Participants presenting with suspected ovarian tumors on ultrasound in our hospital were enrolled between January 2018 and September 2022. Subsequently, a total of 164 patients with pathologically confirmed ovarian malignancies, encompassing both primary and metastatic tumors, were selected for the study. All of the imaging data were stored in the picture archiving and communication system (PACS). The interval between operation and examination (ultrasound and MRI) of these patients did not exceed 120 days. Moreover, the patients had no previous history of ovarian cancer. The exclusion criteria were as follows: 1) the ovarian masses were subclassified correctly by the ADNEX model; 2) no enhanced MRI was performed before surgery.

Clinical information, including Cancer Antigen 125 (CA125), Cancer Antigen 199(CA199), Carcinoembryonic Antigen (CEA), and human epididymis protein 4 (HE-4) test results, with different histopathologic types, were analyzed if detected. In this study, patients whose ovarian tumors was mis-subclassified by the ADNEX model were categorized into two groups: primary ovarian cancer (including both I and II–IV stages) and metastatic ovarian tumor. Their age, menopausal status, and relevant tumor markers were presented separately, and above data between the two groups were compared. All assessments were performed prior to surgery and chemotherapy.

All 164 patients presented with at least one adnexal mass and subsequently underwent transvaginal and transabdominal ultrasonography. The imaging was conducted using EPIQ5 ultrasound machines (Philips Health Systems, Bothell, WA, USA) and Vivid E95 ultrasound machines (GE Healthcare), equipped with a 7.0 MHz–9.0 MHz transvaginal probe and a 3.5-MHz transabdominal probe.

Experienced ultrasonographers preoperatively assessed sonographic tumor morphology according to the IOTA consensus about the terms, definitions, and measurements used to describe the ultrasound features of adnexal tumors in 2000 (21). Multiangle scans were performed to obtain more information about the masses from ultrasound images. For bilateral ovarian masses, the mass with the most complex ultrasound features was included to the ADNEX model. If two masses had similar ultrasound morphologies, the largest mass or the one most easily accessible by ultrasonography was included (13, 21).

We input the variables needed into the ADNEX model paid for from the Apple store. The model includes nine variables: age (years), serum CA125 level (U/mL), type of center (oncology referral center vs. non-oncology center), maximal diameter of the lesion (mm), maximal diameter of the largest solid part (mm), number of papillary projections (0, 1, 2, 3, or more than 3), number of cysts locules (≤10 vs. >10), acoustic shadows (yes or no), and ascites (yes or no) (12). All ADNEX model parameters were logged objectively. In the final diagnosis, the masses were divided into five types: benign ovarian tumors, BOTs, stage I OC, stage II–IV OC, and metastatic cancer to the adnexa.

MRI was performed using a 3.0 Tesla (T) MR superconductor unit (TwinSpeed, GE Medical Systems, Milwaukee, WI, USA). A pelvic phased-array coil was employed for all cases. The following unenhanced sequences were acquired: axial T1-weighted imaging (T1WI) with a time of repetition (TR) of 340 ms and a time of echo (TE) of 10 ms; axial fast spin echo (FSE) T2-weighted imaging (T2WI) with fat saturation, using a TR/TE of 8,000 ms/83 ms; sagittal FSE T2WI with a TR/TE of 8,000 ms/98 ms. An axial DWI was performed with a b value of 1,000 s/mm², utilizing a TR/TE of 3350 ms/67.6 ms. Apparent diffusion coefficient (ADC) maps were automatically generated. Contrast-enhanced T1WI LAVA 2D with fat saturation was conducted in the axial, sagittal, and coronal planes following the injection of gadopentetate dimeglumine (Gd-DTPA, 0.1 mmol/kg of body weight, Magnevist; Bayer Schering, Guangzhou, China) at a rate of 2 mL–3 mL.

