|Year : 2020 | Volume
| Issue : 7 | Page : 1648-1655
Predicting myometrial invasion in endometrial cancer based on whole-uterine magnetic resonance radiomics
Yuqing Han1, Han Xu2, Ying Ming3, Qingwei Liu2, Chencui Huang4, Jingxu Xu4, Jie Zhang2, Yan Li5
1 Department of Radiology, Cheeloo College of Medicine, Shandong Provincial Hospital, Shandong University, Jinan, Shandong, China
2 Department of Radiology, Cheeloo College of Medicine, Shandong Provincial Hospital, Shandong University; Department of Radiology, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, Shandong, China
3 Department of Radiology, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, Shandong, China
4 Department of Research Collaboration, R&D Center, Beijing Deepwise and League of PHD Technology Co., Ltd., Beijing, China
5 Department of Medical Imaging Center, Cheeloo College of Medicine, The Second Hospital, Shandong University, Jinan, Shandong, China
|Date of Submission||21-Sep-2020|
|Date of Decision||20-Oct-2020|
|Date of Acceptance||24-Nov-2020|
|Date of Web Publication||9-Feb-2021|
Department of Radiology, Shandong Provincial Hospital, Cheeloo College of Medicine, Shandong University, 324 Jingwu Weiqi Road; Department of Radiology, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan 250021, Shandong
Department of Medical Imaging Center, The Second Hospital, Cheeloo College of Medicine, Shandong University, 247#Beiyuan Road, Jinan 250033, Shandong
Source of Support: None, Conflict of Interest: None
Objective: The objective of this study was to evaluate whether whole-uterine magnetic resonance imaging (MRI) radiomic features can predict myometrial invasion (MI) depth in endometrial cancer (EC).
Materials and Methods: The preoperative 3.0 T magnetic resonance examinations of EC patients were retrospectively reviewed. Whole-uterus segmentation was performed, and features were extracted based on sagittal T2-weighted imaging (T2WI) and axial diffusion-weighted imaging (DWI). The logistic regression (LR) classifier algorithm was used to establish the radiomic model, which was verified by ten times five-fold cross-validation. The areas under the receiver operating characteristic (ROC) curves (AUCs) were assessed by the DeLong test to compare differences among the models based on different sequences. The LR model was compared with the subjective diagnosis results by the Chi-square test.
Results: Of the 163 EC patients included, 44 had deep myometrial invasion (DMI). The feature consistency of the whole uterus was higher than that of the lesion (P < 0.05). The sagittal T2WI, axial DWI, and combined models had AUCs of 0.76, 0.80, and 0.85 in the validation set, respectively. The DeLong test showed that there were no significant differences in AUCs among the models (P > 0.05). The single-sequence LR models had lower specificity and accuracy than the corresponding subjective diagnostic results (P < 0.05), while the sensitivity was higher (P > 0.05). The combined model included 24 radiomic features, and the accuracy, sensitivity, and specificity were 0.83, 0.77, and 0.85 for DMI, respectively. There was no significant difference compared with subjective diagnosis (P > 0.05).
Conclusion: Whole-uterine MRI radiomic features based on sagittal T2WI and axial DWI show potential in predicting MI in EC.
Keywords: Endometrial cancer, magnetic resonance imaging, myometrial invasion, radiomics
|How to cite this article:|
Han Y, Xu H, Ming Y, Liu Q, Huang C, Xu J, Zhang J, Li Y. Predicting myometrial invasion in endometrial cancer based on whole-uterine magnetic resonance radiomics. J Can Res Ther 2020;16:1648-55
|How to cite this URL:|
Han Y, Xu H, Ming Y, Liu Q, Huang C, Xu J, Zhang J, Li Y. Predicting myometrial invasion in endometrial cancer based on whole-uterine magnetic resonance radiomics. J Can Res Ther [serial online] 2020 [cited 2021 Mar 8];16:1648-55. Available from: https://www.cancerjournal.net/text.asp?2020/16/7/1648/308758
| > Introduction|| |
Endometrial cancer (EC) is one of the most common gynecologic cancers in China, and its incidence is increasing, especially in high urbanization areas. In America, 61,880 estimated new cases and 12,160 estimated deaths from EC were projected to occur in 2019. Total hysterectomy plus bilateral salpingo-oophorectomy is a basic surgical method, and whether to perform lymph node dissection is controversial. The depth of myometrial invasion (MI) is a crucial factor for surgical determination and prognosis and is closely correlated with lymph node metastasis.,, According to the European Society for Medical Oncology guidelines, lymphadenectomy is suggested or recommended for patients with deep myometrial invasion (DMI). Meanwhile, sentinel lymph node biopsy may replace lymphadenectomy.
