Ann Nucl Med DOI 10.1007/s12149-017-1208-x
ORIGINAL ARTICLE
Assessment of intratumor heterogeneity in mesenchymal uterine tumor by an 18F-FDG PET/CT texture analysis Tetsuya Tsujikawa1 · Makoto Yamamoto2 · Kunihiro Shono1 · Shizuka Yamada2 · Hideaki Tsuyoshi2 · Yasushi Kiyono1 · Hirohiko Kimura3 · Hidehiko Okazawa1 · Yoshio Yoshida2
Received: 21 July 2017 / Accepted: 8 September 2017 © The Japanese Society of Nuclear Medicine 2017
Abstract Objective The aim of this study was to retrospectively evaluate the clinical significance of 18F-FDG PET/CT textural features for discriminating uterine sarcoma from leiomyoma. Methods Fifty-five patients with suspected uterine sarcoma based on ultrasound and MRI findings who underwent pretreatment 18F-FDG PET/CT were included. Fifteen patients were histopathologically proven to have uterine sarcoma, 14 patients by surgical operation and one by biopsy, and 40 patients were diagnosed with leiomyoma by surgical operation or in a follow-up for at least 2 years. A texture analysis was performed on PET/CT images from which second- and higher order textural features were extracted in addition to standardized uptake values (SUVs) and other first-order features. The accuracy of PET features for differentiating between uterine sarcoma and leiomyoma was evaluated using a receiver-operating-characteristic (ROC) analysis. Results The intratumor distribution of 18F-FDG was more heterogeneous in uterine sarcoma than in leiomyoma. Entropy, correlation, and uniformity calculated from normalized gray-level co-occurrence matrices and SUV
standard deviation derived from histogram statistics showed greater area under the ROC curves (AUCs) than did maximum SUV for differentiating between sarcoma and leiomyoma. Entropy, as a single feature, yielded the greatest AUC of 0.974 and the optimal cut-off value of 2.85 for entropy provided 93% sensitivity, 90% specificity, and 92% accuracy. When combining conventional features with textural ones, maximum SUV (cutoff: 6.0) combined with entropy (2.85) and correlation (0.73) provided the best diagnostic performance (100% sensitivity, 94% specificity, and 95% accuracy). Conclusions In combination with the conventional histogram statistics and/or volumetric parameters, 18F-FDG PET/CT textural features reflecting intratumor metabolic heterogeneity are useful for differentiating between uterine sarcoma and leiomyoma.
Electronic supplementary material The online version of this article (doi:10.1007/s12149-017-1208-x) contains supplementary material, which is available to authorized users.
Uterine leiomyoma is a common benign uterine tumor of mesenchymal origin that occurs in 25–30% of women older than 30 years of age [1]. In contrast, mesenchymal tumors other than uterine leiomyomas, such as uterine sarcomas, are uncommon [2]. Uterine sarcomas constitute only 3–7% of uterine malignancies, including carcinosarcoma (CS), leiomyosarcoma (LMS), low- or high-grade endometrial stromal sarcoma (L-, H-ESS), and adenosarcoma. Due to the possibility of the conservative treatment of uterine leiomyoma as well as the poor prognosis of patients with uterine sarcoma, with an overall 5-year survival rate of 8–12% for advanced stages [3], the differential diagnosis of uterine sarcoma from
* Tetsuya Tsujikawa awaji@u‑fukui.ac.jp 1
Biomedical Imaging Research Center, University of Fukui, 23‑3 Matsuoka‑Shimoaizuki, Eiheiji‑cho, Fukui 910‑1193, Japan
2
Department of Obstetrics and Gynecology, Faculty of Medical Sciences, University of Fukui, Fukui, Japan
3
Department of Radiology, Faculty of Medical Sciences, University of Fukui, Fukui, Japan
Keywords Mesenchymal uterine tumor · PET · Texture analysis · Differentiation
Introduction
13
Vol.:(0123456789)
leiomyoma is important for effective treatment, particularly in the early phase of the disease. Since a preoperative histological examination of uterine myometrial lesions may present practical difficulties and be associated with significant sampling errors, imaging plays an important role in the evaluation of myometrial lesions. Although magnetic resonance imaging (MRI) is widely accepted as the standard for a differential diagnosis of mesenchymal uterine tumors, no adequate evidence has been provided by large scale or prospective studies and this is partly due to the rareness of uterine sarcoma [4–6]. Intratumor heterogeneity in phenotypic and genetic properties is a widespread phenomenon of critical importance for tumor progression and responses to therapeutic interventions [7]. Positron-emission tomography with 2-[18F]-fluoro-2-deoxy-D-glucose (18F-FDG PET) represents glycolytic activity throughout a tumor and also reflects metabolic heterogeneity associated with cellular and molecular characteristics such as cell proliferation, necrosis, perfusion, and the histological architecture [8, 9]. Interest is increasing in assessing the global and local–regional heterogeneities of the distribution of 18F-FDG with textural feature measurements using a number of mathematical methods that describe the relationships between the gray-level intensity of pixels or voxels and their position within an image [10, 11]. An 18F-FDG PET texture analysis has been used to differentiate malignant and benign bone/soft-tissue lesions [12] and predict treatment responses and patient prognoses in cervix, lung, and esophageal cancers [13–15]. We hypothesized that uterine sarcoma is metabolically more heterogeneous than leiomyoma. The aim of our study was to investigate whether first-order and high-order textural features on 18F-FDG PET images are useful for differentiating between uterine sarcoma and leiomyoma.
