Objective This study aims to systematically evaluate and compare the diagnostic performance of various artificial intelligence approaches, including traditional radiomics, ResNet50, morphological features, Vision Transformer （ViT）, and a ViT model integrated with morphological features （ViT+shape）, in differentiating orbital solitary fibrous tumor （SFT） from two common benign focal masses （cavernous hemangioma and schwannoma）. The study seeks to identify the model with optimal generalizability and, based on this model, to further optimize the combination of MRI sequences, while evaluating the added value of contrast-enhanced imaging in this differential diagnosis. Methods This study retrospectively included 520 patients with surgically and pathologically confirmed focal orbital masses, comprising an internal cohort of 327 patients （69 SFT, 156 cavernous hemangiomas, and 102 schwannomas） and an external cohort of 193 patients （25 SFT, 113 cavernous hemangiomas, and 55 schwannomas）. The study cohort was divided into a training set, an internal test set, and an external test set. We constructed five classification models based on conventional radiomics, ResNet50, shape features （Shape）, ViT, and ViT+Shape, and comprehensively evaluated their generalization performance on the internal test set and the independent external test set. Each model was trained and evaluated for three-class classification under five sequence combinations: T₁WI; T₂WI; the non-contrast combination of T₁WI and T₂WI; contrast-enhanced T₁WI （CE-T₁WI）; and the full-sequence combination of T₁WI, T₂WI, and CE-T₁WI. Model performance was comprehensively assessed using macro-average AUC （AUC_macro）, micro-average AUC （AUC_micro）, per-class precision/recall/F1-score, Cohen's Kappa coefficient, and confusion matrices, with the DeLong test combined with Bonferroni correction used for statistical comparison of AUC between models. Results In the multi-sequence combined analysis of the internal test set, the Shape model achieved the highest AUC_macro （0.856）, followed by ViT+Shape （0.849）, while ResNet50 had the lowest （0.560）. After Bonferroni correction, ResNet50 demonstrated significantly lower AUC for SFT compared with Shape, ViT, and ViT+Shape （DeLong test, Bonferroni correction, all P < 0.05）. In the external test set, the performance of the radiomics model declined substantially, with the AUC_macro dropping to 0.495 and the Kappa coefficient to only 0.044. The recall for cavernous hemangioma was merely 13.3%, and the confusion matrix revealed that 57.5% of cases were misclassified as SFT. In contrast, the ViT+Shape model performed robustly, achieving an AUC_macro of 0.838 and a Kappa coefficient of 0.433. The addition of contrast-enhanced imaging to the triple-sequence combination showed only marginal enhancement in diagnostic performance compared with the non-contrast dual-sequence combination. In the external test set, the ViT+Shape model demonstrated the most significant improvement, with the macro-AUC rising from 0.810 to 0.838 under the triple-sequence combination. Nevertheless, this difference did not reach statistical significance after Bonferroni correction （P > 0.05）. Conclusion The ViT+Shape model, constructed with prior morphological knowledge, demonstrates high accuracy and robustness in differentiating SFT, cavernous hemangioma, and schwannoma, with stable and reliable performance across external validation. Notably, the model achieves comparable diagnostic performance and clinical utility using only non-contrast T₁WI and T₂WI sequences, comparable to the full-sequence protocol that includes contrast-enhanced imaging. This provides a feasible technical approach to reduce unnecessary contrast-enhanced examinations.

Objective To investigate the predictive value of intratumoral and peritumoral CT-based radiomic features for distinguishing benign from malignant Lung-RADS 4X pulmonary nodules, and to develop a combined model integrating CT morphological features to evaluate its diagnostic performance. Methods Preoperative non-contrast CT images of 323 Lung-RADS 4X pulmonary nodules were retrospectively collected. Regions of interest （ROIs） were delineated as intratumoral, intratumoral + 3 mm peritumoral, and intratumoral + 5 mm peritumoral regions. Radiomic features were extracted and radiomics scores （Rad-scores） were constructed. Logistic regression was used to build radiomics models and a combined model. Results The intratumoral model, intratumoral + 3 mm peritumoral model, and intratumoral + 5 mm peritumoral model achieved AUCs of 0.688, 0.714, and 0.696 in the validation set, respectively. The combined model demonstrated the optimal diagnostic performance, with AUCs of 0.882 and 0.835 in the training and validation sets, respectively. Conclusion Radiomics models incorporating peritumoral regions, particularly the 3 mm area, hold significant value in differentiating benign from malignant Lung-RADS 4X pulmonary nodules. Further integration with CT morphological features can enhance the predictive performance of the model.

