Ann Surg Oncol DOI 10.1245/s10434-014-4284-3
ORIGINAL ARTICLE – HEAD AND NECK ONCOLOGY
Intratumor Textural Heterogeneity on Pretreatment PET Images Predicts Response and Survival After Chemoradiotherapy for Hypopharyngeal Cancer
18
F-FDG
Jungsu S. Oh, PhD1, Bung Chul Kang, MD2, Jong-Lyel Roh, MD2, Jae Seung Kim, MD1, Kyung-Ja Cho, MD3, Sang-wook Lee, MD4, Sung-Bae Kim, MD5, Seung-Ho Choi, MD2, Soon Yuhl Nam, MD2, and Sang Yoon Kim, MD2,6 1
Department of Nuclear Medicine, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 2Department of Otolaryngology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 3Department of Pathology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 4Department of Radiation Oncology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 5Department of Internal Medicine, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 6Biomedical Research Institute, Korea Institute of Science and Technology, Seoul, Republic of Korea
ABSTRACT Background. Increasing evidence suggests that intratumor heterogeneity of solid tumors characterized by textural features on 18-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) images is associated with response to chemoradiotherapy (CRT) and survival. The current study aimed to determine whether a similar relationship exists in hypopharyngeal squamous cell carcinoma (HPSCC). Methods. This study investigated 27 patients with HPSCC who underwent cisplatin-based induction chemotherapy followed by definitive CRT underwent pretreatment 18FFDG PET/CT. Standardized uptake value (SUV), metabolic tumor volume (MTV), and textural features (coarseness, busyness, complexity, and contrast) of primary tumors were measured. Patients were classified as
Jungsu S. Oh, and Bung Chul Kang have contributed equally to this paper.
Electronic supplementary material The online version of this article (doi:10.1245/s10434-014-4284-3) contains supplementary material, which is available to authorized users. Ó Society of Surgical Oncology 2014 First Received: 26 September 2014 J.-L. Roh, MD e-mail: [email protected]
nonresponders or responders according to the response evaluation criteria in solid tumors. The capacity of each parameter to classify response was assessed using the Mann–Whitney U test. Cox-proportional hazard models were used to identify variables associated with disease-free survival (DFS) and overall survival (OS). Results. Of 70 patients, 58 (83 %) had complete or partial response after CRT. Responders showed lower maximum SUV (P = 0.037), lower MTV (P = 0.039), lower coarseness (P \ 0.001), and busyness (P = 0.015) compared with nonresponders. Multivariate analysis showed that high coarseness (P = 0.001, hazard ratio [HR] 5.65; 95 % confidence interval [CI] 2.12–15.07) and busyness (P = 0.045; HR 2.56; 95 % CI 1.02–6.42) were independently associated with poor DFS, and that high coarseness (P = 0.013; HR 2.48; 95 % CI 1.21–5.09) was independently associated with poor OS. Conclusion. Abnormal coarseness in baseline 18F-FDG PET scans may be useful for predicting response and survival after CRT in HPSCC patients.
The goal of precision medicine in modern cancer therapy is to select the optimal treatment method and enhance treatment outcomes for individual cancer patients. Accurate prediction of tumor response to treatment is essential to guide therapeutic stratification and personalized cancer therapy. The heterogeneity of tumor cells, as indicated by their distinct morphologic and phenotypic profiles, is regarded as a major obstacle to the
J. S. Oh et al.
success of precision medicine.1 Tumor heterogeneity related to therapeutic resistance is identified by tumor biopsy and histopathology, molecular and genetic profiling, and innovative response assays.1 Increasing evidence suggests that intratumoral spatial heterogeneity can be characterized by textural features on pretreatment fluorine 18-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) images and that these features are associated with response and survival in solid tumors.2–7 As a diagnostic imaging method, 18F-FDG PET/computed tomography (CT) is based on the uptake of tracers for estimation of tumor glucose metabolism. This method is widely used for tumor staging and for evaluating the therapeutic responses of solid tumors, including head and neck squamous cell carcinoma (HNSCC).8–11 Various standardized uptake value (SUV) parameters are predictive of therapeutic response.11,12 However, whether high or low SUVs are predictive of the response to different treatment methods remains controversial. Metabolic tumor volume (MTV) and total lesion glycolysis (TLG), which assess tumor volume and metabolic activity of the entire tumor, respectively, are used as prognostic biomarkers in HNSCC.13,14 However, tumor textural features on baseline 18F-FDG PET images were recently shown to have a higher predictive value for response and survival after chemoradiotherapy (CRT) than SUVs, MTV, or TLG.7 Textural parameters are derived from neighborhood gray-tone difference matrices (NGTDM), which indicate differences between the intensity of each voxel and its neighboring voxels and describe features such as coarseness, contrast, busyness, and complexity.15 This allows differentiation between tumor and normal tissues as well as delineation of target volumes for radiotherapy.4,5 Hypopharyngeal squamous cell carcinoma (HPSCC), a rare type of tumor accounting for approximately 6.5 % of all head and neck cancers, is commonly detected at an advanced stage in patients with regional stage disease.16,17 Nonsurgical laryngeal preservation strategies for HPSCC have been highlighted to preserve voice and swallowing functions.18 The prognosis of HPSCC remains poor despite recent improvements in diagnostic methods and treatment methods, with a 5-year survival rate of 25–40 %.16,17 The poor survival of HPSCC patients is partly due to the common occurrence of second primary cancers and a higher rate of distant metastasis than that of other head and neck cancers.19,20 Therefore, biomarkers that can accurately predict response and survival after treatment in HPSCC patients are needed. We hypothesized that the textural features derived from the NGTDMs of baseline 18F-FDG PET images may be predictive of response to CRT and survival in HPSCC. The current study aimed to compare the predictive and prognostic value of textural parameters in pretreatment 18FFDG PET images with that of SUV or MTV in HPSCC patients who underwent CRT.
PATIENTS AND METHODS Study Population The cancer registry records of consecutive patients with biopsy-confirmed HPSCC treated at our tertiary referral center for cancer patients between January 2006 and December 2011 were reviewed. The inclusion criteria for the study specified previously untreated HPSCC patients without distant metastasis who underwent pretreatment 18 F-FDG PET/CT scanning for initial staging and concurrent CRT. The exclusion criteria ruled out patients who initially underwent definitive surgery, those with a history of head and neck cancer, those with distant metastasis at initial presentation, and those with incomplete treatment or clinical follow-up data. All surviving patients were followed up for at least 12 months. Finally, 70 patients (64 men and 6 women) with a median age of 64 years (range 40–73 years) were eligible for the study. Data obtained from medical records included clinicopathologic variables, treatment regimens, and follow-up evaluation assessing the recurrence and survival of each patient. Tumors were staged according to the American Joint Committee on Cancer (AJCC) staging system.21 The patient clinicopathologic data, treatment methods, CRT response, and follow-up information are described in Table 1. This study was approved by the Institutional Review Board of our institution, and informed consent from each patient was waived because of the study’s retrospective design.