The accurate preoperative assessment of the depth of MI depends on magnetic resonance imaging (MRI), especially T2-weighted (T2W) imaging (T2WI) and contrast-enhanced T1-weighted imaging. Some findings suggested that diffusion-weighted (DW) imaging (DWI) may have good diagnostic accuracy for assessing MI. However, preoperative magnetic resonance (MR) examination has some diagnostic challenges as follows: the uterine cavity is often filled with polypoid tumors; the normal zonal anatomy of the uterus is distorted by leiomyomas, and tumors located in uterine cornu may lead to incorrect evaluations on MI.
The texture features of EC lesions can be used to predict DMI, lymphovascular space invasion, and histologic high-grade tumors. Few studies have been performed based on the whole uterus, which may be easier to mark than marking the lesions. The different depths of tumor MI may cause differences in the thickness of the myometrium, and there may be differences in texture information between the high-signal tumor and the low-signal myometrium. We aimed to extract the radiomic features of the whole uterus using different feature selection algorithms to establish radiomic models and to compare the value of sagittal T2W and axial DW image features and combined radiomic features in predicting MI in EC.
| > Materials and Methods|| |
The respective study was accepted by the Institutional Review Committee, and informed consent was obtained from each patient. Between February 2011 and September 2019, 163 EC patients (including 44 patients with DMI) who underwent 3.0 T MR examinations before surgery in Shandong Provincial Hospital were included. All patients underwent total hysterectomy plus bilateral salpingo-oophorectomy. The inclusion criteria were as follows: (1) the diagnosis was confirmed by surgical histologic findings; (2) pelvic MRI, including conventional MRI and DWI examination, was performed within 2 weeks before surgery; and (3) no neoadjuvant therapy was administered before surgery. The exclusion criteria were as follows: (1) previous history of other malignancies and related treatments; (2) patients with intrauterine devices; and (3) low-quality MR images with severe artifacts.
Magnetic resonance protocols
All MRI studies were performed with 3.0 T MRI systems, including Siemens Magnetom Verio 3.0 T (Erlangen, Germany), Philips Ingenia 3.0 T (Veenpluis, The Netherlands), and Siemens Prisma 3.0 T (Erlangen, Germany). A region of interest (ROI) was segmented around the whole uterus on sagittal T2W turbo spin-echo images and axial DW images with b = 800 s/mm2. The related MR acquisition parameters are listed in [Table 1].
Region of interest segmentation
The sagittal T2W images and axial DW images of EC patients were imported into Deepwise scientific research platform (http://keyan.deepwise.com/; version 1.6). The whole uterus was segmented on each slice as an ROI by radiologist A (Zhang J, with 10 years of experience) to generate a three-dimensional (3D) volume of interest (VOI) [Figure 1]. According to the pathological results, the patients were divided into two groups: DMI absent and DMI present. To assess feature consistency and reproducibility, ROI segmentation was independently performed by radiologist B (Han Y, with 2 years of experience) on 20 randomly selected patients. ROI re-segmentation was independently performed by radiologist A on the same 20 patients after a washout period of 4 weeks. Twenty EC lesions were delineated with the same method to assess consistency. For some EC patients with invasion of adjacent tissues, the whole uterus was delineated as far as possible, including the tumor and the remaining normal uterine outline.
