The HERMES-24 score recently demonstrated high accuracy for outcome prediction after large vessel occlusion (LVO) treatment in late time window patients from randomized clinical trials. In this study, we externally validate the score in a real-world patient cohort. Data from German Stroke Registry patients with LVO treated with endovascular therapy beyond 6 hours from symptom onset or last seen well were used. We performed a complete case analysis, excluding functionally dependent patients (premorbid modified Rankin Scale [mRS] >2/>3 for prediction of mRS ≤2/≤3, respectively). We assessed the HERMES-24 score for 90-day mRS prediction using bootstrap resampling and the c-statistic. The analyzed cohort comprised 2,117 patients (mean age 74 ± 13.3 years; 55.4% female; median admission NIH Stroke Scale (NIHSS) 14 (Q1-Q3: 9-18)). The HERMES-24 score achieved an area under the curve (AUC) of 0.876 (95% CI 0.859-0.889) for mRS ≤2 and 0.856 (95% CI 0.837-0.875) for mRS ≤3. Subgroup analysis for mRS ≤2 prediction showed lower performance in patients with NIHSS <18 (AUC 0.850, 95% CI 0.832-0.870). In our real-world cohort of late time window patients with LVO, the HERMES-24 score showed good discriminative performance, supporting its cautious clinical applicability, considering its lower performance than in trial populations, especially in patients with lower baseline NIHSS scores.

Predicting adult functional outcomes in childhood-onset attention-deficit/hyperactivity disorder using multimodal MRI and machine learning: A prospective follow-up study.

Precision medicine applications depend on elucidating the underlying molecular mechanisms of disease. However, many disorders, such as endometriosis, remain poorly characterised genetically due to data scarcity, positive-unlabelled (PU) imbalance, and the heterogeneous structure of biomedical knowledge. This study aims to develop HetBio-CLiP, a novel heterogeneous graph-based contrastive learning (CL) methodology that improves candidate gene prioritisation across heterogeneous biomedical graphs by explicitly addressing positive-unlabelled learning constraints and structural heterogeneity. HetBio-CLiP (Heterogeneous Biomedical graph Contrastive Learning with Interactions and Positive-Unlabelled learning) integrates multi-relational genomic, variant, and clinical data from real-world patient cohorts. To address the challenges of data scarcity and class imbalance, the graph neural network (GNN) methodology combines heterogeneous graph CL with PU learning. Furthermore, the model incorporates an interpretable GNNShap explainer to provide transparency at both the feature and edge levels, to assess the biological relevance of the predictions. HetBio-CLiP achieved superior performance across key evaluation metrics, attaining the highest Area Under the Curve (AUC) of 0.9489 ± 0.04 and Area Under the Precision-Recall Curve (AUPR) of 0.9401 ± 0.05. This outperformed state-of-the-art baselines such as M-SAGEGraph (AUC 0.9149 ± 0.04) and GAT (AUC 0.9138 ± 0.07). In terms of ranking candidate genes, the model achieved the highest Precision@42 (0.6933 ± 0.06) and TP@42 (29.1 ± 2.4) and second-highest Normalised Discounted Cumulative Gain (NDCG@42) (0.827 ± 0.06). Ablation studies confirmed that the integration of PU and contrastive learning was essential for these performance gains. Biologically, the model successfully prioritised known pathophysiology drivers, with top-ranked validated gene predictions including WNT4, GREB1, and ESR1. HetBio-CLiP presents an effective and interpretable methodology for uncovering candidate gene-disease relationships in data-scarce environments. By combining CL with PU-training, the proposed method addresses critical limitations of current gene prioritisation models and produces biologically meaningful predictions. The model demonstrates a strong balance between ranking accuracy and recall, identifying broader sets of plausible candidates than existing methods. This approach benefits researchers, clinicians, and downstream genomic studies by offering an improved foundation for identifying novel disease-associated genes and guiding future experimental investigation in endometriosis and other complex diseases.

