Mucin-producing tumors of the ovary——preoperative differentiation between metastatic ovarian mucinous carcinoma and primary mucinous malignant tumors

This retrospective multicenter study received approval from each institutional review board, and informed consent was obtained though an opt-out on the website. The study involved 137 patients with histopathologically proved mucin-producing ovarian tumors, diagnosed using surgical specimens from November 2014 to Jane 2023. Seventy-six patients were excluded for the following reasons:1) mucinous adenoma (n = 49); 2) lack of preoperative MRI (n = 24); 3) absence of preoperative diffusion-weighted imaging (DWI) or contrast-enhanced MRI (n = 3) (Fig. 1). The remaining 61 patients included 22 patients with POMTs [age range, 31–66; median age 52 (31, 66) years] and 39 patients with MOMCs [age range, 23–79; median age 49 (38, 56) years]. Their clinical data and preoperative MR images were evaluated.

All MRI examinations were conducted on 1.5 T or 3.0 T MR systems (Discovery MR 750, GE, Healthcare; Avanto, MAGNETOM Verio, Siemens Healthcare) using a dedicated phased-array body coil and spine coil. The conventional pelvic MRI protocol included T1-weighted imaging (T1WI) using spin-echo without fat saturation, fat-suppressed T2-weighted imaging (T2WI) with turbo spin-echo, as well as three-phase axial contrast-enhanced T1WI (30, 60, and 150 s acquisitions) using three-dimensional T1WI with fat suppression after the intravenous administration of 0.2 mmol/kg of gadopentetate dimeglumine at a rate of 2 mL/s. An axial DWI (b = 0, 800 or 1000 s/mm²) sequence was obtained before contrast-agent injection using a single-shot echo planer imaging sequence: repetition time / echo time = 3200 ms / 87 ms, section thickness = 4–5 mm with-out gap, and the number of excitations = 4. All other MR images were obtained as follows: slice thickness 3–4 mm; gap 1–2 mm; field of view 200–250 mm × 340 mm. The total scale time was 20-25 min.

Two readers (C.S.Q. and W.M.R., with 12 and 5 years of gynecological imaging experience, respectively), who were blinded to the patients’ clinical and histopathologic data, independently reviewed the conventional images on PACS (Pathspeed, GE Medical Systems Integrated Imaging Solutions, Prospect, IL). The following MRI features were evaluated: (i) tumor size: maximum diameter; (ii) configuration: cystic lesion; lesion with solid tissue, papillary formations, mural nodules, irregular cyst wall or septations; solid lesion, consists of at least 80% solid tissue with < 20% of lesion volume being cystic; (iii) number of cysts: unicystic; oligocystic, < 10 cysts; multicystic, ≥ 10 cysts; (iv) honeycomb loculi: a densely aggregated numerous loculi of 5–10 mm; (v) stained-glass appearance: different signal intensity of each cystic portion within multilocular cystic lesions on T1WI and T2WI; (vi) fluid–fluid level: appearance in which the nondependent fluid component has a different signal intensity from the dependent fluid component with horizontal delineation; (vii) shading sign: cyst fluid that is hypointense on T2WI (extent of hypointense T2 signal intensity may be homogeneous, variable within the cyst or graduated and dependent) [20]; (viii) size diversity ratio: the size ratio of the smallest/largest loculus on contrast T1WI in oligocystic or multilocular tumors; (ix) signal intensity diversity ratio: the most varying signal intensity on T2WI in oligocystic or multilocular tumors; (x) enhancement of the solid tissue (less than or equal to the myometrium, or greater than the myometrium at 30-40 s post injection) [20]; (xi) enhanced pattern (low-risk: signal intensity lower than that of myometrium at 30 s and 60 s; intermediate -risk: signal intensity lower than that of myometrium at 30 s and higher than that at 60 s; high-risk: signal intensity higher than that of myometrium at 30 s); (xii) apparent diffusion coefficient (ADC) values of the solid components; (xiii) ascites, graded as none or physiological (limited to the Douglas pouch), moderate (limited to the pelvic cavity) or massive ascites (beyond the pelvic cavity). Discrepancies on the categorical features were resolved in consensus. The mean quantitative values of the tumor size and ADC value were adopted by averaging the measurements from two radiologists.