|Figure 1: ROI segmentation and the generated 3D VOI. (a) In the axial diffusion-weighted image, the lesion (thin arrow) showed a high signal intensity and deep myometrial invasion. (b) The 3D VOI of the whole uterus corresponding to A. (c) In the sagittal T2-weighted image, the lesion (thick arrow) showed a slightly higher signal and superficial myometrial invasion. (d) The 3D VOI of the whole uterus corresponding to C. (The red line represents the ROI of the segmented whole uterus.) ROI = Region of interest, 3D VOI = Three-dimensional volume of interest|
Click here to view
B-spline interpolation sampling was used to resample the MR images, voxels were unified as isotropic voxels (2.0 mm × 2.0 mm × 2.0 mm), and the PyRadiomics toolkit (version 2.1.0) was used to extract radiomic features from the ROI. Then, the MRI images were normalized by centering at the mean with standard deviation, and the scale was set to 100. After wavelet transform (high-pass or low-pass filtering in the x, y, and z planes) and Laplacian of Gaussian (log) filter transform (with different λ parameters) were performed on the original images, the radiomic features of these three types of images were extracted, including 234 first-order features, 14 shape features, 286 gray-level co-occurrence matrix features, 208 gray-level size zone matrix features, 208 gray-level run-length matrix features, and 182 gray-level dependence matrix features. A total of 1132 radiomic features were extracted from each of the sagittal T2W and axial DW 3D images. All the radiomic features were standardized by the Z-score method and converted into an eigenvalue with an average of 0 and a standard deviation of 1. After consistency analysis, the features with a consistency of <0.75 were eliminated. Then, the features were examined for the linear dependence relation. When the linear dependent coefficient ρ between any two independent variables was >0.9, one of the features was removed to alleviate the redundancy. The features with high linear dependent coefficients for dependent variables were retained in preference, leaving 189 sagittal T2WI features and 145 axial DWI features.
Establishment and evaluation of machine learning models
First, the radiomic features of single-sequence images (189 sagittal T2WI features and 145 axial DWI features) were used for modeling and dimensionality reduction. The combined features were defined as the combination of retained T2WI and DWI features. Further dimensionality reduction was performed based on the combined features. Multiple feature dimension reduction methods (such as the F-test and L1 regularization) and logistic regression (LR) classifier algorithm were used according to different parameters to establish machine learning models. Ten times five-fold cross-validation followed to verify the stability and reproducibility of the predicted results and to ensure that the performance was not by chance. The diagnostic performance of the LR model was evaluated according to the area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, specificity, and accuracy.
Subjective imaging diagnosis
MI was diagnosed by two senior radiologists (Zhang J and Xu H, with 10 years of experience) based on sagittal T2W images, axial DW images, and a combination of them. A third radiologist (Liu Q, with 20 years of experience) made the judgment if the results were different. Tumor invasion depth ≥50% is defined as DMI present, and myometrium extension <50% is defined as DMI present. The gold standard was the result of surgical pathology.
Intra- and interclass correlation coefficients (ICCs), obtained through reliability analysis, were used to evaluate the feature consistency of the two radiologists' VOI segmentation. The DeLong test was used to evaluate the AUC diagnostic efficacy of different ROC curves. A kappa consistency test was used to compare the subjective diagnostic results of the two radiologists. The McNemar test was used to compare the sensitivity and specificity of the subjective diagnosis among different sequences. The Chi-square test was used to compare the difference in the feature composition ratio of ICC ≥0.75 and ≥0.90 based on two segmentation methods and to compare the diagnostic efficacy of the LR models with the corresponding subjective diagnosis. P < 0.05 was used to determine statistical significance. All statistical analyses were conducted with SPSS Statistics software version 22.0 (IBM, New York, USA) and MedCalc MedCalc Statistical Software version 19.1.2 (MedCalc Software bvba, Ostend, Belgium).
| > Results|| |
Surgical histologic findings
The patient population included 163 women with EC. The median age was 57 years (range, 31–77). An overview of the patient characteristics is shown in [Table 2].
