|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ИНХС РАН |
||
CT radiomic features in differentiation of renal cell carcinoma metastases from pancreatic neuroendocrine tumorsтезисы доклада
- Авторы: Gruzdev I., Lapteva M., Zamyatina K., Tikhonova V., Kondratyev E., Kaldarov A., Proskuryakov I., Podluzhny D., Karmazanovsky G.
- Сборник: EPOS
- Тезисы
- Год издания: 2022
- Место издания: European Congress of Radiology
- DOI: 10.26044/ECR2022/C-16758
- Аннотация: PurposeMetastases to the pancreas occur in only 1–2% of all pancreatic neoplasms [1]. Renal cell carcinoma (RCC) is the most common tumor that metastasizes to the pancreas [2]. The main differential diagnosis of RCC metastases is pancreatic neuroendocrine tumors (PNETs), which are also characteristic of the presence of a hypervascular mass in the pancreatic parenchyma on contrast-enhanced computed tomography (CECT), the main method for diagnosing solid pancreatic masses [3]. Both tumors may form multiple pancreatic masses, including those with various types of genetic syndromes, primarily von Hippel-Lindau [4]. At the same time, solitary metastases of RCC in the pancreas may be the only manifestation of the metastatic process several years after nephrectomy [5]. Texture analysis can improve the accuracy of CT in the preoperative differential diagnosis of RCC metastasis and PNETs. The purpose of the study was to compare texture and CT features between RCC metastases and PNETs.Methods and materialsPatient population: We retrospectively enrolled 81 patients with histologically proven PNETs and 17 patients with RCC pancreatic metastases who underwent preoperative CE MDCT with hypervascular tumors on CT images. If a patient had more than one mass, then all separate lesions larger than 10 mm in diameter were included in the study.Data acquisition: CT protocols varied. The slice thickness ranged from 1 to 5 mm. A mandatory inclusion criterion was the presence of the native and arterial phases of the study.CT image analysis: two radiologists with 4 and 14 years of experience in abdominal imaging independently assessed for composition (totally solid, solid with some cystic areas), homogeneity (homogeneous, inhomogeneous), calcification (present or absent), and for the presence of main pancreatic duct (MPD) dilation (yes/no). For contrast enhancement features evaluation, the radiologists drew a manual ROI in the area with the maximum contrast agent accumulation in the arterial phase of the study and copied it to the corresponding images in the native phase of the study. Also, a round ROI of 1 cm2 was drawn on the normal pancreatic parenchyma at the same slice. Areas of calcification, cystic degeneration, vessels, pancreatic duct were excluded.Mean tumor and normal pancreatic parenchyma attenuation were measured. Lesion-to-parenchyma contrast (LPC) for arterial phase was calculated by the formula:LPC=Ta/Pawhere Ta - tumor attenuation in arterial phase, Pa - pancreatic parenchyma attenuation in the arterial phase.Relative tumor enhancement ratio (RTE) was calculated by the formula:RTEa=(Ta-Tn)/(Pa-Pn)where Ta – tumor attenuation in arterial phase, Tn – tumor attenuation in native phase, Pa – attenuation of normal pancreatic parenchyma in arterial phase, Pn – attenuation of normal pancreatic parenchyma in native phase.CT texture analysis: the LIFEx application (version v5.10, www.lifexsoft.org) was used to extract the texture features of CECT-images [6]. Two radiologists with 4 and 12 years of experience in abdominal imaging drew a manual 3D-ROI covering up the entire tumor volume on the arterial phase (Fig. 1 and Fig.2). After tumor segmentation 52 texture features were automatically calculated.Statistical analysis: For statistical analysis R 4.1.0 (R Foundation for Statistical Computing) was used. Features with low reproducibility according to our previous work were excluded from the analysis [7]. To analyze the reproducibility of CT characteristics when assessed by two radiologists, the type I intraclass correlation coefficient (ICC) for quantitative