Esophageal squamous cell carcinoma (ESCC) is among the top 10 most commonly occurring types of cancer worldwide 1. Although the incidence of esophageal adenocarcinoma is increasing rapidly in Western countries, ESCC remains the predominant histological subtype in East Asia, including China 2. At present, the potentially curative treatment for ESCC is surgery or definitive chemoradiotherapy. Therefore, accurate staging is essential for efficacious treatment planning and prognostication.
18F-fluorodeoxyglucose (18F-FDG) PET is promising in the clinical diagnosis, staging, and follow-up of patients with esophageal cancer 3. Maximum standardized uptake value (SUVmax) is one of the most widely used parameters for the analysis of 18F-FDG PET images and has been shown to be valuable in the prognosis and response prediction to therapy of patients with ESCC. The use of SUVmax was consistent with many reports included in the previous meta-analysis assessing its prognostic significance in esophageal cancer 4,5. However, intratumoral uptake of 18F-FDG is usually not homogenous across the tumor because of necrosis, cell proliferation, microvessel density, and hypoxia 6–8. Therefore, intratumoral heterogeneity may complicate accurate assessment of 18F-FDG uptake.
Recently, PET image texture analysis was proposed to characterize heterogeneity of intratumoral 18F-FDG uptake 9. By quantifying the distribution of gray-level coarseness and regularity within a tumor PET image, textural features could aid in the quantification of intratumoral 18F-FDG distribution. Although textural features are not routinely used in clinical analysis of PET images, increasing evidence suggests its complementary role in diagnosis of and treatment prediction for common cancers. Recently, textural features derived from 18F-FDG PET images have been shown to be useful in predicting treatment responses of head and neck cancer 9, non-small-cell lung cancer 10, and esophageal cancer 11. However, the correlation between the textural features of 18F-FDG PET images and other commonly used semiquantification values (SUVmax, etc.) and tumor characteristics (stage, pathological grade, etc.) has not been well defined for ESCC.
The objective of this study was to analyze the relationship between the whole-tumor heterogeneity assessed by three-dimensional PET image textural features (entropy and energy) and the common metabolic parameter SUVmax and explore their correlations with histological grade, tumor location, and the American Joint Committee on Cancer (AJCC) stage.
Materials and methods
This study was approved by the institutional review board at our institution. Informed consent was waived because of the retrospective nature of the study. The review was conducted on patients with pathologically proven ESCC in Shandong Cancer Hospital and Institute from June 2010 to November 2011. All patients underwent the standard preoperative staging procedures, including physical examination, laboratory tests, ultrasound of the neck and abdomen, computed tomography (CT) scan of the chest and upper abdomen, esophagogastroscopy, barium radiograph, and whole-body 18F-FDG PET/CT scan not earlier than 2 weeks before surgery. All patients with diabetes or inflammatory lung disease and those who received preoperative neoadjuvant chemotherapy or chemoradiation were excluded. At our institution, all patients without metastasis to distant organs or definite direct tumor invasion into adjacent organs on imaging are routinely scheduled for radical esophagectomy and lymphadenectomy, which require resection of at least 12 nodes for accurate N staging and better survival 12. Surgical approach was determined by the locations of the proximal pole of the tumor and of the positive lymph nodes as indicated by preoperative imaging examinations. After surgery, patients were assigned a pathological stage according to the seventh edition of the TNM classification of AJCC 13.
Positron emission tomography/computed tomography image acquisition
The 18F-FDG PET/CT scans, as part of the initial staging process, were performed using an integrated PET/CT scanner (Discovery LS; GE Healthcare, Waukesha, Wisconsin, USA) 1–3 days before surgery following a rigid protocol 14. All patients were asked to fast for at least 6 h, and their height, weight, and blood glucose level were measured. After resting for 60 min, 18F-FDG (370 MBq) of radiopharmaceutical purity more than 95% was injected intravenously. PET/CT images were acquired in two-dimensional mode from the head to the proximal thigh for 5 min/field of view, each covering 14.5 cm, at an axial sampling thickness of 4.25 mm/slice, reconstructed on a 128×128 matrix with CT-derived attenuation correction using the ordered-subset expectation maximization algorithm. The attenuation-corrected PET images, CT images, and fused PET/CT images displayed as coronal, sagittal, and transaxial slices were viewed on a Xeleris workstation (GE Healthcare). For common quantitative analysis of 18F-FDG uptake, SUVmax was assessed. The uptake of each lesion was determined from the slice with the maximal SUV.
