Work

Strepapp: Instant Strep Throat Diagnosis using Deep Learning on Smartphone Oropharyngeal Images

Year: 2026

Type: conference-paper

Abstract: Pharyngitis is a leading cause of pediatric ambulatory visits in the U.S., with 12 million outpatient visits annually. Approximately one-third of these cases are diagnosed as Group A Streptococcus (GAS) pharyngitis or strep throat, for which antibiotic treatment is administered to prevent complications such as rheumatic heart disease. Reliable GAS diagnosis requires a throat swab and microbiologic lab testing, which are not available in telemedicine where antibiotics are frequently overprescribed, contributing to unnecessary overuse and antimicrobial resistance. This paper addresses this clinical problem by developing a real-time, deep learning-based mobile app that integrates clinical symptoms, signs, and epidemiologic factors with photographic pharyngeal images to deliver an instant strep diagnosis. Because no clinical-grade dataset is publicly available for deep learning development, we conducted a multicenter data collection. To achieve this, we developed a data collection platform using a custom mobile app. Currently, this platform has been deployed in nine states across the U.S., and we enrolled more than 4,000 patients aged 3-17 years with confirmed strep lab test results. This dataset was used to develop a multimodal deep learning algorithm to detect strep throat based on throat images and clinical factors. The dataset was split into training, validation, and testing sets. The image preprocessing included a region-of-interest segmentation step using a deep learning model developed in-house. The overall classification strategy incorporating deep learning and statistical analysis yielded an accuracy of 85.5% on the test set. StrepApp reliably identifies low-risk patients, minimizing unnecessary swabbing and testing for suspected pharyngitis.

Source: Design of Medical Devices Conference (DMD)

DOI: 10.1115/dmd2026-1019