MRI features of the tumors were assessed including the following. 1) Tumor shape and tumor size. Maximum diameter of the mass was measured. 2) Tumor T1WI, T2WI, and DWI signal intensity on MR image. The masses were rated as hypointense (lower signal than outer myometrium), isointense (similar signal to outer myometrium), or hyperintense (higher signal than outer myometrium). 3) ADC value. On ADC maps, a circular region of interest (ROI) of at least 1 cm² was placed at targeted areas in the solid components of tumor, by referring to conventional MR images. 4) Patterns of enhancement. The enhancement degree was rated as slight (weaker than muscles), moderate (between muscles and outer myometrium), or intensive (more obvious than outer myometrium). 5) Presence of necrosis or hemorrhage. 6) Composition of the mass. The composition of the mass was rated as solid (mass consists of at least 80% solid tissue), solid-cystic (solid tissue of mass was rated as larger solid portion), and multilocular-cystic (including mass without solid tissue, and solid tissue of mass was rated as papillary projection, mural nodule, irregular septation, and irregular wall). The composition of the mass was according to the O-RADS™ MR Lexicon Categories, Terms and Definitions (22), Table 1 for details. 7) The diagnostic accuracy based on MRI. The MR images of 51 patients were rated by two radiologists.

Radiologists used the Ovarian-Adnexal Reporting and Data System (O-RADS) MRI Risk Stratification System (4) to reassess misclassified malignant ovarian tumors by the ADNEX model. O-RADS MRI 1: No detectable pelvic mass. O-RADS MRI 2: Purely cystic mass, purely endometriotic mass, purely fatty mass, or absence of wall enhancement. O-RADS MRI 3: Absence of solid tissue. O-RADS MRI 4: Lesion with solid tissue enhancing ≤myometrium at 30 s–40 s on non-DCE MRI, with lipid content. Large-volume solid tissue enhancing lesion. O-RADS MRI 5: Lesion with solid tissue enhancing >myometrium at 30 s–40 s on non-DCE MRI, peritoneal, mesenteric, or omental nodularity or irregular thickening with or without ascites.

The histopathological diagnosis of the tumors after surgical removal by laparoscopy or laparotomy was used as a reference standard. Tumors were staged according to the World Health Organization (WHO) classification of tumors (23), and malignant tumors were staged using the International Federation of Obstetrics and Gynecology (FIGO) standards (24).

Statistical analysis was performed using SPSS version 22.0 (IBM Corp, Los Angeles, CA, USA) and MedCalc version 15.2.2 (MedCalc Software, Mariakerke, Belgium) software. The MRI characteristics of the tumors, patients’ clinical features and tumor marker levels were compared using the chi-square test for categorical data and the Mann–Whitney U-test for continuous data. Statistical significance was assumed at P < 0.05 for all comparisons. The kappa coefficients were calculated to assess the interobserver agreement between the two radiologists for imaging parameter analysis. Kappa values of 0.00–0.20, 0.21–0.40, 0.41–0.60, 0.61–0.80, and 0.81–1.00 indicated slight, fair, moderate, substantial, and almost perfect agreement, respectively.

Among the 51 patients, 27 were confirmed to suffer stage I primary ovarian cancer. Within this group, the ADNEX model misdiagnosed four cases as benign tumors, 10 as BOTs, 12 as stage II–IV OC, and 1 as metastatic tumor. Additionally, three cases were confirmed as stage II–IV OC, with one case misdiagnosed as benign tumor and two as BOTs by the ADNEX model. Furthermore, 21 cases were confirmed as metastatic ovarian tumors, among which 8 were misdiagnosed as benign tumors, 6 as BOT, 1 as stage I OC, and 6 as stage II–IV OC.

Among the 51 patients, 30 were diagnosed with primary ovarian cancer, including 5 with high-grade serous ovarian carcinoma (HGSOC), 14 with clear cell carcinoma (CCC), 2 with endometrioid carcinoma (EC), 4 with mucinous carcinoma (MC), 2 with granulosa cell tumor (GCT), 1 with yolk sac tumor (YST), 1 with immature teratoma, and 1 with dysgerminoma. Additionally, 21 patients had metastatic ovarian tumors, with 10 having colorectal cancer, 4 having gastric cancer, 2 having uterine cervical cancer, 3 having endometrial cancer, 1 having breast cancer, and 1 having a low-grade appendiceal mucinous neoplasm (LAMN). Table 2 shows the histopathological types of ovarian malignant tumors misclassified as benign or borderline tumors by the ADNEX model, whereas Table 3 displays histopathological types of stage I ovarian cancers misclassified as more advanced stages and metastatic tumors.

