Gliomas and brain metastases (BMs) on MRI pose significant diagnostic challenges for radiologists. This study aims to develop a multi-task model and a computer-aided diagnosis (CAD) system for the detection and diagnosis of gliomas and BMs. This study enrolled 3909 participants from seven centers, and developed a brain tumor segmentation and classification network (BTSC-Net) and BTSC-CAD with visualization of tumor masks. For detection, BTSC-Net achieved a Dice coefficient of 0.888 and 0.872 on the internal and external test sets, respectively. For diagnosis, BTSC-Net achieved AUCs of 0.941 and 0.933 on the internal and external test sets, respectively. With BTSC-CAD assistance, junior radiologists achieved mean AUC improvements of 4.8% (P < 0.05) for detection and 17.3% (P < 0.001) for diagnosis, along with an average reduction of 64.75 s in reading time. BTSC-CAD significantly improved radiologists' diagnostic accuracy and efficiency.

Disciplines conducting health research rely on discipline-specific workflows that are distinct from data science, the discipline which develops most artificial intelligence (AI)-enabled research tools. Here we reflect on the impact of such differences in workflows and provide practical advice for leveraging AI-enabled research tools in health research. Further, we provide recommendations for how to prepare the next generation of health researchers for the inevitable influx of research integrating AI-enabled tools.

Medical AI education remains fragmented, specialty-skewed, and lacks longitudinal structure, particularly for generalist physicians. Through an integrative review of 23 peer-reviewed articles (2016-2025), we identified three structural gaps: short-term interventions without reinforcement, procedural-field bias, and consistent under-representation of the Affective domain. We present AI-PACE (Psychomotor, Affective, Cognitive, Embedded), a Bloom's Taxonomy-grounded framework organizing AI competencies longitudinally across undergraduate, graduate, and continuing medical education.

OBJECTIVES: This systematic review and meta-analysis are aimed at evaluating the incidence of adverse drug reactions (ADRs) following administration of clinically approved ultrasound contrast agents (UCAs) in adults and children, to assess risks in patients with cardiovascular disease and in pregnancy, and to evaluate the effectiveness of emergency management of severe ADRs. MATERIALS AND METHODS: A PRISMA 2020 systematic review was conducted searching PubMed, Scopus, and Embase. Two reviewers independently screened, extracted data, and assessed quality. Incidence estimates were pooled when feasible, stratified by age group, contrast agent, and administration route. RESULTS: Seventy-four studies encompassing > 1 million adults and > 36,000 children were included, contributing multiple analytic cohorts to the quantitative synthesis. Severe acute ADRs were extremely rare (6 and 16 cases per 100,000 in adults and children, respectively) and absent following endocavitary administration in children. Non-severe acute ADRs occurred in 11 and 8 cases per 10,000 adults and children, respectively. Delayed reactions were very rare (< 1 case per million in adults). No significant safety differences emerged between UCA products. The incidence of ADRs in patients with cardiovascular disease was analogous to the general population. No ADRs were reported in pregnant women. Standard emergency management was effective in almost all serious cases, though rare fatalities occurred. CONCLUSION: UCAs show an excellent safety profile in adults and children, with very rare severe ADRs and few non-severe, typically self-limiting reactions. Strict adherence to recommended emergency management protocols mitigates the remaining risks, supporting safe use across a broad range of clinical indications. PROSPERO REGISTRATION: CRD42023432668. KEY POINTS: Question What is the incidence, type, and severity of acute and delayed ADRs associated with clinically approved UCAs across different patient populations? Findings Severe acute adverse reactions are very rare, and non-severe reactions are rare and self-limiting, with no significant safety differences between adults, children, or patients with cardiovascular disease. Clinical relevant UCAs show an excellent safety profile across populations. These findings support their safe clinical use as reliable alternatives to iodine-based and gadolinium-based contrast agents in routine diagnostic imaging.