Objective To screen independent risk factors for invasiveness of lung adenocarcinoma, and to construct and validate a clinical-radiomics-deep learning multimodal joint prediction model, thereby providing a radiologic basis for precise preoperative risk stratification of lung adenocarcinoma. Methods This study included patient cohorts with ground-glass nodules （GGNs） from two hospitals. The samples were divided into a training set, an internal validation set, and an external test set at an appropriate ratio to ensure sample representativeness and rigor of the validation process. Univariable and multivariable Logistic regression analyses were used to screen independent predictive factors. A clinical model, a radiomics model, a deep transfer learning model, a radiomics-deep transfer learning fusion model, and a clinical-radiomics-deep learning combined model were respectively constructed. Model discrimination was evaluated using ROC curves and the DeLong test, calibration was assessed using calibration curves, clinical net benefit was evaluated using decision curve analysis （DCA）, and a visual nomogram was constructed based on the combined model. Results Age, sex, smoking history, mean CT value, and the proportion of non-solid component were identified as independent risk factors for the invasiveness of lung adenocarcinoma. The combined model showed significant advantages in calibration, clinical net benefit, and visualization, with AUC values of 0.974 （95% CI: 0.961-0.988） in the training set, 0.806 （95% CI: 0.719-0.892） in the internal validation set, and 0.872 （95% CI: 0.799-0.946） in the external test set. The DeLong test demonstrated that the combined model significantly outperformed the clinical model alone in predictive performance. Calibration curves showed good calibration of the combined model across all three cohorts, with high agreement between predicted probabilities and observed risks. DCA confirmed a higher clinical net benefit for the combined model over a wide range of threshold probabilities. The nomogram based on the combined model facilitated individualized visual risk assessment. Conclusions The clinical-radiomics-deep learning combined model significantly outperformed single-modality models in predictive performance, calibration, and clinical utility. The radiomics-deep transfer learning fusion model is preferable for achieving optimal predictive accuracy. Together, these models offer reliable imaging evidence for precise stratification of lung adenocarcinoma invasiveness and individualized management of early-stage lung cancer, demonstrating substantial clinical value.

Objective To investigate the value of a LASSO regression model based on clinical and conventional enhanced CT imaging features in differentiating benign from metastatic mediastinal lymph nodes. Methods A retrospective analysis was conducted on 166 patients with pathologically confirmed mediastinal lymphadenopathy (83 benign, 83 malignant) treated at Guangzhou Chest Hospital from January 2023 to December 2025. Nineteen clinical indicators (age, gender, smoking history, smoking index, clinical symptoms, inflammatory markers, IGRA, CEA, CYFRA21-1, NSE, CA125, CA19-9, etc.) and 13 CT imaging features (lymph node size, morphology, density, enhancement pattern, etc.) were collected. A clinical model, an imaging model, and a combined model were constructed using LASSO regression, with the optimal penalty parameter λ.1se determined by 10-fold cross-validation. Model discrimination was internally validated using the Bootstrap method (200 repetitions). An independent external validation was performed using data from the First Affiliated Hospital of Guangzhou Medical University to assess model generalizability. AUC differences between models were compared using the DeLong test, and a calibration curve and a decision curve were plotted to evaluate the calibration and clinical net benefit of the combined model. Results The combined model retained 18 predictive variables (age, fever, smoking history, smoking index, lymphocyte count, hs-CRP, IGRA, CEA, CYFRA21-1, CA125, maximum short axis diameter of lymph node, long/short diameter ratio, margin, lymph node hilum, density homogeneity, necrosis/cystic change, calcification, and venous phase CT value). Bootstrap validation showed that the combined model achieved a mean AUC of 0.945 (95% CI: 0.920-0.981), a sensitivity of 95.2%, and a specificity of 96.4%. The DeLong test showed that the full-dataset AUC of the combined model (0.975) was significantly higher than that of the clinical model (0.943, P = 0.025) and that of the imaging model (0.834, P < 0.001). External validation demonstrated that the combined model achieved an AUC of 0.881 (95% CI: 0.802-0.960), an accuracy of 84.2%, a sensitivity of 71.1%, and a specificity of 97.4%, confirming good model generalization. The Hosmer-Lemeshow test yielded P = 0.204; the calibration curve demonstrated good agreement between predicted and observed probabilities, and the decision curve indicated that the combined model yielded higher clinical net benefit than either the clinical or imaging model alone. Conclusion The LASSO regression combined model based on clinical and CT imaging features demonstrates excellent diagnostic performance in differentiating benign from metastatic mediastinal lymph nodes and can provide quantitative support for clinical non-invasive decision-making.