Mycophenolate mofetil is widely used in liver transplantation but poses dosing challenges owing to the narrow therapeutic window and high pharmacokinetic variability of its active moiety, mycophenolic acid (MPA). Overexposure increases the risk of adverse effects, whereas underexposure increases the risk of rejection and graft loss. The ImmunoSuppressant Bayesian Dose Adjustment (ISBA) platform estimates the MPA area under the curve (AUC) 0-12h using 3 post-dose samples (20 minutes, 1 hour, 3 hours) and provides dose recommendations to target an AUC 0-12h of 30-60 mg·h/L. The aim of this study was to describe the MPA AUC 0-12h and assess the impact of ISBA-guided dose adjustments. MPA concentrations were measured at each visit, and AUCs were estimated in real time during routine clinical follow-up. All data collected from liver-transplant recipients aged ≥18 years from October 2005 to May 2020 were retrospectively analyzed. Dosing recommendations were adjusted proportionally to reach the AUC target of 45 mg.h/L. A total of 3632 requests involving 1872 patients were analyzed, making this the largest real-world cohort of this population reported to date. Among them, 658 patients benefited from at least 2 successive dose adjustments, allowing for a comparison between the results obtained without any prior use of the online platform and those obtained after the first visit (AUC of the second visit). At the first visit (before any dose recommendation) during the first year after transplantation, the median AUC 0-12h (interquartile range) was 27 (16-39) mg·h/L (n = 422). After dose adjustment recommendation, the median AUC 0-12h increased significantly to 38 (30-49) mg·h/L (n = 205, P < 0.001). The proportion of patients within the therapeutic target range (30-60 mg.h/L) increased from 40% to 61% (+21%, P < 0.001). Many patients were underexposed to MPA before dose adjustment. The ISBA recommendations significantly improved target attainment rates.

Relationship between tibio-fibular overlap ratio and lower limb torsion in an Asian population.

Intratumoral and Peritumoral Habitat Radiomics on Multiparametric MRI for Preoperative Prediction of 1-Year Progression-Free Survival Status in Glioblastoma: A Multicenter Study.

Although spleen stiffness measurement (SSM) via 100 Hz probe shows promise in predicting hepatic decompensation, its prognostic value across different etiological backgrounds remains insufficiently validated. We evaluated its efficacy in patients with compensated advanced chronic liver disease (cACLD) in a Chinese cohort. This retrospective study included 713 cACLD patients. The optimal SSM cut-off was derived using the Fine-Grey competing risk model and validated by bootstrapping. The SSM-based model was compared against established models using time-dependent area under the curve (AUC), C-index, and decision curve analysis (DCA). During a median follow-up of 36.6 months, 28 patients (3.9%) developed decompensation. SSM was an independent predictor (sHR 1.03, 95% CI 1.02-1.05, p < 0.001). An SSM threshold of > 55 kPa was identified and demonstrated excellent stability (bootstrap 95% CI 39.8-57.3 kPa). The SSM > 55 kPa showed comparable discriminative ability to the multivariate NICER model at 1 year (AUC: 0.820 vs. 0.868, p = 0.328) and 2 years (AUC: 0.809 vs. 0.851, p = 0.271). Notably, SSM > 55 kPa significantly outperformed albumin-combined models (LSM-ALB, NICER-ALB) at 1-year prediction (all p < 0.05). DCA revealed that SSM > 55 kPa provided the highest net clinical benefit across all models. Patients with SSM > 55 kPa had a markedly higher decompensation risk (sHR 8.69, 95% CI 4.15-18.23, p < 0.001), with decompensation incidences of 15.2% vs. 2.0% compared to those below the threshold. In cACLD patients, a simple 100-Hz SSM threshold (> 55 kPa) effectively predicts hepatic decompensation with performance rivalling more complex models, offering a practical, non-invasive tool for risk stratification.

Elevated plasma concentrations of glycocalyx degradation products have been associated with organ dysfunction in preeclampsia. We hypothesized that the glycocalyx degradation products syndecan-1, hyaluronic acid, and thrombomodulin, could predict subsequent organ dysfunction in women diagnosed with preeclampsia. This was a prospective, observational cohort study. Women were enrolled at the time of diagnosis, provided no organ dysfunction was present. The primary outcome was a composite of complications based on the core outcome set for preeclampsia research (BJOG 2020). Exposures were plasma concentrations of syndecan-1, hyaluronic acid and thrombomodulin. Receiver operating characteristic curves were constructed for the composite and for single outcomes with ≥ 10 events. Predictive performance was quantified as the area under the curve (AUC) with a 95% confidence interval (CI). None of the glycocalyx degradation products predicted the composite outcome. For single outcomes, syndecan-1 was associated with elevated liver enzymes (AUC of 0.64; 95% CI 0.52-0.76). Hyaluronic acid predicted thrombocytopenia (AUC 0.67; 95% CI 0.52-0.83) and elevated liver enzymes (AUC 0.69; 95% CI 0.57-0.82). Combining the markers in multivariable analyses showed similar results as univariable analyses. When combined with sFlt-1, none of the glycocalyx degradation products improved the predictive performance compared to sFlt-1 alone. In women with a confirmed diagnosis of preeclampsia, plasma concentrations of syndecan-1, hyaluronic acid, and thrombomodulin demonstrated limited predictive ability for the subsequent composite outcome of organ dysfunction. Hyaluronic acid was the most promising marker for the single outcomes thrombocytopenia and elevated liver enzymes.

