Comparing and combining O-RADS ultrasound, serum CA125, and HE4 for the preoperative diagnosis of ovarian tumors

A total of 367 patients were enrolled in this study. Among these, 55.59% (204 out of 367) were diagnosed with benign ovarian tumors, while 44.41% (163 out of 367) were diagnosed with malignant ovarian tumors. The specific histopathological subtypes of the 367 benign and malignant masses are detailed in Table  1 . Table 1 Pathologic results for 367 adnexal masses Pathology No. of lesions Benign 204  Endometrioma 72  Dermoid 32  Serous Cystadeno(fibro)ma 29  Mucinous Cystadeno(fibro)ma 17  Simple cyst 17  Paramesonephric cyst 16  hydrosalpinx 7  Serous mucinous cystadenoma 6  Fibroma、Thecoma 5  Corpus Luteum Cyst 3 Malignant 163  High-Grade Serous Carcinoma 79  Clear Cell Carcinoma 26  Endometrioid Carcinoma 16  Metastasis 7  Mucinous Adenocarcinoma 6  Serous Borderline Tumor 6  Immature Teratoma 4  Mucinous Borderline Tumor 4  Sertoli-Leydig Cell Tumor 3  Adult Granulosa Cell Tumor 3  Atypical Proliferative Endometrioid Tumor 3  Dysgerminoma 2  Yolk Sac Tumor 2  Carcinosarcoma 2 Pathologic results for 367 adnexal masses The age distribution exhibited a significant difference between the two groups. Due to the deviation from normality in the age distribution within the benign group, as indicated by the Shapiro‑Wilk test ( p  < 0.05), the non-parametric Mann-Whitney U test was employed. The analysis revealed that patients in the malignant group were significantly older than those in the benign group [54.0 (46.0, 61.0) years vs. 37.0 (30.0, 47.0) years; Z = -3.85, p  < 0.001]. Additionally, menopausal status varied significantly between the groups ( p  < 0.001), with a higher proportion of postmenopausal women present in the malignant group compared to the benign group [60.7% (99/163) versus 19.6% (44/204)]. Furthermore, the maximum diameter of the lesions was significantly greater in the malignant group than in the benign group [77.0(54.0,115.0) versus 59.0 (47.0,70.8), p  < 0.001]. A strong correlation was identified between O-RADS classification and tumor pathology (χ²=278.335, p  < 0.001). The proportion of malignant tumors increased significantly with higher O-RADS categories: O-RADS 2 (0/118, 0%), O-RADS 3 (5/69, 7.2%), O-RADS 4 (65/85, 76.5%), and O-RADS 5 (93/95, 97.9%).The interobserver agreement for the initial O-RADS classification between two experienced sonographers (10 and 15 years in gynecological tumor diagnosis) showed a weighted kappa of 0.82 (95% CI: 0.76–0.88), indicating almost perfect agreement per Landis and Koch criteria, thus confirming the high reliability and reproducibility of the ultrasound-based O-RADS risk assessment in our study. Malignant tumors more frequently exhibited complex sonographic features: solid (74/163, 45.4%), unilocular-solid (38/163, 23.3%), or multilocular-solid (33/163, 20.2%). In contrast, benign tumors were predominantly unilocular (127/204, 62.3%) or multilocular (64/204, 31.4%) cysts (χ² = 256.3, p  < 0.001). Higher color Doppler flow scores (moderate/marked vascularity) were observed exclusively in the malignant group (54/163, 33.1% and 21/163, 12.9%, respectively), whereas the majority of benign tumors showed no or minimal flow (151/204, 74.0% and 53/204, 26.0%; χ² = 204.1, p  < 0.001). Ascites was significantly more common in malignant cases (43/163, 26.4%) than in benign cases (1/204, 0.5%; χ² = 57.6, p  < 0.001). Serum concentrations of all three biomarkers were markedly elevated in the malignant group compared to the benign group, as determined by the Mann-Whitney U test (all p  < 0.001). Specifically, the median CA125 level was 163.0 U/mL (IQR: 31.0, 447.4) in the malignant group, compared to 21.3 U/mL (IQR: 12.2, 44.7) in the benign group. Similarly, the median HE4 concentration was 88.10 pmol/L (IQR: 48.2, 239.6) in the malignant group, in contrast to 32.3 pmol/L (IQR: 27.1, 37.8) in the benign group. The Risk of Ovarian Malignancy Algorithm (ROMA) index also demonstrated a substantial difference, with a median value of 50.6 (IQR: 11.7, 85.2) in malignant cases, as opposed to 3.4 (IQR: 2.0, 6.1) in benign cases (refer to Table  2 ). Table 2 General clinical characteristics of the 367 patients Characteristic Benign ( n  = 204) Malignant ( n  = 163) p value Age (years) 37.0(30.0, 47.0) 54.0 (46.0, 61.0) < 0.001 Menopausal status < 0.001  yes 40 99  no 164 64 maximum diameter of lesions 59.0 (47.0,70.8) 77.0(54.0,115.0) < 0.001  O-RADS US < 0.001  O-RADS 2 118 0  O-RADS 3 64 5  O-RADS 4 20 65  O-RADS 5 2 93 Morphologic Type < 0.001  Unilocular cyst 127 5  Multilocular cyst 64 13  Unilocular cyst- solid 6 38  Multilocular cyst-solid 1 33  Solid 6 74 Color Score < 0.001  1 151 9  2 53 79  3 0 54  4 0 21 Ascites < 0.001  yes 1 43  no 203 120 CA125 (U/mL) 21.3(12.2,44.7) 163.0(31.0,447.4) < 0.001 HE4 (pmol/L) 32.3(27.1,37.8) 88.1(48.2,239.6) < 0.001 ROMA index(%) 3.4(2.0,6.1) 50.6(11.7,85.2) < 0.001 General clinical characteristics of the 367 patients Stratified analysis by menopausal status revealed distinct patterns in the differentiation of benign and malignant ovarian tumors (Table  3 ). In the overall cohort and the premenopausal subgroup, malignant cases were consistently associated with significantly higher values across all parameters—including age, maximum lesion diameter, O-RADS classification, CA125, HE4, and the ROMA index—compared to benign cases (all p  < 0.001). A notable divergence was observed in the postmenopausal subgroup: while O-RADS classification and the serum biomarkers (CA125, HE4, and ROMA) remained highly significant discriminators (all p  < 0.001) and maximum lesion diameter retained significance ( p  = 0.010), patient age no longer differed significantly between benign and malignant tumors ( p  = 0.743). Table 3 General clinical characteristics in premenopausal and postmenopausal women of 367 patients Characteristic Benign ( n  = 204) Malignant ( n  = 163) p value Premenopausal Age (years) 34.0(29.0, 40.0) 43.5(36.5, 47.0) < 0.001 Maximum Lesion Diameter(mm) 60.0(48.0. 70.0) 97.0(63.0, 140.5) < 0.001  O-RADS US < 0.001  O-RADS 2 100 0  O-RADS 3 50 2  O-RADS 4 13 28  O-RADS 5 1 34 CA125 (U/mL) 22.9(12.6, 45.3) 116.3(32.9, 470.3) < 0.001 HE4 (pmol/L) 31.6(26.9, 38.3) 68.9(39.9, 173.5) < 0.001 ROMA index(%) 2.7(1.9, 4.2) 16.9(4.5, 67.6) < 0.001 Postmenopausal Age (years) 61.5(54.5,67.0) 60.0(56.0, 65.0) 0.743 Maximum Lesion Diameter(mm) 55.5(39.5,73.5) 70.0(48.5, 98.0) 0.01  O-RADS US < 0.001  O-RADS 2 18 0  O-RADS 3 14 3  O-RADS 4 7 37  O-RADS 5 1 59 CA125 (U/mL) 17.6(11.4, 37.2) 170.8(29.9, 511.6) < 0.001 HE4 (pmol/L) 33.6(27.9, 36.9) 110.0(55.8, 324.7) < 0.001 ROMA index(%) 8.8(6.5, 15.1) 65.4(19.8, 91.0) < 0.001 General clinical characteristics in premenopausal and postmenopausal women of 367 patients A forward stepwise (likelihood ratio) binary logistic regression analysis was performed to identify independent predictors of ovarian malignancy. Candidate variables included O-RADS score (entered as a categorical variable with O-RADS 5 as the reference), log-transformed HE4 (natural logarithm), log-transformed CA125 (natural logarithm), and menopausal status.The final model retained O‑RADS score and log-transformed HE4 as independent predictors of malignancy, while log-transformed CA125 and menopausal status were not retained.O‑RADS score showed a strong association with malignancy risk ( p  < 0.001). Compared to O-RADS 5, the odds of malignancy decreased progressively with lower O-RADS score.Higher HE4 levels were also independently associated with increased malignancy risk. The odds ratio for a one-unit increase in ln(HE4) was 8.188 (95% CI: 2.491–26.912, p  = 0.001). (refer to Table  4 ). The model also showed good calibration (Hosmer–Lemeshow test: χ² = 7.68, p  = 