Objective To investigate the differential diagnostic value of a multiparametric model based on gadoxetic acid disodium-enhanced MRI (EOB-MRI) for hepatobiliary phase (HBP) hyperintense hepatocellular carcinoma (HCC) and focal nodular hyperplasia (FNH). Methods Clinical and imaging data of patients with HBP-hyperintense focal liver lesions who underwent EOB-MRI at three hospitals between January 2018 and December 2024 were retrospectively analyzed, including 52 patients with HCC and 81 patients with FNH. Clinical characteristics, diffusion-related quantitative parameters, dynamic enhancement patterns, and HBP morphological features were compared between the two groups. Multivariate Logistic regression analysis was performed to identify independent predictors for diagnosing HBP-hyperintense HCC, and a combined diagnostic model was constructed. Receiver operating characteristic (ROC) curve analysis was used to evaluate the diagnostic performance of each parameter and the combined model, followed by Bootstrap internal validation and calibration assessment. Results The apparent diffusion coefficient (ADC) value [(1.21±0.24)×10⁻³ mm²/s vs. (1.43±0.29)×10⁻³ mm²/s] and relative apparent diffusion coefficient (rADC) value (0.74±0.12 vs. 1.48±0.34) were significantly lower in the HCC group than in the FNH group (both P < 0.001). The frequencies of wash-in and wash-out enhancement, peritumoral hypointense rim, heterogeneous iso-/hyperintensity on HBP, fatty change, and intralesional necrosis/hemorrhage were significantly higher in the HCC group than in the FNH group (all P < 0.05), whereas the frequencies of wash-in and slow-out enhancement, central scar, and doughnut-like hyperintensity were significantly higher in the FNH group than in the HCC group (all P < 0.05). After adjustment for age, sex, and liver cirrhosis, multivariate Logistic regression analysis showed that rADC ≤ 1.08 (OR = 6.75, 95% CI: 2.36-19.30), wash-in and wash-out enhancement (OR = 4.76, 95% CI: 1.69-13.39), peritumoral hypointense rim (OR = 5.99, 95% CI: 1.94-18.51), and heterogeneous iso-/hyperintensity on HBP (OR = 6.80, 95% CI: 2.00-23.10) were independent predictors for diagnosing HBP-hyperintense HCC (all P < 0.05). The combined model incorporating these four parameters achieved the highest diagnostic performance, with an area under the curve (AUC) of 0.912 (95% CI: 0.861-0.963), a sensitivity of 90.4%, a specificity of 88.9%, and an accuracy of 89.5%. Bootstrap internal validation showed a corrected AUC of 0.894, and the calibration curve demonstrated good agreement between the predicted probability and the actual diagnostic outcome (Hosmer-Lemeshow test: χ² = 6.37, P = 0.61). Conclusion The EOB-MRI-based multiparametric model is useful for differentiating HBP-hyperintense HCC from FNH. The combination of rADC value, dynamic enhancement pattern, and HBP morphological features can improve diagnostic performance and provide valuable imaging evidence for preoperative risk stratification and clinical decision-making for such lesions.

