Improving risk stratification of indeterminate adnexal masses on MRI: What imaging features help predict malignancy in O-RADS MRI 4 lesions?

There were 74 women with a median age of 52 years (interquartile range [IQR] 36–61). Median CA-125 level was 26 U/mL (IQR 14–66). All patients had undergone surgery and the adnexal mass was surgically removed by bilateral salpingo-oophorectomy with or without hysterectomy (n = 42 [56.8%]), unilateral salpingo-oophrectomy with or without hysterectomy (n = 19 [25.7%]), unilateral cystectomy (n = 4 [5.4%]), bilateral cystectomy (n = 2 [2.7%]), unilateral oophorectomy with salpingectomy (n = 2 [2.7%]), and unilateral salpingo-oophorectomy with contralateral salpingectomy (n = 5 [6.8%]). The median interval between MRI and surgery was 22 days (IQR 14–42). On pathology, the adnexal mass was malignant in 41 (55.4%) and benign in 33 (44.6%) adnexal masses. Most common diagnoses were borderline tumors (n = 22 [29.7%]) and cystadenofibromas (n=13 [17.6%]). Four (5.4%) patients had high grade serous carcinoma (HGSC). A detailed breakdown of the histopathology is provided in Table 1 . Patients with malignant O-RADS MRI 4 adnexal masses were significantly younger (median [IQR], 47 years [33–57] vs. 57 years [41–66], p = 0.02). CA-125 levels were significantly higher in patients with malignant O-RADS MRI 4 adnexal masses (35 U/mL [17–105] vs. 19 [12–35], p = 0.03); 39% (16/41) of patients with malignant adnexal masses had elevated CA-125 (>35 U/mL) while this was seen in 18.2% (6/33) of patients with benign adnexal masses. Table 2 shows the detailed comparison of MRI features between malignant and benign adnexal masses, and representative images are shown in Figures 2 , 3 , 4 , and 5 . In brief, for both radiologists, the overall size (long- and short-axes) was not significantly different (p = 0.33–0.77). However, the size of solid tissue was significantly greater for both long- and short-axes (p = 0.01–0.04). No difference was seen in the contour of the solid tissue (p = 0.10 and 0.94). Types of solid tissue were also significantly different between malignant lesions and benign adnexal masses (p <0.01 for both readers). Specifically, papillary projection and larger solid portion (i.e., not fitting into other categories) were more common in malignant lesions, while irregular septation and solid lesion (i.e., >80% solid tissue) were more frequently seen in benign lesions. While malignant lesions more commonly showed hyperintense or intermediate T2WI signal intensity (p = 0.01 and ≤0.01); T1WI signal intensity and heterogeneity were not significantly different between malignant and benign lesions (p = 0.12–0.46). Sixty-two patients had DWI performed as part of the MRI protocol. In these patients, malignant masses more commonly had hyperintense SI of the solid tissue (p = 0.01 and 0.03); however, the ADC values of the solid tissue (ADC solid ) were not significantly different (p = 0.11 and 0.15). In the 62 mixed solid/cystic adnexal masses (i.e., excluding adnexal masses that were >80% solid), the type of fluid (simple vs non-simple), number of locules, and ADC values of the fluid (ADC cystic ) were not significantly different (p = 0.09–0.73). When integrating the information from the type of solid tissue and CA-125 levels, patients with either papillary projections or larger solid portion and elevated CA-125 levels were malignant in 83.3% (15/18, radiologist 1) and 81.3% (13/16, radiologist 2). On the other hand, patients with irregular septations and non-elevated CA-125 levels were malignant in 22.2% (2/9, radiologist 1) and 27.3% (3/11, radiologist 2). All 4 patients with HGSC had papillary projections or larger solid portion and/or elevated CA-125. The inter-reader agreement was excellent for overall size (both long- and short- axes), size of solid tissue (long-axis), solid/overall size ratio, type of solid tissue, ADC solid , T2WI signal intensity, and T1WI signal intensity(ICC = 0.883–0.950; k = 0.765–0.860). The agreement was good for the size of solid tissue (short-axis), ADC cystic , fluid type, and the number of locules (0.662–0.683; k = 0.650–0.727). Agreement for DWI signal intensity of solid tissue was fair ( k = 0.528), and those for contour and SI heterogeneity were poor (k = 0.151–0.328). Details of inter-reader agreement are provided in Table 3 .

