Xiaogang Guo

Medical AI has advanced from task-specific models toward foundation models and agent systems that organize heterogeneous clinical evidence, conduct multi-step reasoning, and execute actionable workflows. We argue that clinical readiness requires simultaneously satisfying four principles—evidence sufficiency, reasoning trustworthiness, decision controllability and auditability, and specialty adequacy. These principles reveal two fundamental tensions: model vs. system and generalist vs. specialist, whose intersection yields four research paradigms. Using these principles as analytical anchors, we examine each paradigm's contributions and capability boundaries. We identify four persistent gaps separating current research from clinical readiness: insufficient evidence-sufficiency guarantees, unresolved reasoning trustworthiness, system-level governance deficits, and tension between specialty depth and generalization. We conclude that medical AI maturity depends not on peak task performance but on sustaining equilibrium among these principles.

Point accumulation for imaging in nanoscale topography (PAINT) is a single‐molecule localization microscopy (SMLM) technique that achieves nanometer‐level resolution, enabling direct observation of complex subcellular structures. However, its application in living cells remains limited by the lack of suitable labeling strategies and the challenge of imaging dynamic cellular structures over extended periods. In this study, we introduce PAINTpro, a protein‐based PAINT technique that facilitates live‐cell SMLM imaging and visualizes dynamic structures at the nanoscale. PAINTpro utilizes an engineered reversible bilirubin (BR)‐binding protein, Fluorescent protein for Maximizing Imaging with Nano‐resolution (FluoroMIN), derived from the green fluorescent protein UnaG. FluoroMIN incorporates mutations based on in silico molecular dynamics simulations to accelerate BR exchange, resulting in reversible fluorescence, high single‐molecule signal yield, and prolonged switching lifetime. These properties enable both live‐cell labeling and super‐resolution PAINT imaging with a resolution of ∼50 nm. Moreover, PAINTpro supports long‐term imaging of dynamic structures, providing time‐resolved super‐resolution images of microtubules with a resolution of up to 60 nm. Notably, PAINTpro is readily adaptable from existing protein‐PAINT protocols, offering a powerful tool for super‐resolution visualization of biological processes.

Background and objective: Inflammation, immune dysfunction, malnutrition and severe coronary artery calcification (CAC) are critical cardiovascular risk factors in patients on dialysis. However, their combined prognostic value in dialysis patients with acute coronary syndrome (ACS) remains unclear. This study aims to evaluate the associations of the C-reactive protein-albumin-lymphocyte (CALLY) index (integrating inflammatory, immune and nutritional parameters) and angiographically detected CAC status with major adverse cardiovascular events (MACE) in this population. Methods: This multicenter retrospective cohort study finally included 681 dialysis patients with ACS between January 2015 and June 2021. Patients were categorized into groups based on the median of the CALLY index and the severity of angiographic CAC. The endpoint was the occurrence of MACE, defined as a composite of all-cause death, non-fatal myocardial infarction and non-fatal stroke. Results: During a median follow-up of 23.54 months, 329 cases of MACE occurred. Multivariable Cox regression models demonstrated that a higher CALLY index was independently associated with a reduced risk of MACE [high vs. low CALLY index: hazard ratio (HR) = 0.521; 95% confidence interval (CI): 0.412-0.659], whereas moderate-to-severe CAC significantly increased the risk of MACE [HR = 1.267; 95% CI: 1.008-1.592]. When stratified by both parameters, patients with a higher CALLY index and no-or-mild CAC exhibited the lowest cumulative incidence of MACE [HR = 0.420; 95% CI: 0.302-0.585). Notably, the combined assessment of the CALLY index and angiographically detected CAC status significantly improved the predictive accuracy beyond the Global Registry of Acute Coronary Events (GRACE) score or baseline risk model. Conclusion: In dialysis patients with ACS, a lower CALLY index and moderate-to-severe CAC were both independent risk factors of MACE. Integrating the categorical classification of the CALLY index level and angiographic CAC status significantly improved prognostic performance, providing additional information for cardiovascular risk stratification in this high-risk population.

