Toward automated severe pharyngitis detection with smartphone camera using deep learning networks.

Area of Science:

• Medical imaging and diagnostics
• Artificial intelligence in healthcare
• Infectious disease detection

Background:

• Severe pharyngitis, often caused by streptococcal infections, can lead to serious complications.
• The COVID-19 pandemic highlighted the need for remote diagnostic tools for respiratory illnesses.
• Telemedicine offers a solution for patients experiencing symptoms like sore throat.

Purpose of the Study:

• To develop an automated system for detecting severe pharyngitis.
• To utilize deep learning with self-taken throat images for diagnosis.
• To create a smartphone-based screening tool for pharyngitis.

Main Methods:

• Collected a dataset of 131 pharyngitis and 208 normal throat images.
• Employed CycleGAN for data augmentation with synthetic images.
• Trained and validated ResNet50, Inception-v3, and MobileNet-v2 models using transfer learning.

Main Results:

• CycleGAN-generated images accurately represented pharyngitis features.
• Deep learning models showed improved pharyngitis diagnostic accuracy with augmented data.
• ResNet50 with GAN augmentation achieved an ROC-AUC of 0.988; 4-fold cross-validation yielded 95.3% accuracy and 0.992 ROC-AUC.

Conclusions:

• The developed deep learning model enables rapid, smartphone-based screening for severe pharyngitis.
• This framework facilitates timely diagnosis and reduces transmission risk for upper respiratory symptoms.
• The system enhances diagnostic convenience, particularly relevant during pandemics like COVID-19.