Materials and methods Patients We retrospectively reviewed the medical records of 65 patients with suspected uterine sarcoma based on ultrasound and MRI findings who underwent 18F-FDG PET/CT between April 2008 and August 2013 in our hospital. The following imaging findings were obtained: (1) ultrasound exhibiting enlarged tumors and the characteristic ‘mosaic pattern’ on ultrasonic power Doppler images and/or (2) MRI exhibiting heterogeneous intensities with high signals on MR T1- and T2-weighted images, suggesting uterine sarcoma. Patients were excluded from the analysis if they (1) were previously diagnosed with another malignant disease and (2) had a follow-up duration of less than 2 years. Fifty-five patients were enrolled in this study. Fifteen patients were histopathologically proven to have
13
Ann Nucl Med
uterine sarcoma, 14 patients by surgical operation, and one by biopsy, consisting of 6 CS, 6 LMS, 2 H-ESS, and 1 L-ESS. Forty patients were diagnosed with leiomyoma, 29 patients by surgical operation, and 11 in a follow-up for at least 2 years. PET/CT imaging Whole-body PET scans with 18F-FDG were performed with a combined PET/CT scanner (Discovery LS; GE Medical Systems, Milwaukee, WI, USA), which permitted the simultaneous acquisition of 35 image slices in a 2-dimensional acquisition mode with inter-slice spacing of 4.25 mm. The PET/CT scanner incorporated an integrated four-slice multidetector CT scanner, which was used for attenuation correction. CT scanning parameters were as follows: Auto mA (upper limit, 40 mA; noise index, 20), 140 kV, 5-mm section thickness, 15-mm table feed, and a pitch of 4. After at least 4 h of fasting, patients received an intravenous injection of 185 MBq 18 F-FDG and image acquisition began 50 min after the injection. A whole-body emission scan was performed from the head to the inguinal region with 2 min per bed position (seven to eight bed positions). PET data were reconstructed by the iterative reconstruction method selecting 14 subsets and 2 iterations, a 128 × 128 matrix, voxel size (width, length and height) = 4 × 4 × 4.25 mm, and post-smoothing with an 8-mm Gaussian filter. Reconstructed images were then converted to a semi-quantitative image corrected by the injection dose and subject’s body weight (=standardized uptake value: SUV). The median duration between PET/CT and MRI acquisition was 14 days (range 0–51 days). Image analysis 18
F-FDG PET/CT images were retrospectively interpreted by the consensus of a gynecologic oncologist (MY, with 10 years of experience in gynecologic oncology) and an experienced radiologist (TT, with 12 years of experience in oncologic PET) who had no knowledge of the clinical data. PET images were co-registered to individual MR images. The volume of interest (VOI) was manually placed on the primary tumor by integrating 2D segmentations of each slice to cover the whole tumor using the PMOD 3.6 software package (PMOD Technologies Ltd, Zurich, Switzerland) (Fig. 1). VOIs were recorded and used for a subsequent texture analysis. Prior to textural feature computation, we resampled the VOI voxel intensities using absolute resampling method with fixed bounds and 64 discrete values [16, 17]: [ ] I(x) − lower bound R(x) = round 64 × (1) upper bound − lower bound where R(x) is the voxel intensity after discretization and I(x) is the voxel intensity before discretization. The lower bound
Ann Nucl Med Fig. 1 Sagittal T2-weighted (a), axial T2-weighted MR (b), 18F-FDG PET (c), and co-registered (d) images of a 50-year-old woman with an enlarging uterine leiomyoma. The VOIs were manually placed on the co-registered images according to MR contour to cover the whole tumor
was set to 0 and the upper bound was set to 25 SUV units which corresponded to the maximum intensity over all of the primary tumors included in this study. As a result, the sampling bin width of 0.4 SUV units was used in this study. SUV histogram, normalized gray-level co-occurrence matrix (NGLCM), and neighborhood gray-tone difference matrix (NGTDM) were used in assessments of first-, second-, and higher order textural features, respectively (Table 2). Eighteen different global and textural features were examined. The formulae and brief explanations of second- and higher order textural features measured in this study were shown in a supplemental document. The computation for textural features was performed using the Chang–Gung Image Texture Analysis toolbox implemented under MATLAB 2014b (Mathworks INC., Natick, MA, USA) [18]. Statistical analysis The Mann–Whitney U test was used to assess differences in PET features between uterine sarcoma and leiomyoma. A receiver-operating-characteristic (ROC) analysis was performed to determine optimal cut-off values for discrimination with high accuracy based on the area under the curves
(AUCs). The significance of differences between the AUCs was tested using the pairwise comparison of DeLong et al. [19]. The Mann–Whitney U test, ROC analysis, and comparison of AUCs were performed using MedCalc® (version 17.1; MedCalc Software bvba). To adjust for multiple comparison problems that resulted from an increase in the type-1 error when tested repeatedly, the Bonferroni correction method was used with an adjusted P value threshold of 0.003 (0.05/18).
Results The characteristics of all 55 patients are summarized in Table 1. Since five patients had two leiomyomas suspected of being uterine sarcomas, a total of 45 leiomyomas and 15 sarcomas were analyzed. Among the 18 PET features examined, SUV skewness, tumor volume, total lesion glycolysis (TLG), coarseness, and contrast from NGTDM were not significantly different between uterine sarcoma and leiomyoma, whereas the other 13 features were significantly different between them after the Bonferroni adjustment (Table 2). The intratumor distribution of 18F-FDG was more heterogeneous
13
Ann Nucl Med
Table 1 Patient characteristics Characteristic
Leiomyoma
Sarcoma
Number (n) of patients Age (years) Mean ± SD Range Menopausal status (n) Premenopausal Postmenopausal Types of lesions on histopathological findings (n)a
40
15
48.8 ± 11.0 26–82
59.0 ± 16.0 33–81
28 12 Typical, 11 Degenerated, 22 Follow-up, 12
5 10 Carcinosarcoma, 6 Leiomyosarcoma, 6 High-grade ESS, 2 Low-grade ESS, 1
leiomyoma was entropy. The AUC was 0.974 for entropy, which was better than the AUC of 0.951 for maximum SUV. The optimal cut-off value of 2.85 for entropy provided 93% sensitivity, 90% specificity, and 92% accuracy. The second most accurate single features were correlation and SUV SD, both of which yielded an AUC of 0.971. The third one was uniformity, which yielded an AUC of 0.965. Pairwise comparisons of AUCs did not show any significant differences between entropy, correlation, SUV SD, uniformity, and maximum SUV. When combining conventional features with textural ones in this study, maximum SUV (cutoff: 6.0) combined with entropy (2.85) and correlation (0.73) provided the best diagnostic performance (100% sensitivity, 94% specificity, and 95% accuracy).