A total of 1132 features were extracted from sagittal T2W images, as well as axial DW images, whether delineating the whole uterus or the lesion. For delineating the whole uterus, the ICC results from the reliability analysis are shown in [Figure 2]. The constituent ratios of features with ICC ≥0.75 based on sagittal T2W images and axial DW images were 94.88% (1074/1132) and 80.48% (911/1132), respectively. ICCs ≥0.75 and ≥0.90 represent good consistency and excellent consistency, respectively. For the constituent ratio of features with ICC ≥0.75 and ≥0.90, constituent ratios for delineating the whole uterus were larger than those for delineating the lesion. The difference was statistically significant (P < 0.05) [Table 3]. This may indicate that the method of lesion segmentation is more influenced by the clinical experience of radiologists.
|Figure 2: Intra- and interclass correlation coefficients (the whole uterus). (a) T2WI intraclass correlation. (b) T2WI interclass correlation. (c) DWI intraclass correlation. (d) DWI interclass correlation. T2WI = T2-weighted imaging, DWI = Diffusion-weighted imaging|
Click here to view
|Table 3: Constituent ratio of radiomic features based on different segmentation methods|
Click here to view
Construction and evaluation of machine learning models
Of the 189 T2WI features analyzed, 160 features were removed in the final model by F-test dimension reduction and the LR classifier algorithm. There were 29 T2WI features included. The AUC for identifying DMI was 0.76 (95% confidence interval [CI], 0.683–0.820) [Figure 3]. Based on 145 axial DWI features, the LR model was established by L1 regularization, with a regularization coefficient of 1.0. There were 26 DWI features included. The AUC for identifying DMI was 0.80 (95% CI, 0.731–0.859).
|Figure 3: ROC curves for the different radiomic models in the validation set. The AUCs of the sagittal T2WI model, axial DWI model, and combined model were 0.76, 0.80, and 0.85, respectively. ROC = Receiver operating characteristic, AUC = Area under the ROC curve, T2WI = T2-weighted imaging, DWI = Diffusion-weighted imaging|
Click here to view
The combined LR model based on T2WI and DWI features established by L1 regularization (the regularization coefficient was 0.1) had an AUC of 0.85 (95% CI, 0.785–0.901) in the validation set. Twenty-four radiomic features (7 T2WI features and 17 DWI features) were incorporated, including 6 original image features, 16 wavelet transform image features, and 2 log transform image features [Table 4]. The accuracy, sensitivity, and specificity of the combined model were 0.83, 0.77, and 0.85, respectively [Table 5]. Among the features, the kurtosis feature, a first-order feature of DWI images, was the most important modeling feature. From the kurtosis histogram, a patient without DMI had a high kurtosis of 4.17, while a patient with DMI had a low kurtosis of 2.61 [Figure 4]. The DeLong test showed that there were no statistically significant differences in AUC among the three LR models (P > 0.05).
|Figure 4: Kurtosis histogram of DWI (the whole uterus). The red curve represents the signal intensity distribution curve of a patient without DMI, and the mass of the distribution is concentrated toward a spike near the mean signal intensity. The kurtosis is 4.17. The green curve represents the signal intensity distribution curve of a patient with DMI, with a flatter peak. The kurtosis is 2.61. DWI = Diffusion-weighted imaging, DMI = Deep myometrial invasion|
Click here to view
|Table 4: The 24 radiomic features of the combined logistic regression model|
Click here to view
|Table 5: Diagnostic efficacies of subjective diagnosis and radiomic models for the determination of deep myometrial invasion|
Click here to view
Subjective imaging diagnosis
The kappa values of the two radiologists' subjective judgments on sagittal T2W images, axial DW images, and combined images were 0.686, 0.687, and 0.726, respectively (P < 0.001). The final results of the subjective diagnosis are shown in [Table 5].
There were no statistically significant differences in sensitivity and specificity among T2WI, DWI, and their combination (T2WI vs. DWI, P = 0.134; T2WI vs. both, P = 0.096; DWI vs. both, P = 1.000).
Comparison of subjective imaging diagnosis and logistic regression models
The comparison of the subjective diagnosis results and the LR model diagnosis results is shown in [Table 5]. The specificity and accuracy of the single-sequence LR models were lower than those of the corresponding single-sequence subjective diagnosis, with statistically significant differences (P < 0.05). The sensitivity was higher than that of the corresponding subjective diagnosis, with no statistically significant difference (P > 0.05). This showed that the single-sequence model had a weak diagnostic effect. On the other hand, there were no statistically significant differences in sensitivity, specificity, and accuracy between the combined LR model based on both T2WI and DWI features and subjective diagnosis (P > 0.05), showing similar predictive efficiency.
