Early Detection of Measles through Smartphone-based Recognition of Koplik Spots

Overview

Measles is a highly contagious respiratory infection that can spread before clinical diagnosis, because transmission often occurs during the prodromal phase when symptoms resemble common viral illness. Early identification is essential for preventing community transmission. Clinically, measles typically presents with fever and the “3 Cs” (cough, coryza, conjunctivitis), followed by rash; Koplik spots are a distinctive oral finding that can appear before rash but are transient and therefore frequently missed in real-world screening. To address this gap, we developed a smartphone-based screening web app that combines symptom-based risk scoring with oral-image recognition of Koplik spots. Users complete a self-assessment for fever, cough, coryza, and conjunctivitis and upload one or more inner-cheek images. A TensorFlow/Teachable Machine image classifier estimates the likelihood of Koplik-spot–consistent features for each image; the system then uses the highest likelihood among uploaded images to reduce the chance of missing transient lesions due to limited sampling. The symptom score applies a higher weight to fever because fever is emphasized as a core component of standard measles clinical case descriptions, while the three “C” symptoms are treated with equal weights to reflect their shared role in the classic prodrome. The app outputs image confidence, symptom score, and a conservative risk recommendation and logs results for threshold refinement. This prototype supports feasibility of combined symptom–image screening, but the solution is not yet clinically validated; prospective testing against clinician- or laboratory-confirmed cases is required to quantify accuracy during the prodromal window.