This single-center study was performed after receiving approval from our institutional review board. The requirement for informed consent was waived due to the retrospective study design. The conduct of the study was compliant with the Health Insurance Portability and Accountability Act. The institutional radiology database was searched to identify all radiology reports for pelvic MRIs that were done to characterize indeterminate adnexal masses from April 1, 2021, to August 16, 2022. This study period was selected as we incorporated the O-RADS MRI scoring system as part of our routine radiology reporting practice at our institution beginning in April 2021. We included all patients with an adnexal mass assigned an O-RADS MRI score of 4. Exclusion criteria were: (1) O-RADS MRI score of 4 was inappropriately assigned and (2) lack of histopathological reference standard or >2 years of follow-up to determine whether the adnexal mass was malignant or not. The flowchart for patient selection is shown in Figure 1 . MRI examinations were performed on various 1.5- and 3.0-Tesla scanners at our institution (n = 26) or at outside institutions (n = 48). Due to this, there was variability in the MRI parameters, but in general, the protocols included T1-weighted imaging with and/or without fat saturation, multiplanar T2-weighted imaging, diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC) map and multiphase contrast-enhanced sequences. The technical parameters of the pelvic MRI for characterizing adnexal masses at our institution have been published elsewhere ( 5 ). The MRIs were independently reviewed by two radiologists, BW (body oncologic imaging fellow) and PC (subspecialized gynecologic oncologic radiologist with 5 years of post-residency experience), blinded to the clinical information. The radiologists were aware that the patients had adnexal masses with an O-RADS MRI score of 4 but were otherwise blinded to clinical and pathological data at the time of image interpretation. For each adnexal mass, the following features were assessed according to recommendations from the O-RADS MRI lexicon white paper and previous literature ( 6 , 7 ): (1) overall lesion size (long-axis and short-axis diameters) in any plane, (2) size (long-axis and short-axis diameters), type (“papillary projection”, “mural nodule”, “irregular septation”, “irregular wall”, “larger solid portion” as defined as an enhancing component not fitting into prior 4 categories, or “solid lesion” as defined as a mass that consists of >80% solid tissue), signal intensity, and heterogeneity of the solid component on various MRI sequences, (3) type of fluid (simple vs non-simple) and signal intensity of any fluid component, and (4) the number of locules (1, 2, 3–10, >10). The following information was obtained from the electronic medical records: age, pre-operative serum cancer antigen 125 (CA-125) levels, type of surgery, and pathologic results from the surgical specimens. The reference standard for determining whether the adnexal mass was malignant or not was based on pathology. For the purpose of this study, borderline tumors were categorized as malignant in accordance with previous studies ( 3 , 7 ). Clinical and MRI findings were compared between malignant and benign adnexal masses using the Wilcoxon rank sum test for continuous variables and either the Pearson’s Chi-square test or Fisher’s exact test for categorical variables. Inter-reader agreement between the two radiologists was calculated using the intra-class correlation coefficients (ICC) for continuous variables and the Cohen k statistics for categorical variables and was categorized into the following: ICC and k values 0.00−0.39, poor agreement; 0.40−0.59, fair agreement; 0.60−0.74, good agreement; and 0.75−1.00, excellent agreement ( 8 , 9 ). Statistical analysis was performed using R software (version 4.3.0, Vienna, Austria). P value <0.05 was considered statistically significant.