Treatments All eligible patients underwent induction chemotherapy using two or three cycles of cisplatin plus 5-fluorouracil with or without docetaxel administration before CRT. Patients were treated with radiotherapy using a median of 70 Gy (range 54–78 Gy) and concurrent chemotherapy with high-dose cisplatin (80–100 mg/m2) intravenously infused on days 1, 22, and 43. Radiotherapy consisting of intensity-modulated or three-dimensional conformal radiation was administered in a single daily fraction of 1.8 or 2.0 Gy, 5 days per week for 7–8 weeks. The initial tumor response was assessed 2 months after completion of CRT with the response evaluation criteria in solid tumors (RECIST) using clinical, endoscopic, and CT or magnetic resonance imaging (MRI) examination.22 All patients received regular physical and endoscopic examinations at every clinic visit after treatments. The clinical follow-up visit was scheduled according to departmental protocol: every 1–2 months for the first year, every 2– 4 months for the second and third years, and every 6 months
Textural Features of
18
F-FDG PET in HPSCC
TABLE 1 Patient characteristics (n = 70) Characteristics
Value n (%)
Median: years (range)
64 (40–73)
Sex Male/female
64 (91)/6 (9)
Smoking Never/\20/C20 pack-year
10 (14)/12 (17)/48 (69)
Alcohol drinking Never/social/heavy drinker Performance status scale by ECOG 0/1
16 (23)/26 (37)/28 (40) 68 (97)/2 (3)
Charlson comorbidity index 0/1/C2
56 (80)/12 (17)/2 (3)
Histologic differentiation Well/moderately/poorly differentiated
6 (9)/49 (70)/15 (21)
Clinical TNM stage T1/T2/T3/T4
12 (17)/17(24)/11 (16)/30 (43)
N0/N1/N2/N3
8 (11)/12 (17)/41 (59)/9 (13)
Overall stage 1/2/3/4
3 (4)/3 (4)/6 (9)/58 (83)
Treatment Induction chemotherapy ? CRT
70 (100)
Response to induction chemotherapy CR/PR/SD/PD
6 (9)/48 (69)/4 (6)/12 (17)
Response to concurrent CRT CR/PR/SD/PD
47 (67)/11 (16)/8 (11)/4 (6)
Follow-up information Median follow-up of survivors: months 56 (18–100) (range) Locoregional recurrences
25 (36)
Distant failure
7 (10)
Status at last follow-up visit NED/AD/DOD/DOC
32 (46)/3 (4)/26 (37)/9 (13)
ECOG eastern cooperative oncology group, TNM tumor-nodemetastasis stage, CRT chemoradiotherapy, CR complete response, PR partial response, SD stable disease, PD progressive disease, NED no evidence of disease, AD alive with disease, DOD died of disease, DOC died of other causes
thereafter. Any lesions suggestive of recurrence or a second primary tumor were confirmed by biopsy and specific additional diagnostic tests. Patients with confirmed recurrence were scheduled for salvage or palliative treatment. 18
F-FDG PET/CT Imaging and SUV Harmonization
The 18F-FDG PET/CT scans in this study were performed using a Biograph Sensation 16 or TruePoint 40 system (Siemens Medical Systems, Knoxville, TN, USA)
equipped with a 16- or 40-slice CT, respectively, for both anatomic information and attenuation correction purposes. All patients were required to fast for 6 h or longer and to have a serum glucose concentration lower than 150 mg/dL before receiving the injection of 18F-FDG. Whole-body images were acquired 50–70 min after intravenous injection of 370–688 MBq (10.0–18.6 mCi). Using the spiral mode, 18F-FDG. CT data were collected from the thigh to the skull at 100 mAs and 120 kV with a section width of 5and 0.75-mm collimation. No oral or intravenous contrast medium was used during the CT scans. Three-dimensional craniocaudal PET was performed during an acquisition time of 2.5 min per cradle position for eight bed positions. The PET results were reconstructed using a standard iterative algorithm based on the CT results. Because the intrinsic spatial resolution measured by the national electrical manufacturers association (NEMA) NU-2 test may differ between the PET scanners used in this study, mere standardization of the reconstruction method could not fully resolve the issue of scanner difference. Moreover, the newer-generation scanner (Biograph TruePoint 40), which uses a point spread function (PSF)-based reconstruction method called True-X, had a much higher spatial resolution than the other scanner. We estimated recovery coefficients (i.e., relative SUV ratios over the ideal SUV of 2.5) for lesions of different sizes using the American College of Radiology (ACR)-approved phantom Esser Flangeless Deluxe (Data Spectrum, Hillsborough, NC, USA) filled with 18 F-FDG. The SUVs of several hot cylinders with differing diameters were measured, and recovery coefficient plots were generated, allowing estimation of the optimal smoothing kernel size for matching the different recovery coefficients, as described previously.23,24 In addition, all PET images were resampled into 5 9 5 9 5-mm3 voxels, the largest voxel size between many scanners, to ensure a matched ‘‘image scale’’ between scanners. This is desirable for optimized texture analyses because texture is dependent on the arrangement of intensities in the image, and this can be affected by voxel size (in other words, ‘‘scale’’). Tumor Image Analysis After the between-scanner harmonization, 18F-FDG PET parameters including mean SUV (SUVmean), maximum SUV (SUVmax), peak SUV (SUVpeak),12 MTV,13,14 and texture features 6,7 were measured for each patient. The region of interest (ROI) was manually drawn to enclose three-dimensional coverage of the entire hypermetabolic primary tumor lesion visible on PET images in all three projections (axial, sagittal, and coronal). Metabolic tumor volume was defined as the volume enclosed by an isocontour of SUV2.5.13,25 One board-certified nuclear
J. S. Oh et al.
medicine physician (J.S.K.) with 20 years of clinical experience identified the lesions with increased tracer uptake, and a physician and nuclear medicine physicist (J.S.O.) blinded to patient outcome performed the analyses. The textural features of primary tumors measured were coarseness, busyness, complexity, and contrast derived from three-dimensional matrices and NGTDMs, as previously described.7,15 The NGTDM-based texture analysis method was selected for measuring intratumor heterogeneity because it was considered appropriate for analyzing relatively small tumor lesions in this study. Nodal disease was not included because textural analysis of small metastatic lesions is not reliable with a small number of voxels.6,7 The textural features on baseline 18F-FDG PET images were analyzed using the in-house software, AMC NM Toolkit for Image Quantification of Excellence (ANTIQUE), developed as previously reported.7,15,26 Textural features have shown similar or better reproducibility than simple SUV measurement.27
prognostic value of clinical and imaging parameters for predicting disease-free survival (DFS) and overall survival (OS). Time periods were calculated from the date of CRT completion to the date of a confirmed event (any recurrence or death) or the last clinical follow-up visit. The survival rates associated with each variable plotted using the Kaplan–Meier method were compared using a nonparametric log-rank test. The Cox-proportional hazard regression model was used to evaluate prognostic variables in uni- and multivariate analyses for prediction of DFS and OS. Tests were based on the likelihood-ratio statistic. The estimated hazard ratio (HR) and the 95 % confidence interval (CI) were calculated. Variables with a P value lower than 0.05 in univariate analyses were selected for multivariate analysis. A two-sided P value lower than 0.05 was considered statistically significant. All statistical analyses were performed using IBM SPSS software version 21.0 (IBM, Armonk, NY, USA). RESULTS
Statistical Analysis Response Prediction Mean differences in SUVs, MTV, and textural parameters were compared between the RECIST responders (complete response or partial response) and the nonresponders (stable or progressive disease) using the Mann– Whitney U test. The area under the receiver operating characteristic (ROC) curve (AUC) was calculated for SUV, MTV, and textural parameters to assess sensitivity and specificity for predicting CRT response and to obtain the optimal cutoff values for survival prediction. Uni- and multi-variate analyses were performed to evaluate the
The distribution of the SUVs, MTV, and textural parameters is summarized in Table 2. After the completion of definitive CRT, complete responses were achieved for 47 patients (67 %) and partial responses for 11 patients (16 %), with stable disease was observed in 8 patients (11 %) and progressive disease in 4 patients (6 %) (Table 1). Salvage surgery was performed for 23 patients (33 %) who did not achieve a complete response.
TABLE 2 Comparison of standardized uptake value (SUV), metabolic tumor volume (MTV), and measured textural features for complete or partial response versus stable or progressive disease after concurrent chemoradiotherapy Variables
Total (n = 70) Mean ± SD
P valuea
Mean CRT response Range
Responders
Nonresponders
SUV parameters SUVmean
7.4 ± 5.6
1.1–33.0
7.0 ± 5.0
10.2 ± 7.8
0.056b
SUVmax
11.9 ± 8.7
1.5–50.88
10.9 ± 7.6
16.6 ± 12.1
0.037b
9.8 ± 7.4
1.3–40.8
9.0 ± 6.6
13.9 ± 9.6
0.034b
30.2 ± 42.5
0–295.2
21.2 ± 23.1
74.0 ± 77.6
0.039b
0.043 ± 0.012
0.059 ± 0.017
SUVpeak 3
MTV (cm ) Tumor heterogeneity Coarseness
0.047 ± 0.018
\0.001b
Busyness
0.037 ± 0.018
0.017–0.154
0.035 ± 0.011
0.049 ± 0.036
0.015
Complexity
8,585 ± 2,948
2,706–16,343
8,403 ± 3,056
9,466 ± 9,598
0.258
1.34 ± 0.56
1.04 ± 0.5
0.086
Contrast
1.29 ± 1.19
CRT chemoradiotherapy, SD standard deviation a Assessed using the Mann–Whitney U test b
0.006–0.098
P \ 0.05
0.29–2.48
Textural Features of
18
F-FDG PET in HPSCC
TABLE 3 ROC Curve analyses of standardized uptake value (SUV), metabolic tumor volume (MTV), and measured textural features to predict chemoradiotherapy (CRT) response Variables
Area under ROC curve
95 % CI
P value
SUVmean
0.69
0.54–0.83
0.041
SUVmax
0.70
0.55–0.84
0.033
SUVpeak MTV (cm3)
0.72 0.87
0.59–0.86 0.79–0.96
0.016 \0.001
Coarseness
0.82
0.68–0.94
0.001
Busyness
0.74
0.59–0.87
0.005
Complexity
0.61
0.46–0.76
0.236
Contrast
0.67
0.51–0.83
0.066
ROC receiver operating characteristics
The SUVmax, SUVpeak, and MTV values differed significantly between responders and nonresponders (P \ 0.05). Among the textural features, low coarseness (P \ 0.001) and busyness (P = 0.015) were significantly associated with a high CRT response rate, whereas no significant differences in complexity and contrast were observed between responders and nonresponders. The results from AUC analyses of SUV, MTV, and textural features for the prediction of CRT response are shown in Table 3 and Supplementary Fig. S1. The highest AUC values were observed for MTV (0.87) and tumor coarseness (0.82). No statistically significant differences from the line of reference were observed for SUVmean or the textural parameters of complexity and contrast (P [ 0.05). Survival Prediction The median follow-up period for survivors was 56 months (range 18–100 months). During the follow-up period, 25 patients (36 %) had locoregional recurrences and 7 patients (10 %) had distant failure. At the last followup visit, 32 patients (46 %) were disease free, 3 (4 %) were alive with disease, 26 (37 %) had died of disease, and 9 (13 %) had died of other causes. For all the patients, the 5year DFS rate was 51.4 % (31 events) and the 5-year OS rate was 48.7 % (33 events). The cutoff values were 9.8 cm3 for SUVmax, 7.7 cm3 for SUVpeak, 23.6 cm3 for MTV, and respectively 0.042, 0.047, 8,214, and 1.46 for the textural parameters of busyness, coarseness, complexity, and contrast. To validate the selected cutoff points by ROC curve analysis, we performed dichotomization analysis of patient survival, with various cutoff points including mean and median values. The selected cutoff points by ROC curve analysis were associated with the lowest P value (\0.05) and were therefore considered as the optimal cutoff points.