Objective To investigate the pathological concordance of CT in assessing pseudocapsule integrity in patients with clear cell renal cell carcinoma (ccRCC) of different grades, and to analyze its association with tumor grade and local invasive features. Methods This prospective study included 60 patients with ccRCC who underwent surgical treatment at Xinxiang Central Hospital between January 2021 and January 2024 and were pathologically confirmed after surgery. Among them, 12 patients had grade Ⅰ tumors, 21 had grade Ⅱ tumors, 17 had grade Ⅲ tumors, and 10 had grade Ⅳ tumors. All patients underwent preoperative multiphase contrast-enhanced abdominal CT within 1 month before surgery. Pseudocapsule integrity was evaluated using 0.625 mm thin-section reconstructed CT images. An intact pseudocapsule was defined as a continuous linear enhancing rim at the tumor-renal parenchyma interface, with an interruption length of < 2 mm. An incomplete or absent pseudocapsule was defined as an interruption length of ≥ 2 mm or the absence of a recognizable enhancing rim. Postoperative pathology was used as the reference standard. Kappa analysis was used to evaluate the concordance between CT assessment and pathological findings, as well as the interobserver agreement between the two radiologists. The Chi-Square test or Fisher's exact test was used to compare the incidence of perirenal capsular invasion, inferior vena cava involvement, and renal sinus collecting system involvement among patients with different ccRCC grades. According to CT assessment, patients were further divided into the intact pseudocapsule group and the incomplete/absent pseudocapsule group, and local invasive features were compared between the two groups. Results CT assessment showed strong concordance with pathological findings in the overall cohort and across different ccRCC grade subgroups. Significant differences were observed among different ccRCC grades in perirenal capsular invasion, inferior vena cava involvement, and renal sinus collecting system involvement (all P < 0.05). The incidence of perirenal capsular invasion was higher in the incomplete/absent pseudocapsule group than in the intact pseudocapsule group (66.67% vs. 29.17%). Similarly, the incidence of inferior vena cava involvement was higher in the incomplete/absent pseudocapsule group than in the intact pseudocapsule group (66.67% vs. 25.00%), and the incidence of renal sinus collecting system involvement was also higher in the incomplete/absent pseudocapsule group (63.89% vs. 20.83%). These differences were statistically significant (χ² = 8.109, 10.000, and 10.725, respectively; all P < 0.05). Conclusion CT assessment of pseudocapsule integrity in ccRCC shows good concordance with postoperative pathological findings. Higher-grade ccRCC is associated with a higher incidence of local invasive features. An incomplete or absent pseudocapsule on CT may indicate a greater tendency toward local invasion and may serve as a reference imaging feature for preoperative risk assessment.

Objective To investigate the efficacy of a preoperative MRI-based radiomics model in predicting lymphovascular space invasion (LVSI) in patients with 2018 International Federation of Gynecology and Obstetrics (FIGO) stage ⅠB1-ⅡA1 cervical squamous cell carcinoma (CSCC) who harbored intermediate-risk factors after radical surgery, and to evaluate the value of a pure imaging-based risk stratification strategy combining this model with conventional MRI features for predicting disease-free survival (DFS). Methods This retrospective study included 456 patients with pathologically confirmed CSCC from two medical centers between December 2014 and June 2021. All patients had no high-risk factors (negative lymph nodes, negative surgical margins, and absence of parametrial invasion) on postoperative pathology. A total of 371 patients from center 1 were randomly divided into a training set (n = 260) and an internal validation set (n = 111) at a ratio of 7∶3, while 85 patients from center 2 served as an external validation set. Radiomics features were extracted from preoperative T₂-weighted imaging (T₂WI) and contrast-enhanced T₁-weighted imaging (CE-T₁WI). Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to select features and construct a radiomics score. With postoperative pathological LVSI as the reference standard, the predictive performance of the continuous radiomics score was assessed using the area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis. To construct a completely preoperative risk stratification, the optimal cutoff derived from the Youden index of the radiomics score in the training set was used to categorize patients into imaging-predicted LVSI-positive and LVSI-negative groups. These were then combined with MRI-measured tumor size (≥ 4 cm) and depth of stromal invasion (≥ outer third, i.e., deep stromal invasion). According to the number of the above three intermediate-risk factors met, patients were divided into a group with ≥ 2 intermediate-risk factors and a group with < 2 intermediate-risk factors. DFS was compared between groups using the Kaplan-Meier method. Results The radiomics model achieved AUCs of 0.858, 0.846, and 0.848 for predicting LVSI in the training, internal validation, and external validation sets, respectively. The calibration curve showed good agreement (Hosmer-Lemeshow test P > 0.05), and decision curve analysis indicated favorable clinical net benefit. Both 3-year and 5-year DFS rates were lower in the ≥ 2 intermediate-risk factors group than in the < 2 group (training set: 90.6% vs. 95.5% and 75.8% vs. 91.9%, respectively; hazard ratio [HR] = 2.90, 95% CI: 1.34-6.27, P = 0.007; internal validation set: 85.1% vs. 92.5% and 59.7% vs. 90.9%, HR = 4.27, 95% CI: 1.30-14.03, P = 0.017; external validation set: 80.1% vs. 92.9% and 61.6% vs. 92.9%, HR = 4.51, 95% CI: 1.17-17.46, P = 0.029). Conclusion The preoperative MRI-based radiomics model demonstrates good performance in predicting LVSI status in early-stage CSCC patients with postoperative intermediate-risk factors. The pure imaging-based risk stratification strategy, constructed by combining radiomics-predicted LVSI with MRI morphological features, can effectively identify individuals at high risk of postoperative recurrence within this population, offering a valuable reference for individualized adjuvant treatment decisions.