0.465; Nagelkerke R ² = 0.871). Table 4 Multivariable logistic regression analysis of predictors for ovarian malignancy Variable OR 95% CI p value O‑RADS score (ref: 5) < 0.001  O-RADS 4 0.148 0.031–0.710 0.016 0.017  O-RADS 2–3 0.006 0.001–0.037 < 0.001 ln(HE4) 8.188 2.491–26.912 0.001 Multivariable logistic regression analysis of predictors for ovarian malignancy 0.031–0.710 0.016 The diagnostic efficacy of the evaluated parameters differed markedly between premenopausal and postmenopausal women. In the premenopausal group, the O-RADS score demonstrated superior performance (AUC: 0.971) compared to all serum biomarkers (CA125, HE4, ROMA index; AUC range: 0.798–0.843). Its high negative predictive value (98.68%) underscores its utility in reliably ruling out malignancy. Conversely, in the postmenopausal group, HE4 emerged as the most potent serum marker (AUC: 0.971), exhibiting specificity and positive predictive value exceeding 95%. Notably, the combination of O-RADS and HE4 achieved the most balanced and robust diagnostic profile across both groups (AUC: 0.959, Sensitivity: 94.95%, Specificity: 92.50%, PPV: 96.91%), highlighting the significant added value of integrating imaging classification with a highly specific biomarker, particularly in postmenopausal women. (refer to Table  5 ; Fig.  1 A and B).Internal validation using 1,000 bootstrap resamples yielded an optimism-corrected mean AUC of 0.958 (95% CI: 0.921–0.988), indicating minimal overfitting. Pairwise comparisons of AUCs using the DeLong test are summarized in Table  6 .In premenopausal women, the O-RADS score demonstrated significantly higher AUC than CA125 ( p  < 0.0001), HE4 ( p   0.05).In postmenopausal women, HE4 achieved the highest AUC among serum biomarkers (0.971), significantly outperforming CA125 ( p  < 0.0001) and ROMA ( p  = 0.0016). The O-RADS score (AUC = 0.939) and the combined O‑RADS+HE4 score (AUC = 0.959) were not significantly different from each other ( p  = 0.1111), nor from HE4 alone (O-RADS vs. HE4: p  = 0.1947; combined vs. HE4: p  = 0.5684). O-RADS was significantly better than CA125 ( p  = 0.0020) but showed no statistical difference compared to ROMA ( p  = 0.2110). These findings indicate that HE4 and the combined model are the most robust serum-based tools for risk stratification in postmenopausal women. Table 5 ROC curve analysis of diagnostic performance for O‑RADS US, CA125, HE4, and ROMA index AUC 95% CI Cut-off Sensitivity (%) Specificity (%) PPV (%) NPV (%) Accuracy (%) Premenopausal  O-RADS US 0.971 0.940–0.989 4 96.88 91.46 81.58 98.68 92.98  CA125 (U/mL) 0.798 0.740–0.848 70.75 62.50 87.20 65.56 85.63 80.26  HE4 (pmol/L) 0.838 0.784–0.884 40.45 75.00 81.71 61.54 89.33 79.82  ROMA index(%) 0.843 0.789–0.888 4.92 73.44 82.93 62.67 88.89 80.26 Postmenopausal  O-RADS US 0.939 0.886–0.973 4 96.97 80 92.31 91.43 92.09  CA125 (U/mL) 0.818 0.744–0.878 75.15 65.66 92.50 95.59 52.11 73.38  HE4 (pmol/L) 0.971 0.928–0.992 42.65 92.93 95.00 97.87 84.44 93.53  ROMA index(%) 0.898 0.835–0.943 29.63 68.69 100 100 56.34 77.70  O-RADS US+HE4 0.959 0.911–0.985 4 94.95 92.50 96.91 88.10 94.24 ROC curve analysis of diagnostic performance for O‑RADS US, CA125, HE4, and ROMA index Table 6 Comparison of ROC curves (DeLong test) Comparison Premenopausal ( n  = 228) Postmenopausal ( n  = 139) p value p value O-RADS vs. CA125 < 0.0001 0.0020 O-RADS vs. HE4 < 0.0001 0.1947 O-RADS vs. ROMA 0.0001 0.2110 O-RADS vs. O-RADS+HE4 - 0.1111 HE4 vs. CA125 0.3232 < 0.0001 HE4 vs. ROMA 0.1264 0.0016 O-RADS+HE4 vs. CA125 - 0.0004 O-RADS+HE4 vs. HE4 - 0.5684 Comparison of ROC curves (DeLong test) Fig. 1 ( A ) Comparison of ROC curves for CA125, HE4, the ROMA index, and the O-RADS score in differentiating benign from malignant ovarian tumors in premenopausal women. AUC, area under the curve. ( B ) Comparison of ROC curves for CA125, HE4, the ROMA index, the O-RADS score and the