Positron emission tomography image texture extraction
The PET images were transferred to the research treatment planning system Computational Environment for Radiotherapy Research (CERR) where the intensity values were converted into SUV using the Digital Imaging and Communications in Medicine (DICOM) protocol. Our previous study demonstrated that the tumor length seen on an 18F-FDG PET image with a cutoff value of 2.5 was the closest to the gross tumor length 15. On the basis of this result, the regions equal to or greater than SUV 2.5 were selected to automatically delineate the region of interest (ROI) (Fig. 1a and b). Two clinical oncologists with the help of a specialist radiologist adjusted the regions of interest manually by visually inspecting the primary tumor borders to avoid overlapping on adjacent 18F-FDG-avid structures or lesions. All parameters were subsequently extracted from this delineated volume. All image processing procedures such as ROI segmentation, denoising, and extraction of textural features were performed using a code developed and implemented in-house at MATLAB (Mathworks Inc., Natick, Massachusetts, USA).
The gray-level co-occurrence matrix (GLCM) was used as a statistical method to examine texture that considers the spatial relationship of pixels. Creating a GLCM involved three-dimensional gray-level information and then extracting statistical measures, that is textural features, from this matrix. From each primary tumor we obtained one GLCM. The element of GLCM [P(i, j)] contains the number of incidences having intensity values i and j appearing in two voxels separated by distance (d) in direction (a). In our implementation, d was set to a single voxel size, and a was selected to cover the 13 connected neighborhoods in three-dimensional space (Fig. 1c). Two textural features, entropy and energy, defined by Haralick et al. 16 were computed as below, wherein entropy reflects irregularity and energy reflects uniform distribution of gray levels:
Statistical analysis was performed using SPSS for windows version 19.0 (IBM, Armonk, New York, USA). Pearson’s correlation was used to assess the relationship between the textural features (entropy and energy) of PET images and SUVmax. The correlations between heterogeneity parameters SUVmax and TNM classification, histologic grade, tumor location, and AJCC stage were analyzed using the nonparametric Spearman correlation. For the most significant predictor of AJCC stage, specificity and sensitivity [including 95% confidence intervals (CIs)] for the diagnostic threshold were derived using receiver-operating characteristic (ROC) curves measuring associated areas under the ROC curves. All comparisons were two-sided with a P value less than 0.05 used to indicate statistical significance.
The cohort of this study consisted of 40 consecutive patients who met the criteria for inclusion. Patient characteristics have been summarized in Table 1. Nine patients (six with upper thoracic esophageal cancer and three with middle thoracic esophageal cancer) underwent transthoracic esophagectomy (involving laparotomy, right thoracotomy, and cervical anastomosis) with lymph node dissection in three yields (thoracic, abdominal, and cervical). Twenty-one patients (16 with middle thoracic esophageal cancer and five with lower thoracic esophageal cancer) underwent left thoracotomy and cervical anastomosis with lymph node dissection in two yields (thoracic and abdominal). Ten patients with lower thoracic esophageal cancer underwent extended left thoracophrenotomy through the sixth intercostal space with lymph node dissection in two yields (thoracic and abdominal). On average, 15.8 lymph nodes were dissected in each patient.
Correlation between texture parameter and maximum standardized uptake value
The mean values of SUVmax, entropy, and energy of the primary tumor were 15.3±5.1 (range: 3.2–26.9), 4.611±0.420 (range: 3.378–5.002), and 0.020±0.021 (range: 0.006–0.078), respectively. Linear regression showed a positive correlation between SUVmax and entropy (r=0.582, P<0.001) and a negative correlation between SUVmax and energy (r=−0.409, P=0.009) (Fig. 2).