Objective To characterize the heterogeneity of cervical cancer (CC) using subregional features derived from apparent diffusion coefficient (ADC) maps, and to develop and validate a clinical-radiomics nomogram model for predicting lymphovascular space invasion (LVSI) status. Methods A retrospective analysis was conducted on 215 patients with postoperatively pathologically confirmed cervical cancer from two centers, who were divided into a training set (n = 163) and a validation set (n = 52). Tumor regions were segmented based on ADC maps using the k-means clustering algorithm. Radiomic features were extracted, and three types of radiomic prediction models were constructed: the whole-tumor model, the low-diffusion subregion model, and the high-diffusion subregion model. Four machine learning algorithms were used for model building. Independent clinical risk factors were identified through univariate and multivariate Logistic regression analyses to construct a clinical model. Finally, the optimal radiomic model was selected and combined with clinical high-risk factors to develop a nomogram model. Model performance and clinical net benefit were compared using receiver operating characteristic (ROC) curves, the DeLong test, and decision curve analysis. SHapley Additive exPlanations (SHAP) analysis was used to measure the impact of each feature on model predictions. Results Multivariate analysis revealed that FIGO stage and deep stromal invasion were independent predictors of LVSI positivity (all P < 0.05). The low-diffusion subregion model achieved the highest area under the curve (AUC) values in both the training and validation sets (0.867 and 0.799, respectively). The nomogram model demonstrated the best predictive performance, with an AUC of 0.885, a sensitivity of 0.743, and a specificity of 0.872 in the training set, and an AUC of 0.807, a sensitivity of 0.766, and a specificity of 0.794 in the validation set. Conclusion The nomogram model demonstrated strong diagnostic value for the preoperative prediction of LVSI in patients with CC, and may help guide clinical decision-making.

Objective To evaluate the diagnostic value of diffusion-relaxation correlation spectroscopic imaging (DR-CSI) in distinguishing benign prostatic hyperplasia (BPH) from prostate cancer (PCa). Methods This prospective study enrolled 113 patients, including 73 in the BPH group and 40 in the PCa group. All patients underwent multiparametric prostate MRI (including DR-CSI sequences) to obtain relevant parameters and delineate four micro-compartments. A combined model was constructed using Logistic regression, and diagnostic performance was evaluated using ROC curves and the DeLong test. Results There were significant differences in the ADC, T₂, fA, fC, and fD parameters between the two groups (all P < 0.05). Among the individual parameters, fA demonstrated the best diagnostic performance (AUC = 0.955). The four-compartment combined DR-CSI model achieved an AUC of 0.965, which was significantly superior to the traditional combination of ADC and T₂ (AUC = 0.821, P < 0.05). Conclusion The DR-CSI quantitative parameters can effectively distinguish BPH from PCa. The diagnostic performance of each individual parameter is superior to that of conventional ADC and T₂, and combining multiple parameters can further enhance diagnostic performance.

Objective To systematically evaluate the diagnostic accuracy of lumbar spine bone marrow proton density fat fraction (PDFF) for osteopenia and osteoporosis. Methods The databases including PubMed, Embase, Web of Science, Cochrane Library, CNKI, Wanfang, and VIP were systematically searched from inception to April 17, 2026. The QUADAS-2 tool was used to assess the risk of bias of the included studies. Results A total of 23 studies involving 2355 participants were included. For PDFF distinguishing normal from abnormal bone mass (i.e., osteopenia and osteoporosis combined), the pooled sensitivity and specificity were both 0.81, and the AUC was 0.88. For distinguishing osteoporosis from osteopenia, the pooled sensitivity was 0.72, specificity was 0.77, and AUC was 0.80. For distinguishing osteoporosis from non-osteoporosis (i.e., normal and osteopenia combined), the pooled sensitivity was 0.85, specificity was 0.79, and AUC was 0.89. The P-value for publication bias was > 0.10. Conclusion MRI-measured lumbar spine bone marrow PDFF shows moderate to good diagnostic accuracy for osteopenia and osteoporosis, with high rule-out value. It has the potential to serve as a radiation-free opportunistic screening tool; however, the diagnostic threshold is not yet standardized, and PDFF is not yet a routine clinical examination. In addition, heterogeneity in reference standards and measurement techniques should be considered in clinical application.

