Endometriosis is characterized by ectopic implantation or metaplastic proliferation of endometrial cells, coupled with persistent ectopic growth and inflammation resulting from impaired immune function. Histopathological examination requires the presence of at least two of the following features: endometrioid glands, endometrioid stroma, or evidence of chronic hemorrhage [11]. Histologically, atypical endometriosis was first proposed by Czernobilsky and Morris in 1979 [12], defined as the presence of cytological and structural abnormalities in the epithelium of endometriotic cysts without stromal invasion, characterized by cytological atypia (nuclear stratification, hyperchromasia, and pleomorphism) and architectural atypia (simple or complex hyperplasia). Approximately 0.7% - 2.5% of atypical endometriosis may undergo malignant transformation [13]; among these, architectural atypia is primarily associated with ovarian cancer, and its Ki-67 expression is significantly higher than that of cytological atypia [14]. However, the pathological diagnostic criteria for AE have not yet been standardized, and clinical detection rates vary widely (0.8% - 8.8%) [13][15], with clinicians having insufficient awareness of its malignant transformation risk.

Among all subtypes of ovarian cancer, clear cell carcinoma has the closest association with endometriosis [16]. The WHO defines OCCC as a malignant ovarian tumor composed of clear, eosinophilic, and hobnail cells arranged in tubulocystic, papillary, and solid patterns. Grossly, it typically presents as a unilateral, large (3 - 30 cm in diameter, most commonly 13 - 15 cm) cystic-solid mass, with white, yellow, or brown polypoid nodules visible on the inner cyst wall; the cyst wall shows a typical chocolate-brown color and is often associated with endometriotic lesions [17]. Microscopically, the histopathology mainly manifests as three cell types: the clear cell type, rich in glycogen with clear cytoplasm, which constitutes the main body of solid nodules and is the typical cell of OCCC; the hobnail cell type, with nuclei protruding toward the glandular lumen and scant cytoplasm, showing characteristic hobnail morphology, which is the hallmark pathological cell of OCCC and the key morphological feature for pathological diagnosis; and the eosinophilic cell type, with eosinophilic and red-stained cytoplasm, abundant mitochondria, and often intermingled with clear cells [18].

Molecular pathological studies have further revealed the association between the two conditions. Inactivation of the ARID1A tumor suppressor gene is the initiating core event in the malignant transformation of endometriosis, and its mutation rate in atypical endometriosis and OCCC is significantly higher than that in typical endometriosis. Wiegand et al. [19] found that 46% of OCCC cases harbored ARID1A mutations, and that this mutation and loss of BAF250a protein expression were present in both the tumor and the surrounding atypical endometrial tissue, but not in distant normal endometrium, suggesting that ARID1A mutation is a key event in the early transformation of endometrial lesions to ovarian cancer; studies by Ayhan and Xiao et al. [20][21] have also confirmed this conclusion.

Meanwhile, the “retrograde menstruation—iron overload—oxidative stress—proliferative signal activation” axis also constitutes a core driver of malignant transformation in endometriosis. Retrograde menstruation causes repeated bleeding from ectopic endometrium; macrophages in the peritoneal cavity phagocytose and degrade large numbers of red blood cells, releasing excessive free iron that leads to peritoneal iron overload. Iron overload generates large amounts of reactive oxygen species (ROS) through pathways such as the Fenton reaction, causing persistent oxidative stress, which in turn leads to cellular damage and an increased risk of DNA mutations, and activates the NF-κB and ERK/MAPK pathways, conferring a proliferative advantage on ectopic cells [22]. In addition, activating mutations of PIK3CA, MET gene amplification, RASSF methylation, and loss of PTEN and KRAS collectively drive abnormal cellular proliferation and malignant transformation [23]-[27].

OCCC is a special subtype of ovarian cancer, and its clinical features also differ from those of other subtypes. The mean age at onset of OCCC is 53.5 - 56 years; most cases are unilateral; approximately 60% - 70% are diagnosed at FIGO stage I-II; the mean tumor diameter is 10 - 15 cm; and OCCC is often associated with endometriosis [5][28]-[31]. In addition, patients with OCCC have a higher incidence of venous thromboembolism (VTE), twice that of other ovarian cancer subtypes; approximately 10% are accompanied by paraneoplastic syndromes; and serum CA125 is mostly mildly to moderately elevated [32]. Clinical stage is the most important prognostic factor; the 5-year overall survival rates for stages I, II, III, and IV are 88%, 70%, 33%, and 0%, respectively, and platinum resistance in advanced disease is the main cause of poor prognosis [33]. Liver metastasis is a unique high-risk prognostic factor for OCCC, distinct from high-grade serous carcinoma [34].