Univariate analyses showed that advanced T classification (P = 0.004), high MTV (P \ 0.001), high tumor coarsensess (P \ 0.001), and busyness (P = 0.001) were significantly associated with decreased DFS, whereas high MTV (P = 0.014), coarseness (P = 0.004), and busyness (P = 0.025) were significantly associated with poor OS (Table 4). Multivariate Cox regression analyses showed that the textural parameters of coarseness (P = 0.001; HR 5.65; 95 % CI 2.12–15.07) and busyness (P = 0.045; HR 2.56; 95 % CI 1.02–6.42) were the only independent prognostic factors for DFS and that coarseness (P = 0.013; HR 2.48; 95 % CI 1.21–5.09) was the sole independent prognostic factor for OS (Fig. 1). A high coarseness value was associated with a greater risk of death from HPSCC or all causes than a low coarseness value. DISCUSSION The current study showed that textural features of 18FFDG uptake within PET images of HPSCC patients may predict response to CRT and survival outcomes. Our results supported previous evidence suggesting that SUV parameters and measurements of MTV and TLG are valuable imaging biomarkers predictive of therapeutic response in HNSCC.10–14 A previous study showed that MTV with a cutoff value of 40 mL was the only significant predictor of short-term outcome for patients treated with radiotherapy in terms of complete response and no early recurrence in 64 pharyngeal cancer patients.13 Findings show that TLG derived from baseline 18F-FDG PET scans may be a more reliable predictive factor of response to CRT than SUV in oral SCC.14 In addition to SUV and MTV parameters, our study showed that textural features measured from baseline 18FFDG PET images exhibited the ability to differentiate responders from nonresponders to CRT in HPSCC. Among the textural features derived from NGTDMs, coarseness and busyness were able to predict response to CRT. Two previous studies involving a limited number of patients (n = 10 and 20) indicated that 18F-FDG PET/CT-based textural characterization might help delineate radiotherapy planning volumes in head and neck cancer.4,5 Our study is the first to examine the predictive power of textural parameters for response and survival outcomes in 70 patients with M0 HPSCC. Several textural features of cancer imaging have been shown to predict treatment response or to be associated with survival.2–7 Increased intratumoral heterogeneity in CT or 18F-FDG PET images may be associated with tumor histologic and behavioral differences in terms of tumor cellularity, proliferation, necrosis, angiogenesis, and hypoxia.6,28
3.41 (0.81–14.28)
1.79 (0.89–3.60)
1.82 (0.90–3.69)
Clinical TNM stage (IV)
High SUVmax ([9.8)
High SUVpeak ([7.7)
4.53 (2.01–10.18)
3.39 (1.66–6.92)
1.89 (0.89–3.99)
1.05 (0.52–2.13)
High coarseness ([0.047)
High busyness ([0.042)
High complexity ([8,214)
Low contrast (\1.46)
0.891
0.096
0.001
b
\0.001
b
0.003
b
0.098
0.104
0.094
0.119
0.004
b
0.974
0.478 0.513
0.104
0.706
P valuea
2.56 (1.02–6.42)
5.65 (2.12–15.07)
2.51 (0.98–6.44)
3.33 (0.91–9.19)
Multivariable HR (95 % CI)a
0.045
b
0.001
b
0.055
0.089
P value
1.61 (0.79–3.30)
1.39 (0.71–2.75)
2.20 (1.11–4.38)
2.80 (1.39–5.64)
2.29 (1.18–4.47)
1.44 (0.74–2.81)
1.39 (0.71–2.70)
1.71 (0.60–4.87)
1.18 (0.80–1.72)
1.36 (0.68–2.70)
1.05 (0.46–2.43)
1.08 (0.53–2.21) 1.86 (0.89–3.89)
1.35 (0.47–3.83)
1.47 (0.75–2.88)
Univariable HR (95 % CI)a
Overall survival
b
a
P \ 0.05
Cox-proportional hazards model
HR hazard ratio, CI confidence interval, SUVmax maximum standardized uptake value, MTV metabolic tumor volume
2.94 (1.44–6.00)
High MTV ([23.6 cm )
3
3.69 (1.52–8.97)
1.14 (0.91–2.22)
1.02 (0.42–2.48)
Poorly differentiated
T classification (T3–T4)
1.24 (0.60–2.98) 1.27 (0.62–2.61)
Smoking ([20 pack-year) Heavy drinker
Clinical N stage (N2–N3)
1.14 (0.57–2.29)
2.22 (0.85–5.82)
Male sex
Univariable HR (95 % CI)
Disease-free survival
Age ([65 years)
Variables
TABLE 4 Univariate and multivariate analyses of variables associated with disease-free and overall survival
0.193
0.341
0.025
b
0.004b
0.014b
0.286
0.334
0.313
0.403
0.358
0.904
0.836 0.098
0.575
0.257
P value
1.79 (0.87–3.67)
2.48 (1.21–5.09)
1.82 (0.90–3.66)
Multivariable HR (95 % CI)a
0.111
0.013b
0.094
P value
J. S. Oh et al.
18
F-FDG PET in HPSCC
A
B 1.0
Disease-free survival probability
FIG. 1 Kaplan–Meier curves of disease-free survival (a, b) and overall survival (c, d) according to tumor coarseness and busyness on 18fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) images in the study population (n = 70). P \ 0.05, log-rank test
Disease-free survival probability
Textural Features of
Coarseness Low
0.8 0.6 0.4
High
0.2
p < 0.001 0.0 0
12
24
36
48
60
72
1.0
Busyness
0.8
Low 0.6 0.4
High 0.2
p < 0.001 0.0
84
0
12
Follow-up (months)
C
36
48
60
72
84
D 1.0
Coarseness
Overall survival probability
Overall survival probability
24
Follow-up (months)
0.8
Low 0.6
High
0.4 0.2
p = 0.003
1.0
Busyness
0.8 0.6
Low
0.4
High
0.2
p = 0.021 0.0
0.0 0
12
24
36
48
60
Follow-up (months)
Although various local, regional, and global textural features have been described,6 the complexity of measurements might oversimplify the relationship between tumor biology and heterogeneity. Because the relationship between tumor textural features and tissue characteristics is complex, differentiation between hetero- and homo-geneity is difficult using textural feature measurements derived from structural or functional images in cancer. However, textural features derived from NGTDMs might have the ability to predict response and survival.7 Local textural features including coarseness, contrast, busyness, and complexity have been correlated with the visual perception of texture within images.5,15 Coarseness, which quantifies tumor granularity within an image, has been described as the parameter that best discriminates responders to CRT from nonresponders.6,7 Moreover, it allows differentiation of primary and nodal tumors from normal tissues in head and neck cancer patients.4 Contrast is related to the dynamic range of intensity levels; busyness is related to the rate of intensity changes within an image; and complexity describes high information content such as many sharp edges or lines relevant to two-dimensional images. The three textural parameters of coarseness, contrast, and busyness were able to predict response to CRT and progression-free survival outcomes for patients with non-small cell lung cancer (NSCLC).7 The current study
72
84
0
12
24
36
48
60
72
84
Follow-up (months)
confirmed the ability of these parameters to predict DFS and OS outcomes and to differentiate responders to CRT from nonresponders. Among the SUVs, MTV, and local textural parameters previously reported,3,6,7 coarseness and busyness within the tumor were the only independent predictors for DFS, whereas coarseness was also related to OS and proved to be the only independent predictor of survival among other clinical and imaging risk factors. High MTV and T classification were predictors of long-term DFS in univariate analysis but not in the multivariate Cox regression analysis. The ineffectiveness of these parameters could be attributed to the inclusion of predominantly advanced-stage patients (92 %) and their subsequent selection for CRT rather than surgery for organ preservation. Traditionally, the AJCC tumor-node-metastasis (TNM) staging system used to stratify the disease status of cancer patients is the mainstay of decision making regarding treatment methods.21 Although TNM staging reflects overall disease status, it does not adequately assess tumor burden or biologic activity. Therefore, the outcomes of treatment and the prognosis of HPSCC patients cannot be fully determined with TNM staging alone. In addition, because a high MTV or TLG value in 18F-FDG PET/CT indicates a large tumor volume with metabolic activity, volumetric measurements have been used to predict
