Ultrasonography (IOTA-ADNEX) vs magnetic resonance imaging (ADNEX MR) for diagnosing ovarian cancer in high-risk patients: a prospective study.

This was a prospective, single‐center study of consecutively enrolled patients conducted from May 2021 to December 2022 at Aarhus University Hospital, Aarhus, Denmark, a tertiary referral center for gynecological oncology. All patients in the Central Denmark Region with suspected adnexal malignancy on first‐line ultrasound are referred to this center. The study was a systematic head‐to‐head comparison of the IOTA‐ADNEX model and the ADNEX MR scoring system that included all patients referred for second‐line imaging owing to suspected adnexal malignancy on first‐line ultrasound. The results of the first‐line ultrasound examinations performed at the referring hospitals are described separately in a multicenter prospective study by our group 44 . The present study adheres to the Standards for Reporting Diagnostic Accuracy Studies (STARD) guidelines for reporting diagnostic accuracy studies and the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement for reporting the development and validation of risk‐prediction models (Appendices  S1 and S2 ) 45 , 46 . The protocol was approved by the Central Denmark Region Ethical Committee as a quality improvement initiative (approval number: 1‐10‐72‐1‐21), and the study was registered in the internal registry of the Central Denmark Region and on ClinicalTrials.gov ( NCT05842629 ). All patients provided written informed consent to participate. The study was conducted in accordance with the Declaration of Helsinki 47 and the European Union General Data Protection Regulation (GDPR). Consecutive patients were referred to the onco‐gynecological department at Aarhus University Hospital for further evaluation based on the results of a non‐expert ultrasound examination performed in a general gynecological setting. Clinical suspicion of adnexal malignancy on first‐line examination by non‐expert ultrasound examiners was based on a Risk of Malignancy Index (RMI) ≥ 200 or, after consultation with the onco‐gynecological department, on the suspicion of malignancy according to the ADNEX model (> 5% risk) or subjective ultrasound assessment. The inclusion criterion was referral for suspected new adnexal malignancy. Exclusion criteria were not providing informed consent, failure to attend and complete both examinations, previous or concurrent pelvic malignancy and a pre‐existing diagnosis of malignant disease unrelated to the adnexal mass under investigation (except basal cell carcinoma of the skin), as well as concurrent pregnancy and age < 18 years. The second‐line ultrasound examination and MRI scan were performed on the same day, prior to surgery, biopsy or surgical sampling. An IOTA‐certified medical doctor (I.M.D.H.) with a special interest in gynecological ultrasound performed the ultrasound examinations using a Voluson E10 ultrasound machine (GE Healthcare, Zipf, Austria) with a transvaginal probe (4–9 MHz) and a transabdominal probe (2–8 MHz). At least 10 representative videoclips, with and without Doppler, together with three‐dimensional (3D) volumes of the uterus and adnexa, as well as pelvic and abdominal overview images, were systematically acquired using transvaginal and transabdominal ultrasonography and saved in Astraia software (Astraia GmbH, Munich, Germany). Ultrasound findings were reported using a form for systematic classification of ultrasound features according to the IOTA terminology 23 , 48 (Appendix  S3 ) and were integrated and saved in the patient file. Additionally, the IOTA‐ADNEX risk estimates for probabilities of benign, borderline, Stage‐I cancer, Stage‐II–IV cancer and secondary metastatic tumors were calculated automatically in the form. Previous studies have shown good interobserver agreement for most variables included in the ADNEX model 49 , 50 . Within 1 week after imaging, the recorded videoclips, images and 3D volumes were reviewed by an IOTA‐certified ultrasound examiner (M.D.) with European Federation of Societies for Ultrasound in Medicine and Biology (EFSUMB) Level‐III expertise 51 , who confirmed the original ultrasound assessments without modification. All ultrasound examinations and assessments were performed prospectively, with examiners (I.M.D.H., M.D.) blinded to the results of other imaging modalities and to the reference standard (tumor outcome). Details of the IOTA‐ADNEX model, along with IOTA‐ADNEX formulae, are provided in Appendix  S4 . MRI was carried out using a 1.5‐Tesla (T) or 3‐T scanner (Siemens Healthineers, Erlangen, Germany), with a Body 18 coil. Patients were scanned with an empty bladder, and 1 mL of glucagon was administered intramuscularly prior to initiation of the scan to reduce artifacts caused by intestinal movements. The MRI protocol included T1‐weighted and