Correlation Between FDG Hotspots on Pre-radiotherapy PET/CT and Areas of HNSCC Local Relapse: Impact of Treatment Position and Images Registration Method

Aim: Several series have already demonstrated that intratumoral subvolumes with high tracer avidity (hotspots) in 18F-flurodesoxyglucose positron-emission tomography (FDG-PET/CT) are preferential sites of local recurrence (LR) in various solid cancers after radiotherapy (RT), becoming potential targets for dose escalation. However, studies conducted on head and neck squamous cell carcinoma (HNSCC) found only a moderate overlap between pre- and post-treatment subvolumes. A limitation of these studies was that scans were not performed in RT treatment position (TP) and were coregistred using a rigid registration (RR) method. We sought to study (i) the influence of FDG-PET/CT acquisition in TP and (ii) the impact of using an elastic registration (ER) method to improve the localization of hotpots in HNSCC. Methods: Consecutive patients with HNSCC treated by RT between March 2015 and September 2017 who underwent FDG-PET/CT in TP at initial staging (PETA) and during follow-up (PETR) were prospectively included. We utilized a control group scanned in non treatment position (NTP) from our previous retrospective study. Scans were registered with both RR and ER methods. Various sub-volumes (AX; x = 30, 40, 50, 60, 70, 80, and 90%SUVmax) within the initial tumor and in the subsequent LR (RX; x = 40 and 70%SUVmax) were overlaid on the initial PET/CT for comparison [Dice, Jaccard, overlap fraction = OF, common volume/baseline volume = AXnRX/AX, common volume/recurrent volume = AXnRX/RX]. Results: Of 199 patients included, 43 (21.6%) had LR (TP = 15; NTP = 28). The overlap between A30, A40, and A50 sub-volumes on PETA and the whole metabolic volume of recurrence R40 and R70 on PETR showed moderate to good agreements (0.41–0.64) with OF and AXnRX/RX index, regardless of registration method or patient position. Comparison of registration method demonstrated OF and AXnRX/RX indices (x = 30% to 50%SUVmax) were significantly higher with ER vs. RR in NTP (p < 0.03), but not in TP. For patient position, the OF and AXnRX/RX indices were higher in TP than in NTP when RR was used with a trend toward significance, particularly for x=40%SUVmax (0.50±0.22 vs. 0.31 ± 0.13, p = 0.094). Conclusion: Our study suggested that PET/CT acquired in TP improves results in the localization of FDG hotspots in HNSCC. If TP is not possible, using an ER method is significantly more accurate than RR for overlap estimation.

Aim: Several series have already demonstrated that intratumoral subvolumes with high tracer avidity (hotspots) in 18F-flurodesoxyglucose positron-emission tomography (FDG-PET/CT) are preferential sites of local recurrence (LR) in various solid cancers after radiotherapy (RT), becoming potential targets for dose escalation. However, studies conducted on head and neck squamous cell carcinoma (HNSCC) found only a moderate overlap between pre-and post-treatment subvolumes. A limitation of these studies was that scans were not performed in RT treatment position (TP) and were coregistred using a rigid registration (RR) method. We sought to study (i) the influence of FDG-PET/CT acquisition in TP and (ii) the impact of using an elastic registration (ER) method to improve the localization of hotpots in HNSCC.
Methods: Consecutive patients with HNSCC treated by RT between March 2015 and September 2017 who underwent FDG-PET/CT in TP at initial staging (PET A ) and during follow-up (PET R ) were prospectively included. We utilized a control group scanned in non treatment position (NTP) from our previous retrospective study. Scans were registered with both RR and ER methods. Various sub-volumes (A X ; x = 30, 40, 50, 60, 70, 80, and 90%SUVmax) within the initial tumor and in the subsequent LR (R X ; x = 40 and 70%SUVmax) were overlaid on the initial PET/CT for comparison [Dice, Jaccard, overlap fraction = OF, common volume/baseline volume = A X nR X /A X , common volume/recurrent volume = A X nR X /R X ].
Results: Of 199 patients included, 43 (21.6%) had LR (TP = 15; NTP = 28). The overlap between A 30 , A 40 , and A 50 sub-volumes on PET A and the whole metabolic volume of recurrence R 40 and R 70 on PET R showed moderate to good agreements (0.41-0.64) with OF and A X nR X /R X index, regardless of registration method or patient