Objective To quantitatively evaluate the symptomatic nerve in patients with cervical spondylotic radiculopathy (CSR) using diffusion tensor imaging (DTI) and to observe the intervertebral foramen segments of cervical nerve roots by diffusion tensor tractography (DTT). Methods DTI and DTT were performed in 70 patients and 30 healthy volunteers using a 3.0 T magnetic resonance system. The mean FA values for C₅-C₇ cervical nerve roots were calculated at three segments (proximal, middle, and distal) , and DTT was also performed on C₅-C₇ nerve roots. Assessment of sensory and motor function in patients was undertaken using the modified Japanese Orthopaedic Association (mJOA) scoring system. Receiver operating characteristic (ROC) analysis was conducted to determine the diagnostic efficacy of FA values for cervical nerve root compression. Results DTT could clearly display decreased FA values, and twisted and sparse nerve roots in patients with CSR. Except for the distal segment of C₇ nerve roots, FA values of C₅ to C₇ nerve roots were significantly lower on the affected side than on the contralateral side (all P < 0.05). On the affected side, the differences in FA values among the proximal, middle, and distal segments of the same C₅-C₇ nerve root were significant (all P < 0.05). The mJOA score was significantly correlated with the FA values of the compressed nerves at the proximal segments (R² = 0.402, P < 0.001). ROC curve analysis revealed that FA values could identify nerve root compression at all three segments. Conclusion MR DTI and DTT provide an effective means for quantitatively evaluating the compressed nerve in patients with cervical spondylotic radiculopathy.

Objective To compare the image quality and diagnostic accuracy for vascular stenosis between 3.0 T three-dimensional phase-contrast magnetic resonance angiography (3D PC-MRA) and contrast-enhanced magnetic resonance angiography (CE-MRA) in the vascular imaging of diabetic foot. Methods A total of 44 patients with clinically suspected or confirmed diabetic foot who underwent both CE-MRA and 3D PC-MRA were prospectively enrolled. Two senior radiologists, blinded to the imaging modality, independently assessed the arterial delineation and arterial contrast of the dorsalis pedis artery, lateral tarsal artery, arcuate artery, dorsal metatarsal arteries, and dorsal digital arteries using a 5-point Likert scale, and also evaluated overall image artifacts and diagnostic confidence. Using CE-MRA as the reference standard, the sensitivity, specificity, positive predictive value, and negative predictive value of 3D PC-MRA for detecting different degrees of arterial stenosis were calculated. Subjective image quality scores between the two imaging modalities were compared using the paired Wilcoxon signed-rank test. Weighted Kappa analysis was used to assess interobserver agreement and the agreement in stenosis grading between the two MRA techniques. Results No significant differences were found between 3D PC-MRA and CE-MRA in the scores for arterial delineation and arterial contrast of the dorsalis pedis artery and lateral tarsal artery (P = 0.835 and P = 0.811, respectively). For distal small vessels, including the arcuate artery, dorsal metatarsal arteries, and dorsal digital arteries, 3D PC-MRA yielded significantly higher scores for arterial delineation and arterial contrast than CE-MRA (all P < 0.05). The scores for image artifacts and diagnostic confidence were also significantly better with 3D PC-MRA than with CE-MRA (both P < 0.05). There was a high degree of agreement between 3D PC-MRA and CE-MRA in stenosis grading (weighted Kappa = 0.89, P < 0.05). Using moderate or greater stenosis as the diagnostic threshold, the sensitivity, specificity, positive predictive value, and negative predictive value of 3D PC-MRA for detecting arterial stenosis were 95.53%, 97.53%, 93.96%, and 98.19%, respectively. Conclusion At 3.0T, 3D PC-MRA shows high agreement with CE-MRA in grading foot arterial stenosis in diabetic foot and provides superior image quality for visualizing distal small vessels. It may serve as a contrast-free and radiation-free method for evaluating foot arteries, particularly in diabetic foot patients who are at risk from contrast agent administration or require detailed assessment of distal small vessels.