J. S. Oh et al.
survival in HNSCC.13,14 However, in the current study, analysis of 70 HPSCC patients who underwent CRT indicated that textural parameters assessed by 18F-FDG PET/ CT have superior prognostic value compared to SUVs or MTV, as reported previously in NSCLC.7 The current study had several limitations including its retrospective design, its small number of patients, and its inherent selection biases. Because HNSCCs are histologically identical but clinically heterogeneous entities, they differ in natural course or clinical behavior based on their primary location. Therefore, despite certain limitations, our study had the advantage of including only tumors of a single pathology arising in a single anatomic site in a relatively large cohort treated by CRT. The inclusion of data from different PET/CT scanners could be a potential source of bias for SUV and textural feature analyses. However, this was minimized by the harmonization of SUVs from multiple PET scanners using standard methods previously reported.23,24 Large-scale prospective trials are needed to validate our results and confirm the clinical significance of textural parameters measured by 18F-FDG PET/CT in this setting. In conclusion, intratumoral heterogeneity characterized by textural features extracted from pretreatment 18F-FDG PET images may identify patients at risk for low response rates and poor DFS and OS outcomes. The abnormal textural feature of tumor coarseness may be useful for predicting response and survival after CRT in HPSCC patients. This method could hold promise for personalized medicine and the pretreatment stratification of patients in clinical practice. ACKNOWLEDGMENT This study was supported by Grant No. 2014-0306 from the Asan Institute for Life Science and by Grant No. NRF-2012R1A1A2002039 from the Basic Science Research Program through the National Research Foundation of Korea and the Ministry of Education, Science and Technology, Seoul, Korea (J.-L.R.). CONFLICT OF INTEREST
There are no conflicts of interest.
REFERENCES 1. Cirkel GA, Gadellaa-van Hooijdonk CG, Koudijs MJ, Willems SM, Voest EE. Tumor heterogeneity and personalized cancer medicine: are we being outnumbered? Future Oncol. 2014;10:417–28. 2. Eary JF, O’Sullivan F, O’Sullivan J, Conrad EU. Spatial heterogeneity in sarcoma 18F-FDG uptake as a predictor of patient outcome. J Nucl Med. 2008;49:1973–9. 3. El Naqa I, Grigsby PW, Aptea A, et al. Exploring feature-based approaches in PET images for predicting cancer treatment outcomes. Pattern Recognit. 2009;42:1162–71. 4. Yu H, Caldwell C, Mah K, Poon I, Balogh J, MacKenzie R, et al. Automated radiation targeting in head-and-neck cancer using region-based texture analysis of PET and CT images. Int J Radiat Oncol Biol Phys. 2009;75:618–25.
5. Yu H, Caldwell C, Mah K, Mozeg D. Coregistered FDG PET/ CT-based textural characterization of head and neck cancer for radiation treatment planning. IEEE Trans Med Imaging. 2009;28:374–83. 6. Tixier F, Le Rest CC, Hatt M, et al. Intratumor heterogeneity characterized by textural features on baseline 18F-FDG PET images predicts response to concomitant radiochemotherapy in esophageal cancer. J Nucl Med. 2011;52:369–78. 7. Cook GJ, Yip C, Siddique M, et al. Are pretreatment 18F-FDG PET tumor textural features in non–small cell lung cancer associated with response and survival after chemoradiotherapy? J Nucl Med. 2013;54:19–26. 8. Antoch G, Saoudi N, Kuehl H, et al. Accuracy of whole-body dual-modality fluorine-18-2-fluoro-2-deoxy-D-glucose positron emission tomography and computed tomography (FDG-PET/CT) for tumor staging in solid tumors: comparison with CT and PET. J Clin Oncol. 2004;22:4357–68. 9. Lonneux M, Hamoir M, Reychler H, et al. Positron emission tomography with [18F] fluorodeoxyglucose improves staging and patient management in patients with head and neck squamous cell carcinoma: a multicenter prospective study. J Clin Oncol. 2010;28:1190–5. 10. Scho¨der H, Fury M, Lee N, Kraus D. PET monitoring of therapy response in head and neck squamous cell carcinoma. J Nucl Med. 2009;50(Suppl 1):74S–88S. 11. Moeller BJ, Rana V, Cannon BA, et al. Prospective risk-adjusted [18F] fluorodeoxyglucose positron emission tomography and computed tomography assessment of radiation response in head and neck cancer. J Clin Oncol. 2009;27:2509–15. 12. Wahl RL, Jacene H, Kasamon Y, Loddge MA. From RECIST to PERCIST: evolving considerations for PET response criteria in solid tumors. J Nucl Med. 2009;50(Suppl 1):122S–50S. 13. Chung MK, Jeong HS, Park SG, et al. Metabolic tumor volume of [18F]-fluorodeoxyglucose positron emission tomography/computed tomography predicts short-term outcome to radiotherapy with or without chemotherapy in pharyngeal cancer. Clin Cancer Res. 2009;15:5861–8. 14. Dibble EH, Alvarez AC, Truong MT, Mercier G, Cook EF, Subramaniam RM. 18F-FDG metabolic tumor volume and total glycolytic activity of oral cavity and oropharyngeal squamous cell cancer: adding value to clinical staging. J Nucl Med. 2012;53:709–15. 15. Amadasun M, King R. Textural features corresponding to textural properties. IEEE Trans Systems Man Cybern. 1989;19:1264–74. 16. Carvalho AL, Nishimoto IN, Califano JA, Kowalski LP. Trends in incidence and prognosis for head and neck cancer in the United States: a site-specific analysis of the SEER database. Int J Cancer. 2005;114:806–16. 17. Wycliffe ND, Grover RS, Kim PD, Simental A Jr. Hypopharyngeal cancer. Top Magn Reson Imaging. 2007;18:243–58. 18. Lefebvre JL. Candidates for larynx preservation: the next step? Oncologist. 2010;15(Suppl 3):30–2. 19. Morris LG, Sikora AG, Patel SG, Hayes RB, Ganly I. Second primary cancers after an index head and neck cancer: subsitespecific trends in the era of human papillomavirus-associated oropharyngeal cancer. J Clin Oncol. 2011;29:739–46. 20. Lee DH, Roh JL, Baek S, et al. Second cancer incidence, risk factor, and specific mortality in head and neck squamous cell carcinoma. Otolaryngol Head Neck Surg. 2013;149:579–86. 21. Edge SB, Byrd DR, Compton CC, Fritz AG, Greene FL, Trotti A (eds). AJCC cancer staging manual. 7th ed. New York: Springer; 2010, p. 41–56. 22. Therasse P, Arbuck SG, Eisenhauer EA, et al. New guidelines to evaluate the response to treatment in solid tumors (RECIST guidelines). J Natl Cancer Inst. 2000;92:205–16.
Textural Features of
18
F-FDG PET in HPSCC
23. Lasnon C, Desmonts C, Quak E, Gervais R, Do P, Dubos-Arvis C, Aide N. Harmonizing SUVs in multicentre trials using different generation PET systems: prospective validation in nonsmall cell lung cancer patients. Eur J Nucl Med Mol Imaging. 2013;40:985–96. 24. Boellaard R, O’Doherty MJ, Weber WA, et al. FDG PET and PET/CT: EANM procedure guidelines for tumor PET imaging: version 1.0. Eur J Nucl Med Mol Imaging. 2010;37:181–200. 25. Lee P, Weerasuriya DK, Lavori PW, et al. Metabolic tumor burden predicts for disease progression and death in lung cancer. Int J Radiat Oncol Biol Phys. 2007;69:328–33.
26. Haralick RM, Shanmugam K, Dinstein I. Texture features for image classification. IEEE Trans Syst Man Cybern. 1973;S3:610– 21. 27. Tixier F, Hatt M, Le Rest CC, Le Pogam A, Corcos L, Visvikis D. Reproducibility of tumor uptake heterogeneity characterization through textural feature analysis in 18F-FDG PET. J Nucl Med. 2012;53:693–700. 28. Ganeshan B, Goh V, Mandeville HC, Ng QS, Hoskin PJ, Miles KA. Non–small cell lung cancer: histopathologic correlates for texture parameters at CT. Radiology. 2013;266:326–36.