a
Since five patients had two leiomyomas suspected of being uterine sarcomas, a total of 45 leiomyomas and 15 sarcomas were analyzed
in uterine sarcoma than in leiomyoma. The results of the ROC analysis (Table 2; Fig. 2) revealed that entropy, correlation, and uniformity calculated from NGLCM and SUV standard deviation (SUV SD) derived from histogram statistics showed greater AUCs than did maximum SUV for differentiating between sarcoma and leiomyoma. The most accurate single feature for discriminating sarcoma from
Discussion To the best of our knowledge, this is the first study to demonstrate the usefulness of quantitative 18F-FDG PET/CT textural features reflecting intratumor metabolic heterogeneity for the differential diagnosis of mesenchymal uterine tumors. Uterine sarcomas showed a significantly more heterogeneous 18 F-FDG distribution than leiomyomas. One of the secondorder features, entropy derived from NGLCM, had the best
Table 2 PET feature comparison and ROC analysis for the differentiation between leiomyoma and sarcoma
First order (histogram, global) Maximum SUV Mean SUV SUV SD SUV skewness SUV Kurtosis Total lesion glycolysis Tumor volume Second order (NGLCM, local) Uniformity Contrast Entropy Homogeneity Dissimilarity Correlation Higher order (NGTDM, Local) Coarseness Contrast Busyness Complexity Strength *Mann–Whitney U test
13
Leiomyoma Median (25% tile, 75% tile)
Sarcoma Median (25% tile, 75% tile)
P value*
AUC
Cutoff
3.07 (2.64, 4.07) 1.66 (1.47, 2.02) 0.32 (0.25, 0.42) 0.16 (−0.1, 0.68) 1.49 (1.33, 1.84) 491.1 (135.2, 1240.6) 280.4 (96.9, 659.6)
9.85 (8.94, 14.5) 3.39 (2.75, 4.66) 1.74 (1.35, 2.15) 0.53 (0.29, 1.21) 2.63 (2.14, 3.64) 1110.4 (206.3, 2086.5) 343.5 (34.3, 660.7)
0.001 0.001 0.001 0.044 0.001 0.272 0.519
0.95 0.917 0.971 0.674 0.922 0.594 0.556
6.0 2.2 0.60 0.25 2.1 930 130
0.19 (0.12, 0.25) 0.5 (0.44, 0.66) 1.93 (1.67, 2.34) 0.76 (0.71, 0.79) 0.49 (0.43, 0.6) 0.49 (0.37, 0.61)
0.02 (0.01, 0.05) 2.51 (1.95, 7.06) 4.13 (3.67, 4.64) 0.56 (0.44, 0.61) 1.25 (1.02, 1.85) 0.82 (0.79, 0.88)
0.001 0.001 0.001 0.001 0.001 0.001
0.965 0.947 0.974 0.936 0.946 0.971
0.085 1.45 2.85 0.62 0.95 0.73
0.002 (0.001, 0.007) 3E−06 (8E−07, 1E−05) 412.5 (146.3, 1326.1) 0.0017 (0.0003, 0.0046) 0.051 (0.012, 0.15)
0.003 (0.002, 0.021) 6E−06 (3E−06, 0.0002) 14.6 (1.5, 29.4) 0.047 (0.0073, 0.33) 0.92 (0.32, 3.19)
0.11 0.014 0.001 0.001 0.001
0.638 0.713 0.893 0.893 0.908
0.0024 4E−06 45.0 0.006 0.28
Ann Nucl Med
Fig. 2 ROC curves and AUCs in parentheses demonstrating the ability of maximum SUV, SUV SD, uniformity, entropy, and correlation in distinguishing uterine sarcoma from leiomyoma. The ROC curve and AUC for TLG were also shown as a reference
diagnostic performance for differentiating between uterine sarcoma and leiomyoma. Entropy may be utilized to measure the uncertainty or randomness of intensity distribution, with more uncertainty or randomness implying higher entropy [10, 12]. Correlation calculated from NGLCM was the second most accurate single feature for differentiating between uterine sarcoma and leiomyoma. Correlation quantifies how a voxel correlates with a consecutive voxel according to a spatial relationship [12]. Uniformity was the third most accurate single feature for the discrimination. Since uniformity indicates the degree of the local homogeneous nature of tumor 18F-FDG uptake, sarcoma showed ‘lower’ uniformity than did leiomyoma. It is difficult to differentiate uterine sarcoma from leiomyoma when leiomyoma shows high 18F-FDG uptake and degenerative changes on MR images [20]. Although MRI including diffusion-weighted imaging (DWI) and contrastenhanced (CE) imaging has been used as a standard for the differential diagnosis of uterine sarcoma and leiomyoma, no adequate evidence has been provided by large-scale or prospective studies and this is partly due to the rareness of uterine sarcoma. Overlap was noted in the distribution of the apparent diffusion coefficient (ADC) values calculated from DWI between uterine sarcoma and leiomyoma [4, 5]. The most recent study by Lin et al. showed that CE-MRI provides accurate information on a differential diagnosis and is preferable to DWI [6]. In their consecutive cohort of 8 sarcomas and 25 leiomyomas, CE-MRI yielded a significantly
higher diagnostic accuracy (94%) than DWI (52%). Our study consisted of a larger patient population (15 sarcomas and 45 leiomyomas) and showed that an 18F-FDG PET/CT texture analysis (local entropy) yielded a similar diagnostic accuracy (92%) to CE-MRI. This result was consistent with the previous study by Xu et al. in which local entropy on PET images provided the highest accuracy (AUC = 0.841) for differentiating between malignant and benign bone/ soft-tissue lesions [12]. Their study also showed that the correlation on ‘CT images’ may be an indicator of malignancy; the correlation in malignant tumors was higher than that in benign lesions. The reason for this was considered to be as follows: since perfusion or fibrosis, cellular alignment or other shaped opacities, and structural damage often occur simultaneously along a certain direction in malignant tumors on CT images, there are more similar or correlated Hounsfield Unit values along a certain direction in malignant tumors than in benign lesions [12]. This may be the case in our study regarding metabolic textural features on ‘PET images’, in which uterine sarcoma showed a significantly higher metabolic correlation on PET images than leiomyoma. In our study population, the diagnostic accuracy of firstorder global features such as SUV SD and maximum SUV was similar to that of local entropy, correlation, and uniformity. Previous studies regarding differentiation of uterine tumors showed the usefulness of the conventional 18F-FDG PET parameters combined with the assessments of MR imaging and serum lactate dehydrogenase (LDH) levels [21, 22]. In this study, when combining the conventional features with textural ones, maximum SUV (cutoff: 6.0) combined with entropy (2.85) and correlation (0.73) exceeded the diagnostic performance of entropy only (100% sensitivity, 94% specificity, and 95% accuracy). Although the conventional first-order features are calculated from a distribution histogram and do not have positional information, they are easily available in clinical settings. The combination of textural features and histogram statistics may allow for a more accurate differentiation between uterine sarcoma and leiomyoma. Since a single feature cannot be directly linked to a specific biological process, a combination of textural and conventional features including volumetric parameters may be closely related to underlying physiological processes and may provide more accurate differentiation than each single feature. In the near future, artificial intelligence systems will contribute to automatically contouring tumors, extracting numerous textural features, correlating features and gene expression, and finding potential prognostic models and promising effective therapeutic strategies. A radiogenomics (radiomics) approach will be enhanced with the advancement of artificial intelligence [23]. There is a limitation to the present study. Since textural features are dependent on various factors such as image
13
acquisition, reconstruction, preprocessing, segmentation, and mathematical methods [24], the standardization of a PET texture analysis will be necessary for inter-institutional evaluations in the future.