O-RADS score combine HE4 in differentiating benign from malignant ovarian tumors in postmenopausal women. AUC, area under the curve ( A ) Comparison of ROC curves for CA125, HE4, the ROMA index, and the O-RADS score in differentiating benign from malignant ovarian tumors in premenopausal women. AUC, area under the curve. ( B ) Comparison of ROC curves for CA125, HE4, the ROMA index, the O-RADS score and the O-RADS score combine HE4 in differentiating benign from malignant ovarian tumors in postmenopausal women. AUC, area under the curve To evaluate the clinical utility of the diagnostic models, decision curve analysis (DCA) was performed to assess the net benefit across a range of threshold probabilities for recommending surgery or further diagnostic workup.In premenopausal women (Fig.  2 A), the ORADS score demonstrated the highest net benefit across the entire clinically relevant threshold range (approximately 0.1–0.5). HE4 and ROMA showed comparable net benefit, while CA125 performed slightly better at lower thresholds.In postmenopausal women (Fig.  2 B), decision curve analysis demonstrated that the O-RADS score, the combined O-RADS+HE4 model, and HE4 alone yielded nearly identical net benefit across the entire clinically relevant threshold range (approximately 0.1–0.5). All three models consistently outperformed both the “treat-all” and “treat-none” strategies, confirming their clinical utility for surgical triage. Fig. 2 ( A ) Comparison of DCA curves for CA125, HE4, the ROMA index, and the O-RADS score in differentiating benign from malignant ovarian tumors in premenopausal women. DCA, decision curve analysis. ( B ) Comparison of DCA curves for CA125, HE4, the ROMA index, the O-RADS score and the O-RADS score combine HE4 in differentiating benign from malignant ovarian tumors in postmenopausal women. DCA, decision curve analysis ( A ) Comparison of DCA curves for CA125, HE4, the ROMA index, and the O-RADS score in differentiating benign from malignant ovarian tumors in premenopausal women. DCA, decision curve analysis. ( B ) Comparison of DCA curves for CA125, HE4, the ROMA index, the O-RADS score and the O-RADS score combine HE4 in differentiating benign from malignant ovarian tumors in postmenopausal women. DCA, decision curve analysis

This was a retrospective study conducted in Shanghai First Maternity and Infant Hospital between January 2021 and December 2023. Patients with ovarian tumors who received surgery and obtained pathological results were collected. The inclusion criteria were as follows: (1) Complete and clear ultrasound imaging data obtained within 2 weeks before initial treatment; (2) Serum CA125 and HE4 test results available within 2 weeks before initial treatment; (3) Complete medical history, with no prior radiotherapy, chemotherapy, or other anti-tumor treatments before surgery; (4) Complete postoperative pathological data available. Ultrasound examinations were performed using Voluson E8 and E10 ultrasound diagnostic systems (GE Healthcare, USA). All subjects underwent transvaginal color Doppler ultrasound examination with a probe frequency of 5–9 MHz. Ovarian tumors were scanned in multiple planes via transvaginal ultrasound, and static or dynamic images were stored. If a large mass could not be fully visualized by the transvaginal probe, transabdominal ultrasound was additionally performed using a probe frequency of 3.5–4.5 MHz. Two sonographers specialized in gynecological ultrasound, with substantial experience in gynecological tumor diagnosis (10 and 15 years, respectively), independently reviewed and recorded the ultrasonographic features of the uterus and adnexa for each patient without knowledge of the pathological results. The ovarian tumors were scored according to the O-RADS classification system (categories 0–5) [ 10 ]. If a patient had two or more ovarian tumors, the largest one or the one