2026-10-13

Background: Deep learning-based aneurysm segmentation using 3D time-of-flight magnetic resonance angiography (TOF-MRA) offers promising non-invasive screening, but its accuracy and generalizability remain limited. Purpose: To develop a deep learning model for intracranial aneurysm detection and segmentation on MRA, and evaluate its multicenter performance against the digital subtraction angiography (DSA) gold standard and radiologists of varying experience. Materials And Methods: MRA images from patients suspected of harboring unruptured intracranial aneurysms were retrospectively collected from Jinhua Municipal Central Hospital for model development. An external test set, with DSA as the reference standard, was gathered from Sir Run Run Shaw Hospital, Zhejiang University School of Medicine. Manual segmentation of cerebral arteries and aneurysms was performed using 3D-Slicer software. Performance was evaluated via Dice Similarity Coefficient (DSC), sensitivity, and F1-score, and compared against independent clinical evaluations from radiologists with different experience levels. Results: The internal dataset comprised 1,560 patients (1,554 aneurysms), including 421 patients (522 aneurysms) for testing. The external test set included 208 patients (224 aneurysms). On the internal test set, the model achieved a lesion-level sensitivity of 86.0% (95% CI: 82.8 - 88.7), a precision of 87.4% (95% CI: 84.2, 90.0), and a DSC of 0.88 (95% CI: 0.84, 0.90). On the independent external test set, the model demonstrated robust generalizability with a sensitivity of 83.4% (95% CI: 75.9, 88.0) and a DSC of 0.84 (95% CI: 0.79, 0.87). The model achieved an overall detection F1-score of 86.7, significantly outperforming junior radiologists (P <.001) and was comparable to mid-level radiologists (P >.05), with an average processing time of approximately 72seconds per exam. Conclusion: The proposed deep learning model demonstrated robust multicenter performance for intracranial aneurysm detection and segmentation. Achieving diagnostic accuracy comparable to mid-level radiologists, it holds potential as an efficient assistive tool for MRA interpretation in clinical workflows.

Objective: Expert clinician chart review is the most accurate method to identify patients with infection but is laborious and costly. We aimed to develop and externally validate machine-learning models using structured and unstructured data to identify infection in hospitalized patients compared to chart review. Materials And Methods: Adult inpatient encounters at 2 health systems (2006-2016) were manually annotated for suspected infection on admission. Data from one site were used to develop gradient-boosted machine-learning models. Structured features included labs, vital signs, and nursing assessments. Unstructured clinical notes were converted to concept unique identifiers (CUIs) using the clinical Text Analysis and Knowledge Extraction System. Unimodal and multimodal models were tuned to maximize the area under the receiver operating curve (AUROC) in the derivation site through cross-validation. Model discrimination and calibration were assessed in the external validation site. Results: Of the 3014 included admissions, 677 (22.5%) were classified as having suspected infection on admission. A multimodal model combining n-gram TF-IDF and structured data demonstrated the best performance, significantly outperforming the structured data only model (AUROC 0.88 vs 0.82, P < .001, area under the precision-recall curve 0.71 vs 0.58, P < .001). The most important features came from unstructured data. Discussion And Conclusion: Our results highlight the value of unstructured text data for developing models that accurately replicate chart reviews in identifying patients with infection. By expediting the annotation of infection cases, our models can facilitate epidemiological research and the curation of datasets necessary for developing clinical decision support systems for early sepsis detection.

Background: Thyroid nodules (TNs) with lymphocytic thyroiditis (LT) complicate ultrasound (US) assessment, and few prediction models have been validated in this population. The aim of this study is to develop and validate a US-based deep learning radiomics nomogram (DLRN) model for predicting malignancy in TNs coexisting with LT. Methods: This multicenter hybrid study enrolled 640 patients with pathologically confirmed TNs and concomitant LT. The cohort comprised a training set from Anhui Medical University First Affiliated Hospital (Jixi Road Campus), Anhui Medical University First Affiliated Hospital (High-tech Campus), and Affiliated Hospital of Integration Chinese and Western Medicine with Nanjing University of Traditional Chinese Medicine (retrospective, n=411), an external test set from Yijishan Hospital of Wannan Medical College (retrospective, n=162), and a validation set (prospective, n=67). Deep learning (DL) and handcrafted radiomics features were extracted separately from US images. After feature fusion, the features were Z-score normalized and selected using the Mann-Whitney U test or Chi-squared test, with Fisher's exact test used when appropriate, followed by the least absolute shrinkage and selection operator (LASSO) regression. A DL radiomics (DLR) model was developed based on 10 machine learning algorithms and subsequently incorporated with independent clinical factors through logistic regression to build a DLRN. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC), calibration and decision curve, along with SHapley Additive Explanations (SHAP) analysis for interpretability. Results: Logistic regression was selected as the optimal DLR algorithm, achieving AUCs of 0.90 [95% confidence interval (CI): 0.87-0.93] in the training set, 0.90 (95% CI: 0.84-0.94) in the external test set, and 0.82 (95% CI: 0.71-0.91) in the prospective validation set. The DLRN, refined by integrating independent clinical factors via logistic regression, demonstrated superior discrimination with AUCs of 0.93, 0.94, and 0.92 in the training, external test, and prospective validation sets, respectively. The DLRN model demonstrated robust clinical applicability, as evidenced by calibration and decision curve analyses. SHAP analysis enhanced DLRN interpretability by quantifying feature contributions to individual predictions and overall model output. Conclusions: The DLRN model was designed for the preoperative identification of malignant TNs in LT, thereby facilitating personalized treatment.

