The application value of two-dimensional ultrasound combined with contrast-enhanced ultrasound in the differential diagnosis of benign, borderline, and malignant ovarian epithelial tumors.

As shown in Table  1 , patients with BOTs are young women of childbearing age, whereas patients with malignant epithelial ovarian tumors are typically perimenopausal women. In addition, the CA125 levels in the BOTs group were less elevated than those in the malignant group. Table 1 Clinical characteristics of patients, serum CA125 levels, and pathological classification of ovarian tumors Pathological Benign BOTs Malignant Years 44.96 ± 2.71 36.18 ± 1.58 a 47.96 ± 1.79 b BMI 23.77 ± 1.18 22.94 ± 0.68 23.70 ± 0.64 Pre-menopausal /Post-menopausal 17/6 33/1 a 32/19 b Pathological classification Serous 14 17 32 Mucinous 5 3 2 Seromucinous 4 13 — Clear Cell — — 9 Endometrioid — 1 8 CA125 (IU/ml) < 35 23 16 11 ≥ 35 1 18 40 CA125 (IU/ml) 21.08 ± 4.08 67.34 ± 16.84 a 433.1 ± 99.33 a, b a P <0.05 BOTs compared with Benign, b P <0.05 Malignant compared with BOTs Clinical characteristics of patients, serum CA125 levels, and pathological classification of ovarian tumors Pre-menopausal /Post-menopausal a P <0.05 BOTs compared with Benign, b P <0.05 Malignant compared with BOTs Furthermore, in clinical practice, the normal range for CA125 is 0–35 U/ml, but this rang was limited in distinguishing between BOTs and malignant tumors. In this study, we attempted to use the ROC curve to analyze the optimal cutoff value for differentiating between BOTs and malignant tumors. The results revealed that when the optimal cutoff value was 77.3 U/ml, the corresponding sensitivity was 84.8%, the specificity was 66.7%, and the area under the ROC curve was 0.775. The difference was statistically significant ( P  < 0.05). Pathological examination revealed 23 benign tumors, 34 BOTs, and 51 malignant tumors among the 108 ovarian masses. As shown in Table  2 , by comparing the resistance index (RI) values of the cystic wall and solid areas of the ovarian masses in the three groups of patients, we found that the average RI value was 0.55 ± 0.03 in the benign group, 0.45 ± 0.02 in the BOTs group, and 0.45 ± 0.01 in the malignant group. The difference between the benign and BOTs groups was statistically significant ( P  < 0.05), whereas the difference between the malignant and BOTs groups was not statistically significant. Additionally, we observed that tumors in all three groups were predominantly < 10 cm in maximum diameter, with no statistically significant differences. Table 2 2D-US diagnostic results and Sonographic features of 108 cases of epithelial ovarian tumors (numbers of cases) Pathological Benign BOTs Malignant 2D  Benign  BOTs  Malignant 2 1 0 20 30 7 1 3 44 Color score  1  2  3  4 1 2 0 11 14 9 6 10 11 5 8 31 RI 0.55 ± 0.03 0.45 ± 0.02 a 0.45 ± 0.01 a Septal  Yes/No 11/12 20/14 32/19 Solid component  Yes/No 19/4 25/9 50/1 a, b Papillary  Yes/No 15/8 24/10 15/36 a, b Papillary (number)  0–3  ≥ 4 18 23 41 5 11 10 Maximum diameter  < 10 cm  ≥ 10 cm 16 24 33 7 10 18 O-RADS score  2  3  4  5 0 1 0 0 2 1 18 15 22 5 16 28 a P <0.05 compared with benign; b P <0.05 compared with BOTs 2D-US diagnostic results and Sonographic features of 108 cases of epithelial ovarian tumors (numbers of cases) Benign BOTs Malignant 1 2 3 4 0–3 ≥ 4 < 10 cm ≥ 10 cm 2 3 4 5 a P <0.05 compared with benign; b P <0.05 compared with BOTs Furthermore, we performed O-RADS scoring on the ovarian masses of 108 patients. We found that most patients in each group scored 4 or 5 on O-RADS, with no significant statistical differences between the groups. This may be due to selection bias in case selection, as the patients included in this study were mainly those highly suspected of having BOTs or malignant ovarian masses on 2D-US. On the basis of the quantitative analysis of the time-intensity curves (TIC), we found that compared with those of the normal uterine myometrium, the perfusion patterns of the benign ovarian tumor group were mainly showed late enhancement, low