T2‐weighted sequences, as well as axial DWI with a 6‐mm slice thickness and a b‐value of 1000 s/mm 2 , with calculation of corresponding apparent diffusion coefficient (ADC) values. T2‐weighted sequences (slice thickness, 4–5 mm) were acquired in axial and sagittal planes. An axial 3D T1‐weighted Dixon VIBE sequence was used to scan the small pelvis and adnexal masses. The upper abdomen was included in the scan when the adnexal mass extended outside the small pelvis. No gadolinium‐based contrast agents were administered, in accordance with the protocols of the local department of radiology owing to concerns for patient safety and potential side effects of gadolinium, time and resource limitations and the center's experience of similar diagnostic performance using contrast‐ and non‐contrast MRI. Moreover, this represents a feasible, non‐invasive diagnostic MRI technique. This study employed diffusion‐weighted MRI with the ADNEX MR scoring system, without administration of contrast. In accordance with a previously reported model 30 , true diffusion restriction was defined qualitatively as high signal intensity on high b‐value DWI images and low signal intensity on an ADC map. Signal intensity of the solid tissue on T2‐weighted MRI was defined relative to skeletal muscles. Two radiologists (E.M., B.B.) specializing in pelvic MRI prospectively and independently described the MRI findings systematically. Any discrepancies between their assessments were resolved by adjudication. A standardized form (Appendix  S5 ) was used, in which morphological features and signal intensities, subjective assessment and scoring of the adnexal lesions using the ADNEX MR model (Appendix  S6 ) were recorded and used to estimate risk categories analogous to IOTA‐ADNEX risk estimates. The evaluation was performed blinded to the results of the other diagnostic tests, and, by design, to the tumor outcome (reference standard). Serum levels of CA 125 (analyzed using electrochemiluminescence immunoassay, reported in kU/L) were measured routinely at the time of diagnosis of the adnexal mass (within 14 days of second‐line imaging, before surgery). The reference standard was a histopathological diagnosis of the primary adnexal mass following surgical excision, biopsy or cytology; the latter served as a sufficient reference standard only when it confirmed malignancy originating from the female genitalia. Twenty‐four patients diagnosed with a benign tumor on imaging did not undergo surgery; these patients were managed expectantly owing to a low risk of malignancy determined on both expert ultrasound and expert MRI, together with being asymptomatic and/or the presence of patient comorbidities. The patients who were managed expectantly were followed up with ultrasound and clinical examinations for at least 1 year from the completion of imaging; the tumors of these patients were considered benign if no progression in imaging features or other signs of malignancy were observed within 1 year of follow‐up. All histopathological specimens underwent centralized histopathological review by pathologists specialized in gynecological histopathology. Histopathological outcomes were classified according to the World Health Organization criteria 52 , and the stage of malignancy for primary adnexal malignancy was determined according to the International Federation of Gynecology and Obstetrics (FIGO) classification 53 when surgery was performed without prior chemotherapy. Otherwise, the stage of malignancy was determined by onco‐gynecologists based on available imaging and histopathological results. When more than one histopathological diagnosis was made for the same mass, or if several masses were present, the reference standard was based on the worst tumor outcome in terms of patient prognosis (i.e. first invasive malignancy prioritized if present, followed by borderline ovarian tumors, followed by benign histopathological diagnoses). If several benign lesions were present, the primary/top diagnosis recorded in the histopathology report and electronic medical journal (i.e. A‐diagnosis) was used. The reference standard (tumor outcome) was determined blinded to the results of the ultrasound and MRI examinations (index tests). Ultrasound and MRI data were recorded prospectively. Data collection and management were handled in accordance with the GDPR using Research Electronic Data Capture (REDCap; https://project‐redcap.org/ ), a secure web application for building and managing online surveys and databases 54 , 55 . Index test data were retrieved from the patient files and transferred to the secure REDCap database along with data on patient symptoms, surgeries, treatment decisions, histopathology and follow‐up data. The sample size calculation is described in Appendix  S7 . The study was powered prospectively to detect a difference in specificity using a