In contrast, in the study by So KA et al. [15], patients with AE had a mean age of 34.8 years, a mean cyst diameter of 7.2 cm, and were mostly unilateral. Risk factors for malignant transformation of endometriosis include advanced age (40 - 60 years), cyst diameter > 9 cm, postmenopausal status, and long-term estrogen exposure [35][36]. The following clinical situations should raise suspicion of malignant transformation: age ≥ 45 years, postmenopausal status, disease duration of endometriosis ≥ 10 years, change in pain rhythm (dysmenorrhea transitioning to chronic pelvic pain), cyst diameter ≥ 8 cm, imaging findings suggesting solid or papillary structures with abundant low-resistance blood flow, and concurrent endometrial lesions [37]. These clinical features themselves serve as important clues for differential diagnosis and should be incorporated into a comprehensive assessment system.

Differences in Serum Tumor Markers

Serum tumor markers are also among the important features for the differential diagnosis of the two conditions. CA125 has high sensitivity but low specificity, and may be elevated in endometriosis, pelvic inflammatory disease, and even during menstruation [38]. HE4 has superior specificity to CA125 (93% vs. 78%), is not affected by the menstrual cycle, and is elevated in approximately 69% of patients with clear cell carcinoma [39]-[41]; urinary HE4 can serve as a supplement to serum testing and may help identify patients with primary platinum resistance [42]. The combined detection of CA125 and HE4 constitutes the ROMA (Risk of Ovarian Malignancy Algorithm) index; a study by Suri et al. [43] showed that ROMA is the best diagnostic marker for epithelial ovarian cancer in postmenopausal women, whereas HE4 is the best predictive indicator in premenopausal women.

Smac is a pro-apoptotic protein. Dobrzycka et al. [44] first detected its levels in the serum of patients with ovarian cancer in 2015, and its levels were negatively correlated with tumor stage and pathological grade. A study by Xu XR et al. [45] showed that serum Smac in OCCC was significantly lower than in benign endometriosis; the sensitivity of Smac alone reached 91.67%, and the specificity of HE4 reached 98.75%. The three-marker combination (Smac + HE4 + CA125) had the highest sensitivity (98.33%) but lower specificity (75%), while the combination of Smac and HE4 offered a relatively better balance of sensitivity and specificity. A study by Xu T et al. [46] demonstrated that HE4 > 59.7 pmol/L and CA19-9 > 8.5 U/mL are independent risk factors for malignant transformation of endometriosis, with the combined detection achieving a sensitivity of 72.73% and specificity of 78.41%. In addition, an elevated neutrophil-to-lymphocyte ratio (NLR) is an independent predictor of poor prognosis in OCCC [47].

Current serum marker assays offer some assistance in the differential diagnosis of OCCC and AE, but they also have limitations. CA125 may be elevated in both conditions, with limited discriminatory capacity; HE4 has relatively high specificity but insufficient sensitivity; Smac, as an emerging marker, has high sensitivity, but related studies have small sample sizes and lack multicenter validation. In real clinical scenarios, one should not rely solely on serum markers, but rather integrate them with ultrasound findings and clinical features into a combined model, which is also one of the multimodal diagnostic strategies. A comparison of the clinical and pathological features of OCCC and is atypical ovarian endometrioma presented in Table 1.

Table 1.Comparison of clinicopathological features between OCCC and atypical ovarian endometrioma.

Transvaginal ultrasound (TVU) is the first-line imaging modality for ovarian lesions, with diagnostic sensitivity and specificity both approaching 90% [48]. The typical ultrasound features of OMA include a unilocular cyst, ground-glass echogenicity within the cyst, positive tenderness on probe compression, and rare papillary projections or solid components [49][50]. In contrast, AE presents with atypical features such as multilocular cysts, thickened cyst walls, papillary projections on the cyst wall, and cystic-solid lesions [49][50].

