JMIR Publications

ABSTRACT

Background:

COVID-19 remains a common cause of acute respiratory illness, yet symptom-based triage poorly discriminates it from other febrile conditions. A recently developed artificial intelligence (AI)-powered pharyngeal camera acquires pharyngeal images and clinical data to assist influenza diagnosis; leveraging this workflow, we evaluated an adjunct　AI algorithm (“COVID-19-AI”) that reports High/Medium/Low suspicion to guide whether COVID-19 testing should subsequently be performed.

Objective:

To report diagnostic accuracy outcomes of the COVID-19-AI.

Methods:

We conducted a performance evaluation using a prospectively collected multicenter dataset from 26 Japanese institutions. Patients with suspected influenza or COVID-19 were eligible. The COVID-19-AI algorithm was developed using pharyngeal images combined with routine clinical variables from 2,133 patients, and it produced a three-tier output. Diagnostic performance was assessed in 696 independent patients against RT-PCR-confirmed COVID-19 under two prespecified operating criteria: inclusive (High or Medium = positive; Low = negative) and strict (High = positive; Medium or Low = negative). A subanalysis stratified accuracy by time from symptom onset (12-hour bins to 72 hours).

Results:

Among 696 analyzed participants, 247 were RT-PCR-confirmed COVID-19. Under the inclusive criteria, sensitivity of COVID-19-AI was 93.9% (95% CI 90.4-96.4) and specificity 19.6% (95% CI 16.1-23.5). Under the strict criteria, sensitivity was 24.7% (95% CI 19.6-30.4) and specificity 94.4% (95% CI 92.0-96.3). Across 12-hour onset strata, sensitivity under the inclusive criteria remained ≥92.0% and strict-criteria specificity remained ≥83.3%; no pronounced temporal trend was observed.

Conclusions:

Embedded within AI-powered pharyngeal camera workflows, a three-tier AI suspicion output enables complementary operating behaviors—high sensitivity to rule out COVID-19 (inclusive criteria) and high specificity to diagnose COVID-19 (strict criteria). Performance stability across onset times suggests robustness to symptom chronology.