TonguExpert: A Deep Learning-Based Algorithm Platform for Fine-Grained Extraction and Classification of Tongue...

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Biomedical Imaging

Background:

• Tongue analysis is a promising tool for disease detection and health monitoring, particularly in traditional Chinese medicine.
• Subjectivity and lack of large datasets hinder automated tongue diagnosis.
• Existing deep learning methods struggle with subtle tongue details.

Purpose of the Study:

• To develop an automated deep learning framework for tongue image analysis and phenotype extraction.
• To create a free platform, TonguExpert, for archiving and analyzing tongue images.
• To establish the largest publicly available dataset of tongue images and phenotypes.

Main Methods:

• Developed a deep learning framework for tongue segmentation and phenotype extraction.
• Utilized a dataset of 5992 tongue images from a Chinese population.
• Extracted 773 phenotypes, including global and local features, and predicted labels.

Main Results:

• TonguExpert achieved high accuracy in classifying tongue color (ROC-AUC 0.89-0.99), fissures (0.97), and tooth marks (0.88).
• The model demonstrated good generalization capabilities on external datasets for predicting new phenotypes like greasy coating.
• Released the largest publicly available dataset of tongue images and associated phenotypes.

Conclusions:

• TonguExpert significantly outperforms manual classification, advancing automated tongue diagnosis.
• The platform and dataset facilitate wider clinical adoption and future research in tongue diagnosis.
• This work paves the way for expanded applications of automated tongue analysis in healthcare.