Identifying serum biomarkers that accurately reflect the progression of coronary artery disease (CAD) remains a major challenge. Integrative proteomic and metabolomic profiling can provide novel insights into disease pathogenesis and improve clinical prediction. We conducted a four-phase study. In the discovery phase, serum from 40 patients (controls, stable CAD, and acute coronary syndrome (ACS)) was analyzed using data-independent acquisition (DIA) proteomics and liquid chromatography-tandem mass spectrometry (LC–MS)/gas chromatography–mass spectrometry (GC–MS) metabolomics to identify differentially expressed proteins (DEPs) and metabolites. In the verification phase, selected DEPs were validated by parallel reaction monitoring (PRM) in an independent 40-patient cohort. In the derivation phase, six validated proteins were measured by ELISA in 207 angina patients to assess their association with coronary obstruction (≥ 50% stenosis). In the validation phase, a support vector machine (SVM) model incorporating clinical risk factors and these biomarkers was developed in the derivation cohort and tested in an independent 97-patient cohort. Model performance was evaluated using receiver operating characteristic (ROC) curves and decision curve analysis. Coronary obstruction is defined as ≥ 50% luminal diameter stenosis in at least one major coronary artery on angiography. Proteomic analysis identified 97 DEPs, and metabolomic profiling revealed 322 DEMs (including 289 from LC–MS and 33 from GC–MS analyses). Seven proteins showed consistent changes in both DIA and PRM validation. Among these, thrombospondin-1 (TSP-1) was significantly upregulated in stable CAD compared with controls, while serum amyloid A1 (SAA1) was markedly elevated in ACS compared with stable CAD. In an independent angina cohort ( n = 207), serum levels of TSP-1 and SAA1 were significantly higher in patients with coronary obstruction. Multivariate logistic regression adjusted for conventional cardiovascular risk factors (including age, sex, homocysteine, and other clinical variables) demonstrated that TSP-1 remained independently associated with coronary artery occlusion (OR = 1.424, 95% CI 1.057–1.918, P = 0.020). A support vector machine (SVM) model incorporating conventional clinical risk factors was constructed, and the addition of TSP-1 and SAA1 significantly improved diagnostic performance for CAD severity (AUC = 0.919 in derivation, 0.992 in validation). In conclusion, our dual-omics approach identified novel biomarkers and pathways in CAD progression. The SVM-based prediction model offers a promising non-invasive tool for early CAD detection, potentially reducing unnecessary invasive procedures.

To determine the correlation between weight‐adjusted‐waist index (WWI) with prehypertension and hypertension in the general population. This cross‐sectional study was conducted from September 13, 2023 to November 30, 2023 and included 4885 permanent residents in Zhejiang Province. The association between WWI with prehypertension and hypertension was carefully evaluated using logistic regression models and penalized spline functions. The category‐free analysis was performed to corroborate whether WWI enhanced the ability to estimate prehypertension and hypertension. Multivariable logistic regression indicated that after adjustment for conventional risk factors, each standard deviation increase in WWI levels was associated with a 29% and 50% increased risk of prehypertension (hazard ratio, 1.29; 95% confidence interval, 1.14–1.45) and hypertension (1.50; 1.34–1.68), respectively. Referring to the participants in the lowest quartile of WWI, those in the highest quartile had a significantly increased risk of prehypertension (1.64; 1.17–2.30) and hypertension (2.47; 1.80–3.39). Smoothing curve fitting and trend logistic regression suggested a linear positive correlation, and the robustness of these findings was observed in interaction tests and subgroup analyses. The category‐free analysis suggested that adding WWI levels to the traditional model ultimately improved the predictive ability of prehypertension and hypertension. The results of this study confirmed the significant correlation between WWI with prehypertension and hypertension, independent of conventional risk factors. More importantly, these findings suggested the potential usefulness of WWI levels to improve the risk stratification of hypertension and prehypertension.

Eukaryotic mRNAs typically encode a single functional polypeptide, a principle challenged by the discovery of widespread non-canonical peptide-coding ORFs within 5’UTRs. However, their functional significance at the protein level remains underexplored. Using a four-layered pipeline, we identify 14 human transcripts predominantly transcribed in polycistronic forms, each encoding two conserved proteins. Focusing on the SLC35A4 transcript, we show that its 5’UTR encodes a mitochondrial inner membrane-localized microprotein that we name STREMI ( S LC35A4 s t ress re sponse regulating M ICOS i nteractor). Sharing topology and motifs with the MICOS core subunit MIC10, STREMI regulates mitochondrial cristae morphogenesis in mice and human cells. Additionally, the STREMI-encoding uORF mediates stress-responsive translation of SLC35A4 —a Golgi nucleotide sugar transporter—upregulating its translation during the integrated stress response. Evolutionary analyses indicate that these bicistronic transcripts likely arose through transcriptional readthrough following retroposition. We propose a mechanism of “gene symbiosis” that enables functional partitioning and coordinated translation of protein pairs from bicistronic transcripts.

