Background: Helicobacter pylori plays a central role in the development of gastric cancer, and prediction of H pylori infection by visual inspection of the gastric mucosa is an important function of endoscopy. However, there are currently no established methods of optical diagnosis of H pylori infection using endoscopic images. Definitive diagnosis requires endoscopic biopsy. Artificial intelligence (AI) has been increasingly adopted in clinical practice, especially for image recognition and classification. Objective: This study aimed to evaluate the diagnostic test accuracy of AI for the prediction of H pylori infection using endoscopic images. Methods: Two independent evaluators searched core databases. The inclusion criteria included studies with endoscopic images of H pylori infection and with application of AI for the prediction of H pylori infection presenting diagnostic performance. Systematic review and diagnostic test accuracy meta-analysis were performed. Results: Ultimately, 8 studies were identified. Pooled sensitivity, specificity, diagnostic odds ratio, and area under the curve of AI for the prediction of H pylori infection were 0.87 (95% CI 0.72-0.94), 0.86 (95% CI 0.77-0.92), 40 (95% CI 15-112), and 0.92 (95% CI 0.90-0.94), respectively, in the 1719 patients (385 patients with H pylori infection vs 1334 controls). Meta-regression showed methodological quality and included the number of patients in each study for the purpose of heterogeneity. There was no evidence of publication bias. The accuracy of the AI algorithm reached 82% for discrimination between noninfected images and posteradication images. Conclusions: An AI algorithm is a reliable tool for endoscopic diagnosis of H pylori infection. The limitations of lacking external validation performance and being conducted only in Asia should be overcome. Trial Registration: PROSPERO CRD42020175957; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=175957

Background: Vestibular migraine (VM), the common term for recurrent vestibular symptoms with migraine features, has been recognized in the appendix criteria of the third beta edition of the International Classification of Headache Disorders (ICHD-3β). We applied the criteria for VM in a prospective, multicenter headache registry study. Methods: Nine neurologists enrolled consecutive patients visiting outpatient clinics for headache. The presenting headache disorder and additional VM diagnoses were classified according to the ICHD-3β. The rates of patients diagnosed with VM and probable VM using consensus criteria were assessed. Results: A total of 1414 patients were enrolled. Of 631 migraineurs, 65 were classified with VM (10.3%) and 16 with probable VM (2.5%). Accompanying migraine subtypes in VM were migraine without aura (66.2%), chronic migraine (29.2%), and migraine with aura (4.6%). Probable migraine (75%) was common in those with probable VM. The most common vestibular symptom was head motion-induced dizziness with nausea in VM and spontaneous vertigo in probable VM. The clinical characteristics of VM did not differ from those of migraine without VM. Conclusion: We diagnosed VM in 10.3% of first-visit migraineurs in neurology clinics using the ICHD-3β. Applying the diagnosis of probable VM can increase the identification of VM.

Serum pepsinogen assay (sPGA), which reveals serum pepsinogen (PG) I concentration and the PG I/PG II ratio, is a non-invasive test for predicting chronic atrophic gastritis (CAG) and gastric neoplasms. Although various cut-off values have been suggested, PG I ≤70 ng/mL and a PG I/PG II ratio of ≤3 have been proposed. However, previous meta-analyses reported insufficient systematic reviews and only pooled outcomes, which cannot determine the diagnostic validity of sPGA with a cut-off value of PG I ≤70 ng/mL and/or PG I/PG II ratio ≤3. We searched the core databases (MEDLINE, Cochrane Library, and Embase) from their inception to April 2018. Fourteen and 43 studies were identified and analyzed for the diagnostic performance in CAG and gastric neoplasms, respectively. Values for sensitivity, specificity, diagnostic odds ratio, and area under the curve with a cut-off value of PG I ≤70 ng/mL and PG I/PG II ratio ≤3 to diagnose CAG were 0.59, 0.89, 12, and 0.81, respectively and for diagnosis of gastric cancer (GC) these values were 0.59, 0.73, 4, and 0.7, respectively. Methodological quality and ethnicity of enrolled studies were found to be the reason for the heterogeneity in CAG diagnosis. Considering the high specificity, non-invasiveness, and easily interpretable characteristics, sPGA has potential for screening of CAG or GC.

