To develop and validate a diagnostic framework integrating intralesional (ILN) and perilesional (PLN) radiomics derived from multiparametric MRI (mpMRI) for distinguishing IgG4-related ophthalmic disease (IgG4-ROD) from orbital mucosa-associated lymphoid tissue (MALT) lymphoma. This multicenter retrospective study analyzed 214 histopathologically confirmed cases (68 IgG4-ROD, 146 MALT lymphoma) from two institutions (2019-2024). A LASSO-SVM classifier was optimized through comparative evaluation of seven machine learning models, incorporating fused radiomic features (1,197 features) from ILN/PLN regions. Diagnostic performance was benchmarked against two subspecialty radiologists (10-20 years' experience) using receiver operating characteristics - area under the curve (AUC), precision-recall AUC (PR-AUC), and decision curve analysis (DCA), adhering to CLEAR/METRICS guidelines. The fusion model (FR_RAD) achieved state-of-the-art performance, with an AUC of 0.927 (95% CI 0.902-0.958) and a PR-AUC of 0.901 (95% CI 0.862-0.940) in the training set, and an AUC of 0.907 (95% CI 0.857-0.965) and a PR-AUC of 0.872 (95% CI 0.820-0.924) on external testing. In contrast, subspecialty radiologists achieved lower AUCs of 0.671-0.740 (95% CI 0.630-0.780) and PR-AUCs of 0.553-0.632 (95% CI 0.521-0.664) (all p < 0.001). FR_RAD also outperformed radiologists in accuracy (88.6% vs. 66.2% and 71.3%; p < 0.01). DCA demonstrated a net benefit of 0.18 at a high-risk threshold of 30%, equivalent to avoiding 18 unnecessary biopsies per 100 cases. The fusion model integrating multi-regional radiomics from mpMRI achieves precise differentiation between IgG4-ROD and orbital MALT lymphoma, outperforming subspecialty radiologists. This approach highlights the transformative potential of spatial radiomics analysis in resolving diagnostic uncertainties and reducing reliance on invasive procedures for orbital lesion characterization.

To construct and validate an interpretable machine learning (ML) radiomics model derived from multiparametric magnetic resonance imaging (MRI) images to differentiate between luminal and non-luminal breast cancer (BC) subtypes. This study enrolled 1098 BC participants from four medical centers, categorized into a training cohort (n = 580) and validation cohorts 1-3 (n = 252, 89, and 177, respectively). Multiparametric MRI-based radiomics features, including T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC), and dynamic contrast-enhanced (DCE) imaging, were extracted. Five ML algorithms were applied to develop various radiomics models, from which the best performing model was identified. A ML-based combined model including optimal radiomics features and clinical predictors was constructed, with performance assessed through receiver operating characteristic (ROC) analysis. The Shapley additive explanation (SHAP) method was utilized to assess model interpretability. Tumor size and MR-reported lymph node status were chosen as significant clinical variables. Thirteen radiomics features were identified from multiparametric MRI images. The extreme gradient boosting (XGBoost) radiomics model performed the best, achieving area under the curves (AUCs) of 0.941, 0.903, 0.862, and 0.894 across training and validation cohorts 1-3, respectively. The XGBoost combined model showed favorable discriminative power, with AUCs of 0.956, 0.912, 0.894, and 0.906 in training and validation cohorts 1-3, respectively. The SHAP visualization facilitated global interpretation, identifying "ADC_wavelet-HLH_glszm_ZoneEntropy" and "DCE_wavelet-HLL_gldm_DependenceVariance" as the most significant features for the model's predictions. The XGBoost combined model derived from multiparametric MRI may proficiently differentiate between luminal and non-luminal BC and aid in treatment decision-making. An interpretable machine learning radiomics model can preoperatively predict luminal and non-luminal subtypes in breast cancer, thereby aiding therapeutic decision-making.