with the most complex ultrasonographic features was selected for analysis. In cases of disagreement between the two readers, a final assessment was made by a senior physician.When combining O-RADS with serum biomarkers, the initial O-RADS category was appropriately upgraded or downgraded based on predetermined cutoff values of the biomarkers.According to the pre-defined algorithm, an initial O-RADS category 3 was upgraded to category 4 if the biomarker level exceeded the cutoff, while an initial category 4 was downgraded to category 3 if the biomarker level was below the cutoff. All other categories (2, 5, and the remaining category 3/4 combinations) remained unchanged. Before the assay, the frozen samples were thawed until the room temperature. Serum CA125 levels were measured using the Tumor-Associated Antigen CA125 Detection Kit (Chemiluminescent Immunoassay, CLIA) on a Mindray CL-8000i automated immunoassay analyzer (Mindray, China). Serum HE4 levels were determined using the Human Epididymis Protein 4 (HE4) Detection Kit (Chemiluminescent Immunoassay, CLIA) on the same Mindray CL-8000i analyzer. The Risk of Ovarian Malignancy Algorithm (ROMA) index for each patient was calculated according to the formula established by Moore et al. in the original publication [ 11 ], incorporating the patient’s menopausal status, CA125 level (U/mL), and HE4 level (pmol/L). The ROMA score corresponds to predicted probability and is expressed by a percentage rate. Statistical analysis was performed using IBM SPSS Statistics version 19.0 (IBM Corp., Armonk, NY, USA). Normality of continuous variables was assessed using the Shapiro‑Wilk test. Variables following a normal distribution ( p  > 0.05) are presented as mean ± standard deviation (SD) and compared between two groups using the independent samples t-test. Non‑normally distributed variables are presented as median with interquartile range (IQR; Q1–Q3) and compared using the Mann-Whitney U non-parametric test. Categorical data were expressed as number (percentage) and compared using the Chi-square test or Fisher’s exact probability test, as appropriate. Multivariable binary logistic regression with forward stepwise (likelihood ratio) selection was used to identify independent predictors of ovarian malignancy.Results are reported as odds ratios (ORs) with 95% confidence intervals (CIs). The predictive performance was assessed by plotting receiver operating characteristic (ROC) curves and calculating the area under the curve (AUC). The optimal cutoff value for predicting malignancy was determined based on the maximum Youden’s index. Subsequently, the diagnostic accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated at this optimal cutoff. Pairwise comparisons of AUCs were performed using the DeLong test as implemented in MedCalc software (version 23.3.4), which has a built-in procedure for comparing ROC curves. To evaluate the clinical utility of the diagnostic models, decision curve analysis (DCA) was performed.Decision curve analysis (DCA) was performed using R software (version 4.4.1) with the rmda package (version 1.6). All analyses were stratified by menopausal status (premenopausal vs. postmenopausal).

The diagnostic efficacy of ovarian tumor assessment tools exhibits significant variation contingent upon menopausal status. O-RADS consistently demonstrates high performance across all age groups. In postmenopausal women, HE4 surpasses CA125 in performance and achieves diagnostic accuracy comparable to that of O-RADS. While the addition of HE4 to O-RADS does not statistically enhance discrimination or net benefit beyond O-RADS alone, the integrated model offers clinically valuable bidirectional risk adjustment in cases where ultrasound features are inconclusive. A stratified diagnostic approach that accounts for menopausal status optimizes preoperative decision-making and patient triage.