Sepsis-related myocardial injury (SRMI) is associated with high mortality, but the transportability of prediction models across heterogeneous cohorts remains uncertain. We developed and externally validated a machine learning model for in-hospital mortality prediction in SRMI and evaluated its discrimination, calibration, interpretability, and distributional shift across cohorts. Using MIMIC-IV for development and MIMIC-III and a local cohort for external validation (n = 1,226, 962, and 280, respectively), we compared eight classification algorithms. A two-stage feature-selection strategy, recursive feature accumulation and sequential elimination with incremental feature evaluation, ranked 31 candidate variables to derive a parsimonious predictor set. The selected model was evaluated for discrimination, precision-recall performance, calibration, clinical utility, and risk stratification. Logistic recalibration, population stability index (PSI) analysis, and SHapley Additive exPlanations (SHAP) were used to assess probability alignment, cross-cohort distributional shift, and model interpretability. Feature selection reduced 31 candidate variables to 8 predictors with little loss of cross-validated performance (AUC, 0.819 with 31 variables versus 0.810 with 8 variables). The selected random forest model achieved an AUC of 0.819 (bootstrap 95% CI, 0.787-0.848) and a PR-AUC of 0.702 (0.657-0.751) in MIMIC-III, and an AUC of 0.758 (0.693-0.816) and a PR-AUC of 0.525 (0.435-0.630) in the local cohort. Recalibration reduced the Brier score from 0.162 to 0.147 in MIMIC-III and from 0.233 to 0.196 in the local cohort. PSI indicated minimal distributional shift between MIMIC-IV and MIMIC-III but substantial shift in the local cohort, particularly for SOFA score, respiratory rate, and prothrombin time. SHAP identified prothrombin time, invasive ventilation, lactate, temperature, and urine output as major contributors to model predictions. In two external cohorts, an 8-variable random forest model showed moderate discrimination for mortality prediction in SRMI. The evaluation of external performance, calibration, distributional shift, and SHAP-based explanations provides an assessment of model transportability. Prospective validation is needed before clinical deployment.

Background: Frailty adversely affects survival and quality of life in patients undergoing maintenance hemodialysis (MHD). This study aimed to develop a clinical multimodal model to identify prevalent frailty in MHD patients. Methods: Clinical data from 626 MHD patients across four tertiary hospitals were retrospectively collected, including demographics, clinical characteristics, laboratory indicators, and physical function assessments. Frailty was assessed using the FRAIL scale. Patients from Wuxi (n = 502) were divided into training (n = 352) and internal validation (n = 150) sets, while patients from Yancheng (n = 124) formed the external validation set. Least absolute shrinkage and selection operator (LASSO) and Boruta were used for feature selection. Ten machine learning models were developed and evaluated using the area under the receiver operating characteristic curve (AUC), calibration, Brier score, and decision curve analysis. SHAP was applied for model interpretation. Results: Frailty prevalence was 48.6%. Six predictors were identified: age, grip strength, comorbidities, Charlson comorbidity index (CCI), albumin, and pro-B-type natriuretic peptide (pro-BNP). The extreme gradient boosting (XGBoost) achieved the best external validation performance (AUC=0.917, 95% CI: 0.867-0.967; accuracy=0.847; sensitivity=0.850; specificity=0.844), with good calibration and clinical net benefit. The shapley additive explanations (SHAP) identified pro-BNP as the most influential predictor and revealed nonlinear associations and clinically relevant thresholds. Conclusion: The interpretable XGBoost model effectively identified prevalent frailty in MHD patients, demonstrating strong discrimination, calibration, and clinical applicability, and may support systematic frailty screening and individualized assessment.

Background: Accurate diagnosis of idiopathic pulmonary fibrosis (IPF) and connective tissue disease-associated interstitial lung disease (CTD-ILD), along with assessment of the interstitial lung disease gender-age-physiology (ILD-GAP) index, is crucial for formulating effective therapeutic interventions. The aim of this study was to examine whether multiregion radiomics combined with clinical indicators and quantitative computed tomography (QCT) can enhance the diagnostic accuracy for IPF and CTD-ILD. We also assessed the potential of radiomics as a noninvasive alternative to pulmonary function testing (PFT) for evaluating ILD-GAP staging, a prognostic marker. Methods: This retrospective study included patients diagnosed with IPF and CTD-ILD at two centers. Honeycombing (H), reticulation (R), and ground-glass opacity (GGO) were outlined in representative regions, and radiomics features were then extracted. Multidimensional data included multiregional radiomics, QCT (lesion volume), and clinical indicators. The most appropriate machine learning algorithms were selected to develop models. Ultimately, the logistic regression algorithm was employed to establish a nomogram model for distinguishing IPF from CTD-ILD and a nomogram gender-age-physiology (nomogram-GAP) model for predicting ILD-GAP staging. Model performance was evaluated via the area under the receiver operating characteristic curve (AUC), calibration, and decision curve analysis (DCA). Results: To differentiate IPF from CTD-ILD, we enrolled patients from Center A into training (n=153), test (n=64), and internal validation (n=42) sets, as well as 40 patients from Center B into an external validation set. The nomogram model yielded an AUC of 0.854 in the internal validation set and 0.823 in the external validation set. For ILD-GAP staging prediction, Center A patients were split into training (n=155) and internal validation sets (n=31). The nomogram-GAP model had an AUC of 0.881 in the internal validation set. Calibration and DCA demonstrated that the nomogram model and nomogram-GAP model exhibit strong consistency and clinical utility. Conclusions: Two types of nomograms integrating multiregion radiomics, QCT, and clinical indicators demonstrated promising performance in distinguishing IPF from CTD-ILD and predicting ILD-GAP staging. When PFTs are unreliable or impracticable, the nomogram-GAP model may serve as a reliable marker for assessing ILD-GAP staging.