Accurate preoperative prediction of spread through air spaces (STAS) in primary lung adenocarcinoma (LUAD) is critical for optimizing surgical strategies and improving patient outcomes. To develop a machine learning (ML) based model to predict STAS using preoperative CT imaging features and clinicopathological data, while enhancing interpretability through shapley additive explanations (SHAP) analysis. This multicenter retrospective study included 1237 patients with pathologically confirmed primary LUAD from three hospitals. Patients from Center 1 (n=932) were divided into a training set (n=652) and an internal test set (n=280). Patients from Centers 2 (n=165) and 3 (n=140) formed external validation sets. CT imaging features and clinical variables were selected using Boruta and least absolute shrinkage and selection operator regression. Seven ML models were developed and evaluated using five-fold cross-validation. Performance was assessed using F1 score, recall, precision, specificity, sensitivity, and area under the receiver operating characteristic curve (AUC). The Extreme Gradient Boosting (XGB) model achieved AUCs of 0.973 (training set), 0.862 (internal test set), and 0.842/0.810 (external validation sets). SHAP analysis identified nodule type, carcinoembryonic antigen, maximum nodule diameter, and lobulated sign as key features for predicting STAS. Logistic regression analysis confirmed these as independent risk factors. The XGB model demonstrated high predictive accuracy and interpretability for STAS. By integrating widely available clinical and imaging features, this model offers a practical and effective tool for preoperative risk stratification, supporting personalized surgical planning in primary LUAD management.

The classification of renal cell carcinoma (RCC) histological subtypes plays a crucial role in clinical diagnosis. However, traditional image normalization methods often struggle with discrepancies arising from differences in imaging parameters, scanning devices, and multi-center data, which can impact model robustness and generalizability. This study included 1628 patients with pathologically confirmed RCC who underwent nephrectomy across eight cohorts. These were divided into a training set, a validation set, external test dataset 1, and external test dataset 2. We proposed an "Aortic Enhancement Normalization" (AEN) method based on the lesion-to-aorta enhancement ratio and developed an automated lesion segmentation model along with a multi-scale CT feature extractor. Several machine learning algorithms, including Random Forest, LightGBM, CatBoost, and XGBoost, were used to build classification models and compare the performance of the AEN and traditional approaches for evaluating histological subtypes (clear cell renal cell carcinoma [ccRCC] vs. non-ccRCC). Additionally, we employed SHAP analysis to further enhance the transparency and interpretability of the model's decisions. The experimental results demonstrated that the AEN method outperformed the traditional normalization method across all four algorithms. Specifically, in the XGBoost model, the AEN method significantly improved performance in both internal and external validation sets, achieving AUROC values of 0.89, 0.81, and 0.80, highlighting its superior performance and strong generalizability. SHAP analysis revealed that multi-scale CT features played a critical role in the model's decision-making process. The proposed AEN method effectively reduces the impact of imaging parameter differences, significantly improving the robustness and generalizability of histological subtype (ccRCC vs. non-ccRCC) models. This approach provides new insights for multi-center data analysis and demonstrates promising clinical applicability.

To develop and validate a computed tomography (CT) radiomics-based interpretable machine learning (ML) model for predicting 5-year recurrence-free survival (RFS) in non-metastatic clear cell renal cell carcinoma (ccRCC). 559 patients with non-metastatic ccRCCs were retrospectively enrolled from eight independent institutes between March 2013 and January 2019, and were assigned to the primary set (n=271), external test set 1 (n=216), and external test set 2 (n=72). 1316 Radiomics features were extracted via "Pyradiomics." The least absolute shrinkage and selection operator algorithm was used for feature selection and Rad-Score construction. Patients were stratified into low and high 5-year recurrence risk groups based on Rad-Score, followed by Kaplan-Meier analyses. Five ML models integrating Rad-Score and clinicopathological risk factors were compared. Models' performances were evaluated via the discrimination, calibration, and decision curve analysis. The most robust ML model was interpreted using the SHapley Additive exPlanation (SHAP) method. 13 radiomic features were filtered to produce the Rad-Score, which predicted 5-year RFS with area under the receiver operating characteristic curve (AUCs) of 0.734-0.836. Kaplan-Meier analysis showed significant survival differences based on Rad-Score (all Log-Rank p values <0.05). The random forest model outperformed other models, obtaining AUCs of 0.826 [95% confidential interval (CI): 0.766-0.879] and 0.799 (95% CI: 0.670-0.899) in the external test set 1 and 2, respectively. The SHAP analysis suggested positive associations between contributing factors and 5-year RFS status in non-metastatic ccRCC. CT radiomics-based interpretable ML model can effectively predict 5-year RFS in non-metastatic ccRCC patients, distinguishing between low and high 5-year recurrence risks.

