Smart diagnostics: Leveraging artificial intelligence for accurate at-home COVID-19 testing

Reverse transcriptase polymerase chain reaction (RT-PCR), the reference standard for COVID-19 diagnosis, is time-consuming, resource-intensive, and costly. Rapid antigen tests address these limitations but rely on visual interpretation, which may reduce diagnostic performance. This study evaluated the performance of artificial intelligence (AI) for interpreting at-home COVID-19 lateral flow antigen tests compared with visual interpretation, using RT-PCR as the reference standard. Subjects from COVID-19 treatment trials (N = 99) performed at-home antigen testing, and results were interpreted by both an AI model and visual review. Compared with RT-PCR, AI-assisted interpretation demonstrated 96.97% accuracy, 80% sensitivity, and 100% specificity. Visual interpretation showed 68.82% accuracy, 78.57% sensitivity, and 67.09% specificity. Paired comparison of overall diagnostic accuracy demonstrated a statistically significant difference favoring AI-assisted interpretation (McNemar's test, p < 0.0001). AI-assisted interpretation demonstrated higher accuracy and specificity than visual interpretation, while sensitivity was comparable between methods. These findings support the use of AI-based tools to improve reliability of at-home antigen test interpretation.