All statistical analyses were performed using SPSS software (V26.0; IBM, Armonk, NY) and R software (V3.4.1; Boston, MA, USA). All tests were two-sided, and p < 0.05 was considered significant. Continuous variables with a normal distribution were presented as mean ± standard deviation, while non-normal variables were presented as the median (25th percentile and 75th percentile). Interobserver agreements between the two readers for imaging analysis were assessed using the interclass correlation coefficient (ICC): poor, 0–0.2; fair, 0.2–0.4; moderate, 0.4–0.6; good, 0.6–0.8; excellent, 0.8–1.0. The clinical and MRI findings between the POMTs and MOMCs groups were compared using the independent sample t-test or Mann–Whitney U test for continuous variables and Pearson’s x² test or Fisher’s exact test for categorical variables. Factors with a p value less than 0.05 in univariate analyses were entered into the multivariate models. Multivariate logistic regression analysis was performed using the forward LR elimination method to identify the independent predictors. Receiver operating characteristic (ROC) analyses of both independent and combined significant findings were conducted to assess the ability of the measurements to discriminate between the two groups. The corresponding areas under the ROC curves (AUCs), sensitivities, specificities, and 95% confidence intervals (95% CI) were calculated. A nomogram was built using the predictive model to serve as a graphical representation of the results. The nomogram’s predictive performance was measured using Harrell’s C-index and calibration with 1000 bootstrap samples to decrease the overfit bias [21]. A calibration curve was plotted to analyze the nomogram’s diagnostic performance. The Hosmer–Lemeshow test was used to assess the agreement between the nomogram and true subtype from the calibration curve. Decision curve analysis was carried out to determine the clinical usefulness of the nomogram by quantifying the net benefits under all threshold probabilities.

A total of 61 patients with mucin-producing tumors were included in this study. Twenty-two patients (8 carcinomas, 14 borderline tumors) with 22 masses were proved to have PMOTs, and 39 patients (23 colons; 8 gastric; 2 appendix; 3 cervix; 2 bile duct; 1 pancreas) with 49 masses had MOMCs. Ten (25.6%) of 39 patients with MOMCs were bilateral, and all patients with POMTs were unilateral. Fourteen (35.9%) of 39 patients with MOMCs and the primary carcinomas were discovered synchronously; 25 (64.1%) of 39 patients were metachronous. There were no significant differences in human epididymis 4 (HE4), cancer antigen 125 (CA125), and cancer antigen 199 (CA199) levels between the two groups, except for carcinoembryonic antigen (CEA), which was significantly higher in MOMCs (p = 0.002). Baseline patient’s demographic and pathologic characteristics are summarized in Tables 1 and 2.

All of the imaging characteristics showed good to excellent interobserver agreements. Univariate analysis revealed significant differences in tumor size, configuration, enhanced pattern, the number of cysts, honeycomb sign, stained-glass appearance, size diversity ratio, signal intensity diversity ratio, and ascites between the two groups. Compared with POMTs, MOMCs exhibited more common MRI characteristics: (i) smaller tumor size; (ii) solid lesion; (iii) high-risk enhanced pattern; (iv) moderate ascites; (vi) low locular size and signal intensity diversity ratio (Table 3). MOMCs were less likely to have honeycomb sign and stained-glass appearance (Figs. 2 and 3).

Multivariable analysis revealed that the size diversity ratio (p = 0.005, OR = 1.31) and signal diversity ratio (p = 0.10, OR = 4.01) were significant differentiating MRI features. The optimal cut-off ratios of size and signal diversity were 10 and 1.46, respectively, with sensitivity, specificity, and AUC of 83.0%, 77.8% and 0.856, and 85.1%, 77.8%, and 0.805, respectively (Table 4).

The nomogram that integrated the size diversity ratio and signal diversity ratio to predict POMTs is presented in Fig. 4A. Satisfactory predictive performance of the nomogram was observed, with a C-index value of 0.915 (95%CI: 0.854–0.983). The calibration curve for the nomogram’s POMTs probability showed good agreement between prediction and observation (p < 0.001; Fig. 4B). The ROC curve analysis of the model combined locular size diversity ratio and signal intensity diversity ratio yielded a largest AUC of 0.922, with a sensitivity of 82.3%, and a specificity of 88.9% for distinguishing POMTs from MOMCs (Fig. 4C). The net benefit of the decision curve was greater than the curve assuming all patients had POMTs (Fig. 4D).