Conclusion In combination with the conventional histogram statistics and/or volumetric parameters, 18F-FDG PET/CT textural features reflecting intratumor metabolic heterogeneity are useful for differentiating between uterine sarcoma and leiomyoma. Acknowledgements The authors thank Dr. Yu-Hua Dean Fang, the original developer of the Chang-Gung Image Texture Analysis toolbox, and the staff of the Department of Radiology and Biological Imaging Research Center, University of Fukui, for their clinical and technical support. Compliance with ethical standards Consent to participate Formal consent was not required for this type of retrospective study. Conflict of interest The authors have no conflicts of interest to declare. Funding This study was partly funded by Grants-in-Aid for scientific research from the Japan Society for the Promotion of Science (15H04981, 16K10345, and 16K20181) and Takeda Science Foundation.
References 1. Adams Hillard PJ. Benign disease of the female reproductive tract. In: Berek JS, editor. Berek and Novak’s gynecology. 14th edn. Philadelphia, PA: Lippincott Williams and Wilkins; 2007. pp. 431–504. 2. D’Angelo E, Prat J. Uterine sarcomas: a review. Gynecol Oncol. 2010;116:131–9. 3. Gadducci A. Prognostic factors in uterine sarcoma. Best Pract Res Clin Obstet Gynaecol. 2011;25:783–95. 4. Namimoto T, Yamashita Y, Awai K, Nakaura T, Yanaga Y, Hirai T, et al. Combined use of T2-weighted and diffusion-weighted 3-T MR imaging for differentiating uterine sarcomas from benign leiomyomas. Eur Radiol. 2009;19:2756–64. 5. Sato K, Yuasa N, Fujita M, Fukushima Y. Clinical application of diffusion-weighted imaging for preoperative differentiation between uterine leiomyoma and leiomyosarcoma. Am J Obstet Gynecol. 2014;210:368.e1–8. 6. Lin G, Yang LY, Huang YT, Ng KK, Ng SH, Ueng SH, et al. Comparison of the diagnostic accuracy of contrast-enhanced MRI and diffusion-weighted MRI in the differentiation between uterine leiomyosarcoma/smooth muscle tumor with uncertain malignant potential and benign leiomyoma. J Magn Reson Imaging. 2016;43:333–42. 7. Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med. 2012;366:883–92.
13
Ann Nucl Med 8. 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;38:987–91. 9. 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;62:545–53. 10. El Naqa I, Grigsby P, Apte A, Kidd E, Donnelly E, Khullar D, et al. Exploring feature-based approaches in PET images for predicting cancer treatment outcomes. Pattern Recognit. 2009;42:1162–71. 11. Chicklore S, Goh V, Siddique M, Roy A, Marsden PK, Cook GJ. Quantifying tumour heterogeneity in 18F-FDG PET/CT imaging by texture analysis. Eur J Nucl Med Mol Imaging. 2013:133–40. 12. Xu R, Kido S, Suga K, Hirano Y, Tachibana R, Muramatsu K, et al. Texture analysis on 18F-FDG PET/CT images to differentiate malignant and benign bone and soft-tissue lesions. Ann Nucl Med. 2014;28:926–35. 13. Yang F, Thomas MA, Dehdashti F, Grigsby PW. Temporal analysis of intratumoral metabolic heterogeneity characterized by textural features in cervical cancer. Eur J Nucl Med Mol Imaging. 2013;40:716 – 27. 14. Cook GJ, O’Brien ME, Siddique M, Chicklore S, Loi HY, Sharma B, et al. Non-small cell lung cancer treated with Erlotinib: heterogeneity of 18F-FDG uptake at PET-association with treatment response and prognosis. Radiology. 2015;276:883–93. 15. Tixier F, Le Rest CC, Hatt M, Albarghach N, Pradier O, Metges JP, 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. 16. Orlhac F, Thézé B, Soussan M, Boisgard R, Buvat I. Multiscale texture analysis: From 18F-FDG PET images to histologic images. J Nucl Med. 2016;57:1823–8. 17. Orlhac F, Nioche C, Soussan M, Buvat I. Understanding changes in tumor textural indices in PET: a comparison between visual assessment and index values in simulated and patient data. J Nucl Med. 2017;58:387–92. 18. Cheng NM, Fang YH, Chang JT, Huang CG, Tsan DL, Ng SH, et al. Textural features of pretreatment 18F-FDG PET/CT images: prognostic significance in patients with advanced T-stage oropharyngeal squamous cell carcinoma. J Nucl Med. 2013;54:1703–9. 19. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837–45. 20. Kitajima K, Murakami K, Kaji Y, Sugimura K. Spectrum of FDG PET/CT findings of uterine tumors. AJR Am J Roentgenol. 2010;195:737–43. 21. Yoshida Y, Kurokawa T, Sawamura Y, Shinagawa A, Tsujikawa T, Okazawa H, et al. Comparison of 18F-FDG PET and MRI in assessment of uterine smooth muscle tumors. J Nucl Med. 2008;49:708–12. 22. Nagamatsu A, Umesaki N, Li L, Tanaka T. Use of 18F-fluorodeoxyglucose positron emission tomography for diagnosis of uterine sarcomas. Oncol Rep. 2010;23:1069–76. 23. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006. 24. Hatt M, Tixier F, Pierce L, Kinahan PE, Le Rest CC, Visvikis D. Characterization of PET/CT images using texture analysis: the past, the present… any future? Eur J Nucl Med Mol Imaging. 2017;44:151–65.