Alzheimer's disease and related dementias (ADRD) develop years before symptoms emerge, making early prediction critical. Electronic health records (EHR) offer a scalable alternative to neuroimaging but are challenged by irregular encounters, data sparsity, and limited interpretability. We propose Gated Recurrent Unit with Decay & Attention (GRU-DA), which integrates GRU-D missingness modeling with RETAIN-style attention for interpretable risk monitoring. The model was trained on the University of Texas Physicians EHR (15,172 ADRD cases with 1:10 matched controls) and externally validated in the All of Us cohort. EHR data up to 10 years before diagnosis were used, with random follow-up initiation to reflect real-world encounters. Both GRU-DA and GRU-D outperformed competing models, particularly beyond 5 years of follow-up and achieved AUROC ~ 0.7 after 8.5 years. Prediction performance depended more on data availability than follow-up length: 1 year with 15% data availability (AUROC 0.75, Average Precision 0.5) was comparable to 7.5 years with 10% availability. For individual cases, GRU-DA produced stable risk predictions with some variability in timestep and feature-level attributions across folds. These results demonstrate EHR data can support dynamic ADRD risk monitoring up to 10 years before diagnosis, with effectiveness strongly influenced by data completeness.

PURPOSE: Fever of unknown origin (FUO) is a prolonged fever with unknown etiology after an extensive outpatient or inpatient workup. 18F-Flurodeoxyglucose (FDG) positron emission tomography (PET) with computed tomography (CT) has shown a higher diagnostic yield than CT. This study assesses the diagnostic yield of 18F-FDG PET/CT-guided targeted biopsy in FUO. METHODS: This single-centre retrospective study included FUO patients who underwent 18F-FDG PET/CT-guided biopsy. A distinct focal area of FDG-uptake above the surrounding background within the organ of interest was considered PET-positive and targeted using automated robotic arm-assisted PET/CT-guided biopsy systems. Procedure-related complications were recorded, and histopathological examination was taken as the reference standard. Diagnostic yield and other PET-based parameters were calculated. RESULTS: Data from 165 FUO patients [median age 47 years (IQR, 30-60)] who underwent 18F-FDG PET/CT-guided biopsy between 2017 and June 2025 were analysed. 64 (39%) had undergone a previous procedure with negative or inconclusive results. The most common site of targeted biopsy was the lungs (71/165, 43%). The diagnostic yield of 18F-FDG PET/CT-guided biopsy in FUO patients was 96.4% (159/165), with the most common cause being malignancy (70/165, 42%), followed by infection (58/165, 35%), autoimmune (16/165, 10%) and inflammatory (7/165, 4%) causes. The median SUVmax of the target lesions was 11.8 (IQR, 8.6-16.3; range, 3.3-53.8) with no significant difference among target lesions of different etiologies (p = 0.412). Only six patients (3.6%) had procedure-related complications. CONCLUSION: Robotic arm-assisted 18F-FDG PET/CT-guided biopsy is a safe and accurate procedure in FUO patients having FDG-avid accessible lesions and may help in early diagnosis and management.

The introduction of artificial intelligence into healthcare is a reality. Drawing on literature, this critical reflection uses McCance and McCormack's Person-Centred Practice Framework to examine how artificial intelligence can both advance and undermine the humanisation of care. Through structured dialogue and literature synthesis, we identify opportunities alongside challenges. We argue for context-sensitive implementation that integrates artificial intelligence into everyday practice while safeguarding relational ethics, professional judgement, and core person-centred values.

Biomechanical assessment offers a novel approach for the diagnosis and treatment of corneal ectatic disorders, particularly keratoconus, a blinding eye disease characterized by biomechanical weakening. However, current early diagnosis primarily relies on morphology, such as corneal tomography, which can lead to missed cases. This study utilized a video swin transformer to characterize global spatiotemporal corneal biomechanics from deformation videos. This model captures the entire dynamic process, significantly enhancing the diagnostic accuracy for suspicious cases (AUC 0.9942; accuracy 97.37%; 95% CI: 93.0%, 99.7%). Temporal-importance analysis revealed rebound-phase energy dissipation as a key early marker. Additionally, atomic force microscopy-based nanoindentation and adhesion tests confirmed reduced apparent Young's modulus and increased adhesion force in keratoconus. In conclusion, this study presents a novel, interpretable, in vivo, video-driven framework that supports early differential diagnosis and personalized intervention. This approach has clinical potential for precision medicine.