The correlation between texture parameter and tumor stage
SUVmax, entropy, and energy for each T stage or N stage are listed in Tables 2 and 3. Using Spearman’s correlation analysis, positive correlations were found between SUVmax, entropy, and T stage (r s=0.390, P=0.013; r s=0.693, P<0.001); energy was negatively related to T stage (r s=−0.469, P=0.002)). There were also significant correlations between N stage and SUVmax (r s=0.326, P=0.04), entropy (r s=0.501, P=0.001), and energy (r s=−0.413, P=0.008). In the relatively large T3 subgroup, there were nine patients with lymph metastasis and 10 in stage N0; their entropy was 4.726±0.231 and 4.267±0.486 and energy was 0.011±0.004 and 0.018±0.018, respectively. Analysis of variance demonstrated significant difference between the two categories (entropy: F=7.501, P=0.021; energy: F=5.106, P=0.036). AJCC stage also correlated significantly with entropy (r s=0.634, P<0.001), energy (r s=−0.432, P=0.005), and SUVmax (r s=0.358, P=0.023) (Fig. 3). No significant association was found when all the analyses were repeated for histological grade and tumor location.
Prediction for advanced stage using texture parameter
The ability of entropy to predict advanced-stage esophageal cancer was demonstrated by the ROC curve. Figure 4 shows the ROC curve for the textural feature. An entropy above 4.699 predicted tumors above stage IIb with an area under the ROC curve of 0.789 (P<0.001), sensitivity of 77.8% (95% CI: 52.4–93.6), and specificity of 72.7% (95% CI: 49.8–89.3).
Intratumoral heterogeneity is a well-recognized feature of malignant tumors and is associated with many tumor phenotypes such as cellular morphology, gene expression, metabolism, motility, angiogenesis, and proliferative, immunogenic, and metastatic potential 17. At present, there is increasing interest in using textural features derived from 18F-FDG PET images to quantify the heterogeneity of tumor tracer uptake and predict treatment outcome noninvasively 18. In the present study, we found significant correlations between the new tumor uptake heterogeneity parameter and the commonly used simplistic SUV parameter and tumor stage.
Several analytical approaches to imaging have been used to quantify tumor 18F-FDG uptake heterogeneity, such as textural features 9–11, an elliptic solid mathematical model with homogeneous density 19,20, and cumulative SUV volume histograms 8. Among these methods, textural features were used relatively widely. Although they are not routinely used in clinics, studies have demonstrated that textural features (energy, entropy, contrast, local homogeneity, etc.) derived from baseline 18F-FDG PET images can predict response to treatment in head and neck cancer, non-small-cell lung cancer, and esophageal cancer 9–11. In our study, two well-defined textural features, energy and entropy, were extracted from 18F-FDG PET images. Further studies are warranted to determine the best ways to define the 18F-FDG uptake heterogeneity and explore the correlations among these parameters.
The diagnosis of cancer and determination of its prognosis using 18F-FDG PET images are normally carried out by visual analysis and by using a semiquantified index such as SUVmax or other image-derived parameters such as functional tumor length, tumor volume, and total lesion glycolysis 21. In this study, Pearson’s analysis showed that tumors with higher SUVmax tended to be more heterogenous on 18F-FDG uptake. Therefore, the hypothesis can be made that tumor heterogeneity assessed by PET image textural analysis may have the potential to provide prognostic information analogous to that provided by common PET parameters. In our results, correlations were found between SUVmax and AJCC stage, similar to other similar studies 22. However, there are also studies that have found no significant positive correlation between SUVmax and AJCC stages 23. SUVmax is a measurement of a single pixel with the highest radiotracer concentration within the ROI, which may not reflect the heterogenous nature of the tumor. Therefore, we need to explore more informative and discriminative 18F-FDG uptake parameters.
Precise pretherapeutic staging provides valuable prognostic information and aids in selecting appropriate local (surgical resection and/or radiation therapy) and systemic (chemotherapy) therapeutic measures. 18F-FDG PET/CT plays an important role in clinical staging of esophageal cancer, especially in the detection of metastatic disease and mediastinal lymph node metastases 1. However, it has limited utility in the T staging of esophageal cancers. Our study found significant correlations between textural features and T stage and N stage. Patients with lymph metastasis at T3 stage show high entropy and low energy, which may reflect the tumor biology, such as aggressiveness. In future large sample studies, we will try to explore further the relationship between tumor heterogeneity and status of lymph node metastasis. We also found that only textural features are significant predictors of advanced tumor stage according to the ROC curve. This finding suggests that textural features have the potential to become novel methods for predicting the extent of local tumor invasion, status of lymph node metastasis, and tumor stage.
