Endometrial carcinoma is one of the three major malignant tumors of the female reproductive system, primarily originating from the endometrial epithelium [1] . It commonly occurs in postmenopausal women, with a higher incidence in the age group of 50 to 64 years. Clinically, patients often present with symptoms such as vaginal bleeding, abnormal vaginal discharge, and lower abdominal pain. In contrast, submucosal uterine fibroids are uterine fibroids that protrude into the uterine cavity, typically manifesting as abnormal uterine bleeding, dysmenorrhea, menorrhagia, prolonged menstruation, and potential complications such as anemia and infertility. Therefore, accurate differentiation between endometrial carcinoma and submucosal uterine fibroids is crucial before clinical treatment [2] .

Magnetic resonance imaging (MRI) has been widely recognized as a primary imaging modality for endometrial lesions, particularly in diagnosis and staging [3] . Studies have shown that MRI features, including lesion morphology, signal intensity, functional imaging, and enhancement characteristics, provide essential clinical evidence for distinguishing benign from malignant tumors. However, there is some overlap in imaging features between endometrial carcinoma and submucosal uterine fibroids, making their differentiation challenging [4] - [6] . Additionally, the treatment strategies for these two conditions differ significantly: endometrial carcinoma typically requires total hysterectomy and bilateral salpingo-oophorectomy (with optional pelvic and para-aortic lymphadenectomy), while submucosal fibroids can be managed with hysteroscopic myomectomy [7] [8] . Therefore, it is particularly crucial to explore the application of radiomics of CE-T1WI and DWI in the differential diagnosis of endometrial carcinoma and submucous myoma of the uterus. The three-dimensional imaging diagrams of endometrial carcinoma and submucous myoma of the uterus are shown in Figures 1(A)-(D) .

In recent years, with the continuous development of MRI technology, diagnostic methods based on morphology, functional features, and molecular characteristics have been widely applied. Radiomics increasingly employs techniques such as feature redundancy reduction, dimensionality reduction, and preprocessing, combined with machine learning classification methods, to objectively extract high-throughput features from images. These methods help establish robust and clinically relevant prognostic models, providing quantitative and objective support for tumor detection and treatment decision-making [9] [10] . This study aims to utilize CE-T1WI and DWI radiomics models to effectively differentiate between endometrial carcinoma and submucosal uterine fibroids.

In the arterial phase of CE-T1WI, 24 out of 33 endometrial carcinoma cases showed enhancement lower than the myometrium, 7 cases showed slightly lower enhancement, 1 case showed slightly higher enhancement, and 1 case showed higher enhancement. In contrast, among 30 submucosal fibroid cases, 4 showed lower enhancement, 8 showed slightly lower enhancement, 1 showed equal enhancement, 12 showed slightly higher enhancement, and 5 showed higher enhancement. Statistical analysis revealed significant differences between the two groups in terms of lower enhancement (χ² = 22.4509, P = 0.0000) and slightly higher enhancement (χ² = 13.1146, P = 0.0003).

In the venous phase, 25 endometrial carcinoma cases showed lower enhancement, 6 showed slightly lower enhancement, 2 showed equal enhancement, and none showed slightly higher or higher enhancement. Among submucosal fibroid cases, 16 showed lower enhancement, 3 showed slightly lower enhancement, 4 showed equal enhancement, 6 showed slightly higher enhancement, and 1 showed higher enhancement. The difference in higher enhancement between the two groups was statistically significant (χ² = 5.1583, P = 0.0231). ( Table 1 , Table 2 )

Regarding DWI findings, 25 out of 33 endometrial carcinoma cases showed restricted diffusion, 8 showed partial restriction, and none showed unrestricted diffusion. Among 30 submucosal fibroid cases, none showed restricted diffusion, 9 showed partial restriction, and 21 showed unrestricted diffusion. The difference in diffusion restriction between the two groups was highly significant (χ² = 37.6794, P = 0.0000). Among cases with ADC values recorded, the mean ADC value in the endometrial carcinoma group (0.71 ± 0.22) was significantly lower than that in the submucosal fibroid group (1.34 ± 0.28) (χ² = 7.1828, P = 0.0000). No significant difference was observed in time-signal curve types between the two groups (χ² = 0.6969, P = 0.4107). The imaging features and time-intensity curves (absolute signal intensity) of endometrial carcinoma and submucous myoma of the uterus are shown in Figure 2 and Figure 3 , and the specific data are shown in Tables 3-5 .