An analysis of ultrasound features in 152 cases of OCCC by Pozzati et al. [31] showed: 84.2% were unilateral; the median maximum tumor diameter was 117 mm; solid components were visible in all lesions (median solid nodule diameter 69 mm); 34.9% were unilocular cystic-solid, 41.4% multilocular cystic-solid, and 23.7% solid masses; papillary projections were visible in 38.2% (of which 51.7% had >3 projections and 91.1% showed detectable blood flow); and cyst fluid was most commonly hypoechoic (43.5%). Among these, patients with OCCC pathologically confirmed to have arisen from malignant transformation of endometriosis were younger (median 47.5 years vs. 55.0 years) and more frequently showed ground-glass echogenicity (50.0% vs. 16.5%).

A comparative study of 63 cases of AE and 57 cases of OCCC by Huang KJ et al. [51] showed that patients with OCCC were older (53.11 ± 9.28 years vs. 39.33 ± 7.04 years), had larger cyst diameters (12.68 ± 4.60 cm vs. 7.81 ± 2.81 cm), larger solid components (4.82 ± 3.53 cm vs. 0.93 ± 1.74 cm), and more frequent disappearance of ground-glass echogenicity (68.4% vs. 6.3%). Multivariate analysis showed that age > 47.5 years, cyst diameter > 11.55 cm, solid component > 1.37 cm, and disappearance of ground-glass echogenicity were independent factors suggestive of malignancy.

Among the above ultrasound features, the disappearance of ground-glass echogenicity is a noteworthy sign. Some studies have speculated that locally expressed degradative enzymes within endometriotic cysts gradually break down the cyst fluid, while periodic bleeding replenishes the cyst contents, maintaining the typical ground-glass echogenicity; after perimenopause, fresh bleeding decreases, and the typical ground-glass echogenicity may disappear [52][53]. After adjusting for age, the disappearance of ground-glass echogenicity in individuals under 45 years of age may be an important indicator for differentiating AE from OCCC [51]. From a pathological perspective, the cystic-solid structure of OCCC corresponds to solid echogenic areas within a cystic mass on ultrasound; the cyst fluid appears chocolate-like due to old hemorrhage, explaining the fine punctate or mixed echogenicity on ultrasound; tumor cells rich in glycogen arranged in solid sheets constitute the main body of solid nodules, and abundant eosinophilic hyaline material in the stroma may form sclerotic hyaline nodules, which may appear as hyperechoic spots or calcification-like echoes within solid areas on ultrasound; papillary projections and hobnail cells protruding into the lumen form intracystic wall nodules visible on ultrasound [18]. The main conventional ultrasonographic features differentiating OCCC from atypical ovarian endometrioma are shown in Table 2.

Ultrasound is an important modality for the differential diagnosis of OCCC and AE, with cyst size, solid component diameter, and changes in ground-glass echogenicity being the main differentiating features. However, these signs are not absolute and may overlap between patients with OCCC and AE, particularly in perimenopausal women, where the disappearance of ground-glass echogenicity may merely reflect changes in cyst fluid composition rather than malignant transformation. Therefore, ultrasound should serve as a screening tool, and atypical cases require further evaluation in combination with new ultrasound technologies and MRI.

Table 2.Key conventional ultrasonic features for differentiating OCCC from atypical ovarian endometrioma.

Color Doppler ultrasound can further improve the differential diagnostic performance for the two conditions by assessing the microcirculatory perfusion and hemodynamic features of lesions. Blood flow signals in AE lesions are mostly confined to the cyst wall or septations, presenting as scant punctate or linear signals, and solid components are mostly avascular; in contrast, blood flow in OCCC lesions is predominantly distributed within solid components and wall nodules, manifesting as relatively abundant linear blood flow signals, consistent with tumor neovascularization.