In this retrospective review of 74 O-RADS MRI 4 lesions, we found several radiological and clinical features that discriminated malignant from benign lesions. First, morphological features on MRI provided clues for better identifying malignant adnexal masses even within O-RADS MRI 4 lesions. Not surprisingly, the size of the solid tissue was greater in malignant masses than benign masses, while the overall size of the adnexal mass was not significantly different. This is in line with recent observations focused on fat-containing adnexal masses showing that solid tissue size is a more important factor in determining adnexal malignancy ( 5 ). Nevertheless, considering the wide overlap between malignant and benign lesions, size cannot be used as a standalone criterion. Rather, important consideration may be needed to determine the type of solid tissue. Our study shows that papillary projection and larger solid portions (i.e., not fitting into other categories) were associated with malignancy while irregular septations and solid lesion (i.e., >80% solid) were associated with benign masses. This is in agreement with the known literature – papillary projections are commonly seen in borderline ovarian tumors ( 10 ); within the spectrum of cystic and solid ovarian masses, increasing solid components are known to show higher probability of malignancy ( 11 ); thickened irregular septa is commonly seen in benign cystadenofibromas ( 12 ); and benign fibromas and fibrothecomas (which are typically >80% solid) are known to show heterogeneous enhancement when degenerated ( 13 ). As such, a comprehensive investigation of the solid tissue components may be crucial for optimal risk stratification in patients with indeterminate adnexal masses. We observed that the role of DWI was mixed for differentiating between malignant and benign O-RADS MRI 4 adnexal masses. Specifically, while the solid tissue more commonly showed hyperintense signal intensity on DWI in malignant than in benign masses, there was no significant difference in the ADC values (1.321 vs. 1.424 by R1 and 1.375 vs. 1.446 [x 10 –3 mm 2 /sec] by R2, for malignant vs benign lesions, respectively). This adds to the controversy about whether DWI is helpful for differentiating benign vs malignant adnexal masses. For example, Manganaro et al ( 14 ) observed that ADC values differed between benign and malignant adnexal lesions and proposed that O-RADS 4 lesions with ADC values lower than 0.849 × 10 –3 mm 2 /sec could be upgraded to 5. On the other hand, Thomassin-Naggara et al ( 15 ) did not find significant differences in ADC values. We speculate that there are several factors that attribute to these conflicting results between our and other studies in the literature: (1) heterogeneity of histological subtypes – unlike in prostate cancer (where there is single predominant type of histology), there are numerous types of distinct histologies arising from the adnexa, adding to heterogeneity; (2) characteristics of the patient cohort – prevalence of malignancy in O-RADS MRI 4 lesions in the study by Manganaro et al ( 14 ) was extremely high (93%), deviating from the not only the benchmark estimates provided by O-RADS recommendations (50%), but also that reported in our study (55.4%) and the reported pooled prevalence according to a recent meta-analysis (60%); (3) the intermediate nature of borderline tumors – as in other studies, we included borderline tumors in the category of malignancy but it has been observed that borderline tumors often demonstrate indeterminate ADC values between that of invasive tumors and benign histologies ( 14 ); ( 4 ) technical aspects of MRI – in our study, MRI was performed in different institutions with heterogeneous technical parameters and there are also differences amongst studies with regards to vendor, scanner and technical parameters used to acquire MRI. In future studies, advanced methodologies (e.g., re-calculating ADC values using raw DWI data, normalization using reference organ, or harmonization) could be employed to potentially mitigate this issue. While further exploring the value of DWI is warranted, we additionally need to recognize that the extreme variation in reported ADC cutoff values (1.06–2.15 × 10 –3 mm 2 /sec), even in studies that do report differences between malignant and benign lesions makes it difficult if not impossible to apply it in our daily practice, today ( 16 , 17 ). Characteristics of the cystic component were not helpful for sub-stratifying O-RADS MRI 4 lesions. For instance, the type of fluid (simple vs non-simple) and the ADC value of the cystic component were not significantly different between malignant and benign lesions. On the contrary, a recent study by Assouline et al ( 7 ) found that ADC of the fluid >1.69 × 