enhancement, and synchronous washout, overall presenting a relatively uniform low enhancement. In the solid areas of cystic-solid lesions or within large papillae, punctate enhancement could be observed. In contrast, the BOTs group exhibited mainly late enhancement, low enhancement, and early washout. Additionally, BOTs generally have relatively complete capsules. In the solid areas of cystic-solid lesions and within large papillae, some vessels showed irregular thickening and disorderly branching, presenting a dendritic pattern, with irregular borders showing serrations. There was no statistically significant difference between the two groups in terms of the perfusion phase or enhancement intensity during perfusion, whereas there was a statistically significant difference in the washout phase (χ2 = 5.956, P  < 0.05). The contrast enhancement pattern of malignant tumors mainly manifested as early enhancement, high enhancement, and early washout. The tumor showed overall heterogeneous high enhancement, with rough and irregular borders. Within the solid areas of cystic-solid lesions or within large papillae, there were large and disorderly vessels, with contrast agent enhancement appearing radially from the inside out. Compared with those of the BOTs, the perfusion phase (χ2 = 11.069, P  < 0.05) and enhancement intensity (χ2 = 6.644, P   0.05) (Fig.  2 ; Table  3 ). Fig. 2 Typical imaging features, TIC curves, and pathological results of benign, BOTs, and malignant ovarian epithelial tumors. A 30-year-old woman with a benign tumor. ( A ) 2D-US revealed an anechoic cystic cavity in the right ovary with three papillae on the cyst wall. The largest papillary was 9 mm in height. ( B ) CEUS showed contrast agent perfusion within the papillary region. ( C ) TIC analysis indicated that the papillary region exhibited late enhancement, low enhancement, and synchronous washout, suggesting benignity. ( D ) Pathology results confirmed it as a serous papillary cystadenoma. A 30-year-old woman with a BOTs. ( E ) 2D-US showed a cystic-solid mass in the left ovary with a diameter of 4.8 cm, and the solid area had a diameter of 3.6 cm. ( F ) CEUS showed contrast agent perfusion within the solid area. ( G ) TIC analysis indicated that the solid area exhibited late enhancement, low enhancement, and early washout, suggesting a borderline nature. ( H ) Pathology results confirmed it as a borderline serous-mucinous tumor. Malignant group, a 63-year-old woman. ( I ) 2D-US showed a cystic-solid mass in the right ovary with a diameter of 3.2 cm, and the solid area had a diameter of 1.3 cm. ( J ) CEUS showed contrast agent perfusion within the solid area. ( K ) TIC analysis indicated that the solid area exhibited early enhancement, high enhancement, and early washout, suggesting malignancy. ( L ) Pathology results confirmed it as high-grade serous carcinoma. Red arrow: uterus, yellow arrow: ovarian lesion, green arrow: small intestine Typical imaging features, TIC curves, and pathological results of benign, BOTs, and malignant ovarian epithelial tumors. A 30-year-old woman with a benign tumor. ( A ) 2D-US revealed an anechoic cystic cavity in the right ovary with three papillae on the cyst wall. The largest papillary was 9 mm in height. ( B ) CEUS showed contrast agent perfusion within the papillary region. ( C ) TIC analysis indicated that the papillary region exhibited late enhancement, low enhancement, and synchronous washout, suggesting benignity. ( D ) Pathology results confirmed it as a serous papillary cystadenoma. A 30-year-old woman with a BOTs. ( E ) 2D-US showed a cystic-solid mass in the left ovary with a diameter of 4.8 cm, and the solid area had a diameter of 3.6 cm. ( F ) CEUS showed contrast agent perfusion within the solid area. ( G ) TIC analysis indicated that the solid area exhibited late enhancement, low enhancement, and early washout, suggesting a borderline nature. ( H ) Pathology results confirmed it as a borderline serous-mucinous tumor. Malignant group, a 63-year-old woman. ( I ) 2D-US showed a