Regarding hemodynamic parameters, OCCC lesions are mostly low-resistance, with a resistive index (RI) mostly < 0.4, whereas AE lesions mostly have an RI > 0.6. However, RI may be influenced by factors such as tumor size, lesion location, and measurement technique, limiting its differential diagnostic value. In addition, power Doppler ultrasound is more sensitive in detecting low-velocity blood flow and can identify neovascularization within small solid components, offering supplementary value. Overall, the septations and intracystic solid components of AE cysts are mostly avascular or show only scant blood flow signals; the presence of vascularized solid papillary structures is highly suggestive of malignancy. Furthermore, Doppler interpretation is substantially influenced by operator experience, equipment performance, and parameter settings. In clinical differential diagnosis, caution should also be exercised with benign lesions such as decidualized endometriomas during pregnancy, which may also show hypervascular solid projections internally and can be easily confused with malignant lesions [50].

4.1.1. Technical Principles and Basic Workflow

In recent years, with the rapid development of the integration of artificial intelligence technology and medical imaging, radiomics has emerged as a new field and research hotspot in quantitative assessment techniques for precision medical imaging. The concept of radiomics was first proposed by the American scholar Gillies [54] in 2010, and Lambin [55] further refined its definition in 2012. It refers to a technology that uses computers to extract a large number of quantitative features carried in medical images through high-throughput processing and obtains information reflecting the underlying pathophysiology of lesions through machine learning analysis [54][55]. Ultrasound radiomics applies radiomics methods to ultrasound images, extracting quantitative features such as morphology, texture, and wavelet transform from grayscale ultrasound, and capturing subtle image structures and intrinsic relationships that are difficult to observe with the naked eye [56][57]. Its basic workflow comprises five main steps: medical image acquisition, region of interest (ROI) segmentation, feature extraction, feature selection, and model construction and validation. Under the supervised learning framework, commonly used feature selection methods include LASSO regression and mRMR, and classifiers include support vector machines (SVM), random forests, logistic regression, and convolutional neural networks (CNN) [58][59].

4.1.2. Diagnostic Performance of Radiomics Based on Grayscale Ultrasound

In a large-sample study of 6809 pathologically confirmed cases of endometriosis by Chao et al. [60], a gradient boosting decision tree was used to construct a prediction model, achieving an AUC of 0.942 in the test set, with a sensitivity of 86.8% and specificity of 86.7%, significantly outperforming the traditional logistic regression model (AUC = 0.891); moreover, the AUC remained at 0.876 in a prospective dataset. In a benign-malignant prediction model for adnexal masses developed by Moro et al. [61] based on 2428 patients, the clinical-radiomics combined model constructed using the XGBoost algorithm achieved the best diagnostic performance, with a validation set AUC of 0.89, sensitivity of 0.83, and specificity of 0.81.

Miao et al. [62] used the ConvNeXt deep learning algorithm to differentiate OMA from benign mucinous cystadenoma, enrolling 184 patients; the test set AUC reached 0.90, with sensitivity and specificity both at 90%, and decision curve analysis (DCA) confirmed its clinical utility. Liu et al. [63] developed a machine learning model to differentiate OMA from benign teratoma; the logistic regression model achieved an AUC of 0.981, accuracy of 94.8%, and sensitivity of 95.5%; a nomogram model integrating clinical and radiomics features achieved an AUC of 0.987. The diagnostic performance of this radiomics model was comparable to that of senior physicians and superior to that of junior physicians; moreover, after using it for assisted diagnosis, the performance of junior physicians significantly improved. The above studies have all confirmed the excellent performance of ultrasound radiomics in the differential diagnosis of gynecological diseases.

4.1.3. Prediction Models Combining Radiomics with Clinical Features

Prediction models combining radiomics with clinical features have also demonstrated certain value in the early identification of OCCC. The study by Moro et al. [61] showed that the clinical-radiomics combined model (AUC 0.89) outperformed the radiomics-only model, confirming the value of multimodal information fusion. The nomogram model by Liu et al. [63] (AUC 0.987) also outperformed the radiomics-only model. When imaging data are combined with clinical, genomic, and pathological information, the predictive performance of these models is significantly enhanced. This also applies to the differentiation of OCCC and AE; integrating clinical features such as age, cyst size, and serum markers with ultrasound radiomics features holds promise for constructing more accurate early malignant transformation risk prediction models.

Existing research on ultrasound radiomics in gynecological diseases has established a certain foundation, but studies directly targeting the specific differential diagnosis of OCCC and AE remain very limited. Most studies simply include OCCC within the broad category of ovarian malignant tumors, failing to capture its unique radiomics features. As a special subtype of ovarian cancer, OCCC has certain unique clinical features, and its radiomics features may also differ from those of other types of ovarian cancer. Future studies specifically focusing on the radiomics of OCCC and AE are needed.