McNemar framework, under the hypothesis that MRI would demonstrate increased specificity for detecting malignancy at a fixed sensitivity. A predefined statistical analysis plan was followed. For the primary analysis, the reference standard for each mass was dichotomized as benign or malignant. Malignant lesions were defined as primary invasive adnexal malignancy, primary ovarian borderline tumors or metastases to the adnexa, in accordance with the protocols of the IOTA‐ADNEX and ADNEX MR development studies (Appendices  S4 and S6 ) as well as current international consensus recommendations for the management of ovarian malignancy 56 , 57 . To construct 2 × 2 contingency tables, true‐positive, false‐positive, true‐negative and false‐negative findings were determined by dichotomizing ADNEX MR scores at ≥ 2, ≥ 3, ≥ 4 or = 5, as defined in the original model development study. Likewise, IOTA‐ADNEX risk estimates were dichotomized at ≥ 10%, ≥ 20%, ≥ 30%, ≥ 40%, ≥ 50% or ≥ 75%, with values at or above each threshold classified as malignant. Diagnostic accuracy, sensitivity, specificity, positive predictive value and negative predictive value were calculated for each of the abovementioned thresholds. Receiver‐operating‐characteristics (ROC) curves were constructed to determine the performance of the IOTA‐ADNEX model with or without CA 125, the ADNEX MR model and CA 125 alone for predicting malignancy. For the ordinal ADNEX MR score (categories 2–5), the categories were plotted as categorical points of sensitivity as a function of 1 – specificity on the graph. The area under the ROC curve (AUC) was calculated with 95% CI. Pairwise comparisons of AUCs were performed using DeLong's test 58 . For the IOTA‐ADNEX model, calibration curves were constructed using locally estimated scatterplot smoothing (LOESS), and observed‐to‐expected ratios were calculated 59 , 60 . Misclassified cases were analyzed descriptively. There were no missing data for the 342 included patients. Agreement between ultrasound and MRI was assessed using Cohen's kappa statistics for nominal variables and weighted kappa for ordinal variables. The Landis and Koch scale was followed for interpreting kappa statistics (poor, < 0; slight, 0–0.20; fair, 0.21–0.40; moderate, 0.41–0.60; substantial, 0.61–0.80; almost perfect, 0.81–1.00) 61 . To assess agreement for continuous variables (i.e. lesion size and size of the largest solid component), Bland–Altman plots were computed to detect systematic bias, and the intraclass correlation coefficient (ICC) was used to evaluate consistency between the imaging modalities. ICC was interpreted according to Koo and Li (poor,  0.75–0.90; excellent, > 0.90) 62 . A secondary analysis was performed for the combined use of imaging methods, in which an ADNEX MR score of ≥ 4 or an IOTA‐ADNEX (without CA 125) risk estimate of ≥ 10%, ≥ 20%, ≥ 30%, ≥ 40% or ≥ 50% indicated a positive index test. In a prespecified subgroup analysis, AUCs were evaluated in subgroups defined by sonographic lesion characteristics (solid (≥ 80% solid), yes or no; size of solid component ≥ 2 cm, yes or no; largest diameter ≥ 10 cm, yes or no). Lastly, a sensitivity analysis of the diagnostic performance of IOTA‐ADNEX and ADNEX MR models, excluding the patients managed expectantly, was performed. Statistical analysis was performed using Stata version 17.0 (StataCorp LLC, College Station, TX, USA) and R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria).

A total of 380 patients were referred to Aarhus University Hospital with suspected adnexal malignancy from May 2021 to December 2022 (Figure  1 ). Twelve patients who were unable to attend the examinations owing to logistical constraints or comorbidities or who did not provide consent were excluded. Therefore, 368 patients were eligible for inclusion. Subsequently, 25 (6.8%) patients were excluded because they were unable to complete the systematic MRI scan owing to anxiety ( n  = 9), a pacemaker ( n  = 6), pain ( n  = 3), abdominal circumference ( n  = 1) or other reasons ( n  = 6). We excluded one further patient with only extra‐adnexal carcinomatosis, in whom an adnexal mass was not visualized on either of the second‐line imaging assessments. The final study cohort consisted of 342 women who completed both a systematic MRI scan and an expert ultrasound examination. Patients were referred from secondary referral hospitals in general gynecological settings ( n  = 265), from non‐gynecological hospital departments ( n  = 37), private gynecological clinics ( n  = 21) or following direct contact from a primary care physician ( n  = 19). Tumor outcome was obtained by surgical excision in 252 patients, biopsy in 54 patients, cytology in 12 patients and through follow‐up confirming a benign outcome with no progression in 24 (7.0%) patients who were managed expectantly. Flowchart showing inclusion