| > Discussion|| |
In a meta-analysis of enhanced MRI combined with DWI for subjective diagnosis, the sensitivity and specificity for MI were 0.807 and 0.885, respectively. However, more experienced radiologists may have a higher diagnostic accuracy than unexperienced radiologists. Lymphadenectomy is suggested or recommended for patients with DMI. Systematic lymphadenectomy involves the removal of pelvic and para-aortic lymph nodes to the renal vascular level, which can lead to lymph cyst formation, deep vein thrombosis, and pulmonary embolism development, and approximately 25% of the patients develop lower extremity lymphedema and lower extremity dysfunction.,,
Furthermore, hypoxia-inducible factor-1α expression was significantly correlated with depth of MI. Therefore, accurate DMI prediction before surgery is very important.
Tumor lesions have been segmented in some radiomic studies. For example, Ytre-Hauge et al. confirmed that high tumor entropy based on apparent diffusion coefficient (ADC) maps may predict DMI. They used texture parameters from two-dimensional (2D) ROIs to perform statistical analysis. The depth of MI may be assumed by artificially delineating the boundary of EC lesions. However, the method of lesion segmentation may be more influenced by doctors' clinical experience than that of whole-uterus segmentation [Table 3]. In addition, lesion segmentation cannot be used to analyze the relevant texture information around the tumor. Relevant studies demonstrated that the mean ADC value at the peritumoral zone had a good ability to predict DMI. Xie et al. confirmed that the radiomic features of the whole uterus on ADC maps could better differentiate uterine fibroids and sarcomas than those of the tumor itself or the tumor with a small piece of surrounding tissue. Organ-based radiomic studies have been conducted to predict the recurrence of acute pancreatitis and the aggressiveness of prostate cancer., Therefore, we conducted radiomic research based on the whole uterus.
Kurtosis is a measure of the sharpness of the distribution of values in the image ROI. It represents the peakedness of the distribution. In our research, PyRadiomics kurtosis was not corrected by subtracting 3, yielding a value 3 for normal distributions. We found that the dependent variable MI is negatively correlated with kurtosis [Figure 4]. So the kurtosis of a patient without DMI is >3, which indicates that the signal intensity distribution is more peaked than the normal distribution in the histogram of DW images. The kurtosis of a patient with DMI is <3, which indicates that the overall data distribution is flatter than the normal distribution. That means the mass of the signal intensity distribution is concentrated toward the tails rather than toward the mean. Meanwhile, whole-uterus segmentation can contain the texture information of the interface between the myometrium and EC lesions. The interface information includes its position, the range of the interface (the volume of all voxels in the interface), the signal contrast between the two sides of the interface, the clarity of the interface, and so on. In this study, the interface information in the ROIs (whole uterus) was mainly located in the interface between EC lesions and the myometrium, partly in the EC lesions (such as the interface between necrosis or hemorrhage and EC parenchyma) and partly in the myometrium (such as the interface between leiomyoma, adenomyosis, and the normal myometrium). The junctional zone is often obscure postmenopause. Most of the EC patients were postmenopausal. The interface between the outer myometrium and the junctional zone was not obvious, especially in DW images. This interface information is mainly reflected in the high-frequency components in the wavelet transform images. Therefore, most of the features in this study were from the wavelet transform images and contained high-frequency components in at least one direction [Table 4]. The kurtosis of the original DW images was the most important feature of the combined model.
In this retrospective study, we found that the delay time of contrast-enhanced MRI was not the same, and related radiomic models may not accurately evaluate MI. In addition, Takeuchi et al. indicated that the combination of reduced field of view DWI and T2WI improved the evaluation of DMI. Guo et al. found that fused T2WI-DWI may be a good approach for MI depth assessment. A meta-analysis confirmed that T2WI combined with DWI had a better diagnostic performance than dynamic contrast-enhanced (DCE) MRI or DWI. Therefore, the radiomic features of T2W images and DW images were used to assess MI. It was easier to observe and delineate the uterus in sagittal T2W images than in axial or coronal images, so the former were included in the study. Most of the patients underwent axial DWI scanning with a b value of 800 s/mm2, so we included these images. Moreover, DW images showed a more obvious contrast between EC lesions and the myometrium than ADC images. The interface shown in DW images between EC lesions and the myometrium was clearer than that shown in ADC images. However, the DW images showed less interface in the lesions and in the myometrium than ADC images. The junction zone of menopausal patients is often unclear. At this time, the interface between the myometrium and the junction zone is not obvious, and this interface cannot be displayed on the DW images. Therefore, the interface information in DW images was mainly from the interface between EC lesions and the myometrium. This may contribute to the establishment of the DMI model. Thus, we did not use ADC images for radiomic research in this study.