This study offers a detailed evaluation of the diagnostic performance of the O-RADS US risk stratification system, serum CA125, HE4, and the ROMA index, stratified by menopausal status, in distinguishing between benign and malignant ovarian tumors. The primary findings underscore the significant influence of menopausal status on the effectiveness of diagnostic biomarkers and confirm the superior and consistent utility of the O-RADS system. Moreover, these results support the adoption of a tailored, integrated approach to diagnosis. The O-RADS system exhibited remarkable diagnostic accuracy, achieving an area under the curve (AUC) greater than 0.93 in both cohorts examined, aligning with prior validation studies conducted across diverse populations [ 13 – 15 ]. Recent studies by Pelayo et al. have further confirmed the high diagnostic performance of O-RADS, as well as IOTA simple rules and the ADNEX model, in differentiating benign from malignant adnexal masses [ 16 , 17 ]. In our cohort, O-RADS demonstrated an exceptionally high NPV, particularly in premenopausal women (98.68%), which is of substantial clinical importance. This finding suggests that a low O-RADS category (≤ 3) can provide highly reliable reassurance against malignancy, potentially mitigating unnecessary anxiety and reducing invasive procedures among a predominantly young patient population. Moreover, the uniformity of its optimal cutoff (≥ 4) across different menopausal groups enhances its practicality and facilitates its clinical implementation. Conversely, in patients with higher O-RADS categories (4–5), accurate preoperative risk stratification also helps avoid overtreatment, especially for borderline ovarian tumors (BOTs) in reproductive-age women. In young women with BOTs, preserving fertility is a clinical priority. As demonstrated by Fagotti et al. preoperative ultrasound features can help distinguish invasive from non-invasive lesions, guiding the choice of conservative surgery [ 18 ]. However, the preoperative differentiation of BOTs from malignant tumors remains challenging. Moro et al. reported that borderline serous tumors and non-invasive low-grade serous carcinomas often share overlapping ultrasound features, both typically presenting as cysts with papillary projections [ 19 ]. Similarly, for mucinous tumors, Moro et al. noted that a multilocular cyst with > 10 locules is suggestive of a gastrointestinal-type borderline tumor, whereas 2–10 locules is more typical of benign cystadenoma [ 20 ]. These sonographic clues, although not definitive, can raise clinical suspicion for BOTs and guide the surgical approach toward fertility-sparing procedures when appropriate. In our cohort, the two youngest patients (29 and 32 years) with BOTs successfully underwent fertility-sparing procedures, and neither has recurred. This underscores the value of accurate preoperative risk assessment — particularly the O-RADS classification — in informing surgical decision-making for reproductive-age women, balancing oncological safety with fertility preservation. Our stratified analysis uncovers a critical paradigm shift: the diagnostic hierarchy of serum biomarkers is not static but undergoes significant changes with menopausal status. In premenopausal women, the serum biomarkers CA125, HE4, and ROMA demonstrated moderate and comparable efficacy, with area under the curve (AUC) values ranging from 0.798 to 0.843. This performance was notably inferior to that of the O-RADS system. The suboptimal sensitivity and positive predictive value (PPV) of CA125 in this demographic are well-documented, primarily due to its non-specific elevation in various benign gynecological conditions common among younger women, such as endometriosis and pelvic inflammatory disease [ 21 – 23 ]. Additionally, factors such as smoking and renal function can influence CA125 and HE4 levels, further complicating their interpretation in premenopausal populations [ 22 ]. In postmenopausal women, HE4 has emerged as the leading serum marker, demonstrating an area under the curve (AUC) of 0.971 and a specificity of 95.00%, comparable to that of imaging techniques. This exceptional performance is biologically plausible, as HE4 expression is less affected by the benign gynecological conditions frequently observed in premenopausal women and is more specifically associated with ovarian malignancy, particularly in older populations [ 24 – 25 ]. In contrast, the Risk of Ovarian Malignancy Algorithm (ROMA) index, although achieving 100% specificity and positive predictive value (PPV) in our postmenopausal cohort, exhibited a sensitivity of 68.69% and a negative predictive value (NPV) of 56.34%, which are unacceptably low, thereby limiting its effectiveness as a standalone rule-out test. This observation diverges from some previous studies and underscores the necessity for context-and population-specific validation [ 12 , 26 ]. The stratified analysis disclosed a significant shift in pattern: O-RADS alone predominates in premenopausal women, whereas HE4 demonstrates equivalent efficacy to O-RADS in postmenopausal women. This underscores the necessity for distinct diagnostic strategies tailored to each population, which constitutes a principal conclusion of this study. The principal finding of this study is the complementary diagnostic value offered by integrating the O-RADS system