The institutional review board approved this retrospective study, and the requirement for informed consent was waived for all participants. The inclusion criteria were as follows: 1) performing MRI within preoperative 2 weeks, 2) maximal tumor diameter larger than 1 cm on MR imaging, 3) not having non-gynecological malignancy at the time of their participation in this study, 4) no chemo-radiotherapy and/or immunosuppressive therapy before MRI, 5) complete clinical data, 6) complete MRI protocols and no image quality pitfalls, 7) clinical FIGO stage I~III, and 8) being able to follow up. From November 2013 to August 2018, 62 patients with pathologically confirmed SEC (N = 37, mean age: 62.0 ± 7.2 years old) and EEC (N = 25, mean age: 53.5 ± 8.99 years old) were retrieved from the institutional picture archiving and communication system (PACS) from institution 1. A total of 54 patients (mean age: 55.3 ± 10.1 years old), consisting of 27 SEC samples and 27 EEC samples, from institution 2 between January 2018 and December 2022 were included as an external, independent test datasheet ( Figure 1 ). All included patients were pathologically confirmed by invasive surgery (laparoscopy or laparotomy) from two tertiary centers. FIGO staging, pathological types, immunohistochemical staining results, and laboratory examinations were collected through the hospital information system.

In institution 1, MRI was performed using a 1.5-T MR scanner (Magnetom Avanto, Siemens, Erlangen, Germany) with a phased-array eight-channel sensitivity encoding abdominal coil. Routine MRI protocols used for the assessment of pelvic masses included axial turbo spin-echo (TSE) T1-weighted imaging (T1WI), sagittal TSE T2-weighted imaging (T2WI), and axial/sagittal TSE fat-suppressed T2WI (fs-T2WI). Diffusion-weighted imaging (DWI) using a two-dimensional sequence of echo-planar imaging (EPI) was performed in the axial plane with a parallel acquisition technique by using b = 0, 100, and 800 s/mm². Pelvic-enhanced imaging was acquired at multiple enhancement phases in sagittal and axial planes. Dynamic contrast-enhanced T1WI gradient-echo images were obtained at 30, 60, and 90 s in the axial plane and 120 s in the sagittal plane after injection of gadobutrol (Magnevist; Bayer Schering, Berlin, Germany) at an injection flow of 2 mL/s and a dose of 1 mL/kg body weight.

In institution 2, all scans were performed on a 3.0-T MRI scanner (Siemens Magnetom Skyra, Erlangen, Germany) with a phased-array eight-channel sensitivity encoding abdominal coil. The conventional MRI protocols are the same as institution 1. DWI was performed using b = 0, 500, and 1,000 s/mm². Contrast-enhanced fat-suppressed T1WI volumetric interpolated breath-hold examination was performed at 60 s after injection of gadobutrol (gadopentetate dimeglumine injection; Beijing Beilu Pharmaceutical Co., Ltd., Beijing, China) at a dose of 1 mL/kg body weight. The detailed basic scanning parameters are summarized in the supplementary table.

All MRI characteristics were evaluated by two experienced radiologists (both with more than 10 years of experience in gynecology MR knowledge) on a PACS terminal server referring to a reading consensus. On T1WI, the hypo-, iso-, and hyperintensities were similar for the pelvic fluid, pelvic wall muscle, and subcutaneous fat signal; on T2WI, the hypo-, iso-, and hyperintensities were similar to the pelvic bone, pelvic wall muscle, and endometrium signal; on high b-value DWI images, the low-, intermediate-, and high-signal intensities were similar to the pelvic fluid, myometrium, and endometrium. Apparent diffusion coefficient (ADC) values were manually measured on a commercially available workstation on ADC map images. Signal intensity T2 ratio (SI_T2Ratio) means that the lesion’s signal intensity is divided by the uterus’s signal intensity on T2WI; contrast signal intensity ratio (SI_contrastRatio) means that the lesion’s signal intensity is divided by the uterus’s intensity on the latest phase of contrast images which is approximately 60 s after contrast injection ( Figure 2 ); lesion area ratio (Lesion_areaRatio) means that the largest lesion’s area is divided by the uterus’s area on sagittal T2WI; and lesion volume ratio (Volume_areaRatio) means that the volume of a lesion is divided by the uterus’s volume on sagittal T2WI. All lesions were calculated and measured using the ITK-Snap software by one experienced radiologist and another radiologist will review the results one by one. Once an agreement is reached, the final ROI is defined and documented. If there is a disagreement, negotiation is used to achieve consensus. This collaborative effort helps ensure that the ROI is as accurate and reliable as possible, reducing the likelihood of errors due to individual bias or oversight.

The MRI images were assessed, including 1) the lesion maximum diameter and size, 2) the signal intensity and the ratio of signal intensity, 3) myometrial invasion (less or deep myometrial invasion), and 4) the presence of node enlargement and coexistent etiologies.

The clinical characteristics included the onset age, body mass index (BMI), the serum carbohydrate antigen 125 level (CA125), and the serum human epididymal protein 4 level (HE4). The pathological characteristics included pathological types, lymphatic metastatic node (LNM), LVSI, and Ki-67 expression. All patients were clinically staged according to the 2009 FIGO staging system.