To validate ultrasound-derived fat fraction (UDFF) against histology for diagnosing hepatic steatosis, monitor metabolic changes after metabolic and bariatric surgery, and develop a pre-operative nomogram for predicting early steatosis normalization. A prospective cohort of 160 patients scheduled for metabolic and bariatric surgery was enrolled. Among them, 120 patients with biopsy-confirmed metabolic dysfunction-associated steatotic liver disease (hepatic fat ≥5%) underwent repeated assessments of UDFF, anthropometrics, and biochemical markers pre-operatively and at 30, 90, and ≥180 days post-operatively. Follow-up completion rates were 82.5%(99/120) at 30 days, 76.7% (92/120) at 90 days, and 60.0%(72/120) at ≥180 days. Using data from 92 patients with complete 90-day follow-up, a nomogram was developed via uni-variable and multi-variable logistic regression. Its performance was evaluated by the area under the curve (AUC), calibration, and decision curve analysis. UDFF demonstrated a strong correlation with histological steatosis grade (Spearman's ρ=0.78;p<0.001) and high diagnostic accuracy for ≥S1 steatosis (AUC=0.99). Post-operative monitoring revealed rapid, phased reductions in UDFF, body weight, and triglyceride levels. Baseline weight, UDFF, and triglycerides were identified as independent predictors of steatosis normalization (all p<0.05). The nomogram integrating these predictors exhibited good discrimination (AUC=0.84), calibration, and clinical utility. UDFF is a valid, non-invasive tool for quantifying hepatic steatosis and monitoring treatment response. The developed nomogram provides a practical means for the pre-operative identification of patients at risk for delayed normalization, thereby facilitating personalized surgical management and optimizing follow-up strategies.

To compare the Vitreous Floater Functional Questionnaire (VFFQ-23) to the NEI Visual Function Questionnaire (VFQ-25) in patients with vision degrading myodesopsia (VDM) from vitreous floaters. Composite scores for each questionnaire were compared in asymptomatic controls, patients who chose observation (OBS), and those electing vitrectomy. The impact of limited refractive vitrectomy (LRV) on the composite scores of each questionnaire was determined. A retrospective study comparing questionnaire correlations with management choice (OBS or LRV), and a prospective, non-randomized, interventional case series. 74 subjects were asymptomatic controls; 193 subjects had bilateral floaters: 139 chose OBS, 54 elected LRV surgery. Patients complaining of bilateral "floaters" completed the VFFQ-23 and VFQ-25 questionnaires and composite scores were correlated with visual acuity (VA), contrast sensitivity (CS), quantitative ultrasonography (QUS), and management choice (OBS or LRV). LRV was performed in 54 patients, and all tests were repeated at 6, 12, 24, and 36 months post-operatively. VFFQ-23 (0 to 100), NEI VFQ-25 (0 to 100), VA (Snellen), CS (Freiburg Acuity Contrast Test), QUS (AU), predictive value, Receiver Operator Characteristic areas under the curve (AUC), decision curve analysis (DCA). VFFQ-23 correlated with CS (R = -.619; P < .0001) and QUS (R = -.632; P < .0001). VFFQ-23 correctly identified 89% of patients who chose LRV with a positive predictive value of 73.8% for LRV and 95.3% for OBS, while VFQ-25 had a positive predictive value of only 60%; VFFQ-23 AUC = 0.923, VFQ-25 AUC = 0.570. DCA demonstrated a higher net benefit for VFFQ-23 than both "Treat All" and "Treat None" strategies. After LRV, VFFQ-23 improved 72.2% ± 3.2% (P < .001), while VFQ-25 only improved 7.4% ± 1.8%, (P < .001). The difference-in-differences before and after surgery between VFFQ-23 and VFQ-25 was 31.9 ± 12.0 (P < .001). In patients with vitreous floaters, the VFFQ-23 correlated with CS and QUS. VFFQ-23 was a better predictor of patient management choice (OBS or LRV) and was more sensitive to post-operative improvement than VFQ-25. VFFQ-23 superiority to VFQ-25 was confirmed by DCA. Thus, VFFQ-23 may be more useful in assessing the subjective impact of floaters on visual quality-of-life, in screening/triage, and as an outcome measure of current and future therapies for VDM.