Note: For the comparison of the number of cases with enhancement higher than and lower than the myometrium between the two groups, χ² = 20.0629, P = 0.0000.

Note: The comparison of cases with enhancement higher and lower than the myometrium between the two groups yielded χ² = 7.1795, P = 0.0074.

Note: The overall comparison of diffusion restriction between the two groups yielded χ² = 34.6500, P = 0.0000.

Note: Generally, when the ADC value is <1, the likelihood of malignancy is higher, and when the ADC value is >1, the likelihood of a benign condition is higher. In this study, only 9 cases in the submucosal uterine fibroid group showed partial diffusion restriction, and all had ADC values greater than 1.

Accurate diagnosis of submucosal uterine fibroids and endometrial carcinoma is crucial for developing appropriate treatment strategies. This study demonstrates the significant value of CE-T1WI and DWI radiomics features in their diagnosis.

In terms of CE-T1WI enhancement characteristics, our findings are consistent with previous studies [11] [12] . Endometrial carcinoma, with its abundant but immature neovascularization and high vascular permeability, often shows slower contrast agent uptake in the arterial phase, resulting in lower enhancement compared to the myometrium [13] . In contrast, submucosal fibroids, with their rich and regular vascular supply, exhibit diverse enhancement patterns, sometimes higher or slightly higher than the myometrium [14] . In the venous phase, as contrast distribution stabilizes, endometrial carcinoma typically shows lower enhancement, while submucosal fibroids exhibit more complex enhancement patterns, with a higher proportion of cases showing enhancement above the myometrium [15] .

DWI, as a functional imaging technique, reflects the microscopic motion of water molecules in tissues. Endometrial carcinoma, with its high cellular density and small extracellular space, often shows significant diffusion restriction [16] [17] . ADC values quantitatively reflect the degree of water molecule diffusion, with values > 1 typically indicating benign lesions and values < 1 suggesting malignancy [18] . In this study, the ADC values in the endometrial carcinoma group were significantly lower than those in the submucosal fibroid group, further confirming the diagnostic value of DWI [19] [20] .

However, no significant difference was observed in time-signal curve types between the two groups, possibly due to the relatively small sample size and tumor heterogeneity [21] . Different tumor cell types, growth patterns, and vascular supply may contribute to the diversity of time-signal curves, necessitating larger sample sizes and more detailed subgroup analyses [22] . Additionally, radiomics feature extraction and analysis are influenced by various factors, including scanner field strength, sequence parameters, contrast agent dosage, and injection rate [23] [24] . Although we standardized scanning conditions, subtle differences between scanners may still affect result accuracy [25] . Furthermore, the feature selection and model construction methods used in this study may not be optimal, and future research could explore alternative algorithms and strategies to improve diagnostic accuracy and stability [26] [27] .

In clinical practice, our findings provide new imaging evidence for differentiating submucosal uterine fibroids from endometrial carcinoma. Combining CE-T1WI and DWI radiomics features may improve diagnostic accuracy and reduce unnecessary invasive procedures [28] [29] . However, challenges such as data standardization and feature interpretability remain, requiring further research and exploration [30] .

This study demonstrates that CE-T1WI and DWI radiomics features can effectively differentiate submucosal uterine fibroids from endometrial carcinoma, providing valuable support for clinical diagnosis. However, the limited sample size necessitates further validation with larger cohorts. Optimizing scanning protocols and feature extraction processes will enhance result accuracy and reproducibility, ultimately improving clinical utility.

The small sample size may limit the generalizability of the results. Subtle differences in scanning equipment may affect radiomics feature extraction. The feature selection and model construction methods used may not be optimal, potentially limiting model performance and diagnostic accuracy.

We sincerely thank all individuals and institutions who supported this study. Special thanks to the patients and their families for their cooperation, the hospital imaging department staff for their technical assistance, and the experts who provided valuable guidance.

*Co-first authors.

^#Corresponding author.