ORIGINAL ARTICLE
Assessment of intratumor heterogeneity in mesenchymal uterine tumor by an 18F-FDG PET/CT texture analysis Tetsuya Tsujikawa1 · Makoto Yamamoto2 · Kunihiro Shono1 · Shizuka Yamada2 · Hideaki Tsuyoshi2 · Yasushi Kiyono1 · Hirohiko Kimura3 · Hidehiko Okazawa1 · Yoshio Yoshida2
Received: 21 July 2017 / Accepted: 8 September 2017 © The Japanese Society of Nuclear Medicine 2017
Abstract Objective The aim of this study was to retrospectively evaluate the clinical significance of 18F-FDG PET/CT textural features for discriminating uterine sarcoma from leiomyoma. Methods Fifty-five patients with suspected uterine sarcoma based on ultrasound and MRI findings who underwent pretreatment 18F-FDG PET/CT were included. Fifteen patients were histopathologically proven to have uterine sarcoma, 14 patients by surgical operation and one by biopsy, and 40 patients were diagnosed with leiomyoma by surgical operation or in a follow-up for at least 2 years. A texture analysis was performed on PET/CT images from which second- and higher order textural features were extracted in addition to standardized uptake values (SUVs) and other first-order features. The accuracy of PET features for differentiating between uterine sarcoma and leiomyoma was evaluated using a receiver-operating-characteristic (ROC) analysis. Results The intratumor distribution of 18F-FDG was more heterogeneous in uterine sarcoma than in leiomyoma. Entropy, correlation, and uniformity calculated from normalized gray-level co-occurrence matrices and SUV
standard deviation derived from histogram statistics showed greater area under the ROC curves (AUCs) than did maximum SUV for differentiating between sarcoma and leiomyoma. Entropy, as a single feature, yielded the greatest AUC of 0.974 and the optimal cut-off value of 2.85 for entropy provided 93% sensitivity, 90% specificity, and 92% accuracy. When combining conventional features with textural ones, maximum SUV (cutoff: 6.0) combined with entropy (2.85) and correlation (0.73) provided the best diagnostic performance (100% sensitivity, 94% specificity, and 95% accuracy). Conclusions In combination with the conventional histogram statistics and/or volumetric parameters, 18F-FDG PET/CT textural features reflecting intratumor metabolic heterogeneity are useful for differentiating between uterine sarcoma and leiomyoma.
Electronic supplementary material The online version of this article (doi:10.1007/s12149-017-1208-x) contains supplementary material, which is available to authorized users.
Uterine leiomyoma is a common benign uterine tumor of mesenchymal origin that occurs in 25–30% of women older than 30 years of age [1]. In contrast, mesenchymal tumors other than uterine leiomyomas, such as uterine sarcomas, are uncommon [2]. Uterine sarcomas constitute only 3–7% of uterine malignancies, including carcinosarcoma (CS), leiomyosarcoma (LMS), low- or high-grade endometrial stromal sarcoma (L-, H-ESS), and adenosarcoma. Due to the possibility of the conservative treatment of uterine leiomyoma as well as the poor prognosis of patients with uterine sarcoma, with an overall 5-year survival rate of 8–12% for advanced stages [3], the differential diagnosis of uterine sarcoma from
* Tetsuya Tsujikawa awaji@u‑fukui.ac.jp 1
Biomedical Imaging Research Center, University of Fukui, 23‑3 Matsuoka‑Shimoaizuki, Eiheiji‑cho, Fukui 910‑1193, Japan
2
Department of Obstetrics and Gynecology, Faculty of Medical Sciences, University of Fukui, Fukui, Japan
3
Department of Radiology, Faculty of Medical Sciences, University of Fukui, Fukui, Japan
Keywords Mesenchymal uterine tumor · PET · Texture analysis · Differentiation
Introduction
13
Vol.:(0123456789)
leiomyoma is important for effective treatment, particularly in the early phase of the disease. Since a preoperative histological examination of uterine myometrial lesions may present practical difficulties and be associated with significant sampling errors, imaging plays an important role in the evaluation of myometrial lesions. Although magnetic resonance imaging (MRI) is widely accepted as the standard for a differential diagnosis of mesenchymal uterine tumors, no adequate evidence has been provided by large scale or prospective studies and this is partly due to the rareness of uterine sarcoma [4–6]. Intratumor heterogeneity in phenotypic and genetic properties is a widespread phenomenon of critical importance for tumor progression and responses to therapeutic interventions [7]. Positron-emission tomography with 2-[18F]-fluoro-2-deoxy-D-glucose (18F-FDG PET) represents glycolytic activity throughout a tumor and also reflects metabolic heterogeneity associated with cellular and molecular characteristics such as cell proliferation, necrosis, perfusion, and the histological architecture [8, 9]. Interest is increasing in assessing the global and local–regional heterogeneities of the distribution of 18F-FDG with textural feature measurements using a number of mathematical methods that describe the relationships between the gray-level intensity of pixels or voxels and their position within an image [10, 11]. An 18F-FDG PET texture analysis has been used to differentiate malignant and benign bone/soft-tissue lesions [12] and predict treatment responses and patient prognoses in cervix, lung, and esophageal cancers [13–15]. We hypothesized that uterine sarcoma is metabolically more heterogeneous than leiomyoma. The aim of our study was to investigate whether first-order and high-order textural features on 18F-FDG PET images are useful for differentiating between uterine sarcoma and leiomyoma.