of participants referred for second‐line imaging after detection of adnexal mass suspicious for malignancy on first‐line ultrasound who underwent both ultrasound and magnetic resonance imaging (MRI) examinations for second‐line assessment. *Tumor outcome was obtained by follow‐up confirming a benign outcome with no progression in 24 patients who were managed expectantly. ADNEX, Assessment of Different NEoplasias in the adneXa; ADNEX MR, MRI scoring system for adnexal masses; DWI, diffusion‐weighted imaging; IOTA, International Ovarian Tumor Analysis. The histopathological outcome of the primary adnexal lesion in each patient and the FIGO stage of the malignancies are shown in Table  1 and Table  S1 , respectively. Malignancy was diagnosed in 201 (58.8%) patients, of whom 30 had a borderline ovarian tumor and 14 had metastatic malignancy. Table  2 presents an overview of patient characteristics, IOTA‐ADNEX ultrasound features and ADNEX MR classification according to the reference standard. The median age was 65.6 (interquartile range, 53.4–74.1) years, and most patients were postmenopausal (267/342 (78.1%)). Histopathological outcome (final diagnosis) of primary adnexal lesion in each patient referred for second‐line imaging owing to suspected adnexal malignancy Data are presented as n (%). Followed‐up for at least 1 year with sequential clinical and imaging examinations with no signs of malignancy. Including schwannoma, bizarre leiomyoma and endometriosis, and three cases of uterine leiomyoma. One case also demonstrated ectopic adrenal tissue. Two cases were low grade serous adenocarcinoma. incl., including; STIC, serous tubal intraepithelial carcinoma. Patient characteristics and ultrasound and magnetic resonance imaging (MRI) features of primary adnexal lesion on second‐line imaging, overall and according to histopathological outcome (final diagnosis) Data are given as median (interquartile range), n (%) or n / N (%). Calculated for 297 patients with present solid component. The diagnostic performance of IOTA‐ADNEX (with and without CA 125), ADNEX MR and CA 125 alone is compared in Figure  2 , with ADNEX MR score cut‐offs shown as categorical points. The AUC of IOTA‐ADNEX (with or without CA 125) for distinguishing patients with benign tumors from those with malignant tumors was 0.96 (95% CI, 0.94–0.98), compared with 0.93 (95% CI, 0.91–0.96) for ADNEX MR. The difference in AUC between IOTA‐ADNEX without CA 125 and ADNEX MR was 0.023 (95% CI, –0.002 to 0.048). CA 125 alone had an AUC of 0.79 (95% CI, 0.74–0.84), while IOTA‐ADNEX with or without CA 125 had similar AUCs. The results were consistent in the sensitivity analysis which excluded the 24 patients managed expectantly ( Table S2 ). Receiver‐operating‐characteristic (ROC) curves of second‐line imaging modalities for differentiating between benign and malignant adnexal lesions in patients deemed at high risk of adnexal malignancy on first‐line imaging. ROC curves represent prediction of malignancy using IOTA‐ADNEX with serum CA 125 ( ) and IOTA‐ADNEX without CA 125 ( ) models on ultrasound, categorical ADNEX MR scoring system ( ) on magnetic resonance imaging (MRI) and serum CA 125 alone ( ). Risk of malignancy cut‐offs are indicated for each model: , IOTA‐ADNEX with CA 125; , IOTA‐ADNEX without CA 125; , ADNEX MR categorical scores; , serum CA 125 (35 kU/L). ADNEX, Assessment of Different NEoplasias in the adneXa; ADNEX MR, MRI scoring system for adnexal masses; IOTA, International Ovarian Tumor Analysis. Table  3 presents the diagnostic metrics of IOTA‐ADNEX (without CA 125) and ADNEX MR at various cut‐off values. The estimated diagnostic metrics using IOTA‐ADNEX with CA 125 were similar and are provided in Table  S3 . The sensitivity of IOTA‐ADNEX without CA 125 at a cut‐off of 30% was 96% (95% CI, 92–98%), with a corresponding specificity of 73% (95% CI, 65–80%). Comparable with this, ADNEX MR at a cut‐off score of 4 demonstrated a sensitivity of 96% (95% CI, 92–98%) and a specificity of 76% (95% CI, 68–83%) (Table  3 ). IOTA‐ADNEX without CA 125 at a cut‐off of 10% had a sensitivity of 99% (95% CI, 96–100%) with a specificity of 50% (95% CI, 41–58%). In comparison, ADNEX MR at a cut‐off score of 3 had a sensitivity of 100% (95% CI, 98–100%) with a specificity of 42% (95% CI, 34–50%). Accuracy increased for IOTA‐ADNEX (with or without CA 125) up to a cut‐off of 50% and for ADNEX MR up to a cut‐off score of 4. At a cut‐off of 50%, IOTA‐ADNEX without CA 125 had an accuracy of 89% (95% CI, 85–92%), while ADNEX MR at a cut‐off score of 4 had an accuracy of 87% (95% CI, 83–91%). We found comparable specificity between ADNEX MR at a cut‐off score of 5 and IOTA‐ADNEX without CA 125 at 75%, with higher sensitivity for IOTA‐ADNEX. The observed risks of malignancy at ADNEX MR scores of 3, 4 and 5 were 15.8%, 65% and 97%, respectively (Table  3 ). Estimated diagnostic performance of the IOTA‐ADNEX without CA 125 model and ADNEX MR scoring system for predicting malignant adnexal lesions at selected cut‐offs Data in parentheses are 95% CI. Risk of malignancy at each ADNEX MR score was 0% (*), 15.8% (†), 65% (‡) and 97% (§). Positive (PPV) and negative (NPV) predictive values depend on prevalence. ADNEX MR, magnetic resonance imaging scoring system for adnexal masses; IOTA‐ADNEX, International Ovarian Tumor Analysis – Assessment of Different NEoplasias in the adneXa; LR+, positive likelihood ratio; LR−, negative likelihood ratio. An overview of misclassified cases at different thresholds showed that the majority (65.8% (25/38) with IOTA‐ADNEX cut‐off of 30%; 64.7% (22/34) with an ADNEX MR cut‐off score of 4) of false‐positive cases were benign solid tumors and that most of the missed malignancies were borderline ovarian tumors (55.6% (5/9) with IOTA‐ADNEX cut‐off of 30%; 66.7% (6/9) with an ADNEX MR cut‐off score of 4) or rare tumors (Table  S4 ). Several of the false‐positive cases were characterized by the presence of degeneration or the concomitant presence of two types of benign tumor. Degenerated benign tumors comprised 13/22 false‐positive solid tumors at the ADNEX MR cut‐off score of 4, and 9/18 false‐positive solid tumors at the IOTA‐ADNEX cut‐off of 40%. A stratified analysis (Figure  3 ) showed that ADNEX MR and IOTA‐ADNEX without CA 125 had similar performance regardless of lesion diameter (≥ 10 cm or < 10 cm), presence of a larger solid component (≥ 2 cm or < 2 cm) and whether the lesions were solid (≥ 80% solid tissue) or not. Areas under the receiver‐operating‐characteristics curve (AUCs) for the IOTA‐ADNEX without serum CA 125 ( ) and ADNEX MR ( ) models, with 95% CI, for differentiating benign from malignant adnexal lesions, stratified by lesion morphology on ultrasound examination. ADNEX MR, magnetic resonance imaging scoring system for adnexal masses; IOTA‐ADNEX, International Ovarian Tumor Analysis – Assessment of Different NEoplasias in the adneXa. IOTA‐ADNEX without CA 125 (cut‐off of 10%) and ADNEX MR (cut‐off score of 3) had moderate agreement (κ = 0.59 (95% CI, 0.48–0.70)) for diagnosing a mass as malignant or benign. IOTA‐ADNEX without CA 125 (cut‐off of 40%) and ADNEX MR (cut‐off score of 4) had substantial agreement for classifying masses as malignant or benign (κ = 0.69 (95% CI, 0.61–0.77)). Agreement was also substantial for the presence of ascites (κ = 0.75 (95% CI, 0.67–0.82)) and number of loculi (weighted κ = 0.68 (95% CI, 0.61–0.75)), moderate for the presence of a solid component (κ = 0.46 (95% CI, 0.36–0.57)) and substantial for classifying lesions as solid (κ = 0.71 (95% CI, 0.63–0.79)) (Table  S5 ). Regarding the largest diameter of the lesion, agreement between ultrasound and MRI was classified as excellent (ICC, 0.92 (95% CI, 0.90–0.94)). The ultrasound measurement of a lesion was, on average, 6.7 mm larger than the MRI measurement. Ultrasound and MRI had good agreement on the size of the largest solid component (ICC, 0.78 (95% CI, 0.72–0.83)). The Bland–Altman analysis for the solid component diameter showed a mean deviation of 8.63 (95% CI, 5.12–12.13) mm, with ultrasound measurements being systematically larger than MRI measurements. The combined sensitivity of ultrasound and MRI, using an IOTA‐ADNEX risk cut‐off of 40%, or an ADNEX MR score of at least 4 to indicate a positive test, was 98% (95% CI, 94–99%), with a specificity of 67% (95% CI, 59–75%) (Table  4 ). This combination achieved an accuracy of 85% (95% CI, 81–89%). Combining an ADNEX MR score cut‐off of 4 with IOTA‐ADNEX cut‐off of 10% resulted in a sensitivity of 100% (95% CI, 98–100%) and specificity of 45% (95% CI, 37–54%) for the detection of malignancy. Combined diagnostic performance of ultrasound and magnetic resonance imaging (MRI) using IOTA‐ADNEX and ADNEX MR models for predicting malignant adnexal lesions Data in parentheses are 95% CI. ADNEX MR, MRI scoring system for adnexal masses; IOTA–ADNEX, International Ovarian Tumor Analysis – Assessment of Different NEoplasias in the adneXa; LR+, positive likelihood ratio; LR−, negative likelihood ratio; NPV, negative predictive value; PPV, positive predictive value. IOTA‐ADNEX demonstrated good discriminative ability between specific tumor types in this high‐risk cohort, notably for differentiating between benign and borderline cases, as well as between benign and Stage‐I invasive malignancy (Table  5 ). However, the discriminative performance of IOTA‐ADNEX (with or without CA 125) was lower for distinguishing between borderline and Stage‐I invasive malignancy (AUC, 0.71 (95% CI, 0.56–0.82)). The performance of IOTA‐ADNEX, particularly without CA 125, was also reduced in discriminating between Stage‐II–IV invasive primary adnexal malignancy vs metastasis to the adnexa. The ADNEX MR scoring