10 –3 mm 2 /sec was associated with a 5-fold greater risk of malignancy. Some of this discrepancy can be explained by the fact that their study included adnexal masses with all O-RADS MRI scores ( 1 – 5 ), and we only focused on score of 4. Many benign entities harbor non-simple fluid (e.g., hemorrhagic, endometriotic, fat) whereas many borderline and malignant tumors contain serous fluid demonstrating SI characteristics similar to simple fluid. Excluding such benign entities by limiting our analysis to O-RADS MRI 4 lesions may have diminished the role of ADC for this matter. Regarding the number of locules, both benign and malignant entities that can fit into O-RADS MRI score of 4 may harbor low to high number of locules. For example, borderline tumors and benign lesions like cystadenofibromas have been known to demonstrate various degrees of loculations ( 10 , 12 ). Most of the MRI features assessed in our study demonstrated good to excellent inter-reader agreement and provide support for using these descriptors as recommended by the O-RADS MRI lexicon white paper. However, some of the findings were notably discordant between radiologists. For example, signal intensity heterogeneity and contour showed poor agreement. This can partly be intuitively understood as the determination of such features is subjective in nature (as opposed to size measurements or categorizing solid tissue type). In addition, the experience of the radiologists may be attributed to our findings. Perhaps, additional consensus on definitions or training may be required before these low-agreement features may be considered helpful in the risk stratification of adnexal masses. We also investigated some clinical characteristics of malignant and benign O-RADS MRI 4 lesions. As expected, CA-125 levels were higher in patients with malignant adnexal lesions ( 18 ). In addition, patients with malignant adnexal masses were significantly younger than those with benign masses (median 47 years vs. 57 years, p = 0.02). Although this may be counterintuitive at first, we must understand that the results of our study are from a specific cohort of women with O-RADS MRI 4 lesions (i.e., the prevalence of about half for having malignancy) and that borderline tumors were included in the category of malignancy for analysis. It is well known that borderline tumors are typically diagnosed at least 10 years earlier than invasive tumors, and approximately a third are diagnosed before the age of 40 ( 19 , 20 ). Considering the patient population and how the outcome is determined (e.g., where borderline tumor fits ins) may be crucial for whether our results apply to other institutions. So how can we take advantage of some of these imaging and clinical features to sub-stratify O-RADS MRI 4 lesions (i.e., intermediate risk), which suffer from wide variability in the prevalence of malignancy between studies ( 4 )? Based on the results of this study, we may consider assigning adnexal masses with either papillary projections or larger solid portion and elevated CA-125 levels (e.g., >35 U/mL) to a subcategory of 4 H (intermediate “high” risk), whereas those with irregular septations and non-elevated CA-125 levels can be put into a subcategory of 4 L (intermediate “low” risk). In our cohort, this sub-stratification would have led to a prevalence of malignancy of 83.3% (15/18) and 81.3% (13/16) in 4 H adnexal masses and 22.2% (2/9), and 27.3% (3/11) in 4 L adnexal masses according to both radiologists. Although these scenarios address only about one third of all O-RADS MRI 4 lesions evaluated in this study, this may be a good starting point of the discussion to address the Achilles heel of the current version of O-RADS MRI and, upon validation may help improve risk stratification of women with sonographically indeterminate adnexal masses, ultimately helping them either avoid unnecessary surgical procedures or tailor (i.e., de-escalate) the type and extent of surgery in patients with lower risk lesions or guide those that may need a more dedicated gynecologic oncologic referral in those with higher risk lesions. Also, while 4 L did show a lower prevalence of malignancy, 22.2% and 27.3% (by each radiologist) is still not negligible, and further studies are needed to find factors that can define a subset of patients with an even lower prevalence of malignancy to truly achieve the goal of individualizing management. Our study has some limitations. First, the retrospective design introduces inherent selection bias. Second, there were a relatively small number of patients (N = 74). However, our study focused on a specific cohort of patients – that is, women with only O-RADS MRI 4 lesions, which is where there is the most critical unmet need for accurate risk stratification of indeterminate