Objective This study aimed to investigate the predictive value of MRI-based nomograms, integrated with clinicopathologic factors, for unfavorable progression-free survival (PFS) in patients with recurrent glioblastoma (rGBM) receiving bevacizumab (BEV) combined with chemotherapy. Methods We retrospectively analyzed the clinicopathological and MRI data of 84 patients with rGBM. Using a 6-month PFS cutoff after BEV plus chemotherapy, the patients were stratified into a good prognosis group (PFS > 6 months) and a poor prognosis group (PFS ≤ 6 months). We compared the clinicopathological characteristics, pre-BEV conventional MRI (cMRI) findings, and relative apparent diffusion coefficient (rADC) values between the two groups. Univariate and multivariate Cox regression analyses were performed to identify independent predictors of unfavorable PFS, which were then used to construct predictive models and nomograms. The predictive performance of the models and nomograms was assessed using receiver operating characteristic (ROC) curve analysis, stratified 10-fold cross-validation, leave-one-out cross-validation (LOOCV), calibration curves, and decision curve analysis (DCA). Results There were significant differences in CD34 expression, recurrence pattern, rFLAIR orthogonal values, and rADC values between the poor and good prognosis groups (all P < 0.05). Multivariate analysis demonstrated that a higher rFLAIR orthogonal value [> -0.384; hazard ratio （HR） = 3.01, P = 0.002] and a lower rADC value (< 1.447; HR = 0.16, P = 0.014) were independent risk factors for poor prognosis. Patients with positive CD34 expression exhibited longer PFS than those with negative CD34 expression (HR = 0.30, P < 0.001). ROC analysis revealed that the integrated model incorporating all independent predictors (CD34 expression, rFLAIR orthogonal value, and rADC) achieved the highest predictive performance, with an AUC of 0.888, a sensitivity of 0.780, and a specificity of 0.906. Stratified 10-fold cross-validation and LOOCV confirmed the stable predictive performance of the integrated model, with AUC values of 0.858 and 0.865, respectively. The calibration curve of the combined model showed excellent agreement with the ideal diagonal line. DCA showed that the model provided positive net benefits when the threshold probability was ≥ 15%. Conclusion Negative CD34 expression, rFLAIR orthogonal value > -0.384, and rADC < 1.447 were identified as risk factors for poor PFS in rGBM patients treated with BEV-combined chemotherapy. A multi-metric combination approach can improve the predictive efficacy. Nomograms constructed based on MRI and pathological molecular factors may facilitate the selection of rGBM patients suitable for BEV-combined chemotherapy.

Objective To investigate the differences in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) hemodynamic parameters between endometrial carcinoma ovarian metastasis (EC metastasis) and primary ovarian carcinoma (primary OC), and to construct a machine learning diagnostic model based on DCE-MRI parameters, radiomics features, and serum biomarkers. Methods A total of 310 patients with ovarian tumors (168 with EC metastasis, 142 with primary OC) were retrospectively enrolled. All patients underwent 3.0-T DCE-MRI examination. Quantitative parameters including K^trans, V_e, and TTP were calculated, and 26 radiomic features were extracted. Serum biomarkers (CA125 and HE4) were integrated. Variables were selected using the LASSO regression-random forest algorithm, and a support vector machine (SVM) model was established for differential diagnosis. Model performance was evaluated using a temporally independent test set (n = 62) and five-fold cross-validation. Results The EC metastasis group showed a significantly lower K^trans than the primary OC group [(0.28 ± 0.09) min^-1 vs. (0.46 ± 0.13) min^-1, P < 0.05], while V_e was significantly higher (0.71 ± 0.19 vs. 0.53 ± 0.21, P < 0.05). The SVM model achieved an area under the curve (AUC) of 0.90 (95% CI: 0.83-0.96) in the temporally independent test set, with a sensitivity of 87.1% and a specificity of 85.5%, significantly outperforming any single-parameter model. Conclusion A machine learning model combining DCE-MRI hemodynamic parameters and radiomic features can effectively differentiate EC metastasis from primary OC, providing a reliable noninvasive tool for preoperative differential diagnosis.