ORIGINAL ARTICLE – HEAD AND NECK ONCOLOGY
Intratumor Textural Heterogeneity on Pretreatment PET Images Predicts Response and Survival After Chemoradiotherapy for Hypopharyngeal Cancer
18
F-FDG
Jungsu S. Oh, PhD1, Bung Chul Kang, MD2, Jong-Lyel Roh, MD2, Jae Seung Kim, MD1, Kyung-Ja Cho, MD3, Sang-wook Lee, MD4, Sung-Bae Kim, MD5, Seung-Ho Choi, MD2, Soon Yuhl Nam, MD2, and Sang Yoon Kim, MD2,6 1
Department of Nuclear Medicine, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 2Department of Otolaryngology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 3Department of Pathology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 4Department of Radiation Oncology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 5Department of Internal Medicine, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea; 6Biomedical Research Institute, Korea Institute of Science and Technology, Seoul, Republic of Korea
ABSTRACT Background. Increasing evidence suggests that intratumor heterogeneity of solid tumors characterized by textural features on 18-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) images is associated with response to chemoradiotherapy (CRT) and survival. The current study aimed to determine whether a similar relationship exists in hypopharyngeal squamous cell carcinoma (HPSCC). Methods. This study investigated 27 patients with HPSCC who underwent cisplatin-based induction chemotherapy followed by definitive CRT underwent pretreatment 18FFDG PET/CT. Standardized uptake value (SUV), metabolic tumor volume (MTV), and textural features (coarseness, busyness, complexity, and contrast) of primary tumors were measured. Patients were classified as
Jungsu S. Oh, and Bung Chul Kang have contributed equally to this paper.
Electronic supplementary material The online version of this article (doi:10.1245/s10434-014-4284-3) contains supplementary material, which is available to authorized users. Ó Society of Surgical Oncology 2014 First Received: 26 September 2014 J.-L. Roh, MD e-mail: [email protected]
nonresponders or responders according to the response evaluation criteria in solid tumors. The capacity of each parameter to classify response was assessed using the Mann–Whitney U test. Cox-proportional hazard models were used to identify variables associated with disease-free survival (DFS) and overall survival (OS). Results. Of 70 patients, 58 (83 %) had complete or partial response after CRT. Responders showed lower maximum SUV (P = 0.037), lower MTV (P = 0.039), lower coarseness (P \ 0.001), and busyness (P = 0.015) compared with nonresponders. Multivariate analysis showed that high coarseness (P = 0.001, hazard ratio [HR] 5.65; 95 % confidence interval [CI] 2.12–15.07) and busyness (P = 0.045; HR 2.56; 95 % CI 1.02–6.42) were independently associated with poor DFS, and that high coarseness (P = 0.013; HR 2.48; 95 % CI 1.21–5.09) was independently associated with poor OS. Conclusion. Abnormal coarseness in baseline 18F-FDG PET scans may be useful for predicting response and survival after CRT in HPSCC patients.
The goal of precision medicine in modern cancer therapy is to select the optimal treatment method and enhance treatment outcomes for individual cancer patients. Accurate prediction of tumor response to treatment is essential to guide therapeutic stratification and personalized cancer therapy. The heterogeneity of tumor cells, as indicated by their distinct morphologic and phenotypic profiles, is regarded as a major obstacle to the
J. S. Oh et al.
success of precision medicine.1 Tumor heterogeneity related to therapeutic resistance is identified by tumor biopsy and histopathology, molecular and genetic profiling, and innovative response assays.1 Increasing evidence suggests that intratumoral spatial heterogeneity can be characterized by textural features on pretreatment fluorine 18-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) images and that these features are associated with response and survival in solid tumors.2–7 As a diagnostic imaging method, 18F-FDG PET/computed tomography (CT) is based on the uptake of tracers for estimation of tumor glucose metabolism. This method is widely used for tumor staging and for evaluating the therapeutic responses of solid tumors, including head and neck squamous cell carcinoma (HNSCC).8–11 Various standardized uptake value (SUV) parameters are predictive of therapeutic response.11,12 However, whether high or low SUVs are predictive of the response to different treatment methods remains controversial. Metabolic tumor volume (MTV) and total lesion glycolysis (TLG), which assess tumor volume and metabolic activity of the entire tumor, respectively, are used as prognostic biomarkers in HNSCC.13,14 However, tumor textural features on baseline 18F-FDG PET images were recently shown to have a higher predictive value for response and survival after chemoradiotherapy (CRT) than SUVs, MTV, or TLG.7 Textural parameters are derived from neighborhood gray-tone difference matrices (NGTDM), which indicate differences between the intensity of each voxel and its neighboring voxels and describe features such as coarseness, contrast, busyness, and complexity.15 This allows differentiation between tumor and normal tissues as well as delineation of target volumes for radiotherapy.4,5 Hypopharyngeal squamous cell carcinoma (HPSCC), a rare type of tumor accounting for approximately 6.5 % of all head and neck cancers, is commonly detected at an advanced stage in patients with regional stage disease.16,17 Nonsurgical laryngeal preservation strategies for HPSCC have been highlighted to preserve voice and swallowing functions.18 The prognosis of HPSCC remains poor despite recent improvements in diagnostic methods and treatment methods, with a 5-year survival rate of 25–40 %.16,17 The poor survival of HPSCC patients is partly due to the common occurrence of second primary cancers and a higher rate of distant metastasis than that of other head and neck cancers.19,20 Therefore, biomarkers that can accurately predict response and survival after treatment in HPSCC patients are needed. We hypothesized that the textural features derived from the NGTDMs of baseline 18F-FDG PET images may be predictive of response to CRT and survival in HPSCC. The current study aimed to compare the predictive and prognostic value of textural parameters in pretreatment 18FFDG PET images with that of SUV or MTV in HPSCC patients who underwent CRT.
PATIENTS AND METHODS Study Population The cancer registry records of consecutive patients with biopsy-confirmed HPSCC treated at our tertiary referral center for cancer patients between January 2006 and December 2011 were reviewed. The inclusion criteria for the study specified previously untreated HPSCC patients without distant metastasis who underwent pretreatment 18 F-FDG PET/CT scanning for initial staging and concurrent CRT. The exclusion criteria ruled out patients who initially underwent definitive surgery, those with a history of head and neck cancer, those with distant metastasis at initial presentation, and those with incomplete treatment or clinical follow-up data. All surviving patients were followed up for at least 12 months. Finally, 70 patients (64 men and 6 women) with a median age of 64 years (range 40–73 years) were eligible for the study. Data obtained from medical records included clinicopathologic variables, treatment regimens, and follow-up evaluation assessing the recurrence and survival of each patient. Tumors were staged according to the American Joint Committee on Cancer (AJCC) staging system.21 The patient clinicopathologic data, treatment methods, CRT response, and follow-up information are described in Table 1. This study was approved by the Institutional Review Board of our institution, and informed consent from each patient was waived because of the study’s retrospective design.