Aims: The high recurrence rate after catheter ablation for atrial fibrillation (AF) remains a challenge. Short-term antiarrhythmic drugs are commonly used during the 3-month blanking period, but the choice between amiodarone and dronedarone is unclear. Objectives: This trial directly compared the effectiveness and safety of dronedarone vs. amiodarone administered during the blanking period post-ablation in AF. Methods and results: In this open-label, randomized controlled trial, 280 patients undergoing AF ablation were assigned to receive 12 weeks of treatment with either dronedarone or amiodarone, which was discontinued upon completion of this treatment phase as per protocol. The primary endpoint was atrial tachyarrhythmia recurrence (>30 s) after the blanking period. A total of 280 participants were enrolled, of whom 273 were included in the modified intention-to-treat (mITT) analysis. In the mITT analysis, recurrence rates were not significantly different between the dronedarone and amiodarone groups (23.4% vs. 16.9%, P = 0.184). Secondary efficacy endpoints also showed no significant differences. However, the dronedarone group had a significantly lower incidence of side effects (40.9% vs. 64.0%, P < 0.001), primarily due to fewer cases of QTc-interval prolongation (5.8% vs. 15.4%, P = 0.01) and hypothyroidism (8.8% vs. 26.5%, P < 0.001). The intention-to-treat analysis, which included all enrolled participants, yielded results consistent with those of the mITT analysis. Conclusion: Therapy with dronedarone during the blanking period after AF ablation was associated with a similar recurrence rate but a superior safety profile compared to amiodarone.

BACKGROUND: Premature atrial contractions (PACs) are independently associated with atrial fibrillation, stroke, and heart failure, yet no pharmacological therapy is approved for PAC suppression. Experimental studies have identified a functional cardiac glutamatergic system in which N-methyl-D-aspartate receptors regulate atrial electrophysiology. Preclinical studies show that pharmacological antagonism of N-methyl-D-aspartate receptors with memantine suppresses atrial arrhythmias. METHODS: We conducted an investigator-initiated, phase 2, multicenter, randomized, double-blind, placebo-controlled trial. Symptomatic adults with frequent PACs (≥1000/24 h) were randomly assigned to receive memantine or placebo for 6 weeks. The primary end point was the percentage change in mean 24-hour PAC count from baseline to the end of treatment. The primary analysis was performed in the intention-to-treat population. Prespecified secondary end points included the responder rate (≥50% PAC reduction), percentage change in nonsustained atrial tachycardia burden, and cumulative incidence of new-onset atrial fibrillation. RESULTS: Among 241 patients included in the efficacy analysis, memantine resulted in a greater reduction in PAC count than placebo (between-group difference, 47.1 percentage points; P =0.0045). The responder rate was higher with memantine than with placebo (52.4% versus 23.1%; P P =0.0043) and was associated with a lower cumulative incidence of new-onset atrial fibrillation (4.8% versus 23.9%; P <0.0001). No clinically meaningful differences were observed in electrocardiographic intervals or left ventricular function, and no drug-related serious adverse events occurred. CONCLUSIONS: In patients with frequent symptomatic PACs, memantine reduced atrial ectopy and atrial tachyarrhythmia burden and demonstrated a favorable safety profile. These findings provide proof of concept for a novel, non–ion channel–based therapeutic strategy targeting the cardiac glutamatergic system. REGISTRATION: URL: https://www.clinicaltrials.gov ; Unique identifier: NCT06501638.

Background: In recent years, large language models (LLMs) have undergone swift cycles of refinement and iteration. However, in the realm of clinical medicine, different LLMs' capability of logical reasoning and disease diagnosis needs further investigation. Objective: The aim of our study was to evaluate the performance of 4 different LLMs in the National Health Professional and Technical Qualification Examination in China. Methods: A total of 398 multiple-choice questions of 5 different question types were integrated within the examination with respect to the diagnosis or care of cases. These questions were categorized into different cardiology subspecialties and different clinical disciplines. DeepSeek V3 and R1 were accessed through an application programming interface, while ChatGPT 4o and o1 were queried via its public chat-based interface. We offered the same prompts instructing LLMs to assume the role of a physician and provide answers with explanations at the beginning of each conversation. We assessed different LLMs’ performance by the accuracy in the responses to the multiple-choice questions. For the first 3 examination sections, McNemar test was used to compare the accuracy among the models, with post hoc pairwise comparisons performed using partitions of chi-square method and Bonferroni correction (significance set at P<.008). For the fourth section involving partially credit scoring, one-way ANOVA was performed to compare the mean scores among the models, with statistical significance set at P<.05. Results: Both DeepSeek V3 and R1 showed superior performance in the first 3 sections of the Chinese National Health Professional and Technical Qualification Examination, achieving an overall performance of 93% and 93.6%, respectively. ChatGPT 4o and o1 achieved accuracies of 73.3% and 69%, respectively (all P<.001 compared with DeepSeek V3). For the fourth section, the performance of all 4 LLMs markedly declined compared to their results in the preceding sections. Particularly, in the section of gastroenterology and hematology, DeepSeek V3 achieved the highest accuracy, while R1 ranked first in cardiology and neurology. ChatGPT o1 achieved the highest accuracy in the topic of coronary artery disease, with no statistical significance. Conclusions: DeepSeek V3 and R1 showed remarkable potential in facilitating clinical decision-making in the Chinese professional examination, with both outperforming ChatGPT 4o and o1. Nonetheless, future research should continue evaluating their economic efficiency and susceptibility to hallucination.