system does not provide stratified risk estimates for specific tumor subtypes. Pairwise areas under the receiver‐operating‐characteristics curve (AUCs) of IOTA‐ADNEX model with and without CA 125 for discriminating between subgroups of ovarian tumor outcome IOTA‐ADNEX, International Ovarian Tumor Analysis – Assessment of Different NEoplasias in the adneXa. IOTA‐ADNEX was well calibrated up to a risk of malignancy of approximately 40%. In patients with a risk > 40%, IOTA‐ADNEX underestimated the risk of malignancy in this high‐risk setting. Furthermore, IOTA‐ADNEX without CA 125 was slightly better calibrated than IOTA‐ADNEX with CA 125 (Figure  4, F igure  S1 ). The observed‐to‐expected ratio for IOTA‐ADNEX without CA 125 was 1.032 (95% CI, 0.945–1.128), slightly underestimating the risk. In contrast, for IOTA‐ADNEX with CA 125 it was 0.989 (95% CI, 0.905–1.081), slightly overestimating the risk. Calibration curve of IOTA‐ADNEX without CA 125 model for prediction of malignancy in adnexal lesions ( n  = 342). , ideal; , flexible calibration (locally estimated scatterplot smoothing (LOESS)). IOTA‐ADNEX, International Ovarian Tumor Analysis – Assessment of Different NEoplasias in the adneXa. Our cohort had a larger proportion of postmenopausal patients (78.1%) than in the cohorts used for the development of the ADNEX MR (35.8%) and IOTA‐ADNEX (41.1%) models (Table  S6 ) 26 , 28 . The prevalence of malignancy was higher in our study (58.8%) than in the IOTA‐ADNEX (32.6%), and ADNEX MR (18.8% (reported as 62/329 lesions)) development studies. Imaging findings in our study included more frequent solid components, larger solid components, more papillary projections and a higher frequency of acoustic shadows compared with findings reported in the development studies.

Expert ultrasound using the IOTA‐ADNEX model and expert diffusion‐weighted MRI (without contrast) using the ADNEX MR scoring system demonstrated comparable and high diagnostic performance in distinguishing malignant from benign adnexal masses in high‐risk patients referred for second‐line assessment with suspected adnexal malignancy. MRI and ultrasound were generally in agreement on morphological features. On either modality, most misclassified masses were solid, rare or borderline tumors. IOTA‐ADNEX provided well‐calibrated risk estimates. However, its ability to distinguish between subtypes of malignancy was reduced when differentiating borderline tumors from Stage‐I invasive malignancy. This study included a large, unselected consecutive cohort of patients deemed to be at risk of malignancy by first‐line ultrasound, representative of clinical practice in onco‐gynecological or imaging specialist centers and minimizing selection bias. Standardized, prospectively described imaging features using IOTA criteria for both ultrasound and MRI, combined with the use of internationally validated models with high reproducibility, enabled a systematic, head‐to‐head comparison of the two modalities. Moreover, bias was minimized by blinding the examiners to the findings from other imaging modalities and clinical outcomes. The completeness of the data further strengthens the validity of the study. Our study population included challenging lesions and had a relatively large proportion of borderline tumors compared with the ADNEX MR development study 28 , non‐contrast‐MRI study 30 and IOTA‐ADNEX development study 26 (8.8% borderline tumors in the present study vs 3.6%, 4.0% and 5.7%, respectively). Most benign tumors in our study (71% (100/141)) were described as having a solid component on second‐line ultrasonography. The generalization of our results may be limited to high‐risk settings with experienced examiners of MRI and ultrasound. MRI was performed without contrast following current institutional practice. Our findings apply to non‐contrast MRI; contrast‐enhanced MRI might perform differently. Contrast‐enhanced MRI is resource‐demanding, may cause complications and adverse effects and is contraindicated in certain patient groups 63 . While dynamic contrast‐enhanced MRI may add functional information, previous studies have demonstrated that most adnexal masses can be accurately characterized using non‐contrast sequences, particularly when combining T1‐ and T2‐weighted imaging with DWI 30 , 64 . Furthermore, the contrast enhancement‐curve variable has shown inconsistent performance in validation studies 65 , 66 , 67 . The primary purpose of contrast in MRI is to characterize vascularity, permeability and microcirculation 68 . Likewise, the IOTA‐ADNEX model does not include the functional ultrasound parameter, blood flow with Doppler, owing to subjectivity in interpreting color scores 69 , nor does it incorporate contrast‐enhanced ultrasound (CEUS). Future studies may evaluate how adding these promising tools for vascularity assessment (i.e. color Doppler, CEUS