INTRODUCTION
Head and neck squamous cell cancer carcinomas (HNSCC) are the sixth most common cancer (1,2) with around 800,000 new cases worldwide in 2015. These tumors have a poor prognosis, with a 5-year survival rate < 50% (3), particularly because two thirds of patients are unfortunately diagnosed at advanced stage. In addition to surgery, concurrent chemo-radiotherapy is a standard of care in the curative-intent management of locally advanced tumors (4,5). However, despite improvements in treatment modalities, locoregional failure rates remain high (4,6).
Several studies have suggested that local recurrence (LR) of HNSCC treated with radiotherapy (RT) occurs mainly within the planning target volume (PTV) regardless of radiotherapy technique, suggesting that the radiation dose delivered may be insufficient for local tumor control (7). RT dose escalation is often limited by the tolerance of surrounding tissues and the associated risk of radiation-induced toxicities (8)(9)(10). Therefore, the ability to accurately define and irradiate areas at high risk of recurrence could be useful to guide a boost protocol with the use of modern techniques such as intensity modulated radiotherapy (IMRT) and stereotactic radiotherapy (11,12).
The usefulness of 18Flurorodesoxyglucose positron emission tomography/computed tomography (FDG-PET/CT) for initial staging, therapeutic assessment and recurrence diagnosis in HNSCC is now well established (13)(14)(15). It is also increasingly considered a useful tool in RT to optimize target volume contouring. Indeed, it allows the delineation of target volume boundaries more precisely, with reduction in inter-and intraobserver reproducibility compared to CT (16)(17)(18)(19). In addition, FDG-PET/CT is currently being investigated as a tool to guide radiotherapy dose escalation in order to decrease toxicitities and improve tumor control (20). One of the most important studies in this context is the ongoing multicentric trial ARTFORCE (NCT01504815), which compares a standard dose of 70 Gy with an FDG-PET/CT-based simultaneous integrated boost to areas of high FDG uptake (hotspots) up to a maximum dose of 84 Gy (21).
Recent studies have reported a high risk of LR within FDG hotspots identified on pre-RT PET/CT in lung (22)(23)(24)(25), rectal (26), and esophageal malignancies (27). Nevertheless, two previous studies conducted on HNSCC failed to confirm good correlation between areas of high FDG uptake and preferential sites of local recurrence (28,29). Indeed, we recently found only a modest overlap index (<0.6) between pre-and posttreatment subvolumes in 19 recurrent lesions (28). One possible explanation lies in the lack of reproducibility of the patients positioning between the two scans. Moreover, weight loss and post-therapeutic tissues distortion in HNSCC could also affect anatomical landmarks, making the registration process with a rigid approach more difficult.
Our main objective was to prospectively determine if PET-CT acquisition in the same RT position and image co-registration with an elastic registration method could improve the overlap between FDG hotspots and HNSCC local relapse subvolumes. Therefore, the study aimed to investigate whether a difference existed between (i) RR and ER registration methods and (ii) TP versus NTP patient positioning for PET-CT acquisition. We also sought to define the optimal SUVmax threshold to identify the lowest volume on the initial PET that could be used as a reduced target volume of RT.
A secondary objective of this study was to confirm the prognostic value of initial metabolic tumor burden (metabolic tumor volume = MTV and total lesion glycolysis = TLG) in patients with HNSCC.

Population
Consecutive patients with histologically proven HNSCC treated with RT with or without concomitant systemic treatment referred between March 2015 and September 2017 to our department for FDG-PET/CT were prospectively enrolled in the current study. All patients had FDG-PET/CT before and after treatment in TP. A control group scanned in NTP from our previously published retrospective series was used for comparison (28).

Treatment Modalities
All patients were treated with RT ± chemotherapy according to international guidelines (30). External RT was delivered using volumetric modulated arc therapy (VMAT) on a Truebeam STx accelerator (Varian R , Palo Alto, USA). The gross tumor volume (GTV) was delineated on a planning CT scan after combining the information provided by endoscopy, contrastenhanced diagnostic CT or MRI. The dose to the GTV was 70Gy (2Gy/fraction/day, 5 sessions/week) over 7 weeks +/-concomitant systemic treatment: Cetuximab, Cisplatin or Carboplatin.