4.2.1. Elastography

The concept of ultrasound elastography was first proposed by Ophir et al. [64] in 1991, inspired by clinical palpation. It utilizes differences in stiffness among biological tissues for coded imaging and elasticity measurement. Common modalities include strain elastography (SE), acoustic radiation force impulse imaging (ARFI), and shear wave elastography (SWE) [65]. Among these, SWE is the most widely used. It generates transverse shear waves by transmitting excitation acoustic beams from the probe; utilizing the Mach cone effect, shear waves propagate faster in stiffer tissues and slower in softer tissues, allowing color coding based on tissue stiffness. Young’s modulus values can also be directly calculated using the Young’s modulus formula to reflect tissue stiffness [66][67].

Herek et al. [68] investigated the diagnostic value of transabdominal ultrasound elastography for ovarian lesions, showing that ovarian endometriomas and complex cystic lesions tend to be stiff with high strain ratios; malignant lesions are mostly soft with very low strain ratios; and benign tumors have moderate stiffness, overall stiffer than malignant lesions. A transvaginal elastography study by Ciledag et al. [69] also showed that OMA overall presents as stiff due to old intracystic hemorrhage and fibrous deposition; simple physiological cysts are extremely soft; and complex cystic malignant lesions, due to intracystic solid papillary and infiltrative components, mostly have elastic stiffness between that of simple cysts and OMA.

The combination of elastography and conventional ultrasound can improve differential diagnostic performance, particularly offering supplementary value in determining the benign or malignant nature of solid components. Within the solid components of OCCC, tumor cells are loosely arranged with stromal edema, resulting in soft texture; in contrast, the cyst walls and septations of AE show increased stiffness due to long-term hemorrhagic organization and fibrous tissue proliferation. This difference in stiffness endows ultrasound elastography with certain value in the differential diagnosis of OCCC and AE.

4.2.2. Contrast-Enhanced Ultrasound

Contrast-enhanced ultrasound (CEUS) is a noninvasive imaging technique that applies acoustic contrast agents on the basis of conventional ultrasound and identifies masses by observing microvascular perfusion patterns. CEUS examination includes the assessment of parameters such as tumor enhancement pattern (homogeneous/heterogeneous), degree of enhancement (hypo-/iso-/hyperenhancement), vascular morphology (presence of coarse, disorganized, and irregular vessels within the lesion), and contrast agent washout time.

Generally, there are marked differences in CEUS enhancement patterns between benign and malignant lesions: malignant lesions show fast-in and fast-out enhancement, heterogeneous enhancement, and visible disorganized neovascular features, whereas benign lesions mostly show homogeneous enhancement without abnormal perfusion. Benign ovarian tumors typically show ring enhancement at the cyst wall and papillary projections, whereas ovarian cancer manifests as rapid and heterogeneous enhancement. The solid portions of OCCC are rich in vessels, but abundant hyaline material deposition in the stroma may form hypovascular fibrotic areas, which on contrast imaging may manifest as heterogeneous hypoenhancement or a differential pattern of fast-in and slow-out [18].

A CEUS study of benign, borderline, and malignant ovarian epithelial tumors by Wang et al. [70] showed that the benign group was characterized by late enhancement, low enhancement, and synchronous washout, with overall homogeneous mild enhancement; the borderline group was characterized by late enhancement, low enhancement, and early washout, with relatively intact capsules; the malignant group manifested as early enhancement, high enhancement, and early washout, with overall heterogeneous high enhancement, abundant coarse and disorganized vessels visible within solid areas, and radial enhancement of contrast agent extending outward from the interior of the lesion. A study by Tian Cai et al. [71] found that the combination of IOTA (International Ovarian Tumor Analysis) simple rules with CEUS significantly improved the diagnostic value for adnexal masses compared with IOTA rules alone.

CEUS has important supplementary value in the identification of small solid components and early malignant transformation lesions. For small nodules that cannot be clearly identified by two-dimensional and color Doppler ultrasound, CEUS can detect low-velocity blood flow and display the microvascular perfusion pattern within solid components, providing key information for the identification of early malignant transformation.