Preoperative contrast-enhanced CT images of patients with histopathologically confirmed PCLs at Peking Union Medical College Hospital from June 1, 2014 to May 31, 2023 were retrospectively collected. The data were randomly partitioned at the lesion level into training, validation, and test sets in a 3∶1∶1 ratio. CT images were preprocessed and regions of interest were delineated. Using postoperative pathological results as the reference standard, five deep learning models(ResNet50, DenseNet121, ResNeXt50, EfficientNet-b5, and MobileNetV2) were constructed to extract arteriovenous dual-phase CT imaging features for binary classification of PCLs. The optimal model was selected based on the area under the curve(AUC), accuracy, sensitivity, and specificity. Further comparative analyses were conducted against single-phase CT-based deep learning models, conventional radiomics models, and radiologist interpretations to comprehensively evaluate the performance of the arteriovenous dual-phase CT-based deep learning model in the differential diagnosis of PCLs.

A total of 480 patients with 485 lesions(206 malignant and 279 benign) were ultimately enrolled. The training, validation, and test sets comprised 291, 97, and 97 lesions, corresponding to 288, 96, and 96 patients, respectively. In the validation set, the ResNeXt50 model based on arteriovenous dual-phase CT features achieved an AUC of 0.837(95% CI: 0.748-0.915), an accuracy of 77.32%(95% CI: 67.70%-85.21%), a sensitivity of 80.49%(95% CI: 65.13%-91.18%), and a specificity of 75.00%(95% CI: 61.63%-85.61%). In the test set, the corresponding values were 0.822(95% CI: 0.737-0.904), 73.20%(95% CI: 63.24%-81.68%), 82.93%(95% CI: 67.94%-92.85%), and 66.07%(95% CI: 52.19%-78.19%), demonstrating overall superior performance. Calibration curves indicated that the predicted probabilities of the model were generally consistent with observed outcomes, albeit with certain shortcomings in probability calibration. Decision curve analysis demonstrated that, overall, the use of this model for clinical decision-making conferred a net benefit to patients. Compared with single-phase CT-based models and conventional radiomics models, the ResNeXt50 model incorporating arteriovenous dual-phase CT features exhibited superior overall performance, and its diagnostic performance was comparable to that of radiologists, with good consistency.

The deep learning model based on arteriovenous dual-phase CT imaging features demonstrates diagnostic value in differentiating benign from malignant PCLs and may serve as an adjunctive reference for preoperative clinical assessment. However, its stability and generalizability warrant further validation in larger cohorts and with external datasets.

Patients diagnosed with EC at Peking Union Medical College Hospital from January 2015 to December 2021 were retrospectively enrolled and randomly divided into a training set and a validation set at a ratio of 8∶2. Relevant clinical data were collected, and radiomics features and DL features were extracted from preoperative contrast-enhanced T1-weighted imaging(CE-T1WI), fat-suppressed T2-weighted imaging(fs-T2WI), and diffusion-weighted imaging(DWI) sequences. High-dimensional feature selection and dimensionality reduction were performed sequentially using the recursive feature elimination(RFE) algorithm to generate a radiomics score(Rad-score) and a deep learning score(DL-score), respectively. Multivariate logistic regression was utilized to construct a clinical model, a pure radiomics model, a clinical-radiomics model, and an integrated multimodal model incorporating clinical indicators, Rad-score, and DL-score. Model performance was assessed and compared using area under receiver operating characteristic curve(AUC) and DeLong test.

A total of 509 patients were enrolled in this study, comprising 413 in the training cohort and 96 in the validation cohort. Independent predictors: Multivariate analysis indicated that preoperative fasting blood glucose level, histological grade, lymph node metastasis status, Rad-score, and DL-score were all independent significant predictors of MMRd status in EC patients. The integrated multimodal model demonstrated optimal predictive performance with an AUC of 0.699(95% CI: 0.635-0.763) in the training set, which was superior to the clinical model(AUC=0.629, 95% CI: 0.561-0.697) and the pure radiomics model(AUC=0.641, 95% CI: 0.575-0.706). In the validation set, the integrated model maintained good generalizability, achieving an AUC of 0.655(95% CI: 0.535-0.775), and its diagnostic efficacy was higher than that of the clinical model(AUC=0.578, 95% CI: 0.450-0.705) and the pure radiomics model(AUC=0.611, 95% CI: 0.488-0.734). According to the DeLong test, the incorporation of DL features resulted in the clinicalradiomicsdeep learning model performing better than both the clinicalonly model(P=0.027) and the radiomicsonly model(P=0.044) in the training cohort.

The initially developed clinical-radiomics-deep learning model exhibits a certain predictive potential for the MMRd status in patients with EC. The inclusion of DL features may help complement the limitations of traditional evaluations, offering a preliminary radiological reference for preoperative non-invasive screening. However, given the current diagnostic performance, its overall accuracy and clinical generalizability warrant further validation in multi-center, large-sample external cohort studies.