cystic-solid mass in the right ovary with a diameter of 3.2 cm, and the solid area had a diameter of 1.3 cm. ( J ) CEUS showed contrast agent perfusion within the solid area. ( K ) TIC analysis indicated that the solid area exhibited early enhancement, high enhancement, and early washout, suggesting malignancy. ( L ) Pathology results confirmed it as high-grade serous carcinoma. Red arrow: uterus, yellow arrow: ovarian lesion, green arrow: small intestine Table 3 Comparison of CEUS modes and imaging results among three groups of epithelial ovarian tumors(numbers of cases) CEUS Modes Perfusion phase Enhancement intensity Regression phase Perfusion Coarse disorganized vessels Sum Early Synchronous Late High Equal Low Early Synchronous Late Uniform/Non-uniform Yes/No Pathological classification Benign 5 0 18 3 1 19 23 12 1 7/16 21/2 23 BOTs 7 3 24 11 4 19 34 7 0 3/31 17/17 34 malignant 29 3 19 31 3 17 51 9 0 8/43 27/24 51 χ 2 (a) 2.935 3.029 5.956 4.429 10.533 P (a) 0.231 0.220 0.035 0.035 0.001 χ 2 (b) 11.069 6.644 0.115 0.853 0.071 P (b) 0.004 0.036 0.782 0.356 0.790 χ 2 (c) 8.663 12.051 16.035 2.133 10.236 P (c) 0.013 0.001 < 0.001 0.144 0.001 Pathological diagnosis(numbers of cases) Benign BOTs Malignant sum Results of CEUS Benign 11 2 2 15 BOTs 11 28 9 48 Malignant 1 4 40 45 sum 23 34 51 108 a P <0.05 BOTs VS Benign; b P <0.05 Malignant VS BOTs; c P <0.05 Malignant VS Benign Comparison of CEUS modes and imaging results among three groups of epithelial ovarian tumors(numbers of cases) a P <0.05 BOTs VS Benign; b P <0.05 Malignant VS BOTs; c P <0.05 Malignant VS Benign As shown in Table  4 ; Fig.  3 , the sensitivity, specificity, accuracy, and AUC values of 2D-US combined with CEUS for the diagnosis of benign vs. BOTs, BOTs vs. malignant, and benign vs. malignant epithelial ovarian tumors were higher than those of single diagnostic techniques. Fig. 3 ROC curves for diagnosing benign, BOTs, and malignant ovarian epithelial tumors using 2D-US, CEUS, and their combined application. The ROC curves of 2D-US, CEUS, and their combination in diagnosing ovarian epithelial tumors were analyzed for differentiating between ( A ) benign tumor and BOTs, ( B ) BOTs and malignant tumor, and ( C ) benign and malignant tumor. ROC, receiver operating characteristic, 2D-US, two-dimensional ultrasound; CEUS, contrast-enhanced ultrasound; and BOTs, borderline ovarian tumors ROC curves for diagnosing benign, BOTs, and malignant ovarian epithelial tumors using 2D-US, CEUS, and their combined application. The ROC curves of 2D-US, CEUS, and their combination in diagnosing ovarian epithelial tumors were analyzed for differentiating between ( A ) benign tumor and BOTs, ( B ) BOTs and malignant tumor, and ( C ) benign and malignant tumor. ROC, receiver operating characteristic, 2D-US, two-dimensional ultrasound; CEUS, contrast-enhanced ultrasound; and BOTs, borderline ovarian tumors Table 4 Diagnostic efficacy of 2D-US, CEUS, and combined diagnosis for benign, BOTs, and malignant epithelial ovarian tumors Sen Spe PV+ PV- Accuracy AUC 95%CI Benign VS BOTs 2D 9.1 96.8 66.7 60.0 60.3 0.548 0.394–0.701 CEUS 50.0 93.0 85.0 71.8 75.0 0.720 0.577–0.862 2D + CEUS 52.3 1.0 1.0 75.0 80.3 0.733 0.594–0.873 BOTs VS Malignant 2D 90.9 78.4 81.1 93.6 88.1 0.889 0.812–0.967 CEUS 87.5 81.6 75.7 90.9 84.0 0.822 0.727–0.917 2D + CEUS 1.0 91.7 88.2 1.0 94.9 0.920 0.853–0.986 Benign VS Malignant 2D 66.7 1.0 1.0 97.8 97.8 0.916 0.844–0.987 CEUS 97.1 95.2 84.6 97.6 94.4 0.903 0.829–0.978 2D + CEUS 1.0 1.0 1.0 1.0 1.0 0.968 0.935-1.000 Sen: sensitivity; Spe: specificity; PV+: positive predictive value; PV−: negative predictive value Diagnostic efficacy of 2D-US, CEUS, and combined diagnosis for benign, BOTs, and malignant epithelial ovarian tumors Benign VS BOTs BOTs VS Malignant Benign VS Malignant Sen: sensitivity; Spe: specificity; PV+: positive predictive value; PV−: negative predictive value In this study, we utilized patients’ 2D-US and Doppler ultrasound images, including blood flow signals, RI values, presence of papillary structures, presence