Furthermore, we summarized the pathological classifications of 100 cases that were accurately diagnosed by the model. Among the 100 patients, 91 were diagnosed with primary ovarian cancer, including 83 with HGSOC, 3 with CCC, 3 with EC, 1 with low-grade serous ovarian carcinoma (LGSOC), and 1 with small cell carcinoma of the ovary-pulmonary type (SCCOPT). Additionally, 9 patients had metastatic ovarian tumors, with 4 originating from colorectal cancer, 3 from gastric cancer, 1 from endometrial cancer, and 1 from high-grade appendiceal adenocarcinoma.

All the patients’ relevant clinical indicators are presented in Table 4 . Among the 51 patients mis-subclassified by the ADNEX model, no significant differences were found between the primary and metastatic groups regarding the aforementioned data, except for CEA (P=0.013).

For all MR imaging variables, the interobserver agreement was good (kappa=0.87–0.93; Table 5 ).

MRI findings of 51 malignant ovarian tumors mis-subclassified by the ADNEX model are presented in Supplementary Table 1. The mean diameter of the primary malignant ovarian tumors was 10.29 cm (range: 3.61 cm–26.02 cm), and the metastatic ovarian tumors was 8.58 cm (range: 3.10 cm–20.30 cm). In this study, 42 masses were lobulated (82.35%,42/51), A total of 26 masses were solid-cystic (26/51, 50.98%), with 18 solid masses and 7 multilocular-cystic masses. There was a significant difference between ovarian CCC and other histopathologic subtype tumors with mean ADC values of 1.01×10⁻³ mm²/s (range: 0.68×10⁻³ mm²/s-1.28×10⁻³ mm²/s) and 0.74×10⁻³ mm²/s (range: 0.48×10⁻³ mm²/s-0.99×10⁻³ mm²/s), respectively (P=0.000). A total of 50 masses presented isointense T1, hyperintense T2, and hyperintense DWI signal intensity on MR image (50/51, 98.04%), 33 masses were of intensive enhancement degree (33/51, 64.71%), 16 were moderate, and 2 were of slight enhancement degree. There were 17 masses that had necrosis (17/51, 33.33%), with the majority being HGSOC, ovarian metastases from colorectal and gastric cancers (12/17, 70.59%). There were 19 masses that presented hemorrhage (19/51, 37.25%), with the majority being ovarian CCC (10/19, 52.63%). A total of 46 masses were diagnosed correctly by the radiologists, and the subject diagnostic accuracy was 90.20%. The five cases of MRI diagnostic errors included one case of stage I HGSOC, one case of stage I CCC, and three cases of metastatic ovarian tumors (colorectal cancer, uterine cervical cancer, and LAMN). A total of 35 masses were rated as O-RADS score 5, 15 masses were rated as O-RADS score 4, and 1 mass was rated as O-RADS score 3.

There were more than half masses mis-subclassified as benign or borderline among the 51 cases (30/51,58.82%) by the ADNEX model. The primary factor contributing to these errors can be attributed to an insufficient evaluation of the solid components. However, in MRI, most of masses mis-subclassified as benign or borderline by the ADNEX model have been accurately diagnosed as malignant ovarian tumor (25/30, 83.33%). Most of the 30 masses presented solid and solid-cystic masses with hyperintense DWI signal and intensive enhancement of solid portion on the MR image. A majority of these malignant masses displayed obvious low ADC values (22/30, 73.33%), consistent with previous studies (25, 26). The O-RADS scoring system, based on MRI, can be instrumental in distinguishing malignant ovarian tumors from benign and borderline ovarian tumors. Among the 30 cases in this study group, a significant majority (21/30, 70.00%) received an O-RADS score of 5, suggesting a malignancy risk between 50% and 90%.

When the solid component is too small, it is more likely for the ADNEX model to predict the mass as a benign or borderline tumor. MRI has a higher capability to identify the solid constituents of masses that were not detected by ultrasound, particularly in cases when some special pathological types of ovarian tumors always present multi-cystic mass without apparent large solid portion such as mucinous neoplasms of the ovary (27) ( Figure 2 ).

From 12 cases in this group, we found that over half of these masses were type I epithelial ovarian cancers (EOC), consisting of four CCC, two EC, and one MC (7/12,58.33%). Concurrently, our study revealed that the ADNEX model demonstrated an accuracy of 0.94 (83/88) for HGSOC (the most common types of type II EOC), whereas its accuracy for CCC was markedly lower at 0.17 (3/17).