Background: The evaluation of large language models (LLMs) in medicine has undergone a shift from knowledge-based testing to practice-based assessment, representing an evolution in how we measure artificial intelligence readiness for clinical deployment. While LLMs now routinely exceed human performance on medical licensing examinations, their translation to clinical practice remains poorly characterized. Objective: This systematic review aims to categorize and analyze medical LLM benchmarks, examining performance patterns across different evaluation paradigms and identifying gaps in current assessment methodologies. Methods: The protocol was registered at PROSPERO (CRD420251139729). Four databases (MEDLINE/PubMed, Embase/Ovid, Cochrane Library, and arXiv) were searched from inception to August 31, 2025, using keywords related to clinical medicine benchmarks in LLMs. Studies were included if they (1) investigated clinical medicine benchmarks in LLMs, (2) were published in English, and (3) were available in full-text. Studies were excluded if they evaluated nonmedical domains or lacked benchmark validation. Methodological quality was assessed using the Mixed Methods Appraisal Tool (version 2018) by 2 independent reviewers (κ=0.91). Due to heterogeneity in evaluation metrics preventing meta-analysis, narrative synthesis was conducted using structured categorization of benchmark types. Results: From 3917 screened records, 39 medical LLM benchmarks were identified and categorized into 21 (54%) knowledge-based, 15 (38%) practice-based, and 3 (8%) hybrid frameworks. These benchmarks collectively encompass over 2.3 million questions across 45 languages and 172 medical specialties. Traditional knowledge-based benchmarks show saturation with leading models achieving 84%-90% accuracy on USMLE (United States Medical Licensing Examination)–style examinations, approaching or exceeding average physician performance. However, practice-based assessments reveal performance challenges, with specific benchmarks showing varied results: DiagnosisArena 45.82% (95% CI 42.9%-48.8%), MedAgentBench 69.67% (95% CI 64.2%-74.6%), and HealthBench 60% (95% CI 58.6%-61.3%) success rates, with practice-based benchmarks showing lower performance (45%-69%) compared to knowledge benchmarks (84%-90%). Task-specific analysis revealed differential performance patterns: factual retrieval maintained 85%-93% accuracy, clinical reasoning dropped to 50%-60%, diagnostic tasks achieved 45%-55% success, and safety assessment showed significant gaps at 40%-50% accuracy despite being life-critical. Geographic representation spans 6 continents with 18 (46%) benchmarks, incorporating non-English content. Quality assessment revealed 26% (10/39) of benchmarks had insufficient methodological reporting for complete evaluation. Conclusions: This systematic review provides the first comprehensive analysis quantifying the significant “knowledge-practice gap” in medical artificial intelligence: high performance on knowledge-based examinations (84%-90%) does not translate to clinical competence (45%-69%), with safety assessments at 40%-50%. Our findings provide quantitative evidence for regulators and health systems that examination scores are insufficient and misleading proxies for clinical readiness. This review concludes that autonomous deployment is not currently justifiable and that all evidence-based implementation strategies must mandate practice-oriented validation and robust human-in-the-loop oversight to ensure patient safety. Trial Registration: PROSPERO CRD420251139729; https://www.crd.york.ac.uk/PROSPERO/view/CRD420251139729

Approximately 1% of the general population and 10% of patients with migraine suffer from vestibular migraine (VM). However, this condition remains relatively unknown; therefore, it is often underdiagnosed despite the recent adoption of international diagnostic criteria for VM. The diagnosis of VM is based on the symptoms, degree, frequency, and duration of the vestibular episodes, a history of migraine, the temporal association of migraine symptoms with vestibular episodes in at least 50% of cases, and the exclusion of other causes. Physical examination and laboratory findings are usually normal in patients with VM but can be used to rule out other vestibular disorders with similar symptoms. The pathophysiology of VM remains incompletely understood; however, several mechanisms link the trigeminal system, which is activated during migraine attacks, and the vestibular system. Because few controlled trials have specifically investigated VM, the treatment options for this order are largely the same as those for migraine and include antiemetics for severe acute attacks, pharmacological migraine prophylaxis, and lifestyle changes.

Endoscopic resection is recommended for gastric neoplasms confined to mucosa or superficial submucosa. The determination of invasion depth is based on gross morphology assessed in endoscopic images, or on endoscopic ultrasound. These methods have limited accuracy and pose an inter-observer variability. Several studies developed deep-learning (DL) algorithms classifying invasion depth of gastric cancers. Nevertheless, these algorithms are intended to be used after definite diagnosis of gastric cancers, which is not always feasible in various gastric neoplasms. This study aimed to establish a DL algorithm for accurately predicting submucosal invasion in endoscopic images of gastric neoplasms. Pre-trained convolutional neural network models were fine-tuned with 2899 white-light endoscopic images. The prediction models were subsequently validated with an external dataset of 206 images. In the internal test, the mean area under the curve discriminating submucosal invasion was 0.887 (95% confidence interval: 0.849–0.924) by DenseNet−161 network. In the external test, the mean area under the curve reached 0.887 (0.863–0.910). Clinical simulation showed that 6.7% of patients who underwent gastrectomy in the external test were accurately qualified by the established algorithm for potential endoscopic resection, avoiding unnecessary operation. The established DL algorithm proves useful for the prediction of submucosal invasion in endoscopic images of gastric neoplasms.