A publicly available pharyngitis dataset and baseline evaluations for bacterial or nonbacterial classification

Area of Science:

• Medical imaging
• Artificial intelligence
• Infectious diseases

Background:

• Differentiating bacterial from nonbacterial pharyngitis is clinically challenging due to similar symptoms.
• Overuse of antibiotics for viral infections contributes to antibiotic resistance.
• Novel diagnostic methods are needed for accurate and timely pharyngitis diagnosis.

Purpose of the Study:

• To develop and release the largest publicly available dataset of high-resolution throat images for pharyngitis research.
• To enable the development of AI-driven tools for non-invasive diagnosis of bacterial versus nonbacterial pharyngitis.
• To support advancements in remote healthcare and medical image analysis for infectious diseases.

Main Methods:

• Collected high-resolution throat images from 742 patients with common cold symptoms.
• Recorded detailed clinical data including 20 symptoms, age, gender, and physician diagnoses.
• Utilized deep neural networks for image analysis to distinguish infection types.
• Established three baseline models for bacterial vs. nonbacterial infection differentiation.

Main Results:

• Created the largest publicly available dataset for pharyngitis visual diagnosis.
• Demonstrated the potential of AI and deep learning in analyzing throat images for diagnostic purposes.
• Provided baseline models for future research and development of diagnostic tools.

Conclusions:

• The developed dataset and AI approaches offer a promising avenue for accurate, non-invasive pharyngitis diagnosis.
• This work can significantly reduce unnecessary antibiotic prescriptions and combat antimicrobial resistance.
• Future research can build upon this dataset to refine AI diagnostic capabilities and promote telehealth solutions.