Advanced-stage cancer of the esophagus, defined as stages above IIb, includes tumors that invade regional lymph nodes (N1–3) or local structures (T4 disease). Tumors within this category may be more aggressive. A recent study indicated that the 5-year survival rate of patients with stage I esophageal cancer was ∼60–80%. However, among patients with advanced-stage esophageal carcinoma, this rate dropped to less than 25% 24. The textural parameter demonstrated by the ROC curve may detect advanced carcinoma with higher efficiency if applied early in diagnosis. Hence, patients with heterogenous 18F-FDG uptake status may benefit from more aggressive treatment.
Textural features can also be obtained from two-dimensional CT or MRI images. CT textural features have the potential to provide additional morphological information relating to tumor heterogeneity. Ganeshan et al. 25 found that entropy of CT images correlated with SUVmax and was significantly greater in esophageal cancers of clinical stage III or IV. They also found that tumor heterogeneity assessed by CT textural features was a significant independent predictor of NSCLC survival, whereas SUV was not significantly associated with survival 26. MRI textural parameters were used to define the heterogeneity in vascular features 27. However, anatomical images seen on CT and MRI are not perceived to portray functional and molecular details of solid tumors 28.
The present results are based on a small cohort of patients and are retrospective, which could be a limitation of our study. A prospective study on a large cohort focusing on therapeutic response or prognosis of patients is currently underway.
This study shows that heterogeneity of tumor 18F-FDG uptake quantified by textural features (energy and entropy) is well correlated with SUVmax and tumor stage in ESCC. Heterogeneity of high 18F-FDG uptake has the potential to detect advanced-stage tumors. A prospective study is currently under way to determine whether these heterogeneity parameters can predict treatment response and survival.
The authors thank Dr Juan Yang from the College of Physics and Electronic Science, Shandong Normal University, for the development and implementation of the image analysis code.
This work was supported in part by the Research Fund of Project 81101700 supported by the National Natural Science Foundation of China (NSFC).
Conflicts of interest
There are no conflicts of interest.
1. Jemal A, Bray F, Center MM, Ferlay J, Ward E, Forman D. Global cancer statistics. CA Cancer J Clin. 2011;61:69–90
2. Enzinger PC, Mayer RJ. Esophageal cancer. N Engl J Med. 2003;349:2241–2252
3. Karaosmanoglu AD, Blake MA. Applications of PET-CT in patients with esophageal cancer. Diagn Interv Radiol. 2012;18:171–182
4. Chua S, Dickson J, Groves AM. PET imaging for prediction of response to therapy and outcome in oesophageal carcinoma. Eur J Nucl Med Mol Imaging. 2011;38:1591–1594
5. Pan L, Gu P, Huang G, Xue H, Wu S. Prognostic significance of SUV on PET/CT in patients with esophageal cancer: a systematic review and meta-analysis. Eur J Gastroenterol Hepatol. 2009;21:1008–1015
6. Pugachev A, Ruan S, Carlin S, Larson SM, Campa J, Ling CC, et al. Dependence of FDG uptake on tumor microenvironment. Int J Radiat Oncol Biol Phys. 2005;62:545–553
7. Van Baardwijk A, Bosmans G, Van Suylen RJ, Van Kroonenburgh M, Hochstenbag M, Geskes G, et al. Correlation of intra-tumour heterogeneity on 18
F-FDG PET with pathologic features in nonsmall cell lung cancer: a feasibility study. Radiother Oncol. 2008;87:55–58
8. Van VFH, Cheebsumon P, Yaqub M, Yaqub M, Egbert F, Otto S, et al. Evaluation of a cumulative SUV-volume histogram method for parameterizing heterogeneous intratumoural FDG uptake in non-small cell lung cancer PET studies. Eur J Nucl Med Mol Imaging. 2011;38:1636–1647
9. El NI, Grigsby P, Apte A, Apte A, Kidd E, Donnelly E, et al. Exploring feature-based approaches in PET images for predicting cancer treatment outcomes. Pattern Recognit. 2009;42:1162–1171