Follow-Up
Clinical follow-up was performed as recommended by the National Comprehensive Cancer Network (30). Patients with persistent disease on FDG-PET/CT 3 months after RT completion, and those who, after initial complete response, relapsed within the radiation field during follow-up were pooled together to comprise the LR group. Histological evidence was highly recommended; otherwise evidence of progression on imaging was used to define LR.
The following clinical characteristics were obtained for each patient and considered as variables in univariate analysis: age, sex, tumor location, AJCC stage, systemic treatment modality, RT dose and RT duration.

FDG-PET/CT Imaging
The first FDG-PET/CT (PET A ) was performed for initial staging. The post-therapeutic PET R was defined as either the PET performed at the time of the first evaluation (3 months) in patients showing persistant/progressive disease, or the first PET performed during follow-up (suspected recurrence or systematic surveillance) that demonstrated LR. All indications for FDG PET/CT were reviewed according to guidelines (15, 30) by a multidisciplinary team.
All FDG-PET/CT data were acquired on a Biograph-mCT TM system (Siemens R , Erlangen, Germany) in the same institution. The patients were required to fast for at least 6h before imaging. Scans were performed 60 min after injection of ∼3-4 MBq/kg of FDG (IBA molecular imaging R , Saclay, France). From March 2015, patients were scanned according to their compliance in TP, supine on a rigid board, neck maintained in a semi-rigid headrest. As previously mentioned, patients from our control group (28) were supine without rigid board or headrest (NTP).
PET data were acquired using a whole-body protocol (2 min per step, 200x200 matrix) and reconstructed using an ordered subsets expectation-maximization (OSEM) algorithm (TrueX TM =PSF (point spread function) + time of flight (TOF) OSEM-3D with 4 x 4 x 2 mm voxels. Data were corrected for random coincidences, scatter and attenuation using the CT scan. PET images were smoothed with a Gaussian filter (full-width at half-maximum = 2 mm).
CT data were acquired in the cranio-caudal direction using a whole body protocol. Intravenous iodine contrast agent (1.5 mL/kg) was used for the CT scan unless contraindicated. The CT consisted of a 64-slice multidetector-row spiral scanner with a transverse field of view of 700 mm. The CT parameters were a collimation of 16 x 1.2 mm, pitch = 1, tube voltage and exposure were automatically regulated (CarekV R , CareDose 4D R ) with 120 kV and 80 QrefmAs as basis parameters and CT iterative reconstruction was used (SAFIRE R , strength 5).
The study was approved by our institutional ethics committee (number 2017.CE25). All patients gave written informed consent.

FDG-PET/CT Analysis
All scans were analyzed by the same nuclear medicine physician (BT). The registration and overlap comparisons were then performed using the MIM TM software (MIM TM Software Inc., Cleveland, USA). For each patient, two registration methods (RR and ER) were studied.
For the RR method, the CT of PET A (CT A ) was registred with CT of PET R (CT R ), focusing on the tumor area. Regions of interest were identified and outlined on the CT, using PET images as reference. Manual adjustment was not allowed. The transformations derived from the CT registration process were then reported on PET A images.
The ER method was performed by the VoxAlign R Engine algorithm, a constrained intensity based, free-form registration (31). A rigid registration between CT A and CT R was first performed, followed by ER. The deformable registration matrix was saved, and applied to PET A . These deformations led to an elastic registered CT and PET.
Seven volumes of interest (VOIs) on PET A and 2 VOIs on PET R (metabolic active residual disease or relapse) were respectively defined. On PET A , baseline sub-volumes were delineated using a relative threshold method (A X with x = 30, 40, 50, 60, 70, 80, and 90% of SUVmax). On PET R , thresholds at x = 40 and 70% of SUVmax were respectively used to delineate R 40 and R 70 recurrent sub-volumes. Baseline sub-volumes A X were reported on PET R , and recurrence sub-volumes R X were reported on PET A , to quantify their respective overlaps (Figure 1).
The following quantitative parameters were also collected on PET A for the prognosis analysis: SUVmax, SUVmean, MTV, and TLG (TLG= SUVmean × MTV).