All analyses were performed using SPSS (version 23.0, IBM, Armonk, NY, USA: IBM Corp). Continuous variables were reported as medians and ranges if the data were not normally distributed and were compared with the Kruskal−Wallis test. Normally distributed variables were presented as the mean and standard deviation and were compared with the Student’s t-test. Categorical variables were reported as percentages and compared with the Mann−Whitney U test. The consistency of inter-/intraoperator measurements was also evaluated by the intraclass coefficient test. A value of p <0.05 was considered statistically significant. A predictive nomogram was constructed based on multivariable logistic analysis to quantitatively identify SEC and was evaluated by ROC and decision curve analysis (DCA) (11).

In this study, the onset age of the SEC group (62.0 ± 7.19 years) was higher than that of the EEC group (53.5 ± 8.99 years). There was a statistically significant difference (p > 0.05). The serum CA125 level was higher in the SEC group than in the EEC group, and the difference was statistically significant (123.1 ± 329.6 U/mL vs. 27.2 ± 25.9 U/mL, p = 0.014) ( Figure 3 ). The serum HE4 level was higher in the EEC group than in the SEC group with no statistically significant difference (82.4 ± 41.2 pmol/L vs. 167.9 ± 210.6 pmol/L, p = 0.342) ( Figure 4 ). The Ki-67 level in the SEC group was higher than that in the EEC group with no statistically significant difference (42.9% ± 19.8% vs. 39.0% ± 22.3%, p = 0.557). All EEC cases were low risk (grades 1 and 2) and all SEC cases were high risk (grade 3). The lesions of the SEC group more often invaded the deep myometrium compared to those of the EEC group (17/37, 45.9% vs. 3/22, 12%, p = 0.011), which represented more aggressive biological abilities. The cases with LNM and LVSI both in the SEC and EEC groups had no statistically significant differences compared to the non-LNM and non-LVSI cases (p > 0.05). In both groups, the most common coexistent etiology was myoma (six cases in the EEC group and seven cases in the SEC group). The five cases with co-occurrent endometrial polyps or hyperplasia were observed only in the SEC group. The basic characteristics of the included samples are summarized in Table 1 .

Briefly, most lesions displayed medium- to high-signal intensity on both T2WI (SEC: 34/37, 91.8% and EEC: 25/25, 100%) and DWI (SEC: 37/37, 100% and EEC: 25/25, 100%). The SEC group had an average ADC value of 0.899 ± 0.210 ×10⁻³ mm²/s, whereas the EEC group had an average of 0.857 ± 0.139 × 10⁻³ mm²/s. On contrast-enhanced T1WI, most of them showed mild enhancement at 60 s after contrast injection (SEC: 27/35, 70.3% and EEC: 22/25, 88%), which presented a rapid flush-out effect. The conventional MRI characteristics, i.e., the signal intensity of T2WI, DWI, and contrast-enhanced T1WI and the ADC value, were not statistically significantly different between the two groups (p > 0.05, Table 2 ).

In the quantitative MRI characteristics, the values of SI_T2Ratio, SI_contrastRatio, Lesion_areaRatio, and Volume_areaRatio in the SEC group were 1.35 ± 0.36, 0.77 ± 0.18, 0.25 ± 0.24, and 0.22 ± 0.26, respectively. The SEC group had significantly greater SI_T2Ratio, SI_contrastRatio, and Volume_areaRatio values compared to the EEC group with statistically significant differences (p = 0.026, 0.001, and 0.045, respectively). The highest discriminative index for distinguishing SEC cases from EEC cases was SI_contrastRatio (AUC: 0.7533; 95% CI: 0.627–0.878). Taking the cutoff SI_contrastRatio value of 0.635 for SEC identification, the sensitivity and specificity were 0.8 and 0.6, respectively. The intraclass correlation coefficients for SI_T2Ratio, SI_contrastRatio, Lesion_areaRatio, and Volume_areaRatio were 0.912, 0.866, 0.957, and 0.791, respectively. For interoperator measurements, the consistency values were 0.924, 0.804, 0.983, and 0.851, respectively.

The multilogistic regression test was used with the following 12 clinical and MRI features: onset age, BMI, CA125, HE4, Ki-67, LNM, LVSI, SI_T2Ratio, SI_contrastRatio, Lesion_areaRatio, Volume_areaRatio, and ADC value. The onset age, CA125, SI_T2Ratio, and SI_contrastRatio were statistically significantly different between the SEC group and the EEC group in the training set, which were selected and used to develop a nomogram predictive model for SEC patients ( Figure 5 ). The nomogram achieved an AUC of 0.814 (95% CI: 0.614–0.968), a sensitivity of 1.0, a specificity of 0.60, a positive predictive value of 0.44, and a negative predictive value of 1.0 in the external testing set. The DCAs of the nomogram for the identification of SEC patients in the testing set demonstrated a net benefit when the threshold was in the range between 0.1 and 0.7 ( Figure 6 ).