and contrast‐enhanced MRI) may affect the classification of adnexal tumors. The inclusion of expectantly managed patients ( n  = 24) may have introduced differential verification bias, but excluding these patients did not alter the results. The Ovarian‐Adnexal Reporting and Data System (O‐RADS) represents the American counterpart to ADNEX 70 , 71 , 72 . The O‐RADS ultrasound system 73 , 74 is based on the IOTA datasets, while O‐RADS MRI was developed in an external validation study of the ADNEX MR score 72 . Previous studies show concordance between O‐RADS and IOTA‐ADNEX 72 , 75 , supporting the generalizability of our findings. One meta‐analysis 76 and a recent study in 44 patients 42 comparing the use of O‐RADS ultrasound and MRI systems found similar diagnostic performance for both techniques. A recent systematic review highlighted the absence of head‐to‐head studies of IOTA‐ADNEX and ADNEX MR within the same patients 34 . The review instead relied on indirect comparisons between separate IOTA‐ADNEX and ADNEX MR studies 34 . Both models showed high pooled AUCs (0.97 (95% CI, 0.95–0.98) for IOTA‐ADNEX and 0.98 (95% CI, 0.96–0.99) for ADNEX MR) 34 . The meta‐analysis reported higher sensitivity for IOTA‐ADNEX, consistent with our findings, but higher specificity for ADNEX MR. This may reflect differing MRI and ultrasound study populations. The few existing comparative diagnostic accuracy studies performed in the same patients used subjective assessment and, occasionally, IOTA LR2 39 , IOTA Simple Rules 40 or non‐validated morphological criteria, compared with subjective MRI assessment 36 , 39 , 40 , 77 , 78 . Subjective assessments increase interobserver variability and may limit generalizability compared with standardized methods. One study reported that subjective ultrasound assessment was more specific than subjective MRI assessment 13 . A Chinese study reported that IOTA‐ADNEX, using a 15% risk cut‐off, performed similarly to subjective MRI evaluation 41 . A low ADNEX cut‐off in first‐line triage may be needed to maximize sensitivity and facilitate the detection of early‐stage malignancy 44 , 79 . Hence, reliable second‐line diagnostics are required to improve specificity and limit overtreatment. The choice of second‐line imaging modality may be influenced by examiner availability, reimbursement/health systems and logistical considerations. Ultrasound offers lower costs, shorter examination times and greater accessibility than MRI. Thus, our study may support second‐line confirmatory imaging by an experienced ultrasound examiner using IOTA‐ADNEX. In a recent study, patients were more likely to prefer ultrasound examinations to MRI, but individual preferences varied 33 . The IOTA‐ADNEX cut‐off of 30% had high accuracy and yielded similar sensitivity and specificity to an ADNEX MR score of 4. The good calibration of IOTA‐ADNEX underscores its usefulness for guiding decision‐making, while also considering individual patient circumstances and priorities. MRI performed by radiologists specialized in pelvic MRI using ADNEX MR is an alternative option with similar diagnostic performance. In resource‐limited settings, ultrasound may be the most cost‐effective option. This study did not identify any morphological subgroups in which ADNEX MR demonstrated a superior diagnostic performance. Our findings do not support the previously described recommendation to perform MRI for predominantly solid adnexal lesions, lesions with small solid components or lesions larger than 5–7 cm because of the presumed reduced accuracy of ultrasound for these lesion types 29 . MRI may be indicated when ultrasonography is limited by lesion location or adjacent anatomy 71 , and for evaluating the resectability of advanced malignancies if ultrasound expertise is inadequate for this assessment 80 , 81 . IOTA‐ADNEX achieved excellent diagnostic performance in distinguishing borderline from benign tumors but lacked precision in differentiating between Stage‐I invasive malignancy and borderline tumors, and between metastasis to the adnexa and Stage‐II–IV invasive malignancy. A stratified MRI‐based risk score for distinguishing borderline from invasive tumors has yet to be developed. Our findings highlight the need for improved characterization of borderline tumors, solid tumors and rare tumor types. To conclude, IOTA‐ADNEX and ADNEX MR models performed equally well in diagnosing adnexal malignancy in consecutive patients referred for second‐line imaging in our center. Our findings support using either ultrasound (with IOTA‐ADNEX, with or without CA 125) or MRI (with ADNEX MR), when performed by experienced examiners, for second‐line confirmatory assessment of adnexal masses deemed suspicious at first‐line non‐expert ultrasonography. Further optimization is needed to improve the assessment of solid, rare and borderline tumors.