Cognitive decline is common in End-Stage Renal Disease (ESRD) patients, yet its neural mechanisms are poorly understood. This study investigates structural and functional brain network reconfiguration in ESRD patients transitioning to Mild Cognitive Impairment (MCI) and evaluates its potential for predicting MCI risk. We enrolled 90 ESRD patients with 2-year follow-up, categorized as MCI converters (MCI_C, n=48) and non-converters (MCI_NC, n=42). Brain networks were constructed using baseline rs-fMRI and high angular resolution diffusion imaging, focusing on regional structural-functional coupling (SFC). A Support Vector Machine (SVM) model was used to identify brain regions associated with cognitive decline. Mediation analysis was conducted to explore the relationship between kidney function, brain network reconfiguration, and cognition. MCI_C patients showed decreased network efficiency in the structural network and compensatory changes in the functional network. Machine learning models using multimodal network features predicted MCI with high accuracy (AUC=0.928 for training set, AUC=0.903 for test set). SHAP analysis indicated that reduced hippocampal SFC was the most significant predictor of MCI_C. Mediation analysis revealed that altered brain network topology, particularly hippocampal SFC, mediated the relationship between kidney dysfunction and cognitive decline. This study provides new insights into the link between kidney function and cognition, offering potential clinical applications for structural and functional MRI biomarkers.

We aimed to evaluate the diagnostic performance of artificial intelligence (AI) in detecting coronary artery stenosis and calcified plaque on CT angiography (CTA), comparing its diagnostic performance with that of radiologists. A thorough search of the literature was performed using PubMed, Web of Science, and Embase, focusing on studies published until October 2024. Studies were included if they evaluated AI models in detecting coronary artery stenosis and calcified plaque on CTA. A bivariate random-effects model was employed to determine combined sensitivity and specificity. Study heterogeneity was assessed using I² statistics. The risk of bias was assessed using the revised quality assessment of diagnostic accuracy studies-2 tool, and the evidence level was graded using the Grading of Recommendations Assessment, Development and Evalutiuon (GRADE) system. Out of 1071 initially identified studies, 17 studies with 5560 patients and images were ultimately included for the final analysis. For coronary artery stenosis ≥50%, AI showed a sensitivity of 0.92 (95% CI: 0.88-0.95), specificity of 0.87 (95% CI: 0.80-0.92), and AUC of 0.96 (95% CI: 0.94-0.97), outperforming radiologists with sensitivity of 0.85 (95% CI: 0.67-0.94), specificity of 0.84 (95% CI: 0.62-0.94), and AUC of 0.91 (95% CI: 0.89-0.93). For stenosis ≥70%, AI achieved a sensitivity of 0.88 (95% CI: 0.70-0.96), specificity of 0.96 (95% CI: 0.90-0.99), and AUC of 0.98 (95% CI: 0.96-0.99). In calcified plaque detection, AI demonstrated a sensitivity of 0.93 (95% CI: 0.84-0.97), specificity of 0.94 (95% CI: 0.88-0.96), and AUC of 0.98 (95% CI: 0.96-0.99)." AI-based CT demonstrated superior diagnostic performance compared to clinicians in identifying ≥50% stenosis in coronary arteries and showed excellent diagnostic performance in recognizing ≥70% coronary artery stenosis and calcified plaque. However, limitations include retrospective study designs and heterogeneity in CTA technologies. Further external validation through prospective, multicenter trials is required to confirm these findings. The original findings of this research are included in the article. For additional inquiries, please contact the corresponding authors.