In comparison with type II EOC, type I EOC tended to exhibit a relatively indolent clinical course (28). Type I EOCs are usually detected in their early stages. In our research, type I EOC showed lobulated, large, solid-cystic masses with a large proportion of solid components, which was one of the variables input into the ADNEX model, potentially leading to the sub-classifications of the ADNEX model as stage II–IV OC.

Compared with other types of malignant ovarian tumors, CCC exhibited a higher ADC value, consistent with findings in previous studies (29, 30) ( Figure 3 ). Additionally, the typical features of ovarian CCC and EC were hemorrhage signals found in these masses. It may be due to the abovementioned two types of ovarian cancer are highly associated with endometriosis (31, 32). Radiologists can observe the characteristic hemorrhage signal inside the mass on MRI and assist in diagnosis. According to another study (33), the utilization of morphological characteristics observed on MRI, such as a round mural nodule exhibiting a high “Height-to-Width ratio” and a focal growth pattern, proves to be valuable in differentiating CCC from EC.

Besides the abovementioned types of ovarian tumors, this group also included four cases HGSOC and one YST. MRI provides a more comprehensive evaluation of organs in pelvis and improve the diagnostic accuracy of HGSOC. In the case of the 26-year-old patient diagnosed with stage I YST, characterized by markedly elevated AFP levels, the ovarian mass exhibited significant enhancement along with multiple signal voids, indicative of its hypervascular nature. These findings are consistent with those reported in prior studies (34, 35).

Preoperative differentiation between ovarian cancer, particularly stages II–IV, and secondary cancers of the adnexa remains a challenge, even with the use of the ADNEX model. Research conducted in oncology centers in China and Brazil (14, 15) demonstrated that the AUC for distinguishing stage I ovarian OC from metastasis was 0.81 and 0.64, respectively, whereas the AUC for stage II–IV OC versus metastasis was 0.78 and 0.89. In this study, seven instances of metastatic ovarian cancer were incorrectly classified as primary ovarian cancer by the ADNEX model. Further analysis uncovered three cases of gastric cancer metastasis, two colorectal, one endometrial, and one cervical cancer metastasis among the misclassified cases. The imaging of the gastric cancer metastases revealed solid masses with intense enhancement, necrotic regions, and clear separation from adjacent structures ( Figure 4 ). MRI can provide a superior assessment of the neighboring organs for the latter three tumor types, encompassing the uterine cervix, endometrium, and sigmoid colon. Additionally, the study reported a misclassification of a stage I dysgerminoma as metastatic by the ADNEX model. This mass revealed characteristic fibrovascular septa on MRI, findings that are in agreement with those reported in the study (36).

A previous study (37) has indicated that patients with metastatic ovarian tumors tend to be younger and present with lower levels of CA125 and HE-4 compared with those with primary ovarian tumors. However, in the current study, a significant difference in CEA levels was observed among the 51 patients who were mis-subclassified by the ADNEX model. The absence of significant distinctions in age and CA125, both variables integrated into the model, could potentially account for the model’s inaccuracies in sub-classification.

When ultrasound imaging fails to qualitatively diagnose an ovarian mass, MRI offers the advantage of providing further assessment for “indeterminate adnexal masses at ultrasound”. The MRI characteristics of the mass contribute to the identification of specific ovarian tumor types and offer a comprehensive assessment of ovarian masses and adjacent organ involvement. However, MRI scanning is relatively slow, incurs high costs, and requires considerable time for scheduling. Additionally, MRI tends to produce motion artifacts and is not superior to contrast-enhanced computed tomography in detecting peritoneal metastasis and ascites (38).

There are several limitations in the study. Firstly, the lack of a control study design indicates a need for further research with a prospective design. Secondly, the small sample size could have impacted the results.

In conclusion, our study showed that DWI signals, ADC values, enhancement levels of the solid portion, and characteristic components within the mass on MRI images can provide more supplementary information for malignant ovarian tumors mis-classified by the ADNEX model. We hope this effort will contribute to enhancing the preoperative diagnostic accuracy for patients with malignant ovarian tumors.