Materials and methods Patients We retrospectively reviewed the medical records of 65 patients with suspected uterine sarcoma based on ultrasound and MRI findings who underwent 18F-FDG PET/CT between April 2008 and August 2013 in our hospital. The following imaging findings were obtained: (1) ultrasound exhibiting enlarged tumors and the characteristic ‘mosaic pattern’ on ultrasonic power Doppler images and/or (2) MRI exhibiting heterogeneous intensities with high signals on MR T1- and T2-weighted images, suggesting uterine sarcoma. Patients were excluded from the analysis if they (1) were previously diagnosed with another malignant disease and (2) had a follow-up duration of less than 2 years. Fifty-five patients were enrolled in this study. Fifteen patients were histopathologically proven to have
13
Ann Nucl Med
uterine sarcoma, 14 patients by surgical operation, and one by biopsy, consisting of 6 CS, 6 LMS, 2 H-ESS, and 1 L-ESS. Forty patients were diagnosed with leiomyoma, 29 patients by surgical operation, and 11 in a follow-up for at least 2 years. PET/CT imaging Whole-body PET scans with 18F-FDG were performed with a combined PET/CT scanner (Discovery LS; GE Medical Systems, Milwaukee, WI, USA), which permitted the simultaneous acquisition of 35 image slices in a 2-dimensional acquisition mode with inter-slice spacing of 4.25 mm. The PET/CT scanner incorporated an integrated four-slice multidetector CT scanner, which was used for attenuation correction. CT scanning parameters were as follows: Auto mA (upper limit, 40 mA; noise index, 20), 140 kV, 5-mm section thickness, 15-mm table feed, and a pitch of 4. After at least 4 h of fasting, patients received an intravenous injection of 185 MBq 18 F-FDG and image acquisition began 50 min after the injection. A whole-body emission scan was performed from the head to the inguinal region with 2 min per bed position (seven to eight bed positions). PET data were reconstructed by the iterative reconstruction method selecting 14 subsets and 2 iterations, a 128 × 128 matrix, voxel size (width, length and height) = 4 × 4 × 4.25 mm, and post-smoothing with an 8-mm Gaussian filter. Reconstructed images were then converted to a semi-quantitative image corrected by the injection dose and subject’s body weight (=standardized uptake value: SUV). The median duration between PET/CT and MRI acquisition was 14 days (range 0–51 days). Image analysis 18
F-FDG PET/CT images were retrospectively interpreted by the consensus of a gynecologic oncologist (MY, with 10 years of experience in gynecologic oncology) and an experienced radiologist (TT, with 12 years of experience in oncologic PET) who had no knowledge of the clinical data. PET images were co-registered to individual MR images. The volume of interest (VOI) was manually placed on the primary tumor by integrating 2D segmentations of each slice to cover the whole tumor using the PMOD 3.6 software package (PMOD Technologies Ltd, Zurich, Switzerland) (Fig. 1). VOIs were recorded and used for a subsequent texture analysis. Prior to textural feature computation, we resampled the VOI voxel intensities using absolute resampling method with fixed bounds and 64 discrete values [16, 17]: [ ] I(x) − lower bound R(x) = round 64 × (1) upper bound − lower bound where R(x) is the voxel intensity after discretization and I(x) is the voxel intensity before discretization. The lower bound
Ann Nucl Med Fig. 1 Sagittal T2-weighted (a), axial T2-weighted MR (b), 18F-FDG PET (c), and co-registered (d) images of a 50-year-old woman with an enlarging uterine leiomyoma. The VOIs were manually placed on the co-registered images according to MR contour to cover the whole tumor
was set to 0 and the upper bound was set to 25 SUV units which corresponded to the maximum intensity over all of the primary tumors included in this study. As a result, the sampling bin width of 0.4 SUV units was used in this study. SUV histogram, normalized gray-level co-occurrence matrix (NGLCM), and neighborhood gray-tone difference matrix (NGTDM) were used in assessments of first-, second-, and higher order textural features, respectively (Table 2). Eighteen different global and textural features were examined. The formulae and brief explanations of second- and higher order textural features measured in this study were shown in a supplemental document. The computation for textural features was performed using the Chang–Gung Image Texture Analysis toolbox implemented under MATLAB 2014b (Mathworks INC., Natick, MA, USA) [18]. Statistical analysis The Mann–Whitney U test was used to assess differences in PET features between uterine sarcoma and leiomyoma. A receiver-operating-characteristic (ROC) analysis was performed to determine optimal cut-off values for discrimination with high accuracy based on the area under the curves
(AUCs). The significance of differences between the AUCs was tested using the pairwise comparison of DeLong et al. [19]. The Mann–Whitney U test, ROC analysis, and comparison of AUCs were performed using MedCalc® (version 17.1; MedCalc Software bvba). To adjust for multiple comparison problems that resulted from an increase in the type-1 error when tested repeatedly, the Bonferroni correction method was used with an adjusted P value threshold of 0.003 (0.05/18).
Results The characteristics of all 55 patients are summarized in Table 1. Since five patients had two leiomyomas suspected of being uterine sarcomas, a total of 45 leiomyomas and 15 sarcomas were analyzed. Among the 18 PET features examined, SUV skewness, tumor volume, total lesion glycolysis (TLG), coarseness, and contrast from NGTDM were not significantly different between uterine sarcoma and leiomyoma, whereas the other 13 features were significantly different between them after the Bonferroni adjustment (Table 2). The intratumor distribution of 18F-FDG was more heterogeneous
13
Ann Nucl Med
Table 1 Patient characteristics Characteristic
Leiomyoma
Sarcoma
Number (n) of patients Age (years) Mean ± SD Range Menopausal status (n) Premenopausal Postmenopausal Types of lesions on histopathological findings (n)a
40
15
48.8 ± 11.0 26–82
59.0 ± 16.0 33–81
28 12 Typical, 11 Degenerated, 22 Follow-up, 12
5 10 Carcinosarcoma, 6 Leiomyosarcoma, 6 High-grade ESS, 2 Low-grade ESS, 1
leiomyoma was entropy. The AUC was 0.974 for entropy, which was better than the AUC of 0.951 for maximum SUV. The optimal cut-off value of 2.85 for entropy provided 93% sensitivity, 90% specificity, and 92% accuracy. The second most accurate single features were correlation and SUV SD, both of which yielded an AUC of 0.971. The third one was uniformity, which yielded an AUC of 0.965. Pairwise comparisons of AUCs did not show any significant differences between entropy, correlation, SUV SD, uniformity, and maximum SUV. When combining conventional features with textural ones in this study, maximum SUV (cutoff: 6.0) combined with entropy (2.85) and correlation (0.73) provided the best diagnostic performance (100% sensitivity, 94% specificity, and 95% accuracy).