The differential diagnosis of invasive pulmonary aspergillosis (IPA), pulmonary mucormycosis (PM), bacterial pneumonia (BP) and pulmonary tuberculosis (PTB) are challenging due to overlapping clinical and imaging features. Manual CT lesion segmentation is time-consuming, deep-learning (DL)-based segmentation models offer a promising solution, yet disease-specific models for these infections remain underexplored. We aimed to develop and validate dedicated CT segmentation models for IPA, PM, BP and PTB to enhance diagnostic accuracy. Methods：Retrospective multi-centre data (115 IPA, 53 PM, 130 BP, 125 PTB) were used for training/internal validation, with 21 IPA, 8PM, 30 BP and 31 PTB cases for external validation. Expert-annotated lesions served as ground truth. An improved 3D U-Net architecture was employed for segmentation, with preprocessing steps including normalisations, cropping and data augmentation. Performance was evaluated using Dice coefficients. Results：Internal validation achieved Dice scores of 78.83% (IPA), 93.38% (PM), 80.12% (BP) and 90.47% (PTB). External validation showed slightly reduced but robust performance: 75.09% (IPA), 77.53% (PM), 67.40% (BP) and 80.07% (PTB). The PM model demonstrated exceptional generalisability, scoring 83.41% on IPA data. Cross-validation revealed mutual applicability, with IPA/PTB models achieving > 75% Dice for each other's lesions. BP segmentation showed lower but clinically acceptable performance ( ＞72%), likely due to complex radiological patterns. Disease-specific DL segmentation models exhibited high accuracy, particularly for PM and PTB. While IPA and BP models require refinement, all demonstrated cross-disease utility, suggesting immediate clinical value for preliminary lesion annotation. Future efforts should enhance datasets and optimise models for intricate cases.

The accurate and early evaluation of response to neoadjuvant chemotherapy (NAC) in breast cancer is crucial for optimizing treatment strategies and minimizing unnecessary interventions. While deep learning (DL)-based approaches have shown promise in medical imaging analysis, existing models often fail to comprehensively integrate spatial and temporal tumor dynamics. This study aims to develop and validate a spatiotemporal interaction (STI) model based on longitudinal MRI data to predict pathological complete response (pCR) to NAC in breast cancer patients. This study included retrospective and prospective datasets from five medical centers in China, collected from June 2018 to December 2024. These datasets were assigned to the primary cohort (including training and internal validation sets), external validation cohorts, and a prospective validation cohort. DCE-MRI scans from both pre-NAC (T0) and early-NAC (T1) stages were collected for each patient, along with surgical pathology results. A Siamese network-based STI model was developed, integrating spatial features from tumor segmentation with temporal dependencies using a transformer-based multi-head attention mechanism. This model was designed to simultaneously capture spatial heterogeneity and temporal dynamics, enabling accurate prediction of NAC response. The STI model's performance was evaluated using the area under the ROC curve (AUC) and Precision-Recall curve (AP), accuracy, sensitivity, and specificity. Additionally, the I-SPY1 and I-SPY2 datasets were used for Kaplan-Meier survival analysis and to explore the biological basis of the STI model, respectively. The prospective cohort was registered with Chinese Clinical Trial Registration Centre (ChiCTR2500102170). A total of 1044 patients were included in this study, with the pCR rate ranging from 23.8% to 35.9%. The STI model demonstrated good performance in early prediction of NAC response in breast cancer. In the external validation cohorts, the AUC values were 0.923 (95% CI: 0.859-0.987), 0.892 (95% CI: 0.821-0.963), and 0.913 (95% CI: 0.835-0.991), all outperforming the single-timepoint T0 or T1 models, as well as models with spatial information added (all p < 0.05, Delong test). Additionally, the STI model significantly outperformed the clinical model (p < 0.05, Delong test) and radiologists' predictions. In the prospective validation cohort, the STI model identified 90.2% (37/41) of non-pCR and 82.6% (19/23) of pCR patients, reducing misclassification rates by 58.7% and 63.3% compared to radiologists. This indicates that these patients might benefit from treatment adjustment or continued therapy in the early NAC stage. Survival analysis showed a significant correlation between the STI model and both recurrence-free survival (RFS) and overall survival (OS) in breast cancer patients. Further investigation revealed that favorable NAC responses predicted by the STI model were closely linked to upregulated immune-related genes and enhanced immune cell infiltration. Our study established a novel noninvasive STI model that integrates the spatiotemporal evolution of MRI before and during NAC to achieve early and accurate pCR prediction, offering potential guidance for personalized treatment. This study was supported by the National Natural Science Foundation of China (82302314, 62271448, 82171920, 81901711), Basic and Applied Basic Research Foundation of Guangdong Province (2022A1515110792, 2023A1515220097, 2024A1515010653), Medical Scientific Research Foundation of Guangdong Province (A2023073, A2024116), Science and Technology Projects in Guangzhou (2023A04J1275, 2024A03J1030, 2025A03J4163, 2025A03J4162); Guangzhou First People's Hospital Frontier Medical Technology Project (QY-C04).