Treatments All eligible patients underwent induction chemotherapy using two or three cycles of cisplatin plus 5-fluorouracil with or without docetaxel administration before CRT. Patients were treated with radiotherapy using a median of 70 Gy (range 54–78 Gy) and concurrent chemotherapy with high-dose cisplatin (80–100 mg/m2) intravenously infused on days 1, 22, and 43. Radiotherapy consisting of intensity-modulated or three-dimensional conformal radiation was administered in a single daily fraction of 1.8 or 2.0 Gy, 5 days per week for 7–8 weeks. The initial tumor response was assessed 2 months after completion of CRT with the response evaluation criteria in solid tumors (RECIST) using clinical, endoscopic, and CT or magnetic resonance imaging (MRI) examination.22 All patients received regular physical and endoscopic examinations at every clinic visit after treatments. The clinical follow-up visit was scheduled according to departmental protocol: every 1–2 months for the first year, every 2– 4 months for the second and third years, and every 6 months
Textural Features of
18
F-FDG PET in HPSCC
TABLE 1 Patient characteristics (n = 70) Characteristics
Value n (%)
Median: years (range)
64 (40–73)
Sex Male/female
64 (91)/6 (9)
Smoking Never/\20/C20 pack-year
10 (14)/12 (17)/48 (69)
Alcohol drinking Never/social/heavy drinker Performance status scale by ECOG 0/1
16 (23)/26 (37)/28 (40) 68 (97)/2 (3)
Charlson comorbidity index 0/1/C2
56 (80)/12 (17)/2 (3)
Histologic differentiation Well/moderately/poorly differentiated
6 (9)/49 (70)/15 (21)
Clinical TNM stage T1/T2/T3/T4
12 (17)/17(24)/11 (16)/30 (43)
N0/N1/N2/N3
8 (11)/12 (17)/41 (59)/9 (13)
Overall stage 1/2/3/4
3 (4)/3 (4)/6 (9)/58 (83)
Treatment Induction chemotherapy ? CRT
70 (100)
Response to induction chemotherapy CR/PR/SD/PD
6 (9)/48 (69)/4 (6)/12 (17)
Response to concurrent CRT CR/PR/SD/PD
47 (67)/11 (16)/8 (11)/4 (6)
Follow-up information Median follow-up of survivors: months 56 (18–100) (range) Locoregional recurrences
25 (36)
Distant failure
7 (10)
Status at last follow-up visit NED/AD/DOD/DOC
32 (46)/3 (4)/26 (37)/9 (13)
ECOG eastern cooperative oncology group, TNM tumor-nodemetastasis stage, CRT chemoradiotherapy, CR complete response, PR partial response, SD stable disease, PD progressive disease, NED no evidence of disease, AD alive with disease, DOD died of disease, DOC died of other causes
thereafter. Any lesions suggestive of recurrence or a second primary tumor were confirmed by biopsy and specific additional diagnostic tests. Patients with confirmed recurrence were scheduled for salvage or palliative treatment. 18
F-FDG PET/CT Imaging and SUV Harmonization
The 18F-FDG PET/CT scans in this study were performed using a Biograph Sensation 16 or TruePoint 40 system (Siemens Medical Systems, Knoxville, TN, USA)
equipped with a 16- or 40-slice CT, respectively, for both anatomic information and attenuation correction purposes. All patients were required to fast for 6 h or longer and to have a serum glucose concentration lower than 150 mg/dL before receiving the injection of 18F-FDG. Whole-body images were acquired 50–70 min after intravenous injection of 370–688 MBq (10.0–18.6 mCi). Using the spiral mode, 18F-FDG. CT data were collected from the thigh to the skull at 100 mAs and 120 kV with a section width of 5and 0.75-mm collimation. No oral or intravenous contrast medium was used during the CT scans. Three-dimensional craniocaudal PET was performed during an acquisition time of 2.5 min per cradle position for eight bed positions. The PET results were reconstructed using a standard iterative algorithm based on the CT results. Because the intrinsic spatial resolution measured by the national electrical manufacturers association (NEMA) NU-2 test may differ between the PET scanners used in this study, mere standardization of the reconstruction method could not fully resolve the issue of scanner difference. Moreover, the newer-generation scanner (Biograph TruePoint 40), which uses a point spread function (PSF)-based reconstruction method called True-X, had a much higher spatial resolution than the other scanner. We estimated recovery coefficients (i.e., relative SUV ratios over the ideal SUV of 2.5) for lesions of different sizes using the American College of Radiology (ACR)-approved phantom Esser Flangeless Deluxe (Data Spectrum, Hillsborough, NC, USA) filled with 18 F-FDG. The SUVs of several hot cylinders with differing diameters were measured, and recovery coefficient plots were generated, allowing estimation of the optimal smoothing kernel size for matching the different recovery coefficients, as described previously.23,24 In addition, all PET images were resampled into 5 9 5 9 5-mm3 voxels, the largest voxel size between many scanners, to ensure a matched ‘‘image scale’’ between scanners. This is desirable for optimized texture analyses because texture is dependent on the arrangement of intensities in the image, and this can be affected by voxel size (in other words, ‘‘scale’’). Tumor Image Analysis After the between-scanner harmonization, 18F-FDG PET parameters including mean SUV (SUVmean), maximum SUV (SUVmax), peak SUV (SUVpeak),12 MTV,13,14 and texture features 6,7 were measured for each patient. The region of interest (ROI) was manually drawn to enclose three-dimensional coverage of the entire hypermetabolic primary tumor lesion visible on PET images in all three projections (axial, sagittal, and coronal). Metabolic tumor volume was defined as the volume enclosed by an isocontour of SUV2.5.13,25 One board-certified nuclear
J. S. Oh et al.
medicine physician (J.S.K.) with 20 years of clinical experience identified the lesions with increased tracer uptake, and a physician and nuclear medicine physicist (J.S.O.) blinded to patient outcome performed the analyses. The textural features of primary tumors measured were coarseness, busyness, complexity, and contrast derived from three-dimensional matrices and NGTDMs, as previously described.7,15 The NGTDM-based texture analysis method was selected for measuring intratumor heterogeneity because it was considered appropriate for analyzing relatively small tumor lesions in this study. Nodal disease was not included because textural analysis of small metastatic lesions is not reliable with a small number of voxels.6,7 The textural features on baseline 18F-FDG PET images were analyzed using the in-house software, AMC NM Toolkit for Image Quantification of Excellence (ANTIQUE), developed as previously reported.7,15,26 Textural features have shown similar or better reproducibility than simple SUV measurement.27
prognostic value of clinical and imaging parameters for predicting disease-free survival (DFS) and overall survival (OS). Time periods were calculated from the date of CRT completion to the date of a confirmed event (any recurrence or death) or the last clinical follow-up visit. The survival rates associated with each variable plotted using the Kaplan–Meier method were compared using a nonparametric log-rank test. The Cox-proportional hazard regression model was used to evaluate prognostic variables in uni- and multivariate analyses for prediction of DFS and OS. Tests were based on the likelihood-ratio statistic. The estimated hazard ratio (HR) and the 95 % confidence interval (CI) were calculated. Variables with a P value lower than 0.05 in univariate analyses were selected for multivariate analysis. A two-sided P value lower than 0.05 was considered statistically significant. All statistical analyses were performed using IBM SPSS software version 21.0 (IBM, Armonk, NY, USA). RESULTS
Statistical Analysis Response Prediction Mean differences in SUVs, MTV, and textural parameters were compared between the RECIST responders (complete response or partial response) and the nonresponders (stable or progressive disease) using the Mann– Whitney U test. The area under the receiver operating characteristic (ROC) curve (AUC) was calculated for SUV, MTV, and textural parameters to assess sensitivity and specificity for predicting CRT response and to obtain the optimal cutoff values for survival prediction. Uni- and multi-variate analyses were performed to evaluate the
The distribution of the SUVs, MTV, and textural parameters is summarized in Table 2. After the completion of definitive CRT, complete responses were achieved for 47 patients (67 %) and partial responses for 11 patients (16 %), with stable disease was observed in 8 patients (11 %) and progressive disease in 4 patients (6 %) (Table 1). Salvage surgery was performed for 23 patients (33 %) who did not achieve a complete response.
TABLE 2 Comparison of standardized uptake value (SUV), metabolic tumor volume (MTV), and measured textural features for complete or partial response versus stable or progressive disease after concurrent chemoradiotherapy Variables
Total (n = 70) Mean ± SD
P valuea
Mean CRT response Range
Responders
Nonresponders
SUV parameters SUVmean
7.4 ± 5.6
1.1–33.0
7.0 ± 5.0
10.2 ± 7.8
0.056b
SUVmax
11.9 ± 8.7
1.5–50.88
10.9 ± 7.6
16.6 ± 12.1
0.037b
9.8 ± 7.4
1.3–40.8
9.0 ± 6.6
13.9 ± 9.6
0.034b
30.2 ± 42.5
0–295.2
21.2 ± 23.1
74.0 ± 77.6
0.039b
0.043 ± 0.012
0.059 ± 0.017
SUVpeak 3
MTV (cm ) Tumor heterogeneity Coarseness
0.047 ± 0.018
\0.001b
Busyness
0.037 ± 0.018
0.017–0.154
0.035 ± 0.011
0.049 ± 0.036
0.015
Complexity
8,585 ± 2,948
2,706–16,343
8,403 ± 3,056
9,466 ± 9,598
0.258
1.34 ± 0.56
1.04 ± 0.5
0.086
Contrast
1.29 ± 1.19
CRT chemoradiotherapy, SD standard deviation a Assessed using the Mann–Whitney U test b
0.006–0.098
P \ 0.05
0.29–2.48
Textural Features of
18
F-FDG PET in HPSCC
TABLE 3 ROC Curve analyses of standardized uptake value (SUV), metabolic tumor volume (MTV), and measured textural features to predict chemoradiotherapy (CRT) response Variables
Area under ROC curve
95 % CI
P value
SUVmean
0.69
0.54–0.83
0.041
SUVmax
0.70
0.55–0.84
0.033
SUVpeak MTV (cm3)
0.72 0.87
0.59–0.86 0.79–0.96
0.016 \0.001
Coarseness
0.82
0.68–0.94
0.001
Busyness
0.74
0.59–0.87
0.005
Complexity
0.61
0.46–0.76
0.236
Contrast
0.67
0.51–0.83
0.066
ROC receiver operating characteristics
The SUVmax, SUVpeak, and MTV values differed significantly between responders and nonresponders (P \ 0.05). Among the textural features, low coarseness (P \ 0.001) and busyness (P = 0.015) were significantly associated with a high CRT response rate, whereas no significant differences in complexity and contrast were observed between responders and nonresponders. The results from AUC analyses of SUV, MTV, and textural features for the prediction of CRT response are shown in Table 3 and Supplementary Fig. S1. The highest AUC values were observed for MTV (0.87) and tumor coarseness (0.82). No statistically significant differences from the line of reference were observed for SUVmean or the textural parameters of complexity and contrast (P [ 0.05). Survival Prediction The median follow-up period for survivors was 56 months (range 18–100 months). During the follow-up period, 25 patients (36 %) had locoregional recurrences and 7 patients (10 %) had distant failure. At the last followup visit, 32 patients (46 %) were disease free, 3 (4 %) were alive with disease, 26 (37 %) had died of disease, and 9 (13 %) had died of other causes. For all the patients, the 5year DFS rate was 51.4 % (31 events) and the 5-year OS rate was 48.7 % (33 events). The cutoff values were 9.8 cm3 for SUVmax, 7.7 cm3 for SUVpeak, 23.6 cm3 for MTV, and respectively 0.042, 0.047, 8,214, and 1.46 for the textural parameters of busyness, coarseness, complexity, and contrast. To validate the selected cutoff points by ROC curve analysis, we performed dichotomization analysis of patient survival, with various cutoff points including mean and median values. The selected cutoff points by ROC curve analysis were associated with the lowest P value (\0.05) and were therefore considered as the optimal cutoff points.
