Related Experiment Video
Updated: Aug 11, 2026
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
A 3-Tier AI Model for COVID-19 Triage Using Pharyngeal Images: Algorithm Development and Validation
Sho Okiyama1,2, Tomonori Aoki3, Memori Fukuda2
1Department of Digital Health, Institute of Medicine, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki, 305-8571, Japan, +81 90 6522 4161.
JMIR Formative Research
|July 20, 2026
Summary
An AI algorithm analyzing pharyngeal images and clinical data aids in triaging patients for SARS-CoV-2 testing. The COVID-19-AI tool demonstrates high sensitivity for ruling out COVID-19 and high specificity for infection control, offering a robust triage solution.
Area of Science:
- Artificial Intelligence in Healthcare
- Medical Diagnostics
- Infectious Disease Management
Background:
- SARS-CoV-2 causes acute respiratory illness, but symptom-based triage is unreliable.
- Artificial intelligence (AI)-powered pharyngeal cameras offer a novel approach for diagnosing respiratory infections.
- An AI algorithm, COVID-19-AI, was evaluated to assist in SARS-CoV-2 testing decisions.
Purpose of the Study:
- To assess the diagnostic accuracy of the COVID-19-AI algorithm.
- To evaluate the clinical utility of COVID-19-AI as a triage support tool.
- To determine the effectiveness of AI-driven pharyngeal imaging in guiding SARS-CoV-2 testing.
Main Methods:
- A multicenter prospective study involving 26 Japanese institutions.
- Development of the COVID-19-AI algorithm using pharyngeal images and clinical data from 2133 patients.
- Performance evaluation on 696 independent patients using reverse transcription polymerase chain reaction (RT-PCR) confirmation under inclusive and strict criteria.
Main Results:
- The COVID-19-AI achieved 93.9% sensitivity and 19.6% specificity (inclusive criteria) and 24.7% sensitivity and 94.4% specificity (strict criteria).
- Performance remained stable across different symptom onset times.
- An integrated model combining images and clinical data (AUROC 0.78) outperformed models using single data types, with pharyngeal images being the most influential predictor.
Conclusions:
- The COVID-19-AI's 3-tier output supports both high sensitivity for ruling out COVID-19 and high specificity for infection control.
- The AI tool demonstrates robustness across varying symptom onset times, providing a standardized approach for clinical triage.
- AI-powered pharyngeal imaging integrated with clinical data enhances diagnostic accuracy for SARS-CoV-2, improving triage efficiency.