Automated detection of superior mesenteric artery occlusion on post-contrast CT Using a 3D deep learning model.

To assess the discriminatory ability of vertebra-specific volumetric bone mineral density (vBMD), finite element analysis-derived fracture load (FEA-derived FL), and texture analysis (TA) features for incidental vertebral fractures, and to compare performance between thoracic and lumbar levels. We retrospectively reviewed baseline and follow-up thoracolumbar CT scans from 420 patients and identified 11 patients with incidental vertebral fractures contributing to 20 fractured vertebrae (7 females; mean age 65.5years). For each fractured vertebra, three level-matched control vertebrae from patients without fractures were selected, yielding 58 controls across 29 control patients (total 78 vertebrae). Parameters evaluated include vBMD, FEA-derived FL, and TA features (24 total). Discriminatory ability was assessed using area under the curve (AUC) values. vBMD, FEA-derived FL, and 4 of 24 TA features showed group-wise differences between fractured and control vertebrae groups. AUCs were 0.76 [95% CI 0.55-0.90] (vBMD) and 0.73 [95% CI 0.52-0.90] (FEA-derived FL); selected texture features ranged 0.70-0.72. Region-stratified AUC point estimates were higher in the lumbar than in the thoracic vertebrae, but the 95% CIs were wide/overlapping; comparisons are descriptive. vBMD had the numerically largest AUC point estimate for discriminating fractured from control vertebrae; FEA-derived FL was similar, and selected texture features showed modest discrimination with comparable point estimates across lumbar and thoracic levels, generating the hypothesis of less region dependence. Regional comparisons are descriptive. Findings are exploratory and intended to prioritize candidate measures for validation and future multivariable modeling before any clinical application.

AI-powered methods for psychological assessment in adolescence psychological disorders: A systematic review and meta-analysis.

To develop and externally validate a bio-ecological model for early screening of developmental coordination disorder (DCD) using maternal and environmental risk factors from electronic health records, aimed at improving early detection in children under 5 years. This was a prospective study that examined data from 150 948 preschool children in China. Perinatal and sociodemographic predictors were integrated using logistic regression and random forest algorithms. The model was internally validated on split training and testing subsets and externally validated on an independent clinical sample of 1359 children aged 3 to 10 years, including confirmed diagnoses of DCD. Model performance was evaluated using the area under the curve (AUC), sensitivity, specificity, and accuracy. In the group aged 3 to 5 years, the model achieved an AUC of 0.70, sensitivity of 71.43%, accuracy of 77.61%, and specificity of 78.00%. In the group aged 6 to 10 years, performance was moderate (AUC = 0.58; sensitivity = 54.88%; accuracy = 61.50%; specificity = 62.28%). This bio-ecological model offers a scalable, cost-effective tool to support the early identification of DCD using electronic health record data. It performs well in early childhood and maintains moderate accuracy in older children, supporting its utility for longer-term risk prediction. The model could enhance existing screening systems by enabling earlier triage and intervention. Further validation across diverse health care settings is warranted.

Accurate early prediction of chemotherapy response in non-small cell lung cancer (NSCLC) remains a clinical challenge. This study aimed to evaluate the utility of PET-based radiomic features extracted from [ 18 F]fluorodeoxyglucose ([ 18 F]FDG) PET/computed tomography (CT) scans in predicting treatment response to platinum based chemotherapy in NSCLC patients. An ambispective observational study was conducted on 70 histopathologically confirmed NSCLC patients who underwent [ 18 F]FDG PET/CT imaging before and after chemotherapy at a tertiary cancer center. Radiomic features were extracted from the primary lesion using commercially available texture analysis software, applying a fixed standardized uptake value (SUV) threshold-based volume of interest segmentation. Conventional metabolic parameters [SUVmax, SUVmean, metabolic tumor volume (MTV), total lesion glycolysis (TLG)] and 46 radiomic features (GLRLM, NGLDM, GLCM, shape features) were analyzed. Treatment response was assessed using PERCIST criteria and patients were categorized as responders or non-responders. Statistical analysis included Mann-Whitney U test and receiver operating characteristic analysis to determine discriminatory ability of parameters. Among the radiomic and metabolic parameters evaluated, only a few showed statistically significant differences between responders and non-responders. GLRLM GLNU, GLCM correlation, NGLDM contrast, and shape surface demonstrated fair predictive performance with area under the curve (AUC) values >0.7. Conventional parameters such as SUVmax and TLG did not show statistically significant differences. The optimal cutoff values, sensitivity, specificity, AUCs, and P values of significant features were also obtained. Select radiomic features derived from [ 18 F]FDG PET/CT scans, particularly GLNU, GLCM correlation, and NGLDM contrast, hold promise in predicting response to platinum-based chemotherapy in NSCLC. These findings suggest the potential utility of radiomics in enhancing personalized treatment strategies, although further validation is warranted.

