Artificial Intelligence-Based Triage and Diagnostic Support for Gastric Biopsies: A Multi-Centre Prospective Evaluation Across NHS Wales
In brief
AI detected all 40 gastric cancers in a prospective Welsh biopsy study
Across 848 biopsies, the AI platform identified all 40 cancers and flagged eight lesions missed in routine review, including one hidden adenocarcinoma. In a workflow simulation, it prioritized all cancers, including six in routine backlogs, with false alerts in 4.6% of cases; whether real-world use speeds reporting or treatment remains untested.
- Journal
- Diagnostics (Basel, Switzerland) (Q2)
- Published
- 28 October 2026
- Study design
- Prospective / inception cohort
- Evidence level
- Level 4, Very Low (CEBM 4)
- Authors
- Muhammad Aslam, Angela Phillips, Pearl Huey, Namor Williams, Andrew Powell, Vincent Onyekwelu, et al.
- PMID
- 43028424
- DOI
- 10.3390/diagnostics16213553
Why clinicians should know about it
- Picked for Pathology and Forensic Medicine (top studies of the week, 15 November 2026): Prospective AI diagnostic evaluation for gastric biopsies
Abstract
Background: Histopathology workforce shortages and operational backlogs severely delay gastric cancer reporting. This multi-centre prospective evaluation assessed the diagnostic accuracy, clinical safety-net utility, and workflow triaging efficiency of an artificial intelligence (AI) platform for gastric biopsies across NHS Wales. Methods: A prospective cohort of 848 consecutive gastric biopsy cases across three NHS Health Boards was processed using a commercial AI platform. Diagnostic performance was evaluated against standard-of-care clinical reference diagnoses. Post-analytic cross-tabulation simulated dynamic AI-driven worklist prioritisation to quantify triage safety, false-positive alerts, and turnaround time impact across urgent and routine pathways. Results: For gastric malignancies (n = 40), the AI platform achieved a sensitivity of 100.0% (95% CI: 91.2-100.0%; AUC 0.986) and a specificity of 97.2% (95% CI: 95.8-98.1%). Sensitivities for low-grade dysplasia/adenoma (n = 12) and gastric intestinal metaplasia (n = 98) were 100.0% and 94.9%, respectively. The AI review functioned as a parallel safety net, capturing eight primary lesions missed during routine human evaluation, including one occult adenocarcinoma. In workflow simulation, the platform correctly prioritised 100.0% of cancers, including 6 cancers hidden within routine backlogs, and 97.2% (106/109) of high-risk non-cancer precursors, with false-positive alerts in 4.6% (39/848) of all cases. Conclusions: AI-assisted gastric pathology review demonstrates exceptional diagnostic accuracy, acts as a reliable clinical support tool, and enables effective risk-stratified triage. Deploying AI triage in public healthcare infrastructure offers a scalable solution with the potential to reduce diagnostic backlogs and accelerate treatment for high-risk oncology patients.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.