adnexal masses (i.e., the widest range of malignancy prevalence, 5–90%). Third, all patients had surgical histopathology as the reference standard. Nevertheless, most, if not all patients that have adnexal masses that fit the definition of O-RADS MRI 4 will get surgical management in clinical practice as shown also in our study. Fourth, the inclusion of MRI studies done at and outside of our institution introduced wide variation in protocol and technical parameters, and as such, it limited the feasibility of investigating more quantitative and advanced analysis of DWI and DCE-MRI. Further studies will be needed to identify the role of these functional MRI techniques better. Also, it is noteworthy that minimalistic approaches (e.g., multiphase instead of DCE-MRI using time-signal intensity curve analysis or even omitting usage of intravenous contrast media altogether) have been reported to work well in the grand scheme of things when using O-RADS MRI ( 21 , 22 ). Furthermore, this variability better reflects real-world practice where one radiologist may encounter MRIs done on different scanners and protocols when assessing adnexal masses. Fifth, only 4 (5.4%) of the patients had high grade serous carcinoma. However, this highlights the point that MRI was used as a problem solver for indeterminate adnexal mass. In clinical practice, many of the high grade serous carcinomas do not require MRI for further characterization as they generally present with advance stage of disease and CT is the mainstay imaging modality used to evaluate extent of disease. Fifth, in this study radiologists did not perform subjective assessments of whether the adnexal mass was considered malignant or benign as this was beyond the scope of our study. However, further investigations should be done to compare the diagnostic performance of subjective assessments vs O-RADS MRI and whether subjective assessment can be used to improve O-RADS MRI for example in lesions with a score of 4.

Approximately half of O-RADS MRI 4 adnexal masses were malignant. Various MRI and clinical features were different between malignant and benign masses. Sub-stratifying O-RADS MRI 4 lesions into high and low risk subgroups may be suggested using the type of solid tissue (papillary projection or larger solid portion vs irregular septation) and CA-125 levels.

Ovarian and adnexal masses are common, and ultrasound is typically the first-line imaging modality used in their assessment. Despite improvements in ultrasound technology and its capability to evaluate adnexal masses, up to a third of them remain indeterminate and require further imaging characterization, usually using magnetic resonance imaging (MRI) ( 1 ). Recently, a risk stratification system for evaluating sonographically indeterminate adnexal masses, called “Ovarian-Adnexal Reporting and Data System for Magnetic Resonance Imaging” (O-RADS MRI) has been developed by the American College of Radiology ( 2 ). This aims to help radiologists communicate the risk of malignancy of ovarian and adnexal masses to clinicians and patients in a standardized and structured way by assigning scores of 1 through 5 (where higher scores indicate a greater likelihood of malignancy) based on various MRI features (e.g., signal intensity, solid tissue enhancement), with the goal of prompt oncologic surgical management in women with potential malignancies and avoiding unnecessary surgery in those with benign lesions. Although the overall performance of O-RADS MRI for determining ovarian malignancy has been validated and shown to have good performance – for example, sensitivity of 0.93 and specificity of 0.91 in a multicenter cohort of 1340 women ( 3 ), there are certain limitations that need to be addressed. While O-RADS MRI scores of 1–3 (normal to low risk) and 5 (high risk) typically do not result in patient management dillemas, the adnexal lesions with O-RADS MRI score of 4 (intermediate risk) have been considered the Achilles heel of this system when considering the wide range in the prevalence of malignancy in such lesions. A recent meta-analysis of 12 studies showed that while the pooled prevalence of malignancy of O-RADS MRI 4 lesions is 60%, in line with estimated values at around 50% according to the O-RADS MRI recommendations, the reported prevalence for individual studies range from as low as 5% to as high as 90% ( 4 ). As such, there is a need for additional criteria to sub-stratify O-RADS 4 lesions for better risk stratification and subsequently to identify women who may require more urgent surgical management vs those who can avoid it. Our study aimed to assess various imaging features of adnexal masses that have been assigned a score of O-RADS 4, which may help further differentiate between malignant and benign lesions.