variables and the Cohen kappa statistic (κ) for binary variables were used. The selection of predictors in the binary logistic model was performed in 2 stages: 1) selection of predictors using one-factor logistic models under the conditions of p <0.2; 2) selection of predictors using L2-regularization (LASSO-regression after standardization of variables). The selected predictors were included in a binary logistic regression model without interactions.ResultsWe did not reveal a significant difference in the composition, homogeneity, calcification, presence of dilation of the MPD. We did not find calcification in RCC metastasis, in contrast to the PNETs. RCC metastasis LCR and RTE were significantly higher compared to PNETs (1.85 vs. 1.47 and 2.57 vs. 1.82, respectively, p<0.001). A strong correlation was observed between these features (ρ = 0.93), while in univariate logistic models there were no significant differences in the size of the association with the diagnosis, and therefore, it was decided to include the LCR in the subsequent selection due to higher reproducibility in radiologists assessments.When using LASSO-regression, 4 predictors were selected in the final diagnostic model: LCR, CONVENTIONAL_HUmin, GLCM_Correlation, NGLDM_Coarseness. When assessed by the Youden method and an index of 16%, the sensitivity and specificity of the diagnostic model in the differential diagnosis of PNETs and RCC metastases were 95.8; 62 respectively (AUC = 0.82).ConclusionIn comparison with RCC metastases, PNETs could contain calcifications, have lower LCR and RTE. The combined use of contrast characteristics and texture features allows differentiation between PNETs and RCC metastases by preoperative CECT.References[1] C. Sperti, L. Moletta, G. Patanè, Metastatic tumors to the pancreas: The role of surgery., World J. Gastrointest. Oncol. 6 (2014) 381–92. https://doi.org/10.4251/wjgo.v6.i10.381.[2] T. Ito, R. Takada, S. Omoto, M. Tsuda, D. Masuda, H. Kato, T. Matsumoto, I. Moriyama, Y. Okabe, H. Shiomi, E. Ishida, K. Hatamaru, S. Hashimoto, K. Tanaka, H. Kawamoto, A. Yanagisawa, T. Katayama, S. Yazumi, Analysis of Prognostic Factors in Pancreatic Metastases: A Multicenter Retrospective Analysis, Pancreas. 47 (2018) 1033–1039. https://doi.org/10.1097/MPA.0000000000001132.[3] M.H. Shah, W.S. Goldner, T.R. Halfdanarson, E. Bergsland, J.D. Berlin, D. Halperin, J. Chan, M.H. Kulke, A.B. Benson, L.S. Blaszkowsky, J. Eads, P.F. Engstrom, P. Fanta, T. Giordano, J. He, M.J. Heslin, G.P. Kalemkerian, F. Kandeel, S.A. Khan, W.Z. Kidwai, P.L. Kunz, B.W. Kuvshinoff, C. Lieu, V.G. Pillarisetty, L. Saltz, J.A. Sosa, J.R. Strosberg, C.A. Sussman, N.A. Trikalinos, N.A. Uboha, J. Whisenant, T. Wong, J.C. Yao, J.L. Burns, N. Ogba, G. Zuccarino-Catania, Neuroendocrine and adrenal tumors, version 2.2018 featured updates to the nccn guidelines, JNCCN J. Natl. Compr. Cancer Netw. 16 (2018) 693–702. https://doi.org/10.6004/jnccn.2018.0056.[4] S. Laks, R. van Leeuwaarde, D. Patel, X.M. Keutgen, P. Hammel, N. Nilubol, T.P. Links, T.R. Halfdanarson, A.B. Daniels, A. Tirosh, Management recommendations for pancreatic manifestations of von Hippel–Lindau disease, Cancer. (2021). https://doi.org/10.1002/CNCR.33978.[5] M. Nogueira, S.C. Dias, A.C. Silva, J. Pinto, J. Machado, Solitary pancreatic renal cell carcinoma metastasis, Autops. Case Reports. 8 (2018). https://doi.org/10.4322/ACR.2018.023.[6] C. Nioche, F. Orlhac, S. Boughdad, S. Reuze, J. Goya-Outi, C. Robert, C. Pellot-Barakat, M. Soussan, F. erique Frouin, I. Buvat, Lifex: A freeware for radiomic feature calculation in multimodality imaging to accelerate advances in the characterization of tumor heterogeneity, Cancer Res. 78 (2018) 4786–4789. https://doi.org/10.1158/0008-5472.CAN-18-0125.[7] I.S. Gruzdev, K.A. Zamyatina, V.S. Tikhonova, E.V. Kondratyev, A.V. Glotov, G.G. Karmazanovsky, A.S. Revishvili, Reproducibility of CT texture features of pancreatic neuroendocrine neoplasms, Eur. J. Radiol. 133 (2020). https://doi.org/10.1016/j.ejrad.2020.109371.
- Добавил в систему: Лаптева Мария Георгиевна