AI-driven scan time reduction is rapidly transforming medical imaging with benefits such as improved patient comfort and enhanced efficiency. A Dual Contrastive Learning Generative Adversarial Network (DCLGAN) was developed to predict full-time PET scans from shorter, noisier scans, improving challenges in imaging patients with movement disorders. 18F FP-CIT PET/CT data from 391 patients with suspected Parkinsonism were used [250 training/validation, 141 testing (hospital A)]. Ground truth (GT) images were reconstructed from 15-minute scans, while denoised images (DIs) were generated from 1-, 3-, 5-, and 10-minute scans. Image quality was assessed using normalized root mean square error (NRMSE), peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), visual analysis, and clinical metrics like BPND and ISR for diagnosis of non-neurodegenerative Parkinson disease (NPD), idiopathic PD (IPD), and atypical PD (APD). External validation used data from 2 hospitals with different scanners (hospital B: 1-, 3-, 5-, and 10-min; hospital C: 1-, 3-, and 5-min). In addition, motion artifact reduction was evaluated using the Dice similarity coefficient (DSC). In hospital A, NRMSE, PSNR, and SSIM values improved with scan duration, with the 5-minute DIs achieving optimal quality (NRMSE 0.008, PSNR 42.13, SSIM 0.98). Visual analysis rated DIs from scans ≥3 minutes as adequate or higher. The mean BPND differences (95% CI) for each DIs were 0.19 (-0.01, 0.40), 0.11 (-0.02, 0.24), 0.08 (-0.03, 0.18), and 0.01 (-0.06, 0.07), with the CIs significantly decreasing. ISRs with the highest effect sizes for differentiating NPD, IPD, and APD (PP/AP, PP/VS, PC/VP) remained stable post-denoising. External validation showed 10-minute DIs (hospital B) and 1-minute DIs (hospital C) reached benchmarks of hospital A's image quality metrics, with similar trends in visual analysis and BPND CIs. Furthermore, motion artifact correction in 9 patients yielded DSC improvements from 0.89 to 0.95 in striatal regions. The DL-model is capable of generating high-quality 18F FP-CIT PET images from shorter scans to enhance patient comfort, minimize motion artifacts, and maintain diagnostic precision. Furthermore, our study plays an important role in providing insights into how imaging quality assessment metrics can be used to determine the appropriate scan duration for different scanners with varying sensitivities.