Univariate analyses showed that advanced T classification (P = 0.004), high MTV (P \ 0.001), high tumor coarsensess (P \ 0.001), and busyness (P = 0.001) were significantly associated with decreased DFS, whereas high MTV (P = 0.014), coarseness (P = 0.004), and busyness (P = 0.025) were significantly associated with poor OS (Table 4). Multivariate Cox regression analyses showed that the textural parameters of coarseness (P = 0.001; HR 5.65; 95 % CI 2.12–15.07) and busyness (P = 0.045; HR 2.56; 95 % CI 1.02–6.42) were the only independent prognostic factors for DFS and that coarseness (P = 0.013; HR 2.48; 95 % CI 1.21–5.09) was the sole independent prognostic factor for OS (Fig. 1). A high coarseness value was associated with a greater risk of death from HPSCC or all causes than a low coarseness value. DISCUSSION The current study showed that textural features of 18FFDG uptake within PET images of HPSCC patients may predict response to CRT and survival outcomes. Our results supported previous evidence suggesting that SUV parameters and measurements of MTV and TLG are valuable imaging biomarkers predictive of therapeutic response in HNSCC.10–14 A previous study showed that MTV with a cutoff value of 40 mL was the only significant predictor of short-term outcome for patients treated with radiotherapy in terms of complete response and no early recurrence in 64 pharyngeal cancer patients.13 Findings show that TLG derived from baseline 18F-FDG PET scans may be a more reliable predictive factor of response to CRT than SUV in oral SCC.14 In addition to SUV and MTV parameters, our study showed that textural features measured from baseline 18FFDG PET images exhibited the ability to differentiate responders from nonresponders to CRT in HPSCC. Among the textural features derived from NGTDMs, coarseness and busyness were able to predict response to CRT. Two previous studies involving a limited number of patients (n = 10 and 20) indicated that 18F-FDG PET/CT-based textural characterization might help delineate radiotherapy planning volumes in head and neck cancer.4,5 Our study is the first to examine the predictive power of textural parameters for response and survival outcomes in 70 patients with M0 HPSCC. Several textural features of cancer imaging have been shown to predict treatment response or to be associated with survival.2–7 Increased intratumoral heterogeneity in CT or 18F-FDG PET images may be associated with tumor histologic and behavioral differences in terms of tumor cellularity, proliferation, necrosis, angiogenesis, and hypoxia.6,28
3.41 (0.81–14.28)
1.79 (0.89–3.60)
1.82 (0.90–3.69)
Clinical TNM stage (IV)
High SUVmax ([9.8)
High SUVpeak ([7.7)
4.53 (2.01–10.18)
3.39 (1.66–6.92)
1.89 (0.89–3.99)
1.05 (0.52–2.13)
High coarseness ([0.047)
High busyness ([0.042)
High complexity ([8,214)
Low contrast (\1.46)
0.891
0.096
0.001
b
\0.001
b
0.003
b
0.098
0.104
0.094
0.119
0.004
b
0.974
0.478 0.513
0.104
0.706
P valuea
2.56 (1.02–6.42)
5.65 (2.12–15.07)
2.51 (0.98–6.44)
3.33 (0.91–9.19)
Multivariable HR (95 % CI)a
0.045
b
0.001
b
0.055
0.089
P value
1.61 (0.79–3.30)
1.39 (0.71–2.75)
2.20 (1.11–4.38)
2.80 (1.39–5.64)
2.29 (1.18–4.47)
1.44 (0.74–2.81)
1.39 (0.71–2.70)
1.71 (0.60–4.87)
1.18 (0.80–1.72)
1.36 (0.68–2.70)
1.05 (0.46–2.43)
1.08 (0.53–2.21) 1.86 (0.89–3.89)
1.35 (0.47–3.83)
1.47 (0.75–2.88)
Univariable HR (95 % CI)a
Overall survival
b
a
P \ 0.05
Cox-proportional hazards model
HR hazard ratio, CI confidence interval, SUVmax maximum standardized uptake value, MTV metabolic tumor volume
2.94 (1.44–6.00)
High MTV ([23.6 cm )
3
3.69 (1.52–8.97)
1.14 (0.91–2.22)
1.02 (0.42–2.48)
Poorly differentiated
T classification (T3–T4)
1.24 (0.60–2.98) 1.27 (0.62–2.61)
Smoking ([20 pack-year) Heavy drinker
Clinical N stage (N2–N3)
1.14 (0.57–2.29)
2.22 (0.85–5.82)
Male sex
Univariable HR (95 % CI)
Disease-free survival
Age ([65 years)
Variables
TABLE 4 Univariate and multivariate analyses of variables associated with disease-free and overall survival
0.193
0.341
0.025
b
0.004b
0.014b
0.286
0.334
0.313
0.403
0.358
0.904
0.836 0.098
0.575
0.257
P value
1.79 (0.87–3.67)
2.48 (1.21–5.09)
1.82 (0.90–3.66)
Multivariable HR (95 % CI)a
0.111
0.013b
0.094
P value
J. S. Oh et al.
18
F-FDG PET in HPSCC
A
B 1.0
Disease-free survival probability
FIG. 1 Kaplan–Meier curves of disease-free survival (a, b) and overall survival (c, d) according to tumor coarseness and busyness on 18fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) images in the study population (n = 70). P \ 0.05, log-rank test
Disease-free survival probability
Textural Features of
Coarseness Low
0.8 0.6 0.4
High
0.2
p < 0.001 0.0 0
12
24
36
48
60
72
1.0
Busyness
0.8
Low 0.6 0.4
High 0.2
p < 0.001 0.0
84
0
12
Follow-up (months)
C
36
48
60
72
84
D 1.0
Coarseness
Overall survival probability
Overall survival probability
24
Follow-up (months)
0.8
Low 0.6
High
0.4 0.2
p = 0.003
1.0
Busyness
0.8 0.6
Low
0.4
High
0.2
p = 0.021 0.0
0.0 0
12
24
36
48
60
Follow-up (months)
Although various local, regional, and global textural features have been described,6 the complexity of measurements might oversimplify the relationship between tumor biology and heterogeneity. Because the relationship between tumor textural features and tissue characteristics is complex, differentiation between hetero- and homo-geneity is difficult using textural feature measurements derived from structural or functional images in cancer. However, textural features derived from NGTDMs might have the ability to predict response and survival.7 Local textural features including coarseness, contrast, busyness, and complexity have been correlated with the visual perception of texture within images.5,15 Coarseness, which quantifies tumor granularity within an image, has been described as the parameter that best discriminates responders to CRT from nonresponders.6,7 Moreover, it allows differentiation of primary and nodal tumors from normal tissues in head and neck cancer patients.4 Contrast is related to the dynamic range of intensity levels; busyness is related to the rate of intensity changes within an image; and complexity describes high information content such as many sharp edges or lines relevant to two-dimensional images. The three textural parameters of coarseness, contrast, and busyness were able to predict response to CRT and progression-free survival outcomes for patients with non-small cell lung cancer (NSCLC).7 The current study
72
84
0
12
24
36
48
60
72
84
Follow-up (months)
confirmed the ability of these parameters to predict DFS and OS outcomes and to differentiate responders to CRT from nonresponders. Among the SUVs, MTV, and local textural parameters previously reported,3,6,7 coarseness and busyness within the tumor were the only independent predictors for DFS, whereas coarseness was also related to OS and proved to be the only independent predictor of survival among other clinical and imaging risk factors. High MTV and T classification were predictors of long-term DFS in univariate analysis but not in the multivariate Cox regression analysis. The ineffectiveness of these parameters could be attributed to the inclusion of predominantly advanced-stage patients (92 %) and their subsequent selection for CRT rather than surgery for organ preservation. Traditionally, the AJCC tumor-node-metastasis (TNM) staging system used to stratify the disease status of cancer patients is the mainstay of decision making regarding treatment methods.21 Although TNM staging reflects overall disease status, it does not adequately assess tumor burden or biologic activity. Therefore, the outcomes of treatment and the prognosis of HPSCC patients cannot be fully determined with TNM staging alone. In addition, because a high MTV or TLG value in 18F-FDG PET/CT indicates a large tumor volume with metabolic activity, volumetric measurements have been used to predict