a
Since five patients had two leiomyomas suspected of being uterine sarcomas, a total of 45 leiomyomas and 15 sarcomas were analyzed
in uterine sarcoma than in leiomyoma. The results of the ROC analysis (Table 2; Fig. 2) revealed that entropy, correlation, and uniformity calculated from NGLCM and SUV standard deviation (SUV SD) derived from histogram statistics showed greater AUCs than did maximum SUV for differentiating between sarcoma and leiomyoma. The most accurate single feature for discriminating sarcoma from
Discussion To the best of our knowledge, this is the first study to demonstrate the usefulness of quantitative 18F-FDG PET/CT textural features reflecting intratumor metabolic heterogeneity for the differential diagnosis of mesenchymal uterine tumors. Uterine sarcomas showed a significantly more heterogeneous 18 F-FDG distribution than leiomyomas. One of the secondorder features, entropy derived from NGLCM, had the best
Table 2 PET feature comparison and ROC analysis for the differentiation between leiomyoma and sarcoma
First order (histogram, global) Maximum SUV Mean SUV SUV SD SUV skewness SUV Kurtosis Total lesion glycolysis Tumor volume Second order (NGLCM, local) Uniformity Contrast Entropy Homogeneity Dissimilarity Correlation Higher order (NGTDM, Local) Coarseness Contrast Busyness Complexity Strength *Mann–Whitney U test
13
Leiomyoma Median (25% tile, 75% tile)
Sarcoma Median (25% tile, 75% tile)
P value*
AUC
Cutoff
3.07 (2.64, 4.07) 1.66 (1.47, 2.02) 0.32 (0.25, 0.42) 0.16 (−0.1, 0.68) 1.49 (1.33, 1.84) 491.1 (135.2, 1240.6) 280.4 (96.9, 659.6)
9.85 (8.94, 14.5) 3.39 (2.75, 4.66) 1.74 (1.35, 2.15) 0.53 (0.29, 1.21) 2.63 (2.14, 3.64) 1110.4 (206.3, 2086.5) 343.5 (34.3, 660.7)
0.001 0.001 0.001 0.044 0.001 0.272 0.519
0.95 0.917 0.971 0.674 0.922 0.594 0.556
6.0 2.2 0.60 0.25 2.1 930 130
0.19 (0.12, 0.25) 0.5 (0.44, 0.66) 1.93 (1.67, 2.34) 0.76 (0.71, 0.79) 0.49 (0.43, 0.6) 0.49 (0.37, 0.61)
0.02 (0.01, 0.05) 2.51 (1.95, 7.06) 4.13 (3.67, 4.64) 0.56 (0.44, 0.61) 1.25 (1.02, 1.85) 0.82 (0.79, 0.88)
0.001 0.001 0.001 0.001 0.001 0.001
0.965 0.947 0.974 0.936 0.946 0.971
0.085 1.45 2.85 0.62 0.95 0.73
0.002 (0.001, 0.007) 3E−06 (8E−07, 1E−05) 412.5 (146.3, 1326.1) 0.0017 (0.0003, 0.0046) 0.051 (0.012, 0.15)
0.003 (0.002, 0.021) 6E−06 (3E−06, 0.0002) 14.6 (1.5, 29.4) 0.047 (0.0073, 0.33) 0.92 (0.32, 3.19)
0.11 0.014 0.001 0.001 0.001
0.638 0.713 0.893 0.893 0.908
0.0024 4E−06 45.0 0.006 0.28
Ann Nucl Med
Fig. 2 ROC curves and AUCs in parentheses demonstrating the ability of maximum SUV, SUV SD, uniformity, entropy, and correlation in distinguishing uterine sarcoma from leiomyoma. The ROC curve and AUC for TLG were also shown as a reference
diagnostic performance for differentiating between uterine sarcoma and leiomyoma. Entropy may be utilized to measure the uncertainty or randomness of intensity distribution, with more uncertainty or randomness implying higher entropy [10, 12]. Correlation calculated from NGLCM was the second most accurate single feature for differentiating between uterine sarcoma and leiomyoma. Correlation quantifies how a voxel correlates with a consecutive voxel according to a spatial relationship [12]. Uniformity was the third most accurate single feature for the discrimination. Since uniformity indicates the degree of the local homogeneous nature of tumor 18F-FDG uptake, sarcoma showed ‘lower’ uniformity than did leiomyoma. It is difficult to differentiate uterine sarcoma from leiomyoma when leiomyoma shows high 18F-FDG uptake and degenerative changes on MR images [20]. Although MRI including diffusion-weighted imaging (DWI) and contrastenhanced (CE) imaging has been used as a standard for the differential diagnosis of uterine sarcoma and leiomyoma, no adequate evidence has been provided by large-scale or prospective studies and this is partly due to the rareness of uterine sarcoma. Overlap was noted in the distribution of the apparent diffusion coefficient (ADC) values calculated from DWI between uterine sarcoma and leiomyoma [4, 5]. The most recent study by Lin et al. showed that CE-MRI provides accurate information on a differential diagnosis and is preferable to DWI [6]. In their consecutive cohort of 8 sarcomas and 25 leiomyomas, CE-MRI yielded a significantly
higher diagnostic accuracy (94%) than DWI (52%). Our study consisted of a larger patient population (15 sarcomas and 45 leiomyomas) and showed that an 18F-FDG PET/CT texture analysis (local entropy) yielded a similar diagnostic accuracy (92%) to CE-MRI. This result was consistent with the previous study by Xu et al. in which local entropy on PET images provided the highest accuracy (AUC = 0.841) for differentiating between malignant and benign bone/ soft-tissue lesions [12]. Their study also showed that the correlation on ‘CT images’ may be an indicator of malignancy; the correlation in malignant tumors was higher than that in benign lesions. The reason for this was considered to be as follows: since perfusion or fibrosis, cellular alignment or other shaped opacities, and structural damage often occur simultaneously along a certain direction in malignant tumors on CT images, there are more similar or correlated Hounsfield Unit values along a certain direction in malignant tumors than in benign lesions [12]. This may be the case in our study regarding metabolic textural features on ‘PET images’, in which uterine sarcoma showed a significantly higher metabolic correlation on PET images than leiomyoma. In our study population, the diagnostic accuracy of firstorder global features such as SUV SD and maximum SUV was similar to that of local entropy, correlation, and uniformity. Previous studies regarding differentiation of uterine tumors showed the usefulness of the conventional 18F-FDG PET parameters combined with the assessments of MR imaging and serum lactate dehydrogenase (LDH) levels [21, 22]. In this study, when combining the conventional features with textural ones, maximum SUV (cutoff: 6.0) combined with entropy (2.85) and correlation (0.73) exceeded the diagnostic performance of entropy only (100% sensitivity, 94% specificity, and 95% accuracy). Although the conventional first-order features are calculated from a distribution histogram and do not have positional information, they are easily available in clinical settings. The combination of textural features and histogram statistics may allow for a more accurate differentiation between uterine sarcoma and leiomyoma. Since a single feature cannot be directly linked to a specific biological process, a combination of textural and conventional features including volumetric parameters may be closely related to underlying physiological processes and may provide more accurate differentiation than each single feature. In the near future, artificial intelligence systems will contribute to automatically contouring tumors, extracting numerous textural features, correlating features and gene expression, and finding potential prognostic models and promising effective therapeutic strategies. A radiogenomics (radiomics) approach will be enhanced with the advancement of artificial intelligence [23]. There is a limitation to the present study. Since textural features are dependent on various factors such as image
13
acquisition, reconstruction, preprocessing, segmentation, and mathematical methods [24], the standardization of a PET texture analysis will be necessary for inter-institutional evaluations in the future.