with serum HE4. Our results indicate that this combined model offers the most balanced and robust diagnostic profile, with an area under the curve (AUC) of 0.959, sensitivity of 94.95%, specificity of 92.50%, and a positive predictive value (PPV) of 96.91%. The synergy between these diagnostic tools is mechanistically justified: O-RADS provides a comprehensive assessment of morphological and vascular architecture, while HE4 adds a layer of biologically specific information [ 27 – 29 ]. Nonetheless, a formal statistical comparison using the DeLong test demonstrated that, in postmenopausal women, the combined model did not yield a statistically significant improvement in AUC over O-RADS alone ( p = 0.1111). This outcome likely reflects the already high performance of O-RADS (AUC = 0.939) within this subgroup, thereby offering limited scope for further enhancement. Despite the absence of statistical significance, the combined model exhibited numerically superior specificity (92.50% vs. 80%), indicating potential clinical utility in specific risk contexts. Decision curve analysis (DCA) provided further insights into the clinical utility of the integrated model. Within the postmenopausal cohort, the O-RADS+HE4 model, O-RADS alone, and HE4 alone yielded nearly identical net benefit curves across the clinically relevant threshold range. Although the combined model did not exhibit superior net benefit compared to O-RADS alone, this observation does not detract from its value. It suggests that HE4 alone is already a robust predictor in postmenopausal women (AUC = 0.971), and its integration with O-RADS does not undermine clinical decision-making. Notably, all three models demonstrated a positive net benefit compared to the “treat-all” and “treat-none” strategies, thereby endorsing their application as clinical triage tools. The clinical necessity and impact of this integration are clearly demonstrated by two representative cases from our cohort. In the first case(Fig.  3 ), a lesion initially classified as O-RADS 4 was reclassified to O-RADS 3 based on a normal HE4 level, with subsequent pathology confirming a benign serous cystadenoma. In the second case(Fig.  4 ), a lesion initially deemed as O-RADS 3 was reclassified as high risk due to an elevated HE4 level, resulting in the accurate diagnosis of endometrioid carcinoma. These cases underscore the bidirectional corrective function of the integrated approach, especially in situations where ultrasound features are indeterminate. Fig. 3 A representative case of a benign ovarian tumor in a postmenopausal woman. ( a ) Transvaginal ultrasound image shows a left adnexal cystic lesion measuring 55 mm×39 mm×52 mm. The cyst contains punctate isoechoic, and a papilla (measuring 10 mm×6 mm×9 mm) is visible on the cyst wall. ( b ) CDFI shows that there is no color blood flow signal in the papilla. (blood flow score: 1). Based on the ultrasound features, the lesion was initially classified as O-RADS US category 4 A representative case of a benign ovarian tumor in a postmenopausal woman. ( a ) Transvaginal ultrasound image shows a left adnexal cystic lesion measuring 55 mm×39 mm×52 mm. The cyst contains punctate isoechoic, and a papilla (measuring 10 mm×6 mm×9 mm) is visible on the cyst wall. ( b ) CDFI shows that there is no color blood flow signal in the papilla. (blood flow score: 1). Based on the ultrasound features, the lesion was initially classified as O-RADS US category 4 However, the serum HE4 level was within the normal range (25.1 pmol/L). Integrating this biomarker result, the final assessment was downgraded to O-RADS US category 3.Final pathological diagnosis: (Left mesosalpinx) Papillary serous cystadenoma, confirming the benign nature of the lesion. Fig. 4 A representative case of a malignant ovarian tumor in a postmenopausal woman. ( a ) Transvaginal ultrasound image shows a right ovarian lesion measuring 62 mm×66 mm×55 mm, characterized by poor echo with local wall thickening. ( b ) CDFI shows that there is no color blood flow signal in the cyst (blood flow score: 1). Based on the ultrasound features, the lesion was initially classified as O-RADS US category 3 A representative case of a malignant ovarian tumor in a postmenopausal woman. ( a ) Transvaginal ultrasound image shows a right ovarian lesion measuring 62 mm×66 mm×55 mm, characterized by poor echo with local wall thickening. ( b ) CDFI shows that there is no color blood flow signal in the cyst (blood flow score: 1). Based on the ultrasound features, the lesion was initially classified as O-RADS US category 3 However, the serum HE4 level was markedly elevated (192 pmol/L). Integrating this biomarker result, the final risk assessment was upgraded to O-RADS US category 4. Final pathological diagnosis: (Right ovary) Endometrioid carcinoma. For postmenopausal women, we propose that definitive preoperative assessment can be enhanced through the simultaneous application of O-RADS and HE4. Specifically, the presence of an O-RADS category of 4 or higher, coupled with an elevated HE4 level, provides strong evidence of high-risk malignancy, thereby offering a clear and actionable basis for expedited referral and surgical planning [ 30 ]. Conversely, a normal HE4 level in conjunction with an O-RADS 4 lesion may warrant a less aggressive approach. This nuanced, risk-adapted strategy embodies the true