Developing tongue coating status assessment using image recognition with deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Tongue coating assessment is crucial for diagnosing various health conditions.
• Objective and detailed tongue coating evaluation methods are needed.
• Current methods can be subjective and time-consuming.

Purpose of the Study:

• To develop an image recognition network for automated tongue coating status evaluation.
• To create separate networks for tongue detection and tongue coating classification.
• To validate the accuracy and agreement of the developed networks.

Main Methods:

• Two convolutional neural networks (CNNs) were implemented: You-Only-Look-Once (YOLO) v2 with ResNet-50 for tongue detection, and ResNet-18 for tongue coating classification.
• Digital tongue images from 251 adults and 144 older adults were utilized.
• Tongue coating was graded by panelists on a 7x7 grid, and the tongue coating index (TCI) was calculated for comparison with network outputs.

Main Results:

• The tongue detection network achieved a high intersection over union (IoU) of 0.885±0.081.
• The tongue coating classification network demonstrated strong agreement with expert panel grading (kappa coefficient=0.826) and TCI (intraclass correlation coefficient=0.807).

Conclusions:

• AI-driven image recognition provides a simple and detailed method for assessing tongue coating.
• The developed network shows high accuracy and reliability for clinical applications.
• This technology has the potential to improve diagnostic capabilities related to tongue coating abnormalities.