The AUCs of the validation set based on the single-sequence models were both >0.7. This may demonstrate that single-sequence models have some predictive ability. The specificity and accuracy of the single-sequence models were weaker than those of the corresponding subjective diagnosis results, but the sensitivity was similar. On the other hand, the predictive performance of the combined model based on the combination of two sequence images was similar to that of the subjective diagnosis (P > 0.05). This finding indicates that the combined model may have the possibility of replacing subjective diagnosis. In our article, the sensitivity (0.77) and specificity (0.85) of the combined model in the diagnosis of DMI were similar to the results of Ueno et al. They reported that the sensitivity and specificity of the MRI texture feature model combining T2WI, DWI, ADC, and DCE in the diagnosis of DMI were 0.793 and 0.823, respectively. That study included 58 patients with DMI and 79 without DMI.
This study has some limitations. First, because this is a retrospective study, radiomic research on enhanced MRI was not performed. We will standardize the DCE scan and perform related research in the future. Second, only the sagittal T2W images were included because it was easier to observe and delineate the uterus on these images; we may perform radiomic research with axial and coronal T2W images in the future. Third, this single-center, retrospective study may have selection bias. We will conduct multicenter and prospective studies in the future.
| > Conclusion|| |
Whole-uterine MRI radiomic features based on sagittal T2WI and axial DWI show potential in predicting MI in EC. The predictive performance of this combined model based on the combination of the two sequence images was similar to that of subjective diagnosis. In summary, the MR radiomic features of the whole uterus have certain advantages in the clinical application of predicting the depth of MI in EC.
Financial support and sponsorship
This study was supported by the Primary Research and Development Plan of Shandong Province (Grant no. 2016GSF201095).
Conflicts of interest
There are no conflicts of interest.
| > References|| |
Chen W, Zheng R, Zhang S, Zeng H, Zuo T, Xia C, et al
. Cancer incidence and mortality in China in 2013: An analysis based on urbanization level. Chin J Cancer Res 2017;29:1-10.
Siegel RL, Miller KD, Jemal A. Cancer statistics, 2019. CA Cancer J Clin 2019;69:7-34.
Cai S, Zhang H, Chen X, Wang T, Lu J, Liu X, et al.
MR volumetry in predicting the aggressiveness of endometrioid adenocarcinoma: Correlation with final pathological results. Acta Radiol 2020;61:705-13.
Guo Y, Wang P, Wang P, Gao W, Li F, Yang X, et al
. Myometrial invasion and overall staging of endometrial carcinoma: Assessment using fusion of T2-weighted magnetic resonance imaging and diffusion-weighted magnetic resonance imaging. Onco Targets Ther 2017;10:5937-43.
Yan B, Liang X, Zhao T, Niu C, Ding C, Liu W. Preoperative prediction of deep myometrial invasion and tumor grade for stage I endometrioid adenocarcinoma: A simple method of measurement on DWI. Eur Radiol 2019;29:838-48.
Colombo N, Creutzberg C, Amant F, Bosse T, González-Martín A, Ledermann J, et al
. ESMO-ESGO-ESTRO consensus conference on endometrial cancer: Diagnosis, treatment and follow-up. Int J Gynecol Cancer 2016;26:2-30.
Rossi EC, Kowalski LD, Scalici J, Cantrell L, Schuler K, Hanna RK, et al
. A comparison of sentinel lymph node biopsy to lymphadenectomy for endometrial cancer staging (FIRES trial): A multicentre, prospective, cohort study. Lancet Oncol 2017;18:384-92.
Lee YJ, Moon MH, Sung CK, Chun YK, Lee YH. MR assessment of myometrial invasion in women with endometrial cancer: Discrepancy between T2-weighted imaging and contrast-enhanced T1-weighted imaging. Abdom Radiol (NY) 2016;41:127-35.