Accurate preoperative diagnosis is essential for planning the management of patients with adnexal masses 1 , 2 . When adnexal lesions are suspicious for malignancy on first‐line ultrasound examination, patients should be offered referral for a second‐line assessment to confirm, substantiate or refute the first‐line findings 2 , 3 , 4 , 5 , 6 , 7 . Depending on the protocol of the center, the second‐line examination may be expert ultrasonography or magnetic resonance imaging (MRI) 4 , 8 , 9 , 10 , 11 , 12 , 13 . Lesions deemed malignant based on imaging should be managed in gynecological oncology centers 14 , 15 , whereas benign lesions may be managed conservatively 4 , 16 , 17 , 18 or with minimally invasive intervention to reduce morbidity and preserve fertility 19 , 20 , 21 . Furthermore, preoperative second‐line assessment enables improved patient counseling and may help estimate the time, expertise and resources required for surgery. A previous randomized controlled trial has demonstrated better subsequent clinical management outcomes (i.e. fewer major surgeries and shorter hospital stays) when second‐line ultrasound examinations are performed by ultrasound experts rather than non‐expert examiners 22 . Decision‐making at ultrasound or MRI can be supported by estimating the risk of malignancy using prediction models such as the Assessment of Different NEoplasias in the adneXa (ADNEX) model with ultrasound terms 23 according to the International Ovarian Tumor Analysis (IOTA) group (IOTA‐ADNEX model) 4 , 24 , 25 , 26 , 27 , or the international scoring system based on MRI features with diffusion‐weighted imaging (DWI) (ADNEX MR scoring system) 28 , 29 , 30 . Ultrasound has exceedingly high resolution for the assessment of morphology, whereas MRI provides good soft‐tissue contrast and, with DWI, allows indirect evaluation of tissue microstructure and cellularity. Several guidelines recommend MRI for the second‐line evaluation of masses identified to be at risk of malignancy on the first‐line ultrasound examination 6 , 7 , 31 , 32 . However, MRI is not always available, is costly and resource‐intensive and may cause anxiety and discomfort for patients 33 . A recent systematic review noted the absence of studies directly comparing standardized ultrasound and standardized MRI (i.e. IOTA‐ADNEX vs ADNEX MR models) for differentiating between benign and malignant masses within the same patient population 34 . The few existing studies comparing MRI and ultrasound in the same patient cohort are relatively small, use mainly subjective assessment rather than standardized, reproducible diagnostic models and have yielded conflicting conclusions 13 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 . The main objective of this study was to validate and compare expert ultrasound using the IOTA‐ADNEX model with MRI using the ADNEX MR scoring system, as second‐line imaging modalities for differentiating benign from malignant adnexal lesions.

Appendix S1–S2 STARD (Appendix S1) and TRIPOD (Appendix S2) checklists. Appendix S3 Systematic form for reporting ultrasound findings. Appendix S4 Further details on IOTA‐ADNEX model with formulae for IOTA‐ADNEX with CA 125 and without CA 125. Appendix S5 Systematic form for reporting magnetic resonance imaging findings. Appendix S6 ADNEX MR scoring system. Appendix S7 Sample‐size calculation. Table S1 Classification of primary adnexal malignancy in cohort according to International Federation of Gynecology and Obstetrics (FIGO) stage. Table S2 Sensitivity analysis for diagnostic performance of IOTA‐ADNEX and ADNEX MR in 318 patients with reference standard based on histopathology. Table S3 Estimated diagnostic performance of the IOTA‐ADNEX with CA 125 model and ADNEX MR scoring system for predicting malignant adnexal lesions at selected cut‐offs. Table S4 Misclassified cases (false‐positive and false‐negative cases) when using ultrasound and MRI for predicting malignancy. Table S5 Agreement in risk assessments and characterization of morphological features between ultrasound and magnetic resonance imaging. Table S6 Descriptive statistics of predictor variables in the model development datasets and in current dataset. Figure S1 Calibration curve of IOTA‐ADNEX with CA 125 model for prediction of malignancy in adnexal lesions.