of solid components, and maximum diameter of tumors, as well as CEUS patterns, uniformity of contrast agent perfusion, presence of large and disordered vessels, etc., for modeling. Using pathological diagnosis as the gold standard, the random forest model automatically selected and ranked various influencing factors. As shown in Fig.  4 , the RI, maximum diameter of the lesion, presence of cystic or solid components, presence of papillary structures, and color score are all important characteristics of 2D ultrasound, with a total score of 0.31. The uniformity of contrast agent perfusion, the presence of coarse and disordered blood vessels, and contrast enhancement mode are all important characteristics of CEUS, with a total score of 0.27. This finding indicates that 2D ultrasound is the foundation for disease diagnosis, whereas CEUS is a valuable complement. Additionally, it is necessary to consider the patient’s serological markers, age, BMI, and other general characteristics. For individual features, we found that the importance of contrast enhancement patterns was the highest (22.0%). The model achieved an accuracy of 76.47%, precision of 88.24%, recall rate of 76.47%, and F1 score of 0.76. Fig. 4 Prediction model for benign, BOTs, and malignant ovarian epithelial tumors based on random forest algorithm Prediction model for benign, BOTs, and malignant ovarian epithelial tumors based on random forest algorithm

The precise preoperative diagnosis of the benign or malignant nature of ovarian masses is crucial for determining treatment plans and selecting surgical approaches for patients. However, this binary classification overlooks the importance of BOTs. Epidemiological studies have shown that the incidence of BOTs is increasing annually, and their proportion among malignant tumors is also increasing [ 25 ]. BOTs are more common in young women and are closely associated with the occurrence of infertility [ 26 ]. Additionally, patients with BOTs have a better prognosis and are less likely to undergo radical surgical treatment [ 5 ]. In this study, compared with patients in the benign and malignant tumor groups, those in the BOTs group were significantly younger, with a median age of 35 years, and predominantly childbearing age. CA125 is commonly used as a serum marker for the diagnosis and prognostic assessment of epithelial ovarian malignant tumors [ 27 ]. Currently, the clinical standard sets the threshold at 35 U/ml, where ≥ 35 U/ml is considered positive and < 35 U/ml is considered negative. This method is primarily used to differentiate between benign and malignant tumors, with limited value in distinguishing between BOTs and malignant tumors. In this study, we found that CA125 levels were significantly lower in the benign group, higher in the malignant group, and intermediate in the borderline group, with statistically significant differences. By conducting ROC curve analysis on CA125 levels in the borderline and malignant groups, we obtained high sensitivity and specificity at the optimal cutoff value of 77.3 U/ml. The differential diagnosis of ovarian masses relies mainly on gynecological ultrasound. Previous studies have indicated that the typical ultrasound imaging features of BOTs include cysts containing papillary projections, with detectable blood flow signals within the papillary projections. However, an increasing number of studies have shown that this feature is also found in some benign and malignant epithelial ovarian tumors and cannot be considered a characteristic ultrasound manifestation of BOTs [ 28 , 29 ]. Consistent with those finds, in this study, we found no significant differences in the maximum diameter of the tumors or the presence of septations within the tumors among the three groups. Additionally, the majority of tumors in all three groups were categorized as O-RADS 4–5, with some also falling into categories 2 and 3, with no significant differences between the groups. This may be due to selection bias in case selection in this experiment, where cases included