Overlap Estimation
All potential overlaps between baseline tumor sub-volumes (A 30 to A 90 ) vs. relapse (R 40 and R 70 ) sub-volumes were investigated using five indices [Dice, Jaccard, overlap fraction (OF), common volume divided by the initial volume (A X nR X /A X ) and common volume divided by the compared volume (A X nR X /R X )], as recommended by Calais et al. (22,27) and as applied in our previous study (28).
Index values for each parameter vary between 0 if the volumes are completely disjointed and 1 if the volumes match perfectly in size, shape and location.
A schematic example of the interpretation of overlap indices is represented in Figure 2.

Statistics
The quality of overlap was assessed using Cohen k-test for agreement between investigators as follows: 0-0.2, poor agreement; 0.21-0.40, fair agreement; 0.41-0.60, moderate agreement; 0.61-0.80, good agreement; and 0.81-1.00, very good agreement (32). Comparison of mean overlap in different subgroups was performed using the Wilcoxon or Mann-Whitney U-tests as appropriate. The statistical associations between FDG-PET/CT and  clinical parameters were tested using a repeated measures analysis of variances (ANOVA) and the chi-2 squared test. A p < 0.05 was considered statistically significant. All analyses were performed using XLSTAT (Addinsoft R , Paris, France).

Patients Characteristics
The final cohort included 199 patients (142M/57F). Characteristics of patients are reported in Table 1.  The mean ± SD time of follow-up of the population was 18.7 ± 11.3 months. At last follow-up, 120 (60.3%) with initial complete response remained free of disease (CR) and 43 (21.6%) experienced local relapse (LR). Twenty-nine LR were identified on imaging and confirmed pathologically. Fourteen LR were considered as such based on evidence of local and metastatic progression disease on any imaging procedure or on clinical examination. Thirty-six patients (18.1%) showed distant dissemination (nodal or metastatic) without LR (Figure 3).

PET/CT Parameters
Fifteen patients with LR were scanned in TP and 28 in NTP.
The initial PET/CT parameters are summarized in Table 2. The mean ± SD initial MTV was 7.7 ± 7.8 cc in the entire cohort and 9.1 ± 7.2 cc in patients with LR, wheras mean ± SD initial TLG was 90.6 ± 101.8 g and 105.9 ± 80.6 g, respectively.

Overlap Comparison
A total of 6,020 overlap indices were obtained, i.e., 140 potential overlaps between the baseline PET A VOIs and relapse PET R VOIs in the 43 patients who had LR and using the 2 registration methods (387 VOIs on 86 PET/CT). Two typical examples are shown in Figure 4. Mean overlap index values are reported in Table 3. The Dice, Jaccard and A X nR 70 /A X indices were very low, mostly below 0.20, irrespective of the thresholds used on PET A .

TP vs. NTP
With RR method, the OF(A X nR 70 ), A X nR 40 /R 40 , and A X nR 70 /R 70 indices for SUVmax thresholds of 30-40% were higher in TP subgroup than in the NTP subgroup with a trend toward significance. For example, the OF(A 30 nR 70 ) and A 30 nR 70 /R 70 indices were higher in the TP subgroup (0.59 ± 0.22) than in the NTP (0.38 ± 0.14) subgroup (p = 0.10); and OF(A 40 nR 70 ) and A 40 nR 70 /R 70 were higher in the TP subgroup (0.50 ± 0.22) than in the NTP (0.31 ± 0.13) subgroup (p = 0.094).
With ER method, there was no significant difference between TP and NTP subgroups with the aboved-mentioned best agreement (moderate to good) of OF(A X nR 40 ), A X nR 40 /R 40 , and A X nR 70 /R 70 indices.

Elastic vs. Rigid Registration Method
In the NTP subgroup, the OF(A X nR 40 ) and OF(A X nR 70 ) indices for SUVmax thresholds of 30-50% were significantly higher with ER than those obtained using the RR method (p < 0.03). The A X nR 70 /R 70 index values for SUVmax thresholds of 30-40% were

FIGURE 4 | Histogram of the mean values of OF and AxnRx/Rx index for various SUVmax thresholds to delineate the volumes on PET A (baseline) and PET R at relapse in the 4 subgroups.
significantly higher with ER vs. RR method (p = 0.028). A typical example is shown in Figure 5.
In the TP subgroup, there was no significant difference between the RR and ER methods neither with the aboved-mentioned best agreement (moderate to good) of the OF(A X nR X ), A X nR 40 /R 40 and A X nR 70 /R 70 indices (Figure 6), nor regarding the lowest. There was only one case where the overlaps increased by a factor 3 (Figure 7).