Koplay M, Dogan NU, Erdogan H, Sivri M, Erol C, Nayman A, et al
. Diagnostic efficacy of diffusion-weighted MRI for pre-operative assessment of myometrial and cervical invasion and pelvic lymph node metastasis in endometrial carcinoma. J Med Imaging Radiat Oncol 2014;58:538-46.
Nougaret S, Horta M, Sala E, Lakhman Y, Thomassin-Naggara I, Kido A, et al.
Endometrial cancer MRI staging: Updated guidelines of the European Society of Urogenital Radiology. Eur Radiol 2019;29:792-805.
Ueno Y, Forghani B, Forghani R, Dohan A, Zeng XZ, Chamming's F, et al.
Endometrial carcinoma: MR imaging-based texture model for preoperative risk stratification-a preliminary analysis. Radiology 2017;284:748-57.
Koo TK, Li MY. A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med 2016;15:155-63.
Luomaranta A, Leminen A, Loukovaara M. Magnetic resonance imaging in the assessment of high-risk features of endometrial carcinoma: A meta-analysis. Int J Gynecol Cancer 2015;25:837-42.
Woo S, Kim SY, Cho JY, Kim SH. Assessment of deep myometrial invasion of endometrial cancer on MRI: Added value of second-opinion interpretations by radiologists subspecialized in gynaecologic oncology. Eur Radiol 2017;27:1877-82.
Frost JA, Webster KE, Bryant A, Morrison J. Lymphadenectomy for the management of endometrial cancer. Cochrane Database Syst Rev 2017;10:CD007585.
Watson CH, Lopez-Acevedo M, Broadwater G, Kim AH, Ehrisman J, Davidson BA, et al
. A pilot study of lower extremity lymphedema, lower extremity function, and quality of life in women after minimally invasive endometrial cancer staging surgery. Gynecol Oncol 2019;153:399-404.
Tawadros AIF, Khalafalla MMM. Expression of programmed death-ligand 1 and hypoxia-inducible factor-1α proteins in endometrial carcinoma. J Cancer Res Ther 2018;14(Supplement):S1063-9.
Ytre-Hauge S, Dybvik JA, Lundervold A, Salvesen ØO, Krakstad C, Fasmer KE, et al
. Preoperative tumor texture analysis on MRI predicts high-risk disease and reduced survival in endometrial cancer. J Magn Reson Imaging 2018;48:1637-47.
Deng L, Wang QP, Yan R, Duan XY, Bai L, Yu N, et al
. The utility of measuring the apparent diffusion coefficient for peritumoral zone in assessing infiltration depth of endometrial cancer. Cancer Imaging 2018;18:23.
Xie H, Zhang X, Ma S, Liu Y, Wang X. Preoperative differentiation of uterine sarcoma from leiomyoma: Comparison of three models based on different segmentation volumes using radiomics. Mol Imaging Biol 2019;21:1157-64.
Chen Y, Chen TW, Wu CQ, Lin Q, Hu R, Xie CL, et al.
Radiomics model of contrast-enhanced computed tomography for predicting the recurrence of acute pancreatitis. Eur Radiol 2019;29:4408-17.
Tanadini-Lang S, Bogowicz M, Veit-Haibach P, Huellner M, Pauli C, Shukla V, et al.
Exploratory radiomics in computed tomography perfusion of prostate cancer. Anticancer Res 2018;38:685-90.
Takeuchi M, Matsuzaki K, Harada M. Evaluating myometrial invasion in endometrial cancer: Comparison of reduced field-of-view diffusion-weighted imaging and dynamic contrast-enhanced MR imaging. Magn Reson Med Sci 2018;17:28-34.
Deng L, Wang QP, Chen X, Duan XY, Wang W, Guo YM. The combination of diffusion- and t2-weighted imaging in predicting deep myometrial invasion of endometrial cancer: A systematic review and meta-analysis. J Comput Assist Tomogr 2015;39:661-73.
[Figure 1], [Figure 2], [Figure 3], [Figure 4]
[Table 1], [Table 2], [Table 3], [Table 4], [Table 5]