in CEUS were mainly those highly suspected of BOTs or malignant ovarian masses under 2D-US. Additionally, internal hemorrhage, deposits, necrotic components, and other changes in benign ovarian masses can also manifest as solid-like components or septate changes under 2D-US, which may also lead to a higher O-RADS classification. Given the complexity of ovarian tissue, the diversity of ovarian tumor types, and the fact that some tumors exhibit nonspecific sonographic features, there is a phenomenon where similar images can correspond to different diseases, or different images can correspond to the same disease. This is especially true for borderline ovarian tumors, where the ultrasound characteristics overlap significantly with those of benign tumors and invasive malignant tumors. In this study, CEUS technology was used to dynamically observe the blood flow and microvascular perfusion in the cyst wall, septa, and solid components, as well as to analyze the perfusion patterns to aid in for accurate preoperative diagnosis of epithelial ovarian tumors. With the rapid development of CEUS technology, the accuracy of ultrasound diagnostics has improved significantly [ 16 , 30 ]. CEUS technology is widely regarded as a safe and reliable diagnostic method [ 31 ], as it does not involve ionizing radiation or nephrotoxicity [ 14 ]. Unlike color Doppler, which focuses on the major vascular network within tumors, CEUS pays more attention to the microvascular structures inside tumors, providing a significant advantage in diagnosing ovarian tumors [ 18 ]. In this study, we used CEUS to observe and quantitatively analyze blood flow perfusion within ovarian masses. We found that in the benign tumor group, the papillae, septa, or solid areas exhibited mainly late enhancement, low enhancement, and synchronous washout, indicating overall homogeneous enhancement with generally no coarse or disordered vessels. The BOTs group primarily exhibited late enhancement, low enhancement, and early washout, showing overall slightly heterogeneous enhancement with occasional irregular vessels. In contrast, the malignant tumor group predominantly exhibited early enhancement, high enhancement, and early washout, showing overall heterogeneous enhancement with coarse and disordered vessels visible within. This may be because the papillae or solid components within benign ovarian masses are hemorrhages, deposits, or necrotic organized components rather than true papillae and therefore have few or no blood vessels, resulting in less and slower contrast agent perfusion. On the other hand, the dense areas of new blood vessels within the solid areas or septa of malignant tumors present a sudden increase in the number of new blood vessels, with larger diameters and higher local microvascular density, easily forming vascular rings or networks. These vessels have thinner walls and increased permeability, leading to faster blood flow and greater blood perfusion [ 32 ], resulting in the fast flow in and out of contrast agent. Compared with the pathological results, the concordance rate of 2D-US was 65.4%, whereas the concordance rate of CEUS was 75.9%, which was significantly higher than that of 2D-US. Other studies have also shown that CEUS has high sensitivity and specificity in diagnosing benign and malignant ovarian tumors. For example, Dutta et al. reported a sensitivity of 100% and a specificity of 96.6% [ 33 ]. Through a comprehensive literature review, Ma et al. confirmed a sensitivity of 93% and a specificity of 95% [ 17 ]. Wu et al., through a meta-analysis, confirmed a sensitivity of 89% and a specificity of 91% [ 34 ]. Qiao reported a sensitivity of 96% and a specificity of 91% for diagnosing ovarian cancer [ 31 ]. These findings are generally consistent with the conclusions of our study. In our present study, we found that the sensitivity and specificity of CEUS in diagnosing benign and malignant epithelial ovarian tumors were 97.1% and 95.2%, respectively. When combined with 2D-US, the sensitivity and specificity increased to 