Univariate Analysis
Gender, RT dose, RT duration, use of chemotherapy, baseline SUVmax and SUVmean were not statistically different between patients with complete response (CR), distant relapse (DR) or local relapse (LR). However, patients with CR were significantly younger (p = 0.021), and more often presented with early stage disease (p = 0.003), and laryngeal cancer (larynx, p = 0.027).

DISCUSSION
The rationale for applying the "hotspot" localization concept to HNSCC relies on the overlap of the recurrence sites with the pre-RT biological target volume (BTV). Indeed, Soto et al. reported that LR was included in the pre-treatment FDG BTV in 8/9 patients after RT (33). Based on these findings, recent articles have suggested the use of biological and functional parameters in FDG-PET/CT to identify the radioresistant tumor area (14,34). Thereby, Jeong and al. suggested that FDG-avid tumors require at least 10-30% higher dose than non-FDG avid tumors (14).
Whilst studies on lung (22)(23)(24)(25), rectal (26), and esophageal (27) (29). Therefore, we hypothesized that performing PETs in TP as well as using deformable registration could improve the methodogy of this process and translate into better results. Our current study is the first aimed at identifying tumor areas of high risk of relapse in HNSCC using PET-CT images acquired in TP and registered with an ER method. This current work relies on 5 overlap indices with 2 registration methods on 43 HNSCC patients with local failures.
This study confirmed our first hypothesis that patient positioning remains essential, with improved overlap between LR and initial FDG tumor hotspots subvolumes in patients scanned with radiotherapy head support. In comparison to the control group (28), we noted the best agreement (moderate to good) of OF(A X nR 70 ) and A X nR 70 /R 70 indices for SUVmax threshold of 30 and 40% in the TP group, ranging from 0.50 to 0.61. For example, the OF(A 40 nR 70 ) and A 40 nR 70 /R 70 index values were 0.50 ± 0.22 in the TP group vs. 0.31 ± 0.13 in the NTP group (p = 0.094). Admittedly, we have only shown a trend toward significance; however, the lack of statistical power is probably due to a small number of patients in the TP group (15 vs. 28 in the NTP group).
We have also demonstrated that using an elastic method is preferable for image registration when patients cannot be scanned in TP. The OF(A X nR 40 ), OF(A X nR 70 ), and A X nR 70 /R 70 index values were significantly better with the ER method for SUVmax threshold of 30-50%. For instance, the OF(A 30 nR 70 ) index value was good (0.64 ± 0.15) with ER vs. moderate (0.38 ± 0.14) with RR (p = 0.014). To our knowledge, only two other studies have been conducted using elastic image registration software. The first, conducted by Due and al. (35), on a cohort of 39 HNSCC after IMRT. However, for this study the PET baseline volumes were delineated based on visual assessment without an SUV-based semi-automated method, therefore increasing the risk of inter and intra-observer variability. Furthermore, the authors determined the overlap between subvolumes segmented on a PET BASELINE and a CT RECURRENCEr , but not on a PET RECURRENCE . Shusharina et al. prospectively studied 19 post-RT residual disease of non squamous cell lung carcinoma (NSCLC) and reported the overlap fraction of an initial subvolume defined as the 50%SUVmax threshold and a relapse subvolume defined as the 80%SUVmax threshold. They showed that the obtained OF(A 50 nR 80 ) was excellent (80%) at 2 weeks after treatment and remained good (63%) at 3 months (25).
Nevertheless, despite RT position and ER method, the hotspot on pre-RT PET-CT that is used to guide definition of areas of high risk of recurrence in patients with HSNCC remains large, and would result in a risk of error with regards to dose escalation. Indeed, the only SUVmax threshold to reach a good agreement value was 30%, which is significantly lower than the threshold obtained in previous studies conducted on other primary tumors. Aerts  values (0.61-0.89) for threshold of 30 to 60% were reported. The authors also recommended a 60% SUVmax threshold on PET A to delineate high FDG uptake areas on pre-RT PET/CT for a dose escalation target volume (27).
Two main hypotheses could explain why overlap index values in HNSCC are lower than in lung and esophageal cancers. Firstly, although HNSCC are frequently locally advanced at diagnosis, we noticed that MTV values were smaller in our HNSCC series (9.1 ± 7.2cc) than those reported in esophageal (25.4 ± 16.2cc) or lung (53.7 ± 45.6cc) cancers, when considering the same SUVmax threshold of 40% (22,27). Unlike these abovementioned tumors, head and neck cancers are known to include necrotic areas without any metabolic activity so the MTV is smaller than real tumor volume. Consequently, with smaller MTV, any mismatch during the registration process can lead to a greater overlap error. Second, weight loss and post-therapeutic tissue distortions are probably more important in HNSCC, with displacement of anatomical landmarks and rendering the registration process more difficult, even with an elastic method.