10. Vaidya M, Creach KM, Frye J, Dehdashti F, Bradley JD, El NI. Combined PET/CT image characteristics for radiotherapy tumor response in lung cancer. Radiother Oncol. 2012;102:239–245
11. Tixier F, Le RCC, Hatt M, Albarghach N, Pradier O, Metges JP, et al. Intratumor heterogeneity characterized by textural features on baseline 18
F-FDG PET images predicts response to concomitant radiochemotherapy in esophageal cancer. J Nucl Med. 2011;52:369–378
12. Chen YJ, Schultheiss TE, Wong JY, Kernstine KH. Impact of the number of resected and involved lymph nodes on esophageal cancer survival. J Surg Oncol. 2009;100:127–132
13. Rice TW, Blackstone EH, Rusch VW. 7th edition of the AJCC Cancer Staging Manual: esophagus and esophagogastric junction. Ann Surg Oncol. 2010;17:1721–1724
14. Yu J, Li X, Xing L, Mu D, Fu Z, Sun X, et al. Comparison of tumor volumes as determined by pathologic examination and FDG-PET/CT images of non-small-cell lung cancer: a pilot study. Int J Radiat Oncol Biol Phys. 2009;75:1468–1474
15. Zhong X, Yu J, Zhang B, Mu D, Zhang W, Li D, et al. Using 18
F-fluorodeoxyglucose positron emission tomography to estimate the length of gross tumor in patients with squamous cell carcinoma of the esophagus. Int J Radiat Oncol Biol Phys. 2009;73:136–141
16. Haralick RM, Shanmugam K, Dinstein I. Textural features for image classification. IEEE Trans Syst Man Cybern. 1973;3:610–621
17. Marusyk A, Polyak K. Tumor heterogeneity: causes and consequences. Biochim Biophys Acta. 2010;1805:105–117
18. Tomasi G, Turkheimer F, Aboagye E. Importance of quantification for the analysis of PET data in oncology: review of current methods and trends for the future. Mol Imaging Biol. 2012;14:131–146
19. O’Sullivan F, Roy S, Eary J. A statistical measure of tissue heterogeneity with application to 3D PET sarcoma data. Biostatistics. 2003;4:433–448
20. O’Sullivan F, Roy S, O’Sullivan J, Vernon C, Eary J. Incorporation of tumor shape into an assessment of spatial heterogeneity for human sarcomas imaged with FDG-PET. Biostatistics. 2005;6:293–301
21. Hatt M, Visvikis D, Albarghach NM, Tixier F, Pradier O, Cheze-le RC. Prognostic value of 18
F-FDG PET image-based parameters in oesophageal cancer and impact of tumour delineation methodology. Eur J Nucl Med Mol Imaging. 2011;38:1191–1202
22. Ganeshan B, Skogen K, Pressney I, Coutroubis D, Miles K. Tumour heterogeneity in oesophageal cancer assessed by CT texture analysis: preliminary evidence of an association with tumour metabolism, stage, and survival. Clin Radiol. 2012;67:157–164
23. Chan WK, Mak HK, Huang B, Yeung DW, Kwong DL, Khong PL, et al. Nasopharyngeal carcinoma: relationship between 18
F-FDG PET-CT maximum standardized uptake value, metabolic tumour volume and total lesion glycolysis and TNM classification. Nucl Med Commun. 2010;31:206–210
24. Refaely Y, Krasna MJ. Multimodality therapy for esophageal cancer. Surg Clin North Am. 2002;82:729–746
25. Ganeshan B, Abaleke S, Young RC, Chatwin CR, Miles KA. Texture analysis of non-small cell lung cancer on unenhanced computed tomography: initial evidence for a relationship with tumour glucose metabolism and stage. Cancer Imaging. 2010;10:137–143
26. Ganeshan B, Panayiotou E, Burnand K, Dizdarevic S, Miles K. Tumour heterogeneity in non-small cell lung carcinoma assessed by CT texture analysis: a potential marker of survival. Eur Radiol. 2012;22:796–802
27. Jackson A, O’Connor JP, Parker GJ. Imaging tumor vascular heterogeneity and angiogenesis using dynamic contrast-enhanced magnetic resonance imaging. Clin Cancer Res. 2007;13:3449–3459
28. Basu S, Kwee TC, Gatenby R, Saboury B, Torigian DA, Alavi A. Evolving role of molecular imaging with PET in detecting and characterizing heterogeneity of cancer tissue at the primary and metastatic sites, a plausible explanation for failed attempts to cure malignant disorders. Eur J Nucl Med Mol Imaging. 2011;38:987–991