100%. Additionally, our study revealed that the combined use of 2D-US and CEUS also showed high sensitivity and specificity in distinguishing benign tumors from BOTs and BOTs from malignant epithelial ovarian tumors. Finally, by constructing a predictive model for benign, borderline, and malignant ovarian tumors based on the random forest algorithm, it was found that the CEUS pattern had the largest weight, accounting for 22% of all cases. In summary, the results of this study suggest that for ovarian epithelial tumors suspected to be BOTs or malignant on the basis of 2D-US, a CEUS examination is recommended for further diagnosis. A comprehensive evaluation should also consider the patient’s age, CA125 level, and intratumoral blood flow signals. For ovarian tumors with an O-RADS score of 4–5, if the patient’s CA125 level is mildly elevated and CEUS shows late enhancement, low enhancement, and early washout in women of reproductive age, the diagnosis is more likely to be BOTs. If the CA125 level is normal and CEUS shows late enhancement, low enhancement, and synchronous washout, a benign diagnosis is more likely. Conversely, for perimenopausal women with significantly elevated CA125 level and CEUS findings showing early enhancement, high enhancement, and early washout, a malignant diagnosis is more likely. However, this study was a retrospective analysis, and some patients were subsequently reviewed at other hospitals. Therefore, the study does not have clear information on patients’ later-stage metastasis and recurrence. Additionally, owing to the limitations in sample selection and the number of cases, there are several shortcomings that need further investigation. In the future, the use of both transabdominal and transvaginal 2D-US with CEUS technology can be employed to assess peritoneal implants and metastasis with the aim of guiding preoperative staging and the selection of surgical methods.

Ovarian epithelial tumors (OETs) account for two-thirds of all ovarian tumors, and more than 90% of ovarian cancers are malignant [ 1 , 2 ]. OETs are mainly divided into benign, borderline ovarian tumors (BOTs), and malignant. Accurate preoperative diagnosis is crucial for selecting optimal treatment plans and determining patient prognosis. For benign epithelial ovarian tumors, direct tumor resection is commonly performed, and the prognosis is generally good. BOTs account for 10–20% of all epithelial ovarian tumors [ 3 ], with a 10-year survival rate of up to 97% [ 4 ], and patients rarely require radical surgery [ 5 ]. However, malignant epithelial ovarian tumors require cytoreductive surgery and additional treatments such as radiotherapy and chemotherapy. Therefore, identifying effective methods for the accurate early-stage diagnosis of ovarian tumors is highly important. Ultrasonography, a first line imaging technique, is used to diagnose adnexal masses [ 6 ]. Each mass should be evaluated for the presence of septations, a solid component, and blood flow using Doppler imaging [ 7 ]. In 2014 and 2018, the International Ovarian Tumor Analysis (IOTA) and American College of Radiology introduced the ADNEX and O-RADS models, which utilize two-dimensional ultrasound (2D-US) and provide a straightforward method for distinguishing between benign and malignant ovarian tumors [ 8 – 11 ]. However, there is a high degree of overlap on 2D-US images of BOTs and malignant epithelial ovarian tumors [ 10 , 12 ]. This is particularly true in the differential diagnosis of cystic-solid and solid tumors. Additionally, there are no imaging features indicative of early stage of BOTs, so only 29–69% of borderline ovarian tumors are accurately diagnosed before surgery [ 13 ]. Therefore, another method for accurately distinguishing between benign tumors and BOTs and between BOTs and malignant tumors is urgently needed. Contrast-enhanced ultrasound (CEUS) is commonly considered a dependable and convenient diagnostic imaging method for patients with ovarian tumors and is adept at visualizing small arterial vessels with a