic volume measured on PET₂ as prognostic/predictive indices. In the present study, a predictive value of V_r and V_g (including Ln(V_r) or Ln(1 + V_g)) was demonstrated in multivariate analyses, suggesting that PET₂ could be of clinical interest.

The main feature of the present parametric method presented here is the quantification of tumour response to treatment at the voxel level, possibly more informative than a single value (e.g. SUVmax) calculated over the whole functional volume. Our method seems attractive when monitoring tumour response on serial FDG-PET/CT as oesophageal cancers are hardly visible on the CT images. The present method could become a valuable tool to quantify dissociated responses on multiple tumour sites, e.g. primary tumour and metastatic nodes, provided that all the regions of interest are selected. Multivariate analyses showed that the parametric method added some predictive value to more conventional variables. The association between larger green volumes (i.e. decreased uptake between PET₁ and PET₂) and worse outcome at 3 months appears to be counter-intuitive. Our data suggest that large initial tumour volumes are the most likely (1) to shrink during treatment (explaining the large absolute green volumes) and (2) to shrink only partially, so persistent activity is still visible on PET₂. For example, if a 50-cm³ functional volume at PET₁ decreased to 30 cm³ at PET₂ (i.e. a major response), a large green volume (20 cm³) would be anyway associated to a bad outcome related to the large initial volume. In contrast, a small green volume (e.g. 2 cm³) would be measured on a 5-cm³ initial functional volume with the response of similar amplitude, i.e. a 3-cm³ residual volume at PET₂, also associated to treatment failure despite a relatively small initial volume. Statistically speaking, our limited number of patients precluded us to test for interaction between initial and green volumes. The omission of the green volume from the multivariate analyses did not alter the proportion of correctly classified patients (76% (3 months) and 78% (1 year)), but slightly decreased the area under the ROC curve (from 0.84 to 0.79).

A limitation of the present study is the relatively small number of evaluable patients. The planned sample size was not reached due to accrual slower than anticipated [27]. We could anyway demonstrate that 46 patients with oesophageal SCC could be prospectively recruited in a multicenter setting, all investigators abiding to strict acquisition procedures and analysed together using an innovative approach. The clinical value of FDG-PET/CT during radiotherapy deserves further validation, possibly using the method presented here.

This prospective, multicentre study performed in patients with squamous cell oesophageal cancer evaluated a parametric method to monitor heterogeneous tumour response patterns on serial FDG-PET/CT images acquired during radiotherapy. We demonstrate its feasibility and ability to characterise early metabolic changes and suggest that it provides added prognostic information to conventional variables such as SUVmax and TV.