For our secondary objective, we confirmed that initial MTV and TLG on baseline PET were significantly higher in relapsed patients than locally controlled patients (p = 0.041 and p = 0.046 respectively) and appeared to be a better prognostic marker than SUVmax (p > 0.05). These results are also consistent with previous studies (37,38). Mapelli et al. studied the value of MTV and TLG to predict outcomes in oropharyngeal carcinomas treated by tomotherapy with simultaneaous integrated boost in  FDG-avid tumor subvolumes. They demonstrated that MTV > 4.4cc and TLG > 34.6g were associated with a better 3-year overall survival (p = 0.006 and p = 0.01, respectively) in a series of 41 patients (39). These results are concordant with our findings.
Textural analysis on pre-RT FDG-PET/CT, already recognized as a prognostic factor for survival (40), could be an interesting approach to predict HNSCC local recurrence sites. Beaumont et al. showed that 15 parameters extracted from a voxel to voxel analysis, combining radiomics and spatial location, allowed better prediction of local failure than a regional analysis, with a median area under the receiver-operating curve of 0.71 (41). The published literature to date mainly underlines methods in assessing tumor hypoxia, a well-known factor for RT resistance (42,43). Thureau et al. reported that IMRT dose-painting with pre-RT 18F-misonidazole (F-MISO) PET/CT provided NSCLC radiotherapy plan matching with dose/volume (D/V) objectives and organs at risk (OAR) tolerance (36). Patients with F-MISO positive scans who received an RT boost (70 to 86Gy) tend to have a better overall survival (median 26.5 vs. 15.3 months, p = 0.71) (44).
Our study has some limitations. First, although larger than previous series, the number of included patients (43 LR) is relatively low, contributing to the lack of power and the inability to confirm superiority of TP when the RR method was used (28,29). With regards to the results on the prognostic performance of the PET parameters, despite the inclusion of 199 patients, we acknowledge that the role of FDG-PET/CT in systematic followup to diagnose occult relapse is still not well defined, despite high performance (45,46). Our population lacked homogeneity, with a higher proportion of patients with younger age, AJJC I-II stages, and laryngeal cancers included in the CR group. These variables are correlated with lower risk of LR (47,48). In addition, relapse in FDG-avid lymph nodes at initial staging was not considered. It would be of interest to test the technical feasibility of this process on involved nodes, as these may also benefit for dose escalation, particularly in N3 disease. Finally, we used a PET segmentation method based on different relative SUVmax thresholds. This procedure remains a simple measurement that is easy to perform using commercially available software tools and was utilized in many studies. Van den Bogaard et al. are the only group to utilize an adaptative threshold method based on signal-to-background (26), and reported additional value compared to cancer clinical characterization alone (18,49). However, this technique remains more tedious to implement and requires a PET calibration phase. Combinations of thresholds could lead to over-or under-estimation of overlaps, and other PET segmentation methods, like automatic approaches should also be tested in future. In fact, several studies have suggested that the gradient-based method (50) best estimates the true tumor volume in NSCLC or HNSCC compared to the SUV-based method (51,52). Moreover, the segmentation using the FLAB algorithm (fuzzy locally adaptive Bayesian) (53) is also an interesting model that may improve MTV delineation (54,55). Unfortunately, this patented method is not freely available.

CONCLUSION
This study suggests that treatment position improves correlation between FDG hotspot areas on pre-RT PET/CT and sites of local relapse on post-RT PET/CT. When PET in TP is not possible, the use of an elastic registration method is significantly more accurate than a rigid registration method for overlap estimation. However, we found lower overlap index values (at best moderate to good agreement, with SUVmax thresholds of 30-50%) than those reported in other cancers. Further larger prospective studies are needed to assess other PET segmentation methods.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by CHRU Brest institutional ethic committee (n2017.CE25). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.