diameter of less than 100 micrometers. CEUS is also regarded as a relatively safe technique that is devoid of ionizing radiation and a risk of nephrotoxicity [ 14 , 15 ]. CEUS involves injecting a mixture of contrast agent containing microbubbles into the vein and continuously monitoring the enhancement and washout of the contrast agent within the lesion in real time [ 16 ]. By observing the perfusion phase, enhancement intensity, washout time, uniformity of perfusion, and presence of coarse disordered vessels within the lesion tissue, among other factors [ 17 ], quantitative analysis plays a significant role in identifying ovarian malignancies at their earliest stages [ 18 ]. Previous studies have revealed distinct CEUS patterns in different types of ovarian tumors. For example, benign tumors typically exhibit ring-like enhancement in the cystic wall and papillae [ 19 ], whereas malignant ovarian cancer exhibit rapid heterogeneous enhancement [ 20 ]. However, some studies have presented conflicting results. Furthermore, few studies have confirmed the efficacy of CEUS in distinguishing BOTs and what distinguishes the imaging patterns of BOTs from those of benign and malignant tumors. Carbohydrate antigen 125 (CA125) is one of the most commonly used gynecological tumor markers, especially because of its high sensitivity in diagnosing ovarian epithelial tumors [ 21 ]. However, its specificity for individual diagnosis is relatively low, as elevated levels can also occur in benign conditions such as endometriosis [ 22 ]. Studies have shown that the combined assessment of CA125 and vaginal ultrasound images can improve the accuracy of ovarian tumor diagnosis [ 23 , 24 ]. However, the role of the combined assessment of 2D-US images, CEUS images, and CA125 in the differential diagnosis of benign tumors, BOTs, and malignant ovarian epithelial tumors is still unknown. Therefore, our study retrospectively assessed the diagnostic efficacy of CEUS imaging in characterizing ovarian epithelial tumors, using histological findings as the gold standard reference. Furthermore, the aim of our study was to enhance the differential diagnostic efficacy of epithelial ovarian tumors by integrating CEUS images with 2D-US images and serum CA125 levels.

The flow chart of this study was shown in Fig.  1 . Fig. 1 The flow chart of this study. 2D, two-dimensional ultrasound, CEUS, contrast-enhanced ultrasound The flow chart of this study. 2D, two-dimensional ultrasound, CEUS, contrast-enhanced ultrasound 108 patients with 2D-US suspected BOTs and malignant ovarian epithelial tumors were subjected to CEUS examination in our study from December 2018 to November 2023. The patients’ ages ranged from 19 to 71 years. The average ages of the patients with benign, BOTs, and malignant epithelial ovarian tumors were 44.96 ± 2.71 years (range from 19 to 68 years, the median age was 47 years), 36.18 ± 1.58 years (range from 22 to 63 years, the median age was 35 years), 47.96 ± 1.79 years (range from 21 to 71 years, the median age was 48 years), respectively. This study was approved in advanced by the Ethical Committee of The Peking University People’s Hospital. And at the time the patients were examined, they consented to having their data used for future retrospective research. Transvaginal or transabdominal grayscale and color Doppler sonography were performed with a Nuewa R9 Exp ultrasonic diagnostic apparatus. The frequency of the transvaginal transducer was 3.0–10.0 MHz, and that of the transabdominal transducer was 1.0–5.0 MHz. Typically, transvaginal probes are used for ovarian tumor examination. If the mass is too large to completely view using the transvaginal probe, a transabdominal probe can be used for examination, or both can be used. The maximum diameter of the lesion, whether it is cystic or solid, unilocular or multilocular cystic, whether the cyst wall is smooth or irregular, whether septations are present, if there are solid components, the number and height of the solid components, and the color score of the solid