Conclusion In combination with the conventional histogram statistics and/or volumetric parameters, 18F-FDG PET/CT textural features reflecting intratumor metabolic heterogeneity are useful for differentiating between uterine sarcoma and leiomyoma. Acknowledgements The authors thank Dr. Yu-Hua Dean Fang, the original developer of the Chang-Gung Image Texture Analysis toolbox, and the staff of the Department of Radiology and Biological Imaging Research Center, University of Fukui, for their clinical and technical support. Compliance with ethical standards Consent to participate Formal consent was not required for this type of retrospective study. Conflict of interest The authors have no conflicts of interest to declare. Funding This study was partly funded by Grants-in-Aid for scientific research from the Japan Society for the Promotion of Science (15H04981, 16K10345, and 16K20181) and Takeda Science Foundation.
References 1. Adams Hillard PJ. Benign disease of the female reproductive tract. In: Berek JS, editor. Berek and Novak’s gynecology. 14th edn. Philadelphia, PA: Lippincott Williams and Wilkins; 2007. pp. 431–504. 2. D’Angelo E, Prat J. Uterine sarcomas: a review. Gynecol Oncol. 2010;116:131–9. 3. Gadducci A. Prognostic factors in uterine sarcoma. Best Pract Res Clin Obstet Gynaecol. 2011;25:783–95. 4. Namimoto T, Yamashita Y, Awai K, Nakaura T, Yanaga Y, Hirai T, et al. Combined use of T2-weighted and diffusion-weighted 3-T MR imaging for differentiating uterine sarcomas from benign leiomyomas. Eur Radiol. 2009;19:2756–64. 5. Sato K, Yuasa N, Fujita M, Fukushima Y. Clinical application of diffusion-weighted imaging for preoperative differentiation between uterine leiomyoma and leiomyosarcoma. Am J Obstet Gynecol. 2014;210:368.e1–8. 6. Lin G, Yang LY, Huang YT, Ng KK, Ng SH, Ueng SH, et al. Comparison of the diagnostic accuracy of contrast-enhanced MRI and diffusion-weighted MRI in the differentiation between uterine leiomyosarcoma/smooth muscle tumor with uncertain malignant potential and benign leiomyoma. J Magn Reson Imaging. 2016;43:333–42. 7. Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med. 2012;366:883–92.
13
Ann Nucl Med 8. 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;38:987–91. 9. 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;62:545–53. 10. El Naqa I, Grigsby P, Apte A, Kidd E, Donnelly E, Khullar D, et al. Exploring feature-based approaches in PET images for predicting cancer treatment outcomes. Pattern Recognit. 2009;42:1162–71. 11. Chicklore S, Goh V, Siddique M, Roy A, Marsden PK, Cook GJ. Quantifying tumour heterogeneity in 18F-FDG PET/CT imaging by texture analysis. Eur J Nucl Med Mol Imaging. 2013:133–40. 12. Xu R, Kido S, Suga K, Hirano Y, Tachibana R, Muramatsu K, et al. Texture analysis on 18F-FDG PET/CT images to differentiate malignant and benign bone and soft-tissue lesions. Ann Nucl Med. 2014;28:926–35. 13. Yang F, Thomas MA, Dehdashti F, Grigsby PW. Temporal analysis of intratumoral metabolic heterogeneity characterized by textural features in cervical cancer. Eur J Nucl Med Mol Imaging. 2013;40:716 – 27. 14. Cook GJ, O’Brien ME, Siddique M, Chicklore S, Loi HY, Sharma B, et al. Non-small cell lung cancer treated with Erlotinib: heterogeneity of 18F-FDG uptake at PET-association with treatment response and prognosis. Radiology. 2015;276:883–93. 15. Tixier F, Le Rest CC, Hatt M, Albarghach N, Pradier O, Metges JP, 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. 16. Orlhac F, Thézé B, Soussan M, Boisgard R, Buvat I. Multiscale texture analysis: From 18F-FDG PET images to histologic images. J Nucl Med. 2016;57:1823–8. 17. Orlhac F, Nioche C, Soussan M, Buvat I. Understanding changes in tumor textural indices in PET: a comparison between visual assessment and index values in simulated and patient data. J Nucl Med. 2017;58:387–92. 18. Cheng NM, Fang YH, Chang JT, Huang CG, Tsan DL, Ng SH, et al. Textural features of pretreatment 18F-FDG PET/CT images: prognostic significance in patients with advanced T-stage oropharyngeal squamous cell carcinoma. J Nucl Med. 2013;54:1703–9. 19. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44:837–45. 20. Kitajima K, Murakami K, Kaji Y, Sugimura K. Spectrum of FDG PET/CT findings of uterine tumors. AJR Am J Roentgenol. 2010;195:737–43. 21. Yoshida Y, Kurokawa T, Sawamura Y, Shinagawa A, Tsujikawa T, Okazawa H, et al. Comparison of 18F-FDG PET and MRI in assessment of uterine smooth muscle tumors. J Nucl Med. 2008;49:708–12. 22. Nagamatsu A, Umesaki N, Li L, Tanaka T. Use of 18F-fluorodeoxyglucose positron emission tomography for diagnosis of uterine sarcomas. Oncol Rep. 2010;23:1069–76. 23. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006. 24. Hatt M, Tixier F, Pierce L, Kinahan PE, Le Rest CC, Visvikis D. Characterization of PET/CT images using texture analysis: the past, the present… any future? Eur J Nucl Med Mol Imaging. 2017;44:151–65.