J. S. Oh et al.
survival in HNSCC.13,14 However, in the current study, analysis of 70 HPSCC patients who underwent CRT indicated that textural parameters assessed by 18F-FDG PET/ CT have superior prognostic value compared to SUVs or MTV, as reported previously in NSCLC.7 The current study had several limitations including its retrospective design, its small number of patients, and its inherent selection biases. Because HNSCCs are histologically identical but clinically heterogeneous entities, they differ in natural course or clinical behavior based on their primary location. Therefore, despite certain limitations, our study had the advantage of including only tumors of a single pathology arising in a single anatomic site in a relatively large cohort treated by CRT. The inclusion of data from different PET/CT scanners could be a potential source of bias for SUV and textural feature analyses. However, this was minimized by the harmonization of SUVs from multiple PET scanners using standard methods previously reported.23,24 Large-scale prospective trials are needed to validate our results and confirm the clinical significance of textural parameters measured by 18F-FDG PET/CT in this setting. In conclusion, intratumoral heterogeneity characterized by textural features extracted from pretreatment 18F-FDG PET images may identify patients at risk for low response rates and poor DFS and OS outcomes. The abnormal textural feature of tumor coarseness may be useful for predicting response and survival after CRT in HPSCC patients. This method could hold promise for personalized medicine and the pretreatment stratification of patients in clinical practice. ACKNOWLEDGMENT This study was supported by Grant No. 2014-0306 from the Asan Institute for Life Science and by Grant No. NRF-2012R1A1A2002039 from the Basic Science Research Program through the National Research Foundation of Korea and the Ministry of Education, Science and Technology, Seoul, Korea (J.-L.R.). CONFLICT OF INTEREST
There are no conflicts of interest.
REFERENCES 1. Cirkel GA, Gadellaa-van Hooijdonk CG, Koudijs MJ, Willems SM, Voest EE. Tumor heterogeneity and personalized cancer medicine: are we being outnumbered? Future Oncol. 2014;10:417–28. 2. Eary JF, O’Sullivan F, O’Sullivan J, Conrad EU. Spatial heterogeneity in sarcoma 18F-FDG uptake as a predictor of patient outcome. J Nucl Med. 2008;49:1973–9. 3. El Naqa I, Grigsby PW, Aptea A, et al. Exploring feature-based approaches in PET images for predicting cancer treatment outcomes. Pattern Recognit. 2009;42:1162–71. 4. Yu H, Caldwell C, Mah K, Poon I, Balogh J, MacKenzie R, et al. Automated radiation targeting in head-and-neck cancer using region-based texture analysis of PET and CT images. Int J Radiat Oncol Biol Phys. 2009;75:618–25.
5. Yu H, Caldwell C, Mah K, Mozeg D. Coregistered FDG PET/ CT-based textural characterization of head and neck cancer for radiation treatment planning. IEEE Trans Med Imaging. 2009;28:374–83. 6. Tixier F, Le Rest CC, Hatt M, et al. Intratumor heterogeneity characterized by textural features on baseline 18F-FDG PET images predicts response to concomitant radiochemotherapy in esophageal cancer. J Nucl Med. 2011;52:369–78. 7. Cook GJ, Yip C, Siddique M, et al. Are pretreatment 18F-FDG PET tumor textural features in non–small cell lung cancer associated with response and survival after chemoradiotherapy? J Nucl Med. 2013;54:19–26. 8. Antoch G, Saoudi N, Kuehl H, et al. Accuracy of whole-body dual-modality fluorine-18-2-fluoro-2-deoxy-D-glucose positron emission tomography and computed tomography (FDG-PET/CT) for tumor staging in solid tumors: comparison with CT and PET. J Clin Oncol. 2004;22:4357–68. 9. Lonneux M, Hamoir M, Reychler H, et al. Positron emission tomography with [18F] fluorodeoxyglucose improves staging and patient management in patients with head and neck squamous cell carcinoma: a multicenter prospective study. J Clin Oncol. 2010;28:1190–5. 10. Scho¨der H, Fury M, Lee N, Kraus D. PET monitoring of therapy response in head and neck squamous cell carcinoma. J Nucl Med. 2009;50(Suppl 1):74S–88S. 11. Moeller BJ, Rana V, Cannon BA, et al. Prospective risk-adjusted [18F] fluorodeoxyglucose positron emission tomography and computed tomography assessment of radiation response in head and neck cancer. J Clin Oncol. 2009;27:2509–15. 12. Wahl RL, Jacene H, Kasamon Y, Loddge MA. From RECIST to PERCIST: evolving considerations for PET response criteria in solid tumors. J Nucl Med. 2009;50(Suppl 1):122S–50S. 13. Chung MK, Jeong HS, Park SG, et al. Metabolic tumor volume of [18F]-fluorodeoxyglucose positron emission tomography/computed tomography predicts short-term outcome to radiotherapy with or without chemotherapy in pharyngeal cancer. Clin Cancer Res. 2009;15:5861–8. 14. Dibble EH, Alvarez AC, Truong MT, Mercier G, Cook EF, Subramaniam RM. 18F-FDG metabolic tumor volume and total glycolytic activity of oral cavity and oropharyngeal squamous cell cancer: adding value to clinical staging. J Nucl Med. 2012;53:709–15. 15. Amadasun M, King R. Textural features corresponding to textural properties. IEEE Trans Systems Man Cybern. 1989;19:1264–74. 16. Carvalho AL, Nishimoto IN, Califano JA, Kowalski LP. Trends in incidence and prognosis for head and neck cancer in the United States: a site-specific analysis of the SEER database. Int J Cancer. 2005;114:806–16. 17. Wycliffe ND, Grover RS, Kim PD, Simental A Jr. Hypopharyngeal cancer. Top Magn Reson Imaging. 2007;18:243–58. 18. Lefebvre JL. Candidates for larynx preservation: the next step? Oncologist. 2010;15(Suppl 3):30–2. 19. Morris LG, Sikora AG, Patel SG, Hayes RB, Ganly I. Second primary cancers after an index head and neck cancer: subsitespecific trends in the era of human papillomavirus-associated oropharyngeal cancer. J Clin Oncol. 2011;29:739–46. 20. Lee DH, Roh JL, Baek S, et al. Second cancer incidence, risk factor, and specific mortality in head and neck squamous cell carcinoma. Otolaryngol Head Neck Surg. 2013;149:579–86. 21. Edge SB, Byrd DR, Compton CC, Fritz AG, Greene FL, Trotti A (eds). AJCC cancer staging manual. 7th ed. New York: Springer; 2010, p. 41–56. 22. Therasse P, Arbuck SG, Eisenhauer EA, et al. New guidelines to evaluate the response to treatment in solid tumors (RECIST guidelines). J Natl Cancer Inst. 2000;92:205–16.
Textural Features of
18
F-FDG PET in HPSCC
23. Lasnon C, Desmonts C, Quak E, Gervais R, Do P, Dubos-Arvis C, Aide N. Harmonizing SUVs in multicentre trials using different generation PET systems: prospective validation in nonsmall cell lung cancer patients. Eur J Nucl Med Mol Imaging. 2013;40:985–96. 24. Boellaard R, O’Doherty MJ, Weber WA, et al. FDG PET and PET/CT: EANM procedure guidelines for tumor PET imaging: version 1.0. Eur J Nucl Med Mol Imaging. 2010;37:181–200. 25. Lee P, Weerasuriya DK, Lavori PW, et al. Metabolic tumor burden predicts for disease progression and death in lung cancer. Int J Radiat Oncol Biol Phys. 2007;69:328–33.
26. Haralick RM, Shanmugam K, Dinstein I. Texture features for image classification. IEEE Trans Syst Man Cybern. 1973;S3:610– 21. 27. Tixier F, Hatt M, Le Rest CC, Le Pogam A, Corcos L, Visvikis D. Reproducibility of tumor uptake heterogeneity characterization through textural feature analysis in 18F-FDG PET. J Nucl Med. 2012;53:693–700. 28. Ganeshan B, Goh V, Mandeville HC, Ng QS, Hoskin PJ, Miles KA. Non–small cell lung cancer: histopathologic correlates for texture parameters at CT. Radiology. 2013;266:326–36.