components should be determined using 2D-US imaging. Diagnosis criteria: Two experienced gynecological ultrasound physicians with the title of deputy chief physician or above shall perform O-RADS scoring on 2D-US images before surgery, and the patient with the highest score shall be the final score. CEUS examinations equipped with transvaginal and transabdominal transducers were performed on a Nuewa R9 Exp ultrasonic diagnostic apparatus in all patients. First, we focus on the section of interest, such as the solid part of the tumor and the dense area of the nipple or septum. At the same time, some of the uterine muscle layers can be displayed as a control. Then, a SonoVue contrast agent suspension was prepared, 1.8-2.0 ml was extracted, and the mixtures was injected through the elbow vein. The tube was quickly flushed with 5 ml of physiological saline, and the real-time storage function was activated. The samples were observed and recorded for at least 90 s. During this period, the probe should not be moved arbitrarily, and the patient needs to breathe normally. Finally, the images were quantitatively analyzed using the Q-LAB software provided with the ultrasound instrument, the uterine muscle layer and the area where the contrast agent was first injected were selected, and a time-intensity curve (TIC) was drawn. perfusion pattern of contrast agent, including perfusion time, enhancement intensity and washout phase, morphology of microvasculature, and uniformity of contrast agent perfusion, as well as the presence of large and disordered branching vessels. Due to the large volume of most ovarian lesions, it is usually not possible to use their own normal ovarian tissue as a control. Therefore, in this study, we used autologous normal uterine muscle layer as a control. By comparing the perfusion phase, enhancement intensity, and washout time of the contrast agent in the ovarian mass and the uterine muscle layer, the perfusion phase was divided into early enhancement, synchronous enhancement, and late enhancement. The enhancement intensity was divided into high enhancement, equal enhancement, and low enhancement. The regression phase was divided into early washout, synchronous washout, and late washout. Compared to a single decision tree, the random forest algorithm performs better and can effectively prevent overfitting. In this study, the proportion of the training set was 0.8, the number of trees was set to 70, the maximum depth of the trees was unlimited. The criterion for node splitting was the Gini index, the minimum number of samples required to split a node was 2, and the minimum number of samples for a leaf node was 1. Furthermore, the bootstrapping and out-of-bag data testing were also included. The code for the model is as follows: Model = Random Forest Classifier (criterion = ‘gini’, max_depth = None, min_samples_leaf = l, min_samples_split = 2, n_estimators = 70, bootstrap = True, oob_score = True, r model.fit (x_train, y_train). The “score” in the Fig. 3 represents the importance of each feature in our random forest model. Specifically, it is derived using the Mean Decrease in Impurity (MDI) method. This method calculates the importance of a feature based on the total decrease in Gini impurity (for classification tasks) or mean squared error (for regression tasks) that the feature brings about across all the trees in the random forest. SPSS v.26.0 statistical software was used for data analysis. Quantitative data are expressed as mean ± standard deviation. For qualitative data, 𝜒2 test and rank sum test were performed. Student t test was used to compare the parameters of benign versus BOTs, begin versus malignant ovarian, and BOTs versus malignant ovarian tumor. 𝑃 < 0.05 was considered statistically significant. ROC curve analysis was used to examinate the diagnostic efficacy of CEUS, 2D-US, and their combined application in ovarian epithelial tumors. Constructed the prediction models of ovarian benign, borderline, and malignant using CEUS, 2D-US, et al. based on random forest algorithm.