A decision-support tool for management of advanced epithelial ovarian cancer in a single centre in India (CT PAUSE Nomogram): a prospective Study (2022-2024).

Lung cancer mortality decreases as a result of low-dose computed tomography (LDCT) screening, but suffers from low uptake and high false-positive rates. The impact of integrating genetic risk using a polygenic risk score (PRS) to optimize lung cancer screening remains underexplored. We developed a genome-wide PRS and evaluated its performance in pre- and post-screening contexts. Screening eligibility was assessed using two UK Biobank (UKB) subsets: UKBPLCO (n = 8957; PLCOm2012norace risk ≥2%) and UKBScreeningCriteria (n = 74,024 meeting screening eligibility criteria). To evaluate nodule management, we used a cohort of 669 ever-smokers with PLCO ≥2% and Lung-RADS score ≥3, referred to as SYNERGIQCPLCO_LungRADS. Multivariable Cox models, time-dependent area under the curve (AUC), and decision curve analyses (DCA) evaluated association, discrimination, and clinical net benefit. In UKBPLCO, the PRS was associated with a hazard ratio (HR) of 1.18 per standard deviation. In UKBScreeningCriteria, PRS showed HR of 1.34. Adding PRS to the PLCOm2012 model improved discrimination (AUC: 0.707 vs 0.696; P = 6.85e-27[likelihood ratio test]) and correctly reclassified 9.2% of incident lung cancer cases. Six-year absolute risks stratified by PRS deciles indicate 3.1-fold increase in the top compared to the bottom decile. In SYNERGIQCPLCO_LungRADS, HRPRS was 1.22. In this context, DCA indicated a modest net benefit for decision thresholds between 10% and 30%. PRS in lung cancer reveals context-dependent net benefit across studied populations. Although PRS adds limited value for determining screening eligibility, it may help reclassify borderline individuals and inform decisions regarding closer follow-up or invasive diagnostic procedures for high-risk screened groups.

Prediction of Osteoporosis and Fragility Fracture Risk Using Proximal Humerus CT Value from Chest CT: A Development and Validation Study.

To develop a novel transparent and lightweight machine learning model, the Green Learning (GL), for automated prostate segmentation (PS) and clinically significant prostate cancer (csPCa) detection on magnetic resonance imaging (MRI). Men who underwent 3-T MRI and prostate biopsy (PBx) were identified. MRI was acquired and interpreted according to the Prostate Imaging-Reporting and Data System (PI-RADS), version 2 or 2.1. The GL was created to automate PS and csPCa detection on biparametric MRI. The performance was compared to the standard-of-care radiologists using PI-RADS, and a conventional deep learning (DL) U-Net model as benchmarking. The PS performance was evaluated by the Dice similarity coefficient (DSC). The area under the curve (AUC) for patient-level csPCa detection was assessed. Model size and computational workload, measured by floating point operations (FLOPs), were reported. A total of 602 MRIs were randomly divided for training (N = 483) and testing (N = 119). Overall, 224 patients had csPCa on PBx. The median DSC for PS was higher for GL than U-Net (0.91 vs 0.88, P < 0.001). The AUC for csPCa detection of GL was similar to PI-RADS (0.75 vs 0.76, P = 0.8) and U-Net (vs 0.74, P = 0.3). A combination of GL and PI-RADS showed a higher AUC of 0.81 than PI-RADS alone (P = 0.02). Compared with U-Net, the GL had smaller magnitude parameters (1.21× 106 vs 177× 106) and less computational workload (9.8× 109 vs 1027× 109 FLOPs). A novel GL model fully automatically detects csPCa on prostate biparametric MRI with comparable performance to PI-RADS and DL. Combined with PI-RADS, GL significantly improves csPCa detection.

Who benefits from a filter? Developing clinical prediction models for thrombus dislodgement in hospitalized patients with deep vein thrombosis: A multicenter retrospective study.