added value of the integrated model—not necessarily in terms of a statistically higher area under the curve (AUC), but as a more refined and personalized clinical decision-making tool. We acknowledge that the current study predominantly utilizes traditional statistical methods and ultrasound-based risk stratification tools. Nonetheless, the field of predictive modeling in obstetrics and gynecology is undergoing rapid advancements, propelled by machine learning (ML) and artificial intelligence (AI) methodologies. For example, in the area of ovarian tumor classification, deep learning models, particularly convolutional neural networks (CNNs), have demonstrated exceptional performance in certain studies, often outperforming conventional approaches [ 31 ]. Additionally, the incorporation of generative adversarial networks (GANs) has been proposed as a solution to the challenge of data scarcity for rare tumor subtypes, such as borderline ovarian tumors [ 32 ]. Moreover, the advancement of fully automated systems, exemplified by the ADNEX-AI model that autonomously extracts ultrasound features from images for ovarian cancer risk stratification, underscores a distinct progression towards more objective and scalable diagnostic tools [ 33 ]. These contemporary methodologies, frequently enhanced by ensemble techniques and exhibiting superior calibration and clinical utility through decision curve analysis, contrast with the single-center, manually derived model discussed in our study. Our study is subject to several limitations. The retrospective, single-center design may introduce selection bias and restrict the generalizability of our findings. Although the sample size is considerable, it should be expanded in future multi-center prospective studies to validate the proposed algorithm. An additional limitation is the quasi-complete separation caused by the perfect prediction of O-RADS category 2 (118/118 benign). This limited the stability of bootstrap internal validation, as many resampled models failed to converge. Notably, our conclusions regarding positive and negative predictive values (PPV and NPV) are specific to this high-risk surgical cohort; these metrics may vary in populations with lower disease prevalence, such as those encountered in primary screening or general gynecological practice. Consequently, the reported predictive values should be validated in external, unselected cohorts. Future research should investigate the integration of these tools with emerging biomarkers or advanced imaging techniques, as well as the application of AI-driven approaches, to further refine preoperative diagnosis. Recently, Ronsini et al. reported that systemic inflammatory indices (SIRI and SIR) demonstrated promising diagnostic performance in differentiating ovarian tumor types, outperforming CA125 in terms of specificity and predictive value [ 34 ]. While our study focused on the integration of O‑RADS with traditional serum biomarkers (HE4 and CA125), these findings suggest that combining ultrasound features with inexpensive inflammatory markers could represent a complementary, cost‑effective strategy for preoperative risk assessment, particularly in resource‑limited settings or for early‑stage tumors.

Ovarian cancer (OC), representing malignancies originating from the ovary, ranks as the eighth in terms of both most frequently diagnosed malignancy and leading cause of oncological mortality in women worldwide [ 1 ]. In 2022, ovarian cancer was diagnosed in approximately324,398individuals, and 206,839 died of ovarian cancer worldwide [ 2 ]. Epithelial malignancies comprise 90% of ovarian cancers, with 70% to 80% being high-grade serous ovarian cancers [ 3 ]. Incidental adnexal masses are commonly discovered at ultrasound, CT, and MRI. In premenopausal women undergoing surgical resection of an adnexal mass, the frequency of malignancy is below 10%; In postmenopausal women undergoing surgical resection of an adnexal mass, the frequency is slightly increased, but still less than 15%; Accurate diagnosis is crucial [ 4 ]. US is the first-line imaging modality for assessment of ovarian and adnexal lesions. Over the last several decades, several classification systems have been proposed for the characterization of adnexal masses to improve risk stratification and to triage management options. The European International Ovarian Tumor Analysis (IOTA) proposed Logistic Regression model in 2005 and, subsequently, proposed ADNEX model in 2014, add 3 clinical indicators. IOTA models are used widely in Europe. Then in 2020, with support from the American College of Radiology, a joint team of experts created the O-RADS system. It standardizes key terms and classifies risk into six categories to guide patient management. This was later updated in 2022 [ 5 – 10 ].In recent years, a risk of ovarian malignancy algorithm (ROMA) has been developed by combining serum cancer antigen 125 (CA125), human epididymis protein 4 (HE4), and menopausal status. This algorithm significantly improves the sensitivity and specificity of ovarian cancer diagnosis [ 11 , 12 ].Therefore, this study aims to investigate whether the combination of O-RADS ultrasound classification and tumor marker detection can better predict